letters

LETTERS
VETERANS’ HEALTH CARE
It is not enough to say thank you to the editors of the Journal for the October 2003
cover picture and article on veterans’ health
care, which were well received in the veterans’ community.1 Far too often veterans are
yesterday’s news, even when our country is
engaged in a new conflict.
New York City veterans are currently involved in a struggle to keep the Veterans Affairs (VA) Hospital at First Ave and 23rd St
open. We have the support of several public
officials, such as Congresswoman Carolyn
Maloney, New York City Council Speaker Gifford Miller, Public Advocate Betsy Gotbaum,
and Council Members Margarita Lopez and
Christine Quinn, but it is still an uphill fight.
The VA CARES (Capital Asset Realignment
for Enhanced Services) Commission has visited many VA medical facilities around the
country and has slated several for major
changes. The commission proposes moving
all inpatient care from 23rd St to other VA
hospitals.
The CARES Commission also has plans to
close down the Montrose and Canandaigua
VA facilities in upstate New York. These closings will be devastating to the veterans community as well as to the health care profes-
Letters to the editor referring to a recent
Journal article are encouraged up to 3 months
after the article’s appearance. By submitting a
letter to the editor, the author gives permission
for its publication in the Journal. Letters
should not duplicate material being published
or submitted elsewhere. The editors reserve the
right to edit and abridge letters and to publish
responses.
Text is limited to 400 words and 10 references. Submit on-line at www.ajph.org for
immediate Web posting, or at submit.ajph.org
for later print publication. On-line responses
are automatically considered for print
publication. Queries should be addressed to
the department editor, Jennifer A. Ellis, PhD,
at [email protected]
sionals who staff these facilities. The impact
on teaching hospitals and funding in these
areas is probably viewed as nothing more
than “collateral damage,” but such a change
will affect many lives and probably none for
the better. I am sure that New York veterans
are not alone in being affected by the draconian measures recommended by the
CARES Commission. Should readers want
more information on this subject, I can connect them with other veterans who can offer
much more knowledge than my limited grasp
of this serious situation allows.
The Journal is a welcome, positive, caring
voice in very uncertain times for veterans’
health care. Please keep this issue alive in the
public health community, as we need all the
help we can get.
George McAnanama, BS
About the Author
The author served in the US Army from 1966 to 1968 and
is now with Veterans for Peace, Chapter 34 (New York City).
Requests for reprints should be sent to George
McAnanama, 22 Livingston Ct, Staten Island, NY 103101630 (e-mail: [email protected]).
Reference
1. Fine MJ, Demakis JG. The Veterans Health Administration’s promotion of health equity for racial and
ethnic minorities. Am J Public Health. 2003;93:
1622–1624.
ELIMINATING HEALTH DISPARITIES:
FOCAL POINTS FOR ADVOCACY
AND INTERVENTION
We applaud the Journal’s October 2003 focus
on eliminating health disparities and the University of Pittsburgh’s National Minority
Health Research Summit. Efforts to reduce inequalities need to be derived from multiple
sources. Two recent publications about the
challenges of achieving equality for American
Indians, Alaska Natives, and Latino peoples
show the failings of our past approaches.1,2
The National Institutes of Health’s National
Center on Minority Health and Health Dis-
April 2004, Vol 94, No. 4 | American Journal of Public Health
parities is paving the way for a more aggressive research agenda in this area.3 Public
health professionals need to educate policymakers about the need for support for such
research, as well as for programs that translate research into practice, such as the Centers for Disease Control and Prevention’s Racial and Ethnic Approaches to Community
Health (REACH) 2010.4 REACH is funded
for fiscal year 2004 at only $37 million—the
same amount it received in fiscal year 2003.5
It is clear that government intentions alone
will not elevate the health status of America’s
marginalized and minority peoples. Voluntary
professional associations such as the Sociey
for Public Health Education and the American Public Health Association (APHA) must
call on their members to connect with all levels of society. A central organization can plan
training, promote cultural sensitivity, share information and research, and develop positions that can be used by other groups in
their advocacy efforts. APHA is demonstrating its leadership in this area by designating
“Eliminating Health Disparities” the theme of
National Public Health Week in 2004.6
In 1966, Dorothy Nyswander defined an
open society as “one where justice is the same
for every [person]; where dissent is taken seriously as an index of something wrong or
something needed; where diversity is expected; . . . where the best of health care is
available to all; where poverty is a community disgrace not an individual’s weakness;
[and] where desires for power over [people]
become satisfaction with the use of power for
people.”7(p37) In keeping with this vision, in
2000 the Society for Public Health Education
(SOPHE) commissioned an Open Society
Commission, which resulted in 6 resolutions
aimed at eliminating health disparities and
calling for widespread socioecological change.
Yet stronger organizational commitment from
groups such as SOPHE and APHA is crucial
to helping diverse practitioners eliminate such
continuing practices among health care professionals as racism, stereotyping, bias, discrimination, and cultural and professional in-
Letters | 519

LETTERS
VETERANS’ HEALTH CARE
It is not enough to say thank you to the editors of the Journal for the October 2003
cover picture and article on veterans’ health
care, which were well received in the veterans’ community.1 Far too often veterans are
yesterday’s news, even when our country is
engaged in a new conflict.
New York City veterans are currently involved in a struggle to keep the Veterans Affairs (VA) Hospital at First Ave and 23rd St
open. We have the support of several public
officials, such as Congresswoman Carolyn
Maloney, New York City Council Speaker Gifford Miller, Public Advocate Betsy Gotbaum,
and Council Members Margarita Lopez and
Christine Quinn, but it is still an uphill fight.
The VA CARES (Capital Asset Realignment
for Enhanced Services) Commission has visited many VA medical facilities around the
country and has slated several for major
changes. The commission proposes moving
all inpatient care from 23rd St to other VA
hospitals.
The CARES Commission also has plans to
close down the Montrose and Canandaigua
VA facilities in upstate New York. These closings will be devastating to the veterans community as well as to the health care profes-
Letters to the editor referring to a recent
Journal article are encouraged up to 3 months
after the article’s appearance. By submitting a
letter to the editor, the author gives permission
for its publication in the Journal. Letters
should not duplicate material being published
or submitted elsewhere. The editors reserve the
right to edit and abridge letters and to publish
responses.
Text is limited to 400 words and 10 references. Submit on-line at www.ajph.org for
immediate Web posting, or at submit.ajph.org
for later print publication. On-line responses
are automatically considered for print
publication. Queries should be addressed to
the department editor, Jennifer A. Ellis, PhD,
at [email protected]
sionals who staff these facilities. The impact
on teaching hospitals and funding in these
areas is probably viewed as nothing more
than “collateral damage,” but such a change
will affect many lives and probably none for
the better. I am sure that New York veterans
are not alone in being affected by the draconian measures recommended by the
CARES Commission. Should readers want
more information on this subject, I can connect them with other veterans who can offer
much more knowledge than my limited grasp
of this serious situation allows.
The Journal is a welcome, positive, caring
voice in very uncertain times for veterans’
health care. Please keep this issue alive in the
public health community, as we need all the
help we can get.
George McAnanama, BS
About the Author
The author served in the US Army from 1966 to 1968 and
is now with Veterans for Peace, Chapter 34 (New York City).
Requests for reprints should be sent to George
McAnanama, 22 Livingston Ct, Staten Island, NY 103101630 (e-mail: [email protected]).
Reference
1. Fine MJ, Demakis JG. The Veterans Health Administration’s promotion of health equity for racial and
ethnic minorities. Am J Public Health. 2003;93:
1622–1624.
ELIMINATING HEALTH DISPARITIES:
FOCAL POINTS FOR ADVOCACY
AND INTERVENTION
We applaud the Journal’s October 2003 focus
on eliminating health disparities and the University of Pittsburgh’s National Minority
Health Research Summit. Efforts to reduce inequalities need to be derived from multiple
sources. Two recent publications about the
challenges of achieving equality for American
Indians, Alaska Natives, and Latino peoples
show the failings of our past approaches.1,2
The National Institutes of Health’s National
Center on Minority Health and Health Dis-
April 2004, Vol 94, No. 4 | American Journal of Public Health
parities is paving the way for a more aggressive research agenda in this area.3 Public
health professionals need to educate policymakers about the need for support for such
research, as well as for programs that translate research into practice, such as the Centers for Disease Control and Prevention’s Racial and Ethnic Approaches to Community
Health (REACH) 2010.4 REACH is funded
for fiscal year 2004 at only $37 million—the
same amount it received in fiscal year 2003.5
It is clear that government intentions alone
will not elevate the health status of America’s
marginalized and minority peoples. Voluntary
professional associations such as the Sociey
for Public Health Education and the American Public Health Association (APHA) must
call on their members to connect with all levels of society. A central organization can plan
training, promote cultural sensitivity, share information and research, and develop positions that can be used by other groups in
their advocacy efforts. APHA is demonstrating its leadership in this area by designating
“Eliminating Health Disparities” the theme of
National Public Health Week in 2004.6
In 1966, Dorothy Nyswander defined an
open society as “one where justice is the same
for every [person]; where dissent is taken seriously as an index of something wrong or
something needed; where diversity is expected; . . . where the best of health care is
available to all; where poverty is a community disgrace not an individual’s weakness;
[and] where desires for power over [people]
become satisfaction with the use of power for
people.”7(p37) In keeping with this vision, in
2000 the Society for Public Health Education
(SOPHE) commissioned an Open Society
Commission, which resulted in 6 resolutions
aimed at eliminating health disparities and
calling for widespread socioecological change.
Yet stronger organizational commitment from
groups such as SOPHE and APHA is crucial
to helping diverse practitioners eliminate such
continuing practices among health care professionals as racism, stereotyping, bias, discrimination, and cultural and professional in-
Letters | 519
 LETTERS 
competence. Let each public health professional rise to the challenges of achieving an
open society and reaching our national goal
of eliminating health disparities.
Patricia D. Mail, MPH, PhD, CHES
Sue Lachenmayr, MPH, CHES
M. Elaine Auld, MPH, CHES
Kathleen Roe, DrPH
About the Authors
The authors are members of SOPHE (http://www.sophe.org).
Patricia D. Mail, a member of the APHA Executive Committee, is with the Addictive Behaviors Research Center,
University of Washington, Seattle. Sue Lachenmayr is with
the New Jersey Department of Health and Senior Services,
Trenton. M. Elaine Auld is a member of APHA’s Action
Board. Kathleen Roe is with the Department of Health Science, San Jose State University, San Jose, Calif.
Requests for reprints should be sent to Sue Lachenmayr,
MPH, CHES, Society for Public Health Education, 750
First St, NE, Suite 910, Washington, DC 20002 (e-mail:
[email protected]).
References
1. Bird M, Bowekaty M, Burhansstipanov L, Cochran
PL, Everingham PJ, Suina M. Eliminating Health Disparities: Conversations with American Indians and
Alaska Natives. Santa Cruz, Calif: ETR Associates;
2002.
ERRATA
In: Roubideaux Y, Buchwald D, Beals J, Middlebrook D, Manson S, Muneta B, RithNajarian S, Shields R, and Acton K. Measuring the Quality of Diabetes Care for Older
American Indians and Alaska Natives. Am J Public Health. 2004;94:60–65.
An article was inadvertently published without the following authorial disclaimer and
acknowledgments:
Note [to About the Authors]. Opinions expressed by this article are those of the authors and do not necessarily reflect the views of the Indian Health Service.
Acknowledgments
This study was supported by the National Institute on Aging (grant 1P30AG/NE15292 to S. Manson and
D. Buchwald), the Administration on Aging (grant 90-AM-0757 to S. Manson), and the National Institute of
Mental Health (grant MH143175 to S. Manson).
The authors wish to thank the many Indian Health Service (IHS), tribal, and urban Indian program providers who contributed to the IHS Diabetes Care and Outcomes Audit in 1997.
In: Ellickson PL, Orlando M, Tucker JS, and Klein DJ. From adolescence to young
adulthood: racial/ethnic disparities in smoking. Am J Public Health. 2004;94:293–299.
Some racial/ethnic smoking rates were expressed incorrectly. On page 295 under the
subhead Regular Smoking, the categories “Asians” and “African Americans” were reversed for the rates given. The text should have read:
By the age of 23 years, regular smoking rates among Hispanics and Whites (29% and
32%, respectively) were approximately twice those among Asians (16%) and 1.5 to 1.7
times those among African Americans (19%).
2. Rodriguez-Trias H, Bracho A, Gil RM, et al. Eliminating Health Disparities: Conversations with Latinos.
Santa Cruz, Calif: ETR Associates; 2003.
3. National Institutes of Health. Strategic research
plan and budget to reduce and ultimately eliminate
health disparities. Volume 1, fiscal years 2002–2006.
Available at: http://www.ncmhd.nih.gov/strategicmock/
our_programs/strategic/volumes.asp (PDF file). Accessed December 16, 2003.
The Face of Public Health
4. Rowe KM, Thomas S. REACH 2010: engaging
the circle of research and practice to eliminate health
disparities: an interview with Imani Ma’at. Health Promot Pract. 2002;3:120–124.
5. Centers for Disease Control and Prevention Financial
Management Office. FY2004 Budget Appropriation Information: Funding by Functional Area. Available at:
http://www.cdc.gov/fmo/FY04%20functional%20table
(PDF file). Accessed February 2, 2004.
6. American Public Health Association announces
“call for solutions” to end health care disparities. Available
at: http://www.apha.org/nphw/pressroom/20031211.
cfm. Accessed January 25, 2004.
7. Nyswander DB. The open society: its implications
for health educators. In: Simonds SK, ed. The Philosophical, Behavioral and Professional Bases for Health
Education. Oakland, Calif: Third Party Publishing Co;
1982:29–42. Vol 1 of the SOPHE Heritage Collection
of Health Education Monographs.
520 | Letters
A
Stock No.: 0-87553-033-8
8 minutes ❚ color ❚ 2004
$13.99 APHA Members
$19.99 Nonmembers
plus shipping and handling
NEW
from
APHA
PHA has created a new video that health
advocates can use to explain the important
work carried out every day by the public health
community.
In this moving video, the people of public health
share what they do and how their work improves
and protects the lives of those in their communities. From healthy lifestyles and immunizations to
policy development and global health, you’ll see
the face of public health working for all of us.
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
VID04J4
American Journal of Public Health | April 2004, Vol 94, No. 4
 LETTERS 
competence. Let each public health professional rise to the challenges of achieving an
open society and reaching our national goal
of eliminating health disparities.
Patricia D. Mail, MPH, PhD, CHES
Sue Lachenmayr, MPH, CHES
M. Elaine Auld, MPH, CHES
Kathleen Roe, DrPH
About the Authors
The authors are members of SOPHE (http://www.sophe.org).
Patricia D. Mail, a member of the APHA Executive Committee, is with the Addictive Behaviors Research Center,
University of Washington, Seattle. Sue Lachenmayr is with
the New Jersey Department of Health and Senior Services,
Trenton. M. Elaine Auld is a member of APHA’s Action
Board. Kathleen Roe is with the Department of Health Science, San Jose State University, San Jose, Calif.
Requests for reprints should be sent to Sue Lachenmayr,
MPH, CHES, Society for Public Health Education, 750
First St, NE, Suite 910, Washington, DC 20002 (e-mail:
[email protected]).
References
1. Bird M, Bowekaty M, Burhansstipanov L, Cochran
PL, Everingham PJ, Suina M. Eliminating Health Disparities: Conversations with American Indians and
Alaska Natives. Santa Cruz, Calif: ETR Associates;
2002.
ERRATA
In: Roubideaux Y, Buchwald D, Beals J, Middlebrook D, Manson S, Muneta B, RithNajarian S, Shields R, and Acton K. Measuring the Quality of Diabetes Care for Older
American Indians and Alaska Natives. Am J Public Health. 2004;94:60–65.
An article was inadvertently published without the following authorial disclaimer and
acknowledgments:
Note [to About the Authors]. Opinions expressed by this article are those of the authors and do not necessarily reflect the views of the Indian Health Service.
Acknowledgments
This study was supported by the National Institute on Aging (grant 1P30AG/NE15292 to S. Manson and
D. Buchwald), the Administration on Aging (grant 90-AM-0757 to S. Manson), and the National Institute of
Mental Health (grant MH143175 to S. Manson).
The authors wish to thank the many Indian Health Service (IHS), tribal, and urban Indian program providers who contributed to the IHS Diabetes Care and Outcomes Audit in 1997.
In: Ellickson PL, Orlando M, Tucker JS, and Klein DJ. From adolescence to young
adulthood: racial/ethnic disparities in smoking. Am J Public Health. 2004;94:293–299.
Some racial/ethnic smoking rates were expressed incorrectly. On page 295 under the
subhead Regular Smoking, the categories “Asians” and “African Americans” were reversed for the rates given. The text should have read:
By the age of 23 years, regular smoking rates among Hispanics and Whites (29% and
32%, respectively) were approximately twice those among Asians (16%) and 1.5 to 1.7
times those among African Americans (19%).
2. Rodriguez-Trias H, Bracho A, Gil RM, et al. Eliminating Health Disparities: Conversations with Latinos.
Santa Cruz, Calif: ETR Associates; 2003.
3. National Institutes of Health. Strategic research
plan and budget to reduce and ultimately eliminate
health disparities. Volume 1, fiscal years 2002–2006.
Available at: http://www.ncmhd.nih.gov/strategicmock/
our_programs/strategic/volumes.asp (PDF file). Accessed December 16, 2003.
The Face of Public Health
4. Rowe KM, Thomas S. REACH 2010: engaging
the circle of research and practice to eliminate health
disparities: an interview with Imani Ma’at. Health Promot Pract. 2002;3:120–124.
5. Centers for Disease Control and Prevention Financial
Management Office. FY2004 Budget Appropriation Information: Funding by Functional Area. Available at:
http://www.cdc.gov/fmo/FY04%20functional%20table
(PDF file). Accessed February 2, 2004.
6. American Public Health Association announces
“call for solutions” to end health care disparities. Available
at: http://www.apha.org/nphw/pressroom/20031211.
cfm. Accessed January 25, 2004.
7. Nyswander DB. The open society: its implications
for health educators. In: Simonds SK, ed. The Philosophical, Behavioral and Professional Bases for Health
Education. Oakland, Calif: Third Party Publishing Co;
1982:29–42. Vol 1 of the SOPHE Heritage Collection
of Health Education Monographs.
520 | Letters
A
Stock No.: 0-87553-033-8
8 minutes ❚ color ❚ 2004
$13.99 APHA Members
$19.99 Nonmembers
plus shipping and handling
NEW
from
APHA
PHA has created a new video that health
advocates can use to explain the important
work carried out every day by the public health
community.
In this moving video, the people of public health
share what they do and how their work improves
and protects the lives of those in their communities. From healthy lifestyles and immunizations to
policy development and global health, you’ll see
the face of public health working for all of us.
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]com
Tel: (301) 893-1894
FAX: (301) 843-0159
VID04J4
American Journal of Public Health | April 2004, Vol 94, No. 4
AMERICAN JOURNAL OF
PUBLIC HEALTH

EDITOR’S CHOICE
The Solution Is Injury Prevention
I awoke this morning to the news that another American life was tragically lost because of a preventable injury, and I wondered
why our nation has been so complacent
about this problem. Then I remembered the
retort I received from a policymaker the last
time I sought funding for an injury program:
“Injury is not a public health problem!”
In 2000, injury was responsible for 10% of
health care expenditures—more than $117 billion. Injury is the leading cause of death for
Americans younger than 35 years and is a
leading cause of disability. In this issue of the
Journal, Lynda Doll and Sue Binder of the
National Center for Injury Prevention and
Control lay out the case for injury as one our
nation’s most preventable problems. They
point to the incredible toll extracted by injury
in human suffering as well as the staggering
fiscal costs to society.
Why, then, is it so difficult to convince policymakers that injury is a public health problem worth addressing? Maybe because they
still believe in the accident paradigm. This
line of thinking argues that injuries are an act
of fate, and while it makes sense to be more
careful, injuries will occur despite our best
efforts.
The evidence against the accident paradigm is exhaustive. Automobile-related injury
has been dramatically reduced by a multifaceted effort to make cars safer. By redesigning
automobile brakes, steering columns, sidewall
protections, seat belts, air bags, and a host of
other safety features, we have made the automobile a much safer machine. The addition
of programs to reduce the incidence of impaired driving and promote driver training
have addressed the human aspects of automobile safety.
Pedestrian trauma from motor vehicles is
on the rise and, once again, safety experts are
intervening by developing novel ways to reduce the hazards of walking. On April 7,
2004, the World Health Organization is em-
phasizing the global scope of the problem by
declaring road safety as the theme for World
Health Day.
In my days as an emergency physician, I
was struck by the fact that most of the injuries I treated could have been prevented if
only simple precautions had been taken. My
colleagues and I would give individualized
advice to patients on how to prevent future
injuries while we provided medical care for
their current injuries. Later, when I became
active in public health, I recognized the need
for population-based interventions as tools to
reduce injuries, such as the interventions advanced in this issue of the Journal regarding
motorcycle helmet laws, speed bumps, and
programs to prevent dating violence. These
interventions work. They reduce injuries, save
lives, and prevent disabilities. They also save
money—lots of money. These are marvelous
accomplishments that are largely overlooked
by most policymakers.
At a time when economists predict doubledigit increases in health care expenditures, I
have a solution: injury prevention. At a time
when we are looking for ways to address disparities in health care, I have a solution: injury prevention. At a time when we are looking to protect our children from harm, I have
a solution: injury prevention. The next time
someone challenges us on the propriety of
addressing injury as a public health problem,
point out that if something kills people or
hurts people, it’s our responsibility as public
health leaders to find a solution.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Georges C. Benjamin, MD, FACP
Executive Director, APHA
EDITOR-IN-CHIEF
Mary E. Northridge, PhD, MPH
TECHNICAL DEPUTY EDITOR
Jennifer A. Ellis, PhD
FEATURE EDITOR
Gabriel N. Stover, MPA
ASSOCIATE EDITORS
Mary Bassett, MD, MPH
Don C. Des Jarlais, PhD
Lawrence J. Fine, MD, DrPH
Michael R. Greenberg, PhD
Michael Gross, PhD
Sofia Gruskin, JD, MIA
Deborah Holtzman, PhD, MSW
Said Ibrahim, MD, MPH
Sherman A. James, PhD
Robert Sember
Stella M. Yu, ScD, MPH
Roger Vaughan, DrPH, MS
INTERNATIONAL ASSOCIATE EDITORS
Daniel Tarantola, MD (Geneva, Switzerland)
Cesar Gomes Victora, MD, PhD (Pelotas, Brazil)
DEPARTMENT EDITORS
John Colmers, MPH
Government, Politics, and Law
Elizabeth Fee, PhD, and Theodore M. Brown, PhD
Images of Health
Public Health Then and Now
Voices From the Past
Mary E. Northridge, PhD, MPH
Health Policy and Ethics Forum
Public Health Matters
EDITORIAL BOARD
Kenneth R. McLeroy, PhD (2004), Chair
Frank J. Chaloupka, PhD (2006)
Vanessa Northington Gamble, MD, PhD (2006)
Audrey R. Gotsch, DrPH, CHES (2004)
M. Lyndon Haviland, DrPH (2005)
Michael D. Kogan, PhD (2004)
Linda Young Landesman, DrPH, MSW (2006)
Bruce Lubotsky Levin, DrPh, MPH (2005)
Marsha D. Lillie-Blanton, DrPH (2004)
Kusuma Madamala, MPH (2006)
Gregory Pappas, MD, PhD (2005)
Alan Steckler, DrPH (2006)
Henrie M. Treadwell, PhD (2005)
Terrie F. Wetle, PhD (2004)
Siu G. Wong, OD, MPH (2004)
Ruth E. Zambrana, PhD (2005)
STAFF
Georges C. Benjamin, MD, FACP
Executive Director
Ellen T. Meyer, Director of Publications
Nancy Johnson, MA, Managing Editor
Dave Stockhoff, MA, Production Editor
Noëlle A. Boughanmi, Associate Production Editor
Cynthia Calgaro, Assistant Production Editor
Ashell Alston, Director of Advertising
Susan Westrate, Michele Pryor, Graphic Designers
Heather Wildrick, Editorial Assistant
Jim Richardson, Joan M. Fuller, Reviews and Customer
Service Coordinators
FREELANCE STAFF
Janis Foster, Greg Edmondson,
Gary Norton, Graham Harles, Susan Blanton,
Gretchen Becker, Alisa Guerzon, Copyeditors
Rebecca Richters, Editorial Proofreader
Alison Moore, Chris Filiatreau, Chrysa Cullather,
Proofreaders
Editor’s Choice | 521

EDITORIALS
Injury
Prevention
Research at
the Centers
for Disease
Control and
Prevention
Recognizing the critical public
health burden that unintentional
and violent injuries place on the
United States, Congress mandated in 1992 that the Centers
for Disease Control and Prevention (CDC) create the National
Center for Injury Prevention and
Control (NCIPC). NCIPC was established to coordinate research
and programmatic responses to
the problem of nonoccupational
injuries.
MORBIDITY, MORTALITY,
AND THE COST OF
INJURIES
Violent and unintentional injuries place a severe physical,
emotional, and financial burden
on our communities. Injuries do
not discriminate; they affect all
races and ages. In fact, injuries
are the leading cause of death in
the first 4 decades of life.1 In
2001, the leading causes of
deaths due to injury in the
United States were motor vehicle
crashes, suicides, and falls.
In addition to being a major
cause of death, injuries cause
suffering and disability. Virtually everyone knows of someone whose life has been
changed because of a motor
vehicle crash or other injurycausing event. Each year, Americans make 30 to 40 million
visits to emergency departments
for treatment of injuries.2 Except for teenagers and young
adults, falls are the leading
cause of emergency department
visits; for persons aged 15 to 23
years, motor vehicle crashes
and striking or being struck by
objects predominate.1
522 | Editorials
The economic costs of injuries
impose a significant burden on
society as well. In 2000, the
United States spent $117 billion
treating injuries, accounting for
10% of all medical expenditures
that year.3 The percentage of
total medical expenditures accounted for by injuries in 2000
was comparable to the percentages attributable to other leading
public health issues, such as obesity (9.1%) and smoking
(6.5%–14.4%).
The mission of NCIPC is to
prevent or reduce injuries. To accomplish its goals, NCIPC works
with numerous partners to support injury surveillance, research, and prevention programs
and to disseminate information
that can inform prevention programs and policies. There is
strong evidence of the effectiveness of many preventive interventions, including use of seat
belts4 and bicycle helmets,5 laws
establishing 0.08 blood alcohol
content as the definition for
drunk driving,6 and residential
smoke alarm and fire safety education programs.7 Effective violence prevention strategies include home visitation of new
parents to prevent mistreatment
of children8 and tenant-based
rental assistance programs to
prevent youths from witnessing
or becoming victims of crime.9
Widespread implementation of
such interventions could save
thousands of lives annually. However, despite the progress that
has been made, there is still
much to be learned about preventing unintentional and violent
injuries and about encouraging
the dissemination and adoption
of strategies that have proven
effective.
NCIPC RESEARCH:
GOALS AND STRATEGIES
Through a participatory process, NCIPC and its partners developed an Injury Research
Agenda, which was published in
June 2002.10 The agenda categorizes research needs along a continuum, beginning with risk factor
identification, proceeding through
intervention evaluation, and ending with dissemination research.
While the agenda recognizes
the need for additional descriptive
research, it emphasizes the right
side of the continuum—intervention
and dissemination research.
The Injury Research Agenda
includes 7 broad categories: injuries occurring at home and in
the community; injuries occurring during sports, recreation, or
exercise; transportation injuries;
intimate partner violence, sexual
violence, and child maltreatment;
suicide; youth violence; and
acute care, disability, and rehabilitation. The first 6 emphasize
prevention, while the seventh focuses on improving outcomes
when prevention efforts fail (e.g.,
enhancing systems for emergency treatment).
The agenda includes crosscutting themes such as alcohol
use, parenting and supervision
styles, economic costs, and dissemination of scientific findings.
It also identifies the importance
of building injury-related research infrastructure, which, for
example, will provide ongoing
support for young researchers.
The NCIPC New Investigator
American Journal of Public Health | April 2004, Vol 94, No. 4
 EDITORIALS 
and Dissertation grant awards, as
well as funds for training and
pilot studies, have been established to help fill this need.
Of the research issues raised in
the agenda-setting process, 48
topics were deemed highest priority, each of which will require
10 to 20 studies to address adequately. NCIPC is using its available funds to address these priority topics; researchers wishing to
apply for NCIPC funds can use
the research agenda to anticipate
future funding announcements. In
fiscal year 2003, the NCIPC research budget of $41 million
funded researcher-initiated grants
(37.5% of funds), research centers (29.4%), research cooperative agreements (25.5%), and research contracts (7.6%). At this
level of funding, full implementation of the Injury Research
Agenda may take some time.
However, at CDC it is firmly believed that lives will be saved
and suffering will be reduced
with this funding.
INJURY RESEARCH:
OPPORTUNITIES AND
CHALLENGES
The context in which research
is conducted is changing rapidly.
Injury researchers should be
aware of changing opportunities
and expectations related to accountability for the expenditure
of federal funds, openness during
the conduct and dissemination of
research findings, emphasis on
research addressing linked health
problems, and globalization.
Federal agencies are increasingly being held accountable not
just for the appropriateness of
their expenditures, but also for
describing the impact of their research. Typical measures of research success (such as published
findings in peer-reviewed jour-
nals), while easy to count, do not
answer the fundamental question
“What difference has your research made?” Answers like “It is
too soon to tell” no longer satisfy
policymakers faced with difficult
funding decisions.
While continuing to produce
publications and other products,
scientists should also collect personal stories from people whose
lives have been affected by the
research, as well as information
on specific uses of the findings.
Initial research protocols should
include strategies for disseminating the findings and ensuring
that the next steps in the public
health continuum are taken. Next
steps might include conducting
follow-up research, sustaining a
successful demonstration program, or disseminating study results to policymakers and tracking whether the results effect
policy or legislative change.
The speed and scope of the
global communication infrastructure present an enormous opportunity to injury researchers.
Members of research consortia
have for some time used the Internet to share information. Recently, the public health community received a glimpse of its
potential future when researchers
around the world shared data to
quickly identify the causal organism of severe acute respiratory
syndrome (SARS).11 This unprecedented collaboration provides a model for how researchers can solve complex public
health problems by working together. The application of this
model in nonacute settings needs
to be explored.
The enhanced communication
infrastructure provides opportunities to respond to the public’s
increasing demand for health information. Injury researchers
now have the opportunity to dis-
April 2004, Vol 94, No. 4 | American Journal of Public Health
seminate their findings not just
through academic journals, but
also through Web sites, newspapers, and other venues with
large audiences. Often, researchers are reluctant to reduce their
findings to sound bites that fail
to capture the nuances of their
work. Nevertheless, by quickly
disseminating their findings, researchers can take advantage of
the public’s interest in health and
safety and can provide usable information for both the public
and policymakers.
The Internet has generated interest in the sharing of governmentsponsored research data. New
policy requires researchers receiving federal funds above a
specific amount to make their
data available to others, while at
the same time ensuring the confidentiality of research participants.12 The goal of this policy is
to ensure that data are used as
widely as possible to inform research and program efforts. The
policy may also inspire new collaborations between researchers
and practitioners with shared
interests.
Openness and participation
also apply to the way research is
conducted. This is reflected by
the increasing emphasis placed
on research that incorporates
community partners at all
stages.13 Community participatory research not only increases
the relevancy (external validity)
of the work, it also enhances the
adoption of scientific findings in
practice settings. However, such
research can be very challenging.
Researchers and community
partners with different research
emphases may need to negotiate.
Research outcomes may take
longer, and questions may arise
about the quality or rigor (internal validity) of the research.
Lessons learned from successful,
rigorous community research
should be disseminated to the injury research community and be
used to train students in community research methods and partnership building.
Another prevalent theme in
injury research is the need to account for linkages between various health problems as they
occur among individuals and in
communities. This consideration
often arises in the context of
doing community participatory
research. CDC, including
NCIPC, has tended to fund research in a categorical manner,
with funds provided to address
specific outcomes (e.g., motor
vehicle injuries) or risk factors
(e.g., tobacco use). However, behaviors and other risk factors
are often common to many
types of injuries and public
health problems. For example,
risky alcohol use is a wellknown factor in many types of
injuries, and recent research
shows impulsivity to be a potentially important risk factor for
suicide,14 unintentional injuries,
deaths from motor vehicle
crashes, and drownings. Understanding how to measure and
improve the supervision of
young children is also critical in
addressing child neglect, injuries
due to falls, and drownings.
NCIPC funds for community
participatory research and addressing crosscutting issues have
been limited. CDC, however, has
received funding for extramural,
peer-reviewed prevention research and has directed the
funds toward these purposes,15
thus placing a high priority on increasing its investment in such
research.
While much of CDC’s research, including that of NCIPC,
has focused on interventions directed at individuals, etiological
Editorials | 523
 EDITORIALS 
studies have shown the importance of community-level factors.
For example, community-level
variables associated with violence include poverty, residential
instability, and low neighborhood
collective efficacy. These factors
have an impact beyond what
might be expected from the characteristics of the individuals living
within the community.16 Using
these data, NCIPC hopes to stimulate research on the effectiveness of modifying communitylevel factors to reduce violent
outcomes.
A final theme in injury prevention is the need to be an active
participant in a global community that is committed to injury
prevention and control. While
most CDC investments in global
health target infectious diseases,
CDC and others recognize the
growing importance of preventing and controlling noncommunicable diseases and conditions.
Worldwide, injuries—whether
they result from road traffic, suicide, falls, interpersonal violence,
or war—take an enormous toll on
lives. Motor vehicle crashes alone
are anticipated to become the
third leading cause of disabilityadjusted life years (DALYs) by
2010, up from ninth place in
1990.17 As we increase our ability to control infectious and nutritional causes of child death
and illness, we observe that injuries make up a greater proportion of DALYs among young children. For example, in Southeast
Asia, unintentional injuries are
now the fifth leading cause of
DALYs among children younger
than 5 years.18
Emphasizing global efforts to
reduce road traffic deaths and injuries is particularly crucial at
this time. The injury burden of
road traffic crashes is steadily increasing worldwide; globally, for
524 | Editorials
men aged 15 to 44 years, road
traffic injuries rank second only
to HIV/AIDS as the leading
cause of illness and premature
death.18 In recognition of this
burden, the World Health Organization has dedicated World
Health Day 2004 to road traffic
safety.19 On April 7, 2004, and
in the weeks that follow, events
around the world will draw attention to road traffic crashes
and potential solutions. The
United Nations will also stress
the need for public health and
transportation agencies to work
together to address this problem.
CONCLUSION
Worldwide, injuries remain a
leading cause of death and suffering. While effective interventions exist for some injury-related
problems, more research is
needed to better understand how
successful interventions can be
incorporated into practice settings. For other injury issues,
much more research is needed.
The CDC Injury Research
Agenda lays out NCIPC research
priorities for the next several
years. As researchers develop
and implement the ideas presented in the agenda, they
should take into account the
changing research scene and
their role in reducing the public
health burden worldwide, and
they should avail themselves of
opportunities to make an even
greater impact.
Lynda Doll, PhD,
Sue Binder, MD
About the Authors
The authors are with the National Center
for Injury Prevention and Control, Centers
for Disease Control and Prevention, Atlanta, Ga.
Requests for reprints should be sent to
Lynda Doll, PhD, National Center for In-
jury Prevention and Control, 4770 Buford
Hwy, Mail Stop K-02, Atlanta GA
30341-3724 (e-mail: [email protected]).
This editorial was accepted January 7,
2004.
References
1. National Center for Injury Prevention and Control. Web-Based Injury Statistics Query and Reporting System
(WISQARS). 2001. Available at: http://
www.cdc.gov/ncipc/wisqars. Accessed
December 31, 2003.
2. Bonnie RJ, Fulco CE, Liverman
CT, eds. Reducing the Burden of Injury:
Advancing Prevention and Treatment.
Washington, DC: National Academies
Press; 1999.
3. Finkelstein E, Fiebelkorn I, Corso
P, Binder S. Medical expenditures attributable to injuries—United States, 2000.
MMWR Morb Mortal Wkly Rep. 2004;
52:1–9.
4. Dinh-Zarr TB, Sleet DA, Shults
RA, et al. Reviews of evidence regarding interventions to increase use of
safety belts. Am J Prev Med. 2001;21
(suppl 4):48–65.
5. Thompson RS, Rivara FP, Thompson DC. A case–control study of the
effectiveness of bicycle safety helmets.
N Engl J Med. 1989;320:1361–1367.
6. Shults RA, Elder RW, Sleet DA, et
al. Reviews of evidence regarding interventions to reduce alcohol-impaired
driving [published correction appears in
Am J Prev Med. 2002;23:72]. Am J Prev
Med. 2001;21(4 suppl):66–88.
7. Mallonee S, Istre GR, Rosenberg
M, et al. Surveillance and prevention of
residential-fire injuries. N Engl J Med.
1996;335:27–31.
8. Hahn RA, Bilukha OO, Crosby A,
et al. First reports evaluating the effectiveness of strategies for preventing violence: early childhood home visitation. Findings from the Task Force on
Community Preventive Services.
MMWR Morb Mortal Wkly Rep. 2003;
52(RR-14):1–9.
agnosis. A multicentre collaboration to
investigate the cause of severe acute
respiratory syndrome. Lancet. 2003;
361:1730–1733.
12. Centers for Disease Control and
Prevention CDC/ATSDR policy on releasing and sharing data. 2003. Available at: http://www.cdc.gov/od/foia/
policies/sharing.htm. Accessed January
2, 2004.
13. Israel BA, Schulz AJ, Parker EA,
Becker AD. Review of communitybased research: assessing partnership
approaches to improve public health.
Annu Rev Public Health. 1998;19:
173–202.
14. Silverman MM, Simon TR, eds.
Houston case–control study of nearly
lethal suicide attempts. Suicide LifeThreatening Behav. 2001;32:1–84.
15. Centers for Disease Control and
Prevention. Community-based participatory prevention research grants. 2003.
Available at: http://www.phppo.cdc.gov/
od/oser/PRGRants.asp. Accessed January 2, 2004.
16. Sampson RJ, Raudenbush SW,
Earls F. Neighborhoods and violent
crime: a multilevel study of collective
efficacy. Science. 1997;277:918–924.
17. Disease burden measured in disability-adjusted life years. Available at:
http://www.who.int/msa/mnh/ems/
dalys/table.htm. Accessed January 1,
2004.
18. Krug E, ed. Injury: A Leading
Cause of the Global Burden of Disease.
Geneva, Switzerland: World Health Organization; 1999.
19. World Health Day 2004: road
safety. Available at: http://www.who.
int/world-health-day/2004/en. Accessed January 28, 2004.
9. Anderson LM, Shinn C, St. Charles J,
et al. Community interventions to promote healthy social environments: early
childhood development and family
housing. A report on recommendations
of the Task Force on Community Preventive Services. MMWR Morb Mortal
Wkly Rep. 2002;51(RR-1):1–8.
10. National Center for Injury Prevention and Control. CDC injury research
agenda. 2002. Available at: http://www.
cdc.gov/ncipc/pub-res/research_agenda/
agenda.htm. Accessed January 2, 2004.
11. World Health Organization Multicentre Collaborative Network for Severe
Acute Respiratory Syndrome (SARS) Di-
American Journal of Public Health | April 2004, Vol 94, No. 4
 FIELD ACTION REPORT 
Integrating the Environment, the Economy, and
Community Health: A Community Health Center’s
Initiative to Link Health Benefits to Smart Growth
| Peter V. McAvoy, JD, MS, Mary Beth Driscoll, BA, and Benjamin J. Gramling, BS
The Sixteenth Street Community Health Center (SSCHC) in Milwaukee, Wis, is making a difference in the livability of surrounding neighborhoods and the overall health of the families it serves.
SSCHC is going beyond traditional health care provider models
and working to link the environment, the economy, and community health through urban brownfield redevelopment and
sustainable land-use planning.
In 1997, SSCHC recognized that restoration of local air and
water quality and other environmental conditions, coupled with
restoring family-supporting jobs in the neighborhood, could have
a substantial impact on the overall health of families. Recent
events indicate that SSCHC’s pursuit of smart growth strategies
has begun to pay off.
A HOLISTIC HEALTH
APPROACH
OPERATING AS 1 OF 15
federally qualified, communitybased health centers in Wisconsin, the Sixteenth Street Community Health Center (SSCHC) has
for more than 34 years relied on
a place-based mission in offering
primary health care to families
living in Milwaukee’s Near SouthSide neighborhood, which primarily comprises low-income Latinos. SSCHC’s Department of
Environmental Health was created in 1997 to address environmental factors that affect health,
including deteriorating lead paint
in housing and poor air and
water quality. The department
was charged with achieving a
healthy environment within its
service area through restoring
abandoned, environmentally contaminated industrial sites; attract-
ing high-quality investment; and
creating family-supporting jobs to
increase the prosperity of the
low-income families it serves,
thereby increasing constituents’
ability to pay for quality health
care, nutritious food, and suitable
housing.
This approach complements
the ongoing development of
“smart growth” plans by Wisconsin municipalities as required by
the Wisconsin State Legislature.
Wisconsin’s “smart growth” legislation offers financial assistance
to municipalities for long-range
planning that links transportation
and land-use policies to quality
of life in both urban and rural
settings. Additional information
regarding Wisconsin’s “smart
growth” legislation can be found
at http://www.doa.state.wi.us/olis.
SSCHC’s Department of
Environmental Health program
promotes sustainable develop-
February 2004, Vol 94, No. 2 | American Journal of Public Health
ment which will create a viable
alternative to the sprawling
suburban development that has
come to characterize southeastern Wisconsin. Sustainable
development features the reuse
of existing buildings and land
(including brownfields), conserving residential neighborhoods,
maintaining local community
character, promoting the health
of the community, and protecting
the environment for future
generations.
THE CENTER OF IT ALL:
MILWAUKEE’S MENOMONEE
RIVER VALLEY
SSCHC’s service area includes
the Menomonee River Valley, a
1500-acre collection of properties that is adjacent to downtown
Milwaukee and Lake Michigan
and surrounded by the most
densely populated neighborhoods in Wisconsin. This valley
was the center of Wisconsin’s industrial production for a century,
employing more than 50 000
people at its peak. Many of the
workers lived in neighborhoods
bordering the valley and either
walked to work or rode a trolley.
Over the last 25 years, many
industrial manufacturers have either closed or relocated. With
the loss of nearby jobs, many
family breadwinners are forced
to commute an hour or more to
jobs in surrounding suburbs. Because few mass-transit alternatives are available, the few workers who own cars must join
countless other commuters using
the region’s interstate highways.
Local and regional transportation
patterns, coupled with industrial
and environmental factors, are
associated with high rates of
asthma and respiratory illness.1
In addition, poor land stewardship, non–point-source pollution,
and contaminated harbor sediments resulted in poor water
quality, which contributed to
KEY FINDINGS
• A community health center
can link restoration of the
local environment, creation
of good family-supporting
jobs, and public health.
• Visioning exercises help residents visualize how a revitalized area can look and
function.
• These visioning and design
events have served as a catalyst for achieving high-quality,
well-designed redevelopment.
• Redevelopment of industrial
brownfield sites may be an
alternative to suburban and
exurban sprawl.
McAvoy et al. | Peer Reviewed | Field Action Report | 525
 FIELD ACTION REPORT 
FIGURE 1—Participants in the 1999 charrette, or visioning and
design workshop, were charged with designing ways to bring highquality investors and family-supporting jobs back to the community
and to reverse the Menomonee River Valley’s historical
environmental abuse.
beach closings (1 of every 4 days
in 2003) and ongoing fishconsumption advisories.2
BEYOND POLICY
The success of the SSCHC-led
1999 Sustainable Development
Design Charrette for Milwaukee’s
Menomonee River Valley (Figure
1) provided visions that fueled
the need to develop site-specific
land-use plans that would accommodate SSCHC’s sustainability
and “smart growth” objectives.
The charrette, or visioning workshop, involved over 140 local design professionals from the public
and private sector. In collaboration with these and other partners, SSCHC hosted the 2002
Menomonee River Valley National Design Competition: Natural Landscapes for Living Communities, which focused on a
140-acre parcel within the
Menomonee River Valley (Figure
2). This property historically supported 5000 employees of the
Chicago, Milwaukee, St. Paul &
Pacific railroad company (the Milwaukee Road), but it employed
increasingly fewer persons in the
second half of the 20th century,
a trend typical of many
Menomonee Valley enterprises.
Barriers to redeveloping this site
are characteristics shared by
most Menomonee Valley properties: poor access, decrepit buildings, impaired soils and groundwater, and low property values.
To obtain ideas regarding the
revitalization of the Menomonee
Valley, SSCHC held the aforementioned design competition
sponsored in part by the National
Endowment for the Arts. A field
of 25 teams was narrowed to 4
finalists, each with experience in
dealing with environmental contamination, landscape architecture, and natural landscaping, as
well as storm-water and floodmanagement techniques. A
group of technical advisors outlined design elements for the set
of problems to which the teams
responded and evaluated the
final designs for technical merit
before a jury of national and
local experts determined the winning plan.
Held over 6 months, the competition process included opportunities for public input. The win-
526 | Field Action Report | Peer Reviewed | McAvoy et al.
ning submission included an industrial park that will provide
family-supporting jobs for surrounding neighborhoods while
significantly adding to the city’s
property tax base. The master
plan allowed for the integration
of natural and open-space elements into the industrial park, including a community green, a
storm-water park that will prevent
water pollution, and Wisconsin’s
new multi-use Hank Aaron State
Trail. In addition, the plan called
for restoration of ecological
systems within the affected segment of the Menomonee River,
a step which will address the valley’s and the surrounding neighborhoods’ lack of recreational
open space.
FINDINGS
The 1999 design charrette
and the 2002 national design
competition have been critical in
providing residents of Milwaukee
with a vision of how a revitalized
Menomonee River Valley could
look and function. They combine
to illustrate an exercise in moving from conceptual analysis and
brainstorming to real-world plan-
FIGURE 2—The abandoned railroad yard (top photo) was the focus
of the 2002 Menomonee River Valley National Design Competition:
Natural Landscapes for Living Communities, which attracted nationally
recognized, award-winning teams from around the United States,
the United Kingdom, Germany, and Canada. The winning master plan
by Wenk Associates of Denver, Colorado (bottom photo), provides both
development space for new jobs and recreational opportunities for the
community in a setting where environmental damage has been repaired.
American Journal of Public Health | February 2004, Vol 94, No. 2
 FIELD ACTION REPORT 
TABLE 1—The Menomonee
Valley Benchmarking Initiative’s
Economic, Environmental, and
Community Indicators for
Tracking Progress in the
Sixteenth Street Community
Health Center’s Sustainability
Objectives3
Economy
Business activity
Employment
Commercial/industrial property
Infrastructure and access
•
•
•
•
Environment
Water quality
Air quality
Land cover and habitat
Flora and fauna
•
•
•
•
Community
Housing
Crime
Arts and events
Health
•
•
•
•
ning and implementation of sustainable redevelopment practices.
Their outcomes, and the widespread media coverage they received, have served as a catalyst
for achieving high-quality, welldesigned redevelopment that will
ensure that people of the adjoining neighborhoods and surrounding communities are reconnected
to the valley through new jobs
and recreational opportunities.
The momentum that has built
around the 1999 design charrette and the national design
competition of 2002 has also
spurred the successful cleanup
of 3 contaminated industrial
sites totaling 38 acres. In addition, the City of Milwaukee, in
collaboration with the civic nonprofit organization Menomonee
Valley Partners, Inc, has initiated
a master planning, cleanup, and
redevelopment effort for a col-
Contributors
Menomonee Valley Sustainable Design Guidelines
Site design
Construction and
demolition
P. V. McAvoy edited and revised the
report. M. B. Driscoll wrote the report.
B. J. Grambling reviewed and revised
the report.
Acknowledgments
Building design and
energy use
Indoor environmental
quality
Materials and
resources
Operations and
maintainence
FIGURE 3—Sustainable guidelines for the Menomonee Valley, which
identify sustainability objectives for developers and property owners
and offer practical suggestions for achieving high performance
green architecture.
The Sixteenth Street Community Health
Center’s sustainable development efforts
have received major support from the
US Environmental Protection Agency,
the Joyce Foundation, the Brico Fund,
the Milwaukee Metropolitan Sewerage
District, the Greater Milwaukee Foundation’s Lenore T. Zinn Environmental
Fund and the Geerda A. Debelak Fund,
the Forest County Potawatomi Community Foundation, the Wisconsin Department of Natural Resources, and the Wisconsin Energy Corporation Foundation
Inc, along with contributions from multiple individual donors
References
lection of properties on the east
end of the Menomonee Valley,
which will add an additional 20
acres of development space. It is
estimated that several thousand
new jobs will be created in the
valley once all properties are
fully redeveloped.
To determine an employment
baseline, a study conducted by
SSCHC and the University of
Wisconsin–Milwaukee, under the
auspices of the Menomonee
Valley Benchmarking Initiative
(Table 1), calculated the number
of jobs located in the Menomonee
Valley during 2002 to be 9451
(which includes 7961 full-time
positions). This study will be
replicated in 2004 and in subsequent years to measure increased
employment opportunities and
long-term progress toward other
economic, environmental, and
social indicators of sustainability
objectives in the valley.3
A community health center
can make a difference in the
livability of surrounding neighborhoods and the overall health
of the families it serves by going
beyond traditional health care
provider models and working to
link the environment, the econ-
April 2004, Vol 94, No. 4 | American Journal of Public Health
omy, and community health.
SSCHC and its partners are
working to establish measurable
standards for private-sector, sustainable development by developing sustainable design guidelines (Figure 3) and marketing
the Menomonee Valley to investors committed to “smart
growth” principles.
Although redevelopment of
Milwaukee’s Menomonee River
Valley began only recently, the
area is already undergoing significant change. Ultimately, success
will be achieved when the valley’s environment is cleaned up,
new family-supporting jobs
located close to housing are created and held by neighborhood
residents, and the health and
livability of neighborhoods
surrounding the valley are
substantially improved.
1. United States Environmental Protection Agency. Smog—Who Does it
Hurt? What You Need to Know About
Ozone and Your Health. Washington,
DC: US Environmental Protection
Agency; 1999.
2. Bannerman RT, Owens DW,
Dodds RB, Hornewer NJ. 1993. Sources
of Pollutants in Wisconsin Stormwater.
Water Science Technology. 28:241–59.
3. Menomonee Valley Benchmarking Initiative State of the Valley Report; Sixteenth Street Community
Health Center and the University of
Wisconsin–Milwaukee’s Center for
Urban Initiatives and Research. Available at: http://www.mvbi.org. Accessed March 8, 2004.
About the Authors
The authors are with the Department of
Environmental Health, Sixteenth Street
Community Health Center, Milwaukee, Wis.
Requests for reprints should be sent to
Mary Beth Driscoll, Department of Environmental Health, Sixteenth Street Community Health Center, 1337 S Cesar
Chavez Drive, Milwaukee, WI 53204
(e-mail: [email protected]).
This report was accepted April 3, 2003.
McAvoy et al. | Peer Reviewed | Field Action Report | 527
 PUBLIC HEALTH THEN AND NOW 
Policies of Inclusion
Immigrants, Disease, Dependency, and
American Immigration Policy at the Dawn
and Dusk of the 20th Century
The racial politics of immigration have punctuated national
discussions about immigration
at different periods in US history, particularly when concerns
about losing an American way of
life or American population have
coincided with concerns about
infectious diseases.
Nevertheless, the main theme
running through American immigration policy is one of inclusion. The United States has historically been a nation reliant on
immigrant labor and, accordingly, the most consequential
public policies regarding immigration have responded to disease and its economic burdens
by seeking to control the behavior of immigrants within our
borders rather than excluding
immigrants at our borders.
| Amy L. Fairchild, PhD, MPH
CARVED IN STONE IN A PILLAR
adorning the National Archives
in Washington, DC, is a line from
Shakespeare that has captured
my imagination: “What is past is
prologue.” Day after day I read
this phrase as I entered the
archives, reviewing the records of
the US Public Health Service
(PHS) relating to the immigrant
medical inspection that was required at the nation’s borders beginning in 1891. Whether boxes
and boxes of records impaired
my judgment, whether I was
swayed by its inexorable logic, or
whether it simply felt true for the
ways in which we have thought
about the intersection of immigration and disease, I became a
firm believer: “What is past is
prologue.”
But the problem is that historians have provided us with multiple prologues, and this can trip
us up when we try to make the
past speak to the present. Despite widespread assumptions
about the exclusionary nature of
American immigration policy,
the history of immigration at the
beginning and end of the 20th
century is in fact a history of
inclusion.
The period 1924 to 1965 is
set off by 2 landmark pieces of
528 | Public Health Then and Now | Peer Reviewed | Fairchild
legislation: the Immigration Act
of 1924, which made national
origin the basis for admission
into the United States, and the
Immigration Act of 1965,1 which
eliminated the national origins
systems and at the peak of the
Civil Rights Movement restored
what President Lyndon Johnson
called “the basic principle of
American democracy.”2 But in
placing undue emphasis on the
racially restrictive nature of policy between 1924 and 1965, it
becomes too easy to view all
policies—past and present—
through a lens of exclusion.3
I had been led to explore the
early history of medical inspection by a contemporary policy
disaster that occurred while I
was working in the Policy Unit of
the New York State Department
of Health’s AIDS Institute in the
early 1990s. At that time, the US
detention of some 200 Haitian
immigrants infected with HIV at
the naval base at Guantanamo
Bay, Cuba, reached its climax. I
found it very easy, using the Immigration Act of 1924 and histories of eugenics and scientific
racism as a lens, to tell a story of
racial restrictions masquerading
as public health policy. It was a
story, I argued, emblematic of
our larger immigration policy.4
Conversely, I was prepared to see
the historical origins of immigrant medical inspection as the
story of public health used for
racial demarcation and exclusion.
This interpretation is very much
in keeping with how social historians have traditionally viewed
the relationship between immigration and disease.5
In this article, I compare the
broad intentions of US policy fundamentally concerned with managing the economic burden of
disease in 2 periods: the Progressive Era, in which medical inspection sought to control the consequences of disease not by turning
immigrants back but by introducing them to industrial values and
expectations regarding work and
dependency, and the current era
of immigration and welfare reform. Although the racial politics
of immigration have typically
framed our understanding of Progressive Era policy, in discussing
that period I consider the day-today practice of immigrant medical inspection and the ways it
was shaped by industrial demands. For the present era, in
which individuals certainly live
within the constraints of both federal and state policy on a day-to-
American Journal of Public Health | April 2004, Vol 94, No. 4
 PUBLIC HEALTH THEN AND NOW 
day basis, I focus on the politics
and policy at the broadest level,
considering the provisions of the
1996 Personal Responsibility and
Work Opportunity Reconciliation
Act (PRWORA, or Personal Responsibility Act), which excluded
illegal and legal immigrants from
many public benefits such as
Medicaid.
There are, of course, important differences between the 2
eras: immigrants, particularly illegal immigrants, fuel the service
industry and highly skilled immigrants fill the ranks of the information sector in the current era
of globalization, whereas in the
late 19th and early 20th centuries immigrants joined the unskilled industrial labor force.
Likewise, the changing position
and power of organized labor potentially gives different meaning
to the notion of inclusion in the
2 different eras. Nonetheless, at
both the dawn and dusk of the
20th century, I argue, the most
consequential public policies responded to disease and its economic burdens by seeking to
control the behavior of immigrants within our borders rather
than excluding immigrants at our
borders. That this theme of inclusion marks 2 such different eras
underscores its enduring significance in American public policy.
INCLUSION AS
BACKDROP:
1882 TO 1924
There coexisted in Progressive Era America 2 models—
interconnected by questions of
race and labor—of citizenship,
one characterized by fitness for
civic participation and the other
by fitness for industrial participation. The political movement
to restrict immigrants from
southern and eastern Europe
“
At both the dawn and dusk of the 20th century, the
most consequential public policies responded to disease and
its economic burdens by seeking to control the behavior of
immigrants within our borders rather than excluding immigrants
at our borders. That this theme of inclusion marks
2 such different eras underscores its enduring
significance in American public policy.
prioritized questions of fitness
for self-government and emphasized the inherent genetic and
intellectual racial inferiority of
the new immigrant streams.6 In
the 1890s, Senator Henry Cabot
Lodge and his Immigration Restriction League connected the
literacy test with protection of
American character and citizenship in the 1890s.7 The literacy
test promised to restrict the
entry of “beaten men from
beaten races” with “none of the
ideas and aptitudes” necessary
for democratic self-rule.8
Critically, however, expansive
notions of racial restriction stemming from civic concerns did not
find their way into actual immigration legislation until well after
the turn of the century.9 Even
the literacy test—as it was finally
passed in 1917 over President
Woodrow Wilson’s veto—required
only that immigrants be able to
read in any language, including
Hebrew and Yiddish. With its
entry into the control of immigration in the 1880s, Congress remained legislatively focused on
the immigrant as industrial participant, aligning itself with business and against labor and the
nation’s proponents of scientific
racism. It was a model of industrial fitness, then, that would
shape US immigration policy
during the Progressive Era.10
April 2004, Vol 94, No. 4 | American Journal of Public Health
Between the ends of the Civil
War and World War I, the
United States was transformed
from a society of artisans, who
largely controlled the pace of
production, into the world’s leading industrial power. By the
1880s and 1890s, mechanization swelled the ranks of the unskilled labor force.11 In the face
of the changing nature of production and the changing composition of the work force, industry took advantage of an
opportunity to assert control and
authority. A new cadre of scientific and industrial managers undertook the tasks of redefining
ability to work in terms of segmented tasks rather than supervision of a product from start
to finish. By 1911, Fredrick
Winslow Taylor’s Principles of
Scientific Management, first published in the 1890s, profoundly
shaped the way the nation
thought about how to organize
work efficiently. Industrial leaders saw scientific management as
a process for removing “the
manager’s brain” from “under
the workman’s cap.”12 While scientific racists were concerned
with ensuring that the nation’s
inhabitants remained “well
born,” those concerned with the
labor half of the equation insisted that this was not enough:
the worker “must be trained
”
Fairchild | Peer Reviewed | Public Health Then and Now | 529
 PUBLIC HEALTH THEN AND NOW 
right as well as born right.”13 Industry, therefore, was interested
in worker discipline.14
Dramatic changes in industrial
production and management not
only allowed the unprecedented
expansion of American industry
but also generated great economic fragility. The phenomenon
of unemployment introduced a
new dimension into defining and
managing a necessarily fluid industrial workforce while at the
same time providing a compelling rationale for disciplining
those deemed destined to destitution. As workers increasingly
located in urban areas and the
labor supply swelled to accommodate the demands of a rapidly
growing industrial power, hundreds of thousands of industrial
workers became “utterly dependent upon their industrial earnings to survive.”15
But workers could not rely
on industrial earnings. Many
“minor” recessions and depressions accompanied the 6 “major”
economic downturns that the nation experienced from 1870 to
1921.16 In this kind of work
economy, sickness could mean
the difference between survival
and destitution.17 While most of
the laboring class relied primarily
on the resources of family and
friends rather than public or private charity or relief organizations during lean times,18 illness,
rather than the nature of the
economy, was viewed as the
“outstanding problem which led
to dependency.”19 In this context,
then, government in addition to
industry had an interest in controlling the worker. Thus, when
the federal government began to
exercise its congressional authority over immigration in 1882,20
it sought not to restrict immigration but rather to control it by
preventing the entry of those
530 | Public Health Then and Now | Peer Reviewed | Fairchild
who could not support themselves as well as “convicts, lunatics, and idiots.”21
Control rather than restriction
would characterize immigration
policy for the next 4 decades.
With the immigration law of
1891, the federal government
created the machinery for federal
officials to inspect and exclude
immigrants. The law required
medical officers of the PHS to
issue a medical certificate to all
immigrants suffering from a
“loathsome or a dangerous contagious disease.”22 Loathsome
and dangerous contagious
diseases—also known as class A
conditions—included trachoma
(also known as granular conjunctivitis), an infectious eye condition that could lead to blindness;
favus, a fungal infection of the
scalp and nails; venereal diseases; parasitic infections; and tuberculosis.23 A subset of class A
conditions included mental conditions such as insanity, feeblemindedness, imbecility, idiocy,
and epilepsy.
In 1903, the PHS created a
new category of class B diseases
or conditions—those rendering
the immigrant “likely to become
a public charge.”24 Class B conditions included hernia, valvular
heart disease, pregnancy, poor
physique, chronic rheumatism,
nervous afflictions, malignant diseases, deformities, senility and
debility, varicose veins, and poor
eyesight.25 But in the context of
industrial-era America, not only
class B conditions affecting ability to earn a living but also the
loathsome and dangerous contagious diseases took on economic
meaning in the hands of the
PHS, which defined contagious
immigrant diseases as “essentially
chronic.” Chronic, debilitating
disease represented the permanent inability of an immigrant to
function in society; it represented
dependency.26
At the core of the industrial
economy were the dual principles of disciplining and discarding the laboring body. It was not
simply the case that the worker
bound for dependency had to be
barred at the nation’s threshold;
rather, at the nation’s threshold,
all workers had to learn the rules
and expectations of industrial society. Immigrant laborers had to
understand that they were expected to remain fit throughout
the inevitable spells of unemployment that they would be required to weather. The message
was clear to Bridget Fitzgerald,
who came from Ireland in 1921
at age 18: “You know what you
needed then mostly? I’ll tell you.
Strong and healthy, that you
won’t become a public charge,
because then, I mean, you go
right back.”27
While federal immigration law
sent roughly 79 000 immigrants
home for diseases or defects, it
brought all 25 million arriving
immigrants—particularly those
traveling in steerage or third
class who would join the ranks of
laborers—under the scrutiny of
the PHS.28 The assembly line of
flesh and bone developed to defend the nation from diseased
immigrants served as the inaugural event in the life of the new
labor force. Immigrant medical
examination centered on the
“line,” which became shorthand
for techniques and procedures
for quickly examining thousands
of immigrants. In the context of
immigrant medical inspection, it
represented a direct and meaningful analogy to the industrial
assembly line and is central to a
story of inclusion.
Ellis Island, where roughly
70% of immigrants entered the
United States, set the standard
American Journal of Public Health | April 2004, Vol 94, No. 4
for examination on the “line.”
After an arriving ship passed the
quarantine inspection in New
York Harbor,29 Immigration Service and PHS immigrant examiners boarded and examined all
first- and second-class passengers. PHS officers transferred
steerage or third-class passengers
to Ellis Island by barge. Proceeding one after the other and lugging heavy baggage, prospective
immigrants entered the oftencongested immigration station
and proceeded slowly through a
series of gated passageways resembling cattle pens. The winding passage leading toward the
PHS officers who waited at the
end ensured that each could witness the inspection of dozens of
immigrants ahead. As they
reached the end of the line, immigrants slowly filed past one or
more PHS officers who, at a
glance, quickly surveyed them
for a variety of serious and
minor diseases and conditions, finally turning back their eyelids
with their fingers or a buttonhook to check for trachoma.30
“Were they ready to enter? Or
would they be sent back?” wondered each immigrant with faces
“taut, eyes narrowed” throughout
the process.31
Manny Steen, who immigrated
from Ireland in 1925, kept the
moment of entry at Ellis Island
fresh in his memory for nearly 7
decades, describing it as “the
worst memory I have of Ellis Island.” He remembered that “doctors were seated at a long table
with a basin full of potassium
chloride and you had to stand in
front of them, follow me, and
they’d ask you and you had to
reveal yourself. . . . Right there in
front of everyone! I mean, it
wasn’t private.”32 His memory of
the humiliating nature of the examination was shared by Enid
Photo courtesy fo the National Park Service, US Department of Interior.
 PUBLIC HEALTH THEN AND NOW 
Griffiths Jones, inspected at Ellis
Island in 1923 at age 10: “And
we went to this big, like an open
room, and there were a couple of
doctors there, and then they tell
you, ‘Strip.’ And my mother had
never, ever undressed in front of
us. In those days nobody ever
would. She was so embarrassed.
And it was all these other, all nationalities, all people there.”33
Steen and Jones described not an
examination but a public spectacle. Even the more intensive examination of the estimated 10%
to 20% that the PHS “turned off
the line” was also a public
event,34 as illustrated in photographs depicting the intensive examination of men and women at
Ellis Island sometime after the
turn of the century (photos this
page and next).
Power, wrote Foucault, “must
be spectacular, it must be seen
by all almost as its triumph.” The
April 2004, Vol 94, No. 4 | American Journal of Public Health
spectacle of inspection on “the
line” represented a “ritual recoding” to be “repeated as often as
possible.”35 The inspections represented emersion in a particular,
routinized, ordered set of exercises or motions—waiting in line,
moving in unison, stepping up to
the medical inspector, moving
forward, stepping up to the immigrant inspector, answering questions. In this fashion, they were
introduced to the repetitive, monotonous habits of industrial
order. For 14-year-old Bessie
Kriesberg, the process impressed
upon her the imperative “to obey
the rules.”36 It was one of many
reinforcing moments in the new
immigrant’s life. Ellis Island was,
in the words of Michael La Sorte,
part of “a seamless continuity”
that began overseas “and ended
somewhere in America.”37
Public health in the Progressive Era was, of course, not
Women undergoing secondary
medical examination at Ellis Island.
Fairchild | Peer Reviewed | Public Health Then and Now | 531
Photo courtesy fo the National Park Service, US Department of Interior.
 PUBLIC HEALTH THEN AND NOW 
Jewish immigrants undergoing the
secondary medical examination at
Ellis Island. As also reflected in the
previous photograph, the secondary
examination was conducted in a
group setting in which immigrants
witnessed the examination of others.
solely about inclusion or absorption of immigrants into the national workforce, as historians
such as Howard Markel, Alan
Kraut, and Nayan Shah have
powerfully demonstrated.38 In
the case of the immigrant medical examination, when groups of
immigrants failed to conform to
societal expectations about the
industrial worker, the examination worked to exclude those
groups at the nation’s borders on
the understanding that they were
not racially fit for industrial
labor. Disease was instrumental
in rationalizing these exclusions,
and the medical examination
served as a flexible tool to
achieve higher exclusion rates in
regions of the country receiving
greater shares of “undesirable”
immigrants. Consequently, immigrants faced considerable medical obstacles to entry and higher
rates of medical certification and
exclusion at the nation’s Pacific
532 | Public Health Then and Now | Peer Reviewed | Fairchild
Coast and Mexican border immigration stations.39 Nonetheless,
given the industrial context, the
terms of inclusion must provide a
backdrop to such exclusionary
endeavors.
EXCLUSION AS
BACKDROP: 1924 TO
1965
The backdrop does, however,
change in 1924 with the national
origins quota system, which was
explicitly racially exclusive. The
Immigration Act of 1924 capped
immigration at 150 000 per year
and restricted immigration to 2%
of the population of each “race”
recorded in the US census of
1890, representing a deliberate
attempt to dramatically limit immigration from southern and
eastern Europe.40 A very vocal
segment of the nation’s political
and intellectual elite viewed the
legislation as an important means
to stem a threatening tide of
physically, genetically, and intellectually inferior southern and
eastern European immigration.
The threat of “inferior races”
and disease informed some of
the most exclusionary policies
from 1924 to 1965. Emily Abel,
for example, describes how the
fears of contagion and dependency enabled public health officials to use tuberculosis as a tool
for repatriating Mexican immigrants and citizens in the West.
As Abel convincingly argues, a
growing consensus regarding
Mexicans’ lack of entitlement to
US citizenship made health officials emphasize the economic
consequences of tuberculosis as a
chronic disease.41 Although Herbert Hoover’s policy of repatriation (ostensibly voluntary, but
often viewed by immigrants as
mandatory) was abandoned
under Franklin Delano Roosevelt, Roger Daniels explains
that “there was nothing even approaching a New Deal for immigration.” Indeed, under Roosevelt, racial exclusions were
extended to Filipinos in 1934.42
US refugee policy during
World War II—or rather the absence of formal policy, epitomized in 1939 when the United
States turned back the St. Louis,
most of whose 933 passengers
were Jewish refugees—stands as
the greatest testament to the exclusionary practices of the United
States after 1924. But even if
woefully inadequate, informal
presidential directives resulted in
some quarter of a million
refugees reaching the United
States during and after the war.43
That some effort was made to
expand America’s immigration
policy during World War II underscores the fact that just as the
Progressive Era was not characterized entirely by inclusion, nei-
American Journal of Public Health | April 2004, Vol 94, No. 4
 PUBLIC HEALTH THEN AND NOW 
ther was the period from 1924
to 1965 characterized entirely
by exclusion.
Many have cited the Immigration Act of 1965 as abruptly
ending the exclusionary era, fundamentally altering the face of
immigration, and, indeed, causing a near-catastrophic rise in immigration.44 No doubt the Immigration Act contributed to the
rising tide of immigration and its
increasing proportion of Asians
and Latin Americans, but it was
actually the 1924 restriction legislation that ushered in the profound, though unintended,
changes in the sources of immigration.45 As Daniels explains,
the relatives of immigrants already in the United States and
immigrants from Latin America
and Canada were not subject to
the numerical limitation under
the quota law; immigration in
these categories increased from
about 10% of total immigration
in the period before World War I
to approximately 45% of immigration by 1930, considerably altering immigration patterns.46
Figure 1 shows that the sources
of immigration began to shift
after the turn of the century,
with the proportion of Europeans
steadily declining after 1900.47
Figure 2 further shows that while
immigration fell off dramatically
after 1924, it quickly began a
steady increase, which was disrupted by depression in 1929
and war in 1940 (we see a similar pattern following World War
I). By 1947, immigration resumed at a level we would have
expected had there been no depression or war.
It was this dip in immigration
during World War II that helped
to create severe labor shortages
and that prompted the United
States to ease restrictive immigration policy, chiefly admitting
Mexican and Chinese laborers.
The Displaced Persons Act of
1948 created a national refugee
policy, resulting in the admission
of some 400 000 persons by
1952. Subsequently, the
Refugee Act of 1953 allotted
an additional 214 000 nonquota
visas; it sought primarily to protect, in true Cold War spirit,
those seeking to escape communism, but it also extended admission to Asian and Middle
Eastern immigrants.48
America’s increasingly permissive stance on immigration was
not limited to refugees. 1952
also saw the passage of the
McCarran–Walter Act, which, in
repealing previous immigration
laws, not only expanded the
classes of aliens subject to exclusion and deportation and made
it easier to accomplish both,
but it also reduced barriers to
skilled immigration and family
reunification and ended the
policy of Asian exclusion. The
McCarran–Walter Act was, to be
sure, not intended as a liberal
measure to increase immigration
to the United States. The backdrop to this legislation was most
decidedly exclusion: Patrick
McCarran sponsored the Internal
Security Act of 1950 that prohibited the immigration of communists and fascists. The bill ultimately retained the national
origins system, but, as Robert Divine has observed, as “an act of
conservatism rather than intolerance.”49 A vast amount of
immigration fell outside of the
quota and nearly 3.5 million
immigrants—many of them
Asian—subsequently entered.
Just as the doors to the nation
did not decisively slam shut in
1924, neither did they dramatically swing open in 1965.
Rather, the period from 1924
through 1965 represented
April 2004, Vol 94, No. 4 | American Journal of Public Health
“
The inspections represented emersion in a
particular, routinized, ordered set of exercises
or motions—waiting in line, moving in
unison, stepping up to the medical
inspector, moving forward, stepping
up to the immigrant inspector,
answering questions.
decades of gradual intended and
unintended change in response
to immigration legislation that at
some moments sought even
tighter restrictions on the basis of
race but at others pursued more
tolerant policies when the economy and humanitarianism demanded them.
INCLUSION AS
BACKDROP, EXCLUSION
AS VEIL: WELFARE AND
IMMIGRATION REFORM
IN THE 1990S
It is the centrality that questions of disease and dependency
would once again take in the
1990s, and how the nation
would respond to them legislatively, that enables us to draw
an analogy between the opening
and closing decades of the 20th
century. The AIDS crisis raised
deep concerns that immigrants
with HIV would swell the Medicaid rolls, causing a collapse of
our hospital and medical systems. These were the concerns
that made Guantanamo Bay
possible and that fostered a
broader immigration policy banning the immigration of individuals with HIV.
The new restrictions on HIV
and immigration were part and
parcel of a growing concern regarding the economic burden
”
that immigrants placed on society. While in 1986 the Immigration Reform and Control Act legalized an unprecedented 2.7
million illegal immigrants living
in the United States,50 the Immigration Act of 1990 attempted to
reverse the flow of immigrants
who might not be self-sufficient,
raising the nation’s immigration
ceiling by providing an unlimited
number of visas to relatives of
US citizens but reducing the allocation of visas for unskilled immigrants and raising the total
visa quota.51
But it was not immigration
policy that most clearly expressed the new concerns regarding disease and dependency; it
was welfare policy. The most
sweeping policy measure affecting immigrants and welfare was
the Personal Responsibility Act of
1996—a policy currently undergoing reauthorization. One of the
distinguishing features of the initial legislation was its withdrawal
of many public benefits from
legal immigrants and “undeserving” citizens. With some exceptions, the law barred immigrants
from receiving Supplemental Security Income and food stamps
until they became citizens.52 Immigrants could not receive cash
assistance,53 Medicaid, Social
Services Block Grant services,
and other federal means-tested
Fairchild | Peer Reviewed | Public Health Then and Now | 533
 PUBLIC HEALTH THEN AND NOW 
programs for 5 years after arrival.54 The law barred illegal immigrants from all but a few selected in-kind, noncash services
typically involving emergency
care or vaccination.55 The income and financial resources of
an immigrant’s sponsor—typically
a family member who had to be
a citizen or lawful permanent
resident—were “deemed” available to any immigrant applying
for benefits.56
Welfare reform’s provisions regarding immigrants resonated
with the exclusionary leitmotif
running through 20th-century
immigrant policy. The 1996 Personal Responsibility Act as
passed promised some $54.1
billion in savings over 6 years,
with $23.8 billion (44%) to be
achieved primarily at the expense of immigrants, both legal
and illegal.57 Because immigrants
are more likely to engage in em-
ployment that carries no health
benefits,58 it was conceivable that
welfare reform would provide a
strong disincentive to legal and illegal immigration. Welfare reform’s clear ties to California’s
Proposition 187—which passed
with nearly 60% of the vote in
1994 and denied a variety of
public services, including public
education, to illegal immigrants—
underscored the extent to which
it was initially viewed as an exclusionary immigration measure.59 Bob Dole, for instance,
the 1996 Republican presidential
candidate, reasoned that “if kids
can’t go to school, the parents
will go home.”60
In part, an enormous Hispanic
voter backlash would cause Republicans gradually to alter their
rhetoric and pitch welfare reform
not as an immigration control
measure but rather as sound social policy.61 The rhetorical
Not Specified
Oceania
100
Africa
Central/South America
90
80
Caribbean
Canada and Newfoundland
70
Mexico
Percentage
60
50
40
Asia
Europe
30
20
10
0
0
0
1
1
1
1
1–
90
1
1
1–
91
1
1
1–
92
1
1
1–
95
–1
1
94
1
1
98
1
1–
97
–1
1
1
96
1
0
0
99
0
0
97
96
95
94
–1
1
93
0
0
0
0
93
92
91
–1
1
89
0
0
90
89
–1
1
88
1
1
98
00
–2
–1
1
99
1
Years
Source. 2000 Statistical Yearbook of the Immigration and Naturalization Service. Available at http://uscis.gov/graphics/shared/aboutus/
statistics/imm00yrbk/imm2000list.htm. Accessed February 25, 2004.
FIGURE 1—Immigration by region, expressed as a percentage of total immigration, 1891 through 2000.
534 | Public Health Then and Now | Peer Reviewed | Fairchild
about-face, however, was not
purely strategic. Welfare reform
was motivated by complex and
deeply rooted sentiments in the
United States regarding humanitarianism and its limits, order
and discipline, enforcing an
“ideal” family structure, citizenship and its entitlements and obligations, and the labor market.62
Senator Bob Bennett of Utah
stated that the issue for children
was to create conditions “so
[they] can learn and be productive citizens.” Senator Daniel
Patrick Moynihan, who ultimately proved a strong opponent
of the bill, also saw the potential
for welfare reform to send a message about the expectations of
citizens: “We expect of you what
we expect of ourselves and our
own loved ones: that you will do
your share in taking responsibility for your life and the lives of
the children you bring into the
world.”63 The themes of citizenship, discipline, and family were
not intended to resonate only for
immigrants but for all of the
working class.
The theme of promoting discipline within and regulation of
the labor market was reflected
not only in welfare policy but
also in immigration policy, although exclusion remained a key
contrapuntal element marking
the debates. The US House of
Representatives, in an amendment to the immigration reform
bill that sought to increase the
number of Border Patrol officers,
increase workplace immigration
inspections, and restrict food
stamps to immigrants,64 voted in
March 1996 to deny public education benefits to illegal immigrant children.65 Speaker of the
House Newt Gingrich argued
that “There is no question that
offering free taxpayer goods to
illegals attracts more illegals.” He
American Journal of Public Health | April 2004, Vol 94, No. 4
 PUBLIC HEALTH THEN AND NOW 
in which the nation had no reciprocal obligations.73
CONCLUSION: TERMS
OF INCLUSION
Just as it would be a mistake
to deny the exclusionary elements of public policy touching
immigrants during the Progressive Era or the countervailing
currents of immigration policy in
the restrictive decades between
1924 and 1965, it would also
be foolhardy to ignore or minimize these elements of recent
policy.74 HIV exclusion has
dropped out of public discussion,
but the events of September 11,
2001, have reinvigorated the
metaphors and language of
disease—infections, terrorist cells,
eradication—and renewed interest in exclusion. The advent of
Sudden Acute Respiratory Syndrome (SARS) no doubt will re-
inforce such interest. We find
ourselves poised on the border
between greater inclusion and
further restriction.75 In the period since the September 11th
attacks, Congress has passed
measures tightening control of
the borders and intensifying the
scrutiny and surveillance of immigrants.76 In March 2002, both
the House and Senate overwhelmingly passed legislation to
increase the number of immigration investigators and inspectors
and to establish a surveillance
system for people entering with
student visas. President Bush has
signed “modern,” “smart border”
agreements with Canada and
Mexico aimed at further limiting
the flow of illegal immigrants,
drugs, and terrorists, without
slowing the flow of goods.77
In this context, it would be easy
to draw analogies to the antiimmigration rhetoric that surfaced
2 000 000
1 800 000
1 600 000
Hart–Celler Act
Displaced Persons Act
Immigration Act
1 400 000
Total Immigration
McCarran–Walter Act
Refugee
1 200 000
1 000 000
800 000
600 000
400 000
200 000
0
1891
1894
1897
1900
1903
1906
1909
1912
1915
1918
1921
1924
1927
1930
1933
1936
1939
1942
1945
1948
1951
1954
1957
1960
1963
1966
1969
1972
1975
1978
1981
1984
1987
1990
1993
1996
1999
concluded: “It is wrong for us to
be the welfare capital of the
world.”66 Although the bill carried the strong support of Bob
Dole,67 incumbent President Bill
Clinton threatened to veto the
measure. Subsequently, the Senate, with the backing of conservative Texas Senators Phil
Gramm and Kay Bailey Hutchinson, successfully blocked the
amendment.68
Clinton signed the immigration
reform bill on September 30,
1996, without measures denying
either public education to illegal
immigrants or federal treatment
funds to legal immigrants infected with HIV/AIDS. Also
missing were provisions to deport
legal immigrants who received
more than a year’s worth of federal benefits within a 7-year period.69 While the immigration reform bill remained “one of the
most sweeping efforts by Congress in years to control illegal
immigration,”70 also absent from
the final legislation were provisions to reduce legal immigration
by 30%, as the Immigration Reform Commission had urged, and
provisions to increase substantially the number of Department
of Labor workplace inspectors to
investigate and penalize employers for hiring illegal immigrants.71
In refusing to limit immigration
or create disincentives for businesses to hire illegal immigrants,
the legislation thus made a powerful statement about the centrality of the contributions of both
legal and illegal immigrants to
the US economy.72 Coming on
the heels of the Personal Responsibility Act, which dramatically
limited the social obligations of
the nation to these immigrants, it
clearly defined the terms of inclusion: immigrants entered a social contract in which they must
make economic contributions but
Year
Source. 2000 Statistical Yearbook of the Immigration and Naturalization Service. Available at http://uscis.gov/graphics/shared/aboutus/
statistics/imm00yrbk/imm2000list.htm. Accessed February 25, 2004.
FIGURE 2—Total immigration to the United States, 1891 through 1999.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Fairchild | Peer Reviewed | Public Health Then and Now | 535
 PUBLIC HEALTH THEN AND NOW 
in the mid-1990s.78 But emerging
debates and even policies in the
aftermath of 9/11 are not likely
easily to derail the broader inclusionary impulses characterizing
American immigration policy in
the current period. Thus, while
one interpretation of the welfare
and immigration reform measures
is that it sends the message “NonYankees Go Home,”79 we have to
look through this veil at the larger
backdrop.
In March 2002, the US House
of Representatives, in support of
negotiations between the Bush
administration and Mexico’s President Vicente Fox, approved a
measure to make it easier for illegal immigrants to gain legal status
in the United States.80 After encountering opposition from Democratic Senator Robert Byrd,
the measure was excluded from
May’s reconciliation legislation,
which hardened immigration enforcement laws. Despite this setback, the proposal had carried
broad bipartisan support in both
the House and Senate. It also enjoyed wide community and business support.81 Indeed, Democratic Senate Majority Leader Tom
Daschle reintroduced the measure on May 9, 2002.82 Most recently, President Bush put forward a guest-worker proposal
that would allow illegal immigrants to obtain renewable 3-year
work permits that many critics
have derided as exploitative.83
It is important, then, to appreciate the exact terms of inclusion
as well as the extent of support
behind inclusion, which includes
not only Republicans, Democrats, and employers—who may
or may not have a stake in improving the terms of inclusion
for immigrants—but also organized labor.
The AFL-CIO, in February
2000, began to urge the legal-
ization and unionization of illegal Mexican immigrants, representing a dramatic shift in a policy position forged during the
Progressive Era.84 At the beginning of the century, unions saw
the new immigrant laborer as
living outside the craftsman’s
ethic of collective behavior; as a
contemporary labor lyric put it,
“There were no men invited
such as Slavs and ‘Tally Annes,’/
Hungarians and Chinamen with
pigtail cues and fans.”85 This
“dangerous class” of unskilled
labor was perceived as “inadequately fed, clothed, and
housed” and, accordingly, it
“threatened the health of the
community.”86 This was a period, after all, in which labor
struggled not only to organize
but to organize against incredible odds: beginning in the
1880s, corporations gained
recognition as “persons” sharing
in constitutional rights in a context in which the courts increasingly reduced the rights of citizenship to “unfettered liberty of
contract.” The courts consistently ruled that regulation of
wages and work conditions represented a fetter on this liberty.
Conversely, the courts consistently prohibited labor boycotts
and strikes.87
The change in the position of
organized labor was a reaction to
its stagnating membership levels
and a political economy that had
changed dramatically since the
dawn of the 20th century.88 Although it was decidedly weak
early in the century, during and
after World War II the power of
unions was considerably strengthened and by century’s end the
national AFL-CIO membership of
some 16 million had been relatively stable for decades.89 But
immigrants potentially promise
increased power and position for
536 | Public Health Then and Now | Peer Reviewed | Fairchild
organized labor. California’s local
unions, drawing on a large and
largely illegal immigrant labor
force, added some 132 000
members in 1999.90 Thus, the
17.7 million immigrants in the
United States91—perhaps 7 million of them illegal—have provided a powerful incentive for a
switch in organized labor’s longheld positions.92
Characterized by the North
American Free Trade Agreement, which was strongly opposed by organized labor,93 globalization also brought into
question the terms of employment not only in this nation but
also in the less-developed nations
where business might locate or
relocate production.94 In this
global context, while immigration
policy can shape whether the nation has an information economy
versus a labor or service-sector
economy, as well as the distribution of wealth within it,95 exclusion no longer represents a viable alternative for controlling
labor conditions or opportunities
within the nation, as it might
have earlier in the century, when
the United States was the leading
industrial producer and American business thrived on a large,
highly mobile, and responsive
unskilled labor force.
But while the new position of
labor offers hope for altering
the terms of inclusion—a hope
not possible early in the 20th
century—the trajectory of welfare reform must give us pause.
In the summer of 2002, the
House passed a version of a
reauthorization bill, still excluding immigrants,96 that increased
work requirements from 20 to
40 hours per week with no exemptions for women with children aged younger than 6
years.97 House Republicans
argued—and some Democrats
agreed—that the PRWORA was
a stunning success, dramatically
reducing the welfare caseloads
despite rising immigration and
unemployment.98
Absent was any suggestion
that increasingly strict requirements provided disincentives to
immigration: legislators viewed
welfare reform as a means of creating and training productive citizens. Mark Foley, a Republican
representative from Florida, argued, “My grandmother came
from Poland, she was a maid at
the Travel Lodge Motel, she
worked hard all her life. All she
wanted to be is a good citizen
and an honest, God-fearing person of this country.” He saw the
bill as “preparing our citizens”
and would-be citizens “for the future of this country and its economy.”99 It was not simply that if
a lifetime of menial, low-wage
work was good enough for
Foley’s grandmother, then certainly it was good enough for
today’s immigrants—it was good
for today’s workers regardless of
their immigration status. President Bush has begun touting welfare reform as ending “the culture of dependency that welfare”
had created for people who
should properly be thought of as
“citizens of this country, with
abilities and aspirations” and not
“charges of the state.”100 Given
that the Senate is now controlled
by Republicans, the House reauthorization seems likely to pass
sometime in 2004.101
Thus, as important as it will be
in this era to remain alert to the
nation’s policies of exclusion, particularly when they turn on questions of race or nationality, a
focus on exclusion at the borders
can obscure a critical analysis of
the terms of inclusion that we set
not only for immigrants but for
all workers. How the nation sets
American Journal of Public Health | April 2004, Vol 94, No. 4
 PUBLIC HEALTH THEN AND NOW 
those terms, and the extent to
which they reflect either suspicion and distrust of immigrants
as a potential burden or a recognition of mutual obligations between workers and society, will
differ as the economic base of
the nation changes, as the position and power of organized
labor alters, and, of course, as
the sources and levels of immigrant shift. As we make decisions
about immigration, we must view
immigration reform along with
welfare reform as being fundamentally concerned with engineering the economic structure
of American society, as being
fundamentally about the nature
of inclusion. This is a prologue
that does not deny the exclusionary impulses within American
immigration history and policy,
but that can refocus our attention
in the present.
About the Author
The author is with the Center for the
History and Ethics of Public Health, Department of Sociomedical Sciences, Mailman School of Public Health, Columbia
University, New York, NY.
Requests for reprints should be sent to
Amy Fairchild, PhD, MPH, Mailman
School of Public Health, Columbia University, 722 W 168th St, 9th Floor, New
York, NY 10032 (e-mail: [email protected]
edu).
This article was accepted May 20,
2003.
Acknowledgments
This work was originally supported by
dissertation grants from the National Endowment for the Humanities and the
National Science Foundation.
The author thanks Gerard Carrino,
David Rosner, and Ronald Bayer as well
as the anonymous reviewers for their
close readings and helpful comments.
Endnotes
1. While the national origins system
was eliminated, the use of quotas continued. Total hemispheric quotas capped
at 290 000 per year. In 1976, hemispheric caps were abandoned and each
country was allotted a quota of 20 000,
and the Refugee Act of 1980 excluded
refugees from the preference system.
Roger Daniels, “Two Cheers for Immigration,” in Debating American Immigration, 1882–Present, ed. Roger Daniels
and Otis L. Graham (New York: Rowman & Littlefield Publishers Inc, 2001),
37, 41, 78.
2. Quoted in Daniels, “Two Cheers
for Immigration,” 43.
3. Rudolph J. Vecoli, “Ethnicity: A
Neglected Dimension of American History,” in The State of American History,
ed. Herbert J. Bass (Chicago: Quadrangle Books, 1970), 73.
4. Amy L. Fairchild and Eileen A.
Tynan, “Policies of Containment: Immigration in the Era of AIDS,” American
Journal of Public Health 84 (1994):
2011–2022.
5. Howard Markel, Quarantine! East
European Jewish Immigrants and the New
York City Epidemics of 1892 (Baltimore:
Johns Hopkins University Press, 1997);
Matthew Frye Jacobson, Whiteness of a
Different Color: European Immigrants and
the Alchemy of Race (Cambridge, Mass:
Harvard University Press); Alan M.
Kraut, Silent Travelers: Germs, Genes,
and the Immigrant Menace (New York:
Basic Books, 1994); Kenneth Ludmerer,
Genetics and American Society: A Historical Appraisal (Baltimore: Johns Hopkins
University Press, 1972).
6. It was such a notion of political citizenship that in part motivated Chinese
exclusion in 1882. Congress deemed
the Chinese unfit for democratic selfrule and barred the Chinese not only
from entry into the United States but
also from naturalization. The 14th
Amendment, ratified in 1868, made citizenship and equal protection under the
law a constitutional birthright. The Naturalization Law of 1870 subsequently
affirmed the right of persons of African
descent to naturalization and their right
to vote. But it also denied citizenship to
first-generation Asian immigrants. Eric
Foner, The Story of American Freedom
(New York: W. W. Norton & Company,
1998), 112. The Immigration Act of
1917 and subsequent Supreme Court
decisions in Ozawa v the United States
(1922) and The United States v Thind
(1923) reaffirmed such exclusion on the
grounds that Asians were not “White.”
See also Ian F. Haney-Lopez, White by
Law: The Legal Construction of Race
(New York: New York University Press,
1996).
7. John Higham, Strangers in the
Land: Patterns of American Nativism,
1860–1925 (New Brunswick, NJ: Rutgers University Press, 1994), 103.
8. Francis A. Walker, “Restriction of
Immigration,” Atlantic Monthly, June
1896, p. 828.
April 2004, Vol 94, No. 4 | American Journal of Public Health
9. Higham, Strangers in the Land,
43–44, 48–49, 73, 99–100, 112,
129–130, 202, 203–204, 221.
Reports of the Immigration Commission,
Abstract of the Report on Immigrants as
Charity Seekers, Vol. 2
10. Likewise, the movement for Chinese exclusion was not rooted exclusively in questions of fitness to participate in American public life.
Congress—in rare concordance with
organized labor, which saw Chinese
workers as severely depressing White
wages—responded to a perceived threat
that the Chinese posed to industrial civilization. Gwendolyn Mink, Old Labor
and New Immigrants in American Political Development: Union, Party, and State,
1875–1920 (Ithaca, NY: Cornell University Press, 1986), 90–91, 96;
Alexander Saxton, The Indispensable
Enemy: Labor and the Anti-Chinese
Movement in California (Berkeley: University of California Press, 1971). The
1882 Exclusion Act did not bar the
entry of all Chinese immigrants; it targeted laborers only.
20. The Supreme Court had ruled in
1849 that control of immigration, as a
matter of “foreign commerce,” fell
within the authority of Congress, but
the Immigration Act of 1882 is generally regarded as the first federal effort
to assert its authority. Higham, Strangers
in the Land, 356, note 19; Benjamin
Klebaner, “State and Local Immigration
Regulation in the United States Before
1882,” International Review of Social
History 3 (1958): 269–295.
11. David Montgomery, The Fall of
the House of Labor: The Workplace, the
State, and American Labor Activism,
1865–1925 (Paris: Cambridge University Press, 1987), 24, 25.
12. William D. Haywood and Frank
Bohn, Industrial Socialism (Chicago,
n.d.), 25, quoted in Montgomery, Fall of
the House of Labor, 45.
13. Fredrick Winslow Taylor, Principles
of Scientific Management (Norcross, Ga:
Engineering and Managing Press, 1911),
viii–xi, 50.
14. Montgomery, Fall of the House of
Labor, 252.
15. Alexander Keyssar, Out of Work:
The First Century of Unemployment in
Massachusetts (Cambridge, England:
Cambridge University Press, 1986), 25.
16. During these years, the economy
faltered 13 times, meaning that each
decade brought 4 years of depression
or recession. Keyssar, Out of Work, 47.
17. Unemployment Committee of the
National Federation of Settlements, Case
Studies of Unemployment (Philadelphia:
University of Pennsylvania Press, 1931),
71.
18. Lizabeth Cohen, Making a New
Deal: Industrial Workers in Chicago,
1919–1939 (Cambridge, England:
Cambridge University Press, 1990), 57.
19. Commissioner-General’s Annual Report (Washington, DC: Government
Printing Office, 1898), 2. See also memorandum abstracting information in
“The Alien as Charity Seeker,” Children’s Bureau, US Department of
Labor, Vol. IV, No. 29, October 1927;
“Aliens and Charity,” Immigration History Research Center, University of Minnesota, St. Paul; US Senate, Reports of
the Immigration Commission, Abstracts of
21. Higham, Strangers in the Land,
43–44; John Higham, “Origins of Immigration Restriction, 1882–1897: A Social Analysis,” Mississippi Valley Historical Review 39 (June 1952): 79–80.
22. The Public Health Service (PHS)
was created in 1798 as the United
States Marine Hospital Service under
the jurisdiction of the Treasury Department, where it remained until 1939. Its
initial function was to provide medical
care to merchant marines. Although I
refer to it consistently as the PHS here,
it was renamed several times and was
not known as such until 1912. Ralph
Chester Williams, MD, The United States
Public Health Service, 1798–1950
(Washington, DC: Government Printing
Office, 1951). Officers of the Immigration Service made the final decisions regarding whether immigrants would be
deported for disease, although deportation of immigrants with class A diseases
was mandatory. Although I refer to it as
the Immigration Service, it, too, was renamed and reorganized several times
throughout its history. Darrel H. Smith,
The Bureau of Immigration (Baltimore:
Johns Hopkins Press, 1926); Darrel H.
Smith, The Bureau of Naturalization (Baltimore: Johns Hopkins Press, 1926).
23. Barbara Bates, Bargaining for Life:
A Social History of Tuberculosis,
1876–1938 (Philadelphia: University of
Pennsylvania Press, 1992), 16–18;
Sheila M. Rothman, Living in the Shadow
of Death: Tuberculosis and the Social Experience of Illness in American History
(New York: Basic Books, 1994), 13–15.
24. Bureau of Public Health and Marine-Hospital Service, Book of Instructions for the Medical Inspection of Immigrants (Washington, DC: Government
Printing Office, 1903), 5, 10–11.
25. Amy L. Fairchild, Science at the
Borders: Immigrant Medical Inspection
and the Making of the Modern Industrial
Labor Force (Baltimore: Johns Hopkins
University Press, 2003), 32.
26. Fairchild, Science at the Borders, 14.
27. Joan Morrison and Charlotte Fox
Zabusky, American Mosaic: The Immi-
Fairchild | Peer Reviewed | Public Health Then and Now | 537
 PUBLIC HEALTH THEN AND NOW 
grant Experience in the Words of Those
Who Lived It (New York: E. P. Dutton,
1980), 42. As an employee, she described herself as “a ‘useful girl.’ ”
28. On average, 4.4% of all immigrants were certified annually from
1909 to 1930, peaking at more than
8.0% in 1918 and 1919, although only
about 11% were ever deported. The
medical deportation rate for medical
causes never exceeded 1%. Fairchild,
Science at the Borders, 4–5. The immigrant medical inspection was designed
for processing third-class or steerage
passengers. Although an officer might
occasionally send a first-class passenger
for closer examination, he searched
primarily not for physical but for social
aberration: “If a passenger is seen in
the first cabin, but his appearance
stamps him as belonging in the steerage or second cabin, his examination
usually follows.” Letter from Assistant
Surgeon General H.D. Geddings to Surgeon General, November 16, 1906,
RG90, Central File, 1897 to 1923,
Box 36, File No. 219, National
Archives and Records Administration,
College Park, Md.
29. A. H. Doty, “The Use of the Clinical Thermometer as an Aid in Quarantine Inspection,” Medical Record, 1 November 1902, p. 690; A. H. Doty,
“Modification of Present Port Inspection,” American Public Health Association
Reports 21 (1906): 260.
30. Letter from Assistant Surgeon General H. D. Geddings to the Surgeon General, November 16, 1923, RG90, Central File, 1897 to 1923, Box 36, File
No. 219, National Archives and Records
Administration, College Park, Md; E. H.
Mullan, “The Medical Inspection of Immigrants at Ellis Island,” Medical Record,
27 December 1913, p. 1168.
31. Quoted in Irving Howe, World of
Our Fathers (New York: Galahad Books,
1976), 43.
32. Paul Sigrist, interview with Manny
Steen, March 22, 1991, Ellis Island Oral
History Project.
33. Janet Levine, interview with Enid
Griffiths Jones, April 18, 1993, Ellis Island Oral History Project.
34. There is no precise data to support
this estimate. Allan Kraut cites this figure, which appears sporadically in the
PHS records. No doubt, during some of
the peak immigration years before the
war, a far smaller percentage was
turned off the line; likewise, during the
war, when immigration levels were very
low and the PHS experimented with the
utility of conducting a more intensive
medical examination, the percentage
was higher.
35. Michel Foucault, Discipline and
Punish: The Birth of the Prison (New
York: Vintage Books, 1979), 34, 111.
36. Bessie Kriesberg, Hard Soil, Tough
Roots: An Immigrant Woman’s Story
(Jericho, NY: Exposition Press Inc,
1973), 138, 139.
37. Michael La Sorte, La Merica: Images of Italian Greenhorn Experience
(Philadelphia: Temple University Press,
1985), 48.
38. Markel, Quarantine!; Kraut, Silent
Travelers; Nayah Shah, Contagious Divides: Epidemics and Race in San Francisco’s Chinatown (Berkeley: University
of California Press, 2001); Emily Abel,
“From Exclusion to Expulsion: Mexicans
and Tuberculosis in Los Angeles,
1914–1940,” Bulletin of the History of
Medicine 77 (Winter 2003): 823–849.
39. Fairchild, Science at the Borders,
chap 4. See also Shah, Contagious Divides, and Abel, “From Exclusion to Expulsion.”
40. Robert A. Divine, American Immigration Policy, 1924–1952 (New York:
Da Capo Press, 1957, 1972); Mae Ngai,
“The Architecture of Race in American
Immigration Law: A Reexamination of
the Immigration Act of 1924,” Journal of
American History 86 (June 1999), available at http://www.historycooperative.
org/journals/jah/86.1//ngai.html, accessed January 16, 2004.
41. Abel, “From Exclusion to Expulsion”; Ngai, “The Architecture of Race.”
42. Daniels, “Two Cheers for Immigration,” 26–27; Benicio Catapusan, “Filipino Immigrants and Public Relief in
the United States,” Sociology and Social
Research 23 (1939): 546–554; Ronald
Takaki, Strangers From a Different Shore:
A History of Asian Americans (New
York: Penguin Books, 1989).
43. Daniels, “Two Cheers for Immigration,” 26–27, 30.
44. Peter Brimelow, Alien Nation:
Common Sense About America’s Immigration Disaster (New York: Random
House, 1995); Roy Beck, The Case
Against Immigration: The Moral, Economic, Social, and Environmental Reasons for Reducing US Immigration Back
to Traditional Levels (New York: W. W.
Norton & Company, 1996), 69–70.
45. Divine, American Immigration Policy, 167.
46. Daniels, “Two Cheers for Immigration,” 24–29; Kitty Calavita, “US Immigration Policymaking: Contradictions,
Myths, and Backlash,” in Regulation of
Immigration: International Experiences,
ed. Anita Bocker, Kees Groenendijk,
Tetty Havinga, and Paul Minderhood
(Amsterdam: Het Sphinhuis Publishers,
1998), 141.
47. See also Christopher Jencks, “Who
538 | Public Health Then and Now | Peer Reviewed | Fairchild
Should Get In? Part II,” New York Review of Books, 20 December 2001,
available at www.nybooks.com/articles/
14942, accessed April 18, 2002.
48. Daniels, “Two Cheers for Immigration,” 29–35, 38; E. P. Hutchinson, Legislative History of American Immigration
Policy 1798–1965 (Philadelphia: University of Pennsylvania Press, 1981),
264–265.
49. Divine, American Immigration Policy, 161, 164–176, 190; Hutchinson,
Legislative History, 297–312.
50. Diane Lindquist, “In Search of
Amnesty; Call to Legalize Workers
Gives Mexicans Hope,” San Diego
Union-Tribune, 29 May 2000, p. 1;
Robin Gerber, “Labor’s Welcome
Change of Course on Immigration,” Baltimore Sun, 29 February 2000, p. 23.
51. Nancy Foner, From Ellis Island to
JFK: New York’s Two Great Waves of Immigration (New Haven, Conn: Yale University Press and Russell Sage Foundation, 2000), 11, 249 (note 73);
National Research Council, The New
Americans: Economic, Demographic, and
Fiscal Effects of Immigration (Washington, DC: National Academy Press,
1997), 29. In the late 1990s, Congress
authorized the Immigration and Naturalization Service to issue more temporary visas to highly skilled workers.
52. In 1997, Congress restored Supplemental Security Income eligibility to immigrants residing in the United States
before welfare reform was passed; Balanced Budget Act of 1997, P. L.
105–33. In 1998 and 2000, food
stamp benefits were restored to some
immigrants, their children, the disabled,
and the elderly; William E. Gibson, “Immigrants to US Discover Welcome Mat
Is Out,” Seattle Times, 5 September
1999, p. 9.
53. Exceptions to cash assistance included programs such as the National
School Lunch Act, Head Start, and
emergency medical assistance. The
Temporary Assistance for Needy Families Block Grant (TANF) replaced Aid to
Families With Dependent Children
(AFDC). States may provide cash assistance under TANF, but they can provide vouchers or services in lieu of
cash.
54. States could continue to withhold
cash benefits from nonexempted immigrants even after this 5-year period; P. L.
104–193, Title I, section 402(a)(1)(B)(ii)
and Title IV, sections 401–403,
411–412.
55. States largely controlled the manner in which immigrants who arrived
before 1996 were covered by meanstested programs. Immigrants, therefore,
remained eligible for some benefits
even within 5 years of arrival in states
like California and in New York City.
56. Exceptions apply to programs such
as emergency medical assistance. The
practice of “deeming” originated in
1980 with Supplemental Security Income benefits. Deeming, as a means of
limiting access to welfare, was extended
to programs such as AFDC before being
incorporated into Personal Responsibility Act in 1996. George J. Borjas,
Heaven’s Door: Immigration Policy and
The American Economy (Princeton, NJ:
Princeton University Press, 1999), 119.
57. Immigrant Policy News 3 (August
1996): 1; Wendell Primus, “Immigration
Provisions in the New Welfare Law,”
Focus 18 (Fall/Winter 1996–1997):
14–18; R. Y. Kim, “Welfare Reform and
‘Ineligibles’: Issue of Constitutionality
and Recent Court Rulings,” Social Work
46 (2001): 315.
58. Marsha Lillie-Blanton and Julie
Hudman, “Untangling the Web:
Race/Ethnicity, Immigration, and the
Nation’s Health,” American Journal of
Public Health 91 (2001): 1736–1738.
Texas almost immediately implemented
the PRWORA cutbacks without additional state compensation.
59. Dave Lesher, “Deadlock on Prop.
187 Has Backers, Governor Fuming,”
Los Angeles Times, 8 November 1997,
p. 1; Patrick J. McDonnell, “Judge’s Final
Order Kills Key Points of Prop. 187,”
Los Angeles Times, 19 March 1998, p.
3; Patrick J. McDonnell and Ken Ellingwood, “Immigration—the Big Issue of
’94—Disappears From ’98 Debate,” Los
Angeles Times, 23 October 1998, p. 3;
Patrick J. McDonnell, “Davis Won’t Appeal Prop. 187 Ruling, Ending Court
Battles,” Los Angeles Times, 29 July
1999, p. 1; Evelyn Nieves, “California
Class Off Effort to Carry Out Immigrant
Measure,” New York Times, 30 July
1999, p. 1.
60. Greg McDonald, “Dole Scorns ‘Liberals’ Over School Issue,” Houston
Chronicle, 25 September 1996, p. 13;
Marc Lacey, “Immigration Debate
About to Resurface,” Los Angeles Times,
26 May 1996, p. 18.
61. McDonnell and Ellingwood, “Immigration”; McDonnell, “Davis Won’t Appeal”; Elaine S. Povich, “Courting Hispanics,” Newsday, 21 April 2002, p. 4.
62. Michael Katz, In the Shadow of the
Poorhouse: A Social History of Welfare
in America (New York: Basic Books,
1996), xi.
63. Congressional Record, Personal Responsibility and Work Act of 1995 Conference Report, Vol. 141, No. 206 (21
December 1995): S19098, S19089.
64. Eric Schmitt, “GOP Seems Ready
American Journal of Public Health | April 2004, Vol 94, No. 4
 PUBLIC HEALTH THEN AND NOW 
to Drop Political Fight,” New York Times,
18 September 1996, p. 6.
Emily Abel, “From Exclusion to Expulsion.”
65. Marc Lacey and Patrick J. McDonnell, “House Votes to OK Bans on Illegal Immigrant Schooling,” Los Angeles
Times, 21 March 1996, p. 1.
75. Bill McAllister, “Immigration Law
Unlikely to Soften; Mexican Workers’
Status Can’t Change for Safety Reasons,
Tancredo Asserts,” Denver Post, 26 September 2001, p. 1; Cindy Rodriguez,
“In the Week of Attacks, Proposals to
Toughen Laws Are Expected,” Boston
Globe, 19 September 2001, p. 26.
66. Ibid. See also Marc Lacey, “Immigration Debate About to Resurface,” Los
Angeles Times, 26 May 1996, p. 18.
67. Schmitt, “GOP Seems Ready.”
68. McDonald, “Dole Scorns ‘Liberals’
Over School Issue.”
69. Marc Lacey, “Toned Down Bill on
Immigration Passes in House,” Los Angeles Times, 29 September 1996, p. 12;
“Non-Yankees Go Home: That’s the
Message of the Punishing New Immigration and Welfare Reform Laws,” Newsday, 2 October 1996, p. 36.
70. The final legislation denied
refugees the right to counsel, raised the
income requirements of people seeking
to sponsor immigrants to 25% over the
poverty level, and doubled the size of
the Border Patrol; McDonald, “Dole
Scorns ‘Liberals’ Over School Issue.”
71. Lacey, “Toned Down Bill”; “NonYankees Go Home.”
72. James Flanigan, “New Bill on Immigration is Borderline at Best,” Los Angeles Times, 29 September 1996, p. 1.
73. Even illegal immigrants, though
formally excluded, have a role and
function within the US economy, from
the laundry industry in Chicago to agriculture in the West and Southwest.
Louis Uchitelle, “INS Is Looking the
Other Way as Illegal Immigrants Fill
Jobs,” New York Times, 9 March 2000,
p. 1; Mark Bixler, “Immigration Deal
Could Bolster Unions; Many Illegals
Who Want to Join Labor Organizations
Fear Deportation,” Atlanta Journal and
Constitution, 2 September 2001, p. 2;
Christopher Parks and Henry Tricks, “Illicit Angels of America’s Economic Miracle,” London Financial Times, 23 February 2000, p. 10; Michael Riley,
“Increase in Immigration Arrests Leaves
Town’s Workforce Depleted,” Denver
Post, 14 April 2002, p. 18.
74. Mae Ngai, “Legacies of Exclusion:
Illegal Chinese Immigration During the
Cold War Years,” Journal of American
Ethnic History 18 (Fall 1998): 3–35;
Mae Ngai, “The Strange Career of the
Illegal Alien: Immigration Restriction
and Deportation Policy in the United
States, 1921–1965,” Law and History
Review 21 (Spring 2003): 69–108; Natalia Molina, Contested Bodies and Cultures: The Politics of Public Health and
Race Within Mexican, Japanese, and Chinese Communities in Los Angeles,
1879–1939 [PhD Dissertation] (Ann
Arbor: University of Michigan, 2000);
76. Robert Pear, “Bill on Border Security and Immigration Passes in House,”
New York Times, 9 May 2002, p. 34;
Diana Jean Schemo, “Officials to Speed
Start of New Student Visa Tracking System,” New York Times, 11 May 2002, p.
12. The effects of September 11th may
be further exacerbated by commentaries such as Pat Buchanan’s Death of
the West: How Dying Populations and
Immigrant Invasions Imperil Our Country
and Civilization (New York: St. Martin’s
Press, 2002).
77. Patty Reinert, “Border Fix Won’t
Be Quick,” Houston Chronicle, 27 March
2002, p. 3; Elisabeth Bumiller, “White
House Announces Security Pact with
Mexico,” New York Times, 22 March
2002, p. 18; Carla Baranauckas, “Bush
Signs Border-Security Measure,” New
York Times, 15 May 2002, p. 20.
78. Peter Brimelow, “Time for a
Change? Enough! America Is Drowning
in a Sea of Immigrants,” Atlanta Journal
and Constitution, 30 April 1995, p. 1G.
79. “Non-Yankees Go Home.”
80. Robert Pear, “House Passes Immigrant Bill to Aid Mexico,” New York
Times, 13 March 2002, p. 1; Robert
Pear, “Immigrant Bill,” New York Times,
17 March 2002, p. 2; Mark Bixler,
“House Votes to Ease Immigration
Rule,” Atlanta Journal and Constitution,
13 March 2002, p. 3A.
81. Associated Press, “Border Security
Bill Sent to Bush,” New York Times, 9
May 2002; Pear, “Bill on Border Security”; Pear, “House Passes Immigrant
Bill.”
82. The bill is still in the Senate Judiciary Committee.
83. David Abraham, “American Jobs
but not the American Dream,” New
York Times, 9 January 2004, p. A19.
84. “AFL-CIO Calls for New Direction
in US Immigration Policy to Protect
Workers, Hold Employers Accountable
for Exploitative Working Conditions,”
February 16, 2000, available at
www.aflcio.org/mediacenter/prsptm/pr
02162000d.cfm, accessed January 16,
2004.
85. Michael McGovern, Labor Lyrics,
and Other Poems (Youngstown, Ohio,
1899), 27–28, quoted in Montgomery,
Fall of the House of Labor, 25.
April 2004, Vol 94, No. 4 | American Journal of Public Health
86. Edith Abbott, “The Wages of Unskilled Labor in the United States,” Journal of Political Economy 13 (1905): 324.
87. In its 1886 ruling in the Wabash
case, the Supreme Court recognized
corporations as “persons” protected
under the 14th Amendment to the Constitution. The ruling also prohibited
states from regulating interstate commerce, giving sole regulatory authority
to the federal government.
88. Steven Greenhouse, “In US Unions,
Mexico Finds an Unlikely Ally on Immigration,” New York Times, 19 July 2001,
p. A1.
89. Eric Brazil, “Unions Widen Their
World; Ex-Foes of Undocumented
Workers Now See Them as Membership Targets,” San Francisco Chronicle, 2
September 2001, p. W1.
90. Lindquist, “In Search of Amnesty.”
91. It is not clear how labor will view
foreign workers who are not either legal
or illegal immigrants. Two federal programs from the early 1990s helped to
expand the labor force without visibly
expanding the immigrant population.
The H1-B visa system was established
in 1990 to permit businesses to sponsor
highly skilled foreign workers if domestic workers were not available. H1-B
visa holders are not technically immigrants, but most are likely to remain
permanently. In a declining economy,
these workers are increasingly viewed
as “serfs” who take American jobs and
drive down wages, and measures have
been introduced to reduce drastically
the number of H1-B visas the Immigration and Naturalization Service may
issue. Alan B. Krueger, “Work Visas Are
Allowing Washington to Sidestep Immigration Reform,” New York Times, 25
May 2000, p. C2; Tom Condon, “Vulnerable Workers Become Victims of
Corporate Greed,” Hartford Courant, 13
October 2002, p. B1; Allan Masri, “No
Shortage of Trained American Engineers,” Los Angeles Times, 8 February
2003, p. 26; Diane E. Lewis, “Congress
Asked to Review IT Field; Engineers
Group Upset Over H1-B Visas, Job
Losses,” Boston Globe, 23 July 2002, p.
D2; Pamela R. Winnick, “Visa Versa,”
Pittsburgh Post-Gazette, 3 May 2002, p.
C10. Other individuals are allowed to
work in the United States if granted
“temporary protected status,” or TPS.
US Citizenship and Immigration Services, “What Is Temporary Protected
Status,” available at http://uscis.gov/
graphics/howdoi/tps.htm, accessed January 16, 2004.
92. Brazil, “Unions Widen Their
World.”
93. Lindquist, “In Search of Amnesty.”
Common Bonds,” updated April 12,
2002, available at http://www.
aflcio.org/aboutaflcio/magazine/
commonbonds.cfm, accessed January
14, 2004.
95. Borjas, Heaven’s Door, xiv–xvi, 5,
8–16, 62–104.
96. Congressional Record, House of
Representatives, May 16, 2002, p.
H2515–H2590; Robert Pear, “GOP
Dispute Delays Vote on Welfare Bill,”
New York Times, 16 May 2002, A20. A
Democratic substitute bill as well as
planned amendments that were overruled would have restored benefits to
legal immigrants along with providing
an additional $5 billion for child care,
given the increased work requirements;
Congressional Record, p. H2559. See
also Robert Pear, “House Passes a Welfare Bill With Stricter Rules on Work,”
New York Times, 17 May 2002, A1, and
Hillary Rodham Clinton, Letter to the
Editor, New York Times, 16 May 2002,
A24.
97. Daniel Altman, “Welfare Bill’s
Tougher Love May Backfire,” New York
Times, 19 May 2002, p. 4.
98. Robert Pear, “Federal Welfare
Roll Shrinks, but Drop Is Smallest
Since ’94,” New York Times, 21 May
2002, p. A12.
99. Congressional Record, May 16,
2002, p. H2515, and May 16, 2002,
p. H2540 and H2545. The bill’s detractors advanced a different vision of
society and repeatedly argued that the
bill failed to lift people out of poverty.
They underscored that reducing the
welfare rolls and combating poverty
were hardly equivalent, thereby questioning the very terms of welfare reform’s success. See, for example, the
pointed remarks of Representative
Lynn Woolsey, a Democrat from California and former welfare recipient, as
well as those of Jose Serrano, a New
York Democrat; Congressional Record,
May 16, 2002, p. H2538. See also the
editorial of Douglass McKinnon, former
recipient of welfare and former press
secretary to former senator Bob Dole,
“The Welfare Washington Doesn’t
Know,” New York Times, 21 May 2002,
p. A21.
100. Ann McFeatters, “Success Stories
Highlight New Welfare Push; Welfareto-Work Law Hailed at White House,”
Pittsburgh Post-Gazette, 15 January
2003, p. A10; Editorial, “Unworkable
Welfare,” Boston Globe, 16 January
2003, p. A10.
101. Senator Max Baucus, “What’s the
Next Phase of Welfare Reform?” Roll
Call, 8 December 2003 (Policy Briefings Section).
94. James B. Parks, “Recognizing Our
Fairchild | Peer Reviewed | Public Health Then and Now | 539
 IMAGES OF HEALTH 
Factory Injuries and Progressive Reform
| Elizabeth Fee and Theodore M. Brown
Source. Prints and Photographs Collection, History of Medicine Division, National Library of Medicine.
THIS COLORED WOOD
engraving, circa 1886, captures
an all-too-common scene in late19th-century America: a woman
injured during the course of factory work. Scenes like this—and
others far more horrific—inspired
muckraking journalists and the
national labor reform movement
of the Progressive Era.1 In 1907,
the popular author Arthur B.
Reeve wrote, “To unprecedented
prosperity . . . there is a seamy
side of which little is said. Thousands of wage earners, men,
women, and children, [are]
caught in the machinery of our
record breaking production and
turned out cripples. . . . Other
thousands [are] killed outright. . . .
540 | Images of Health | Fee and Brown
How many there [are] none can
say exactly for we [are] too busy
making our record breaking production to count the dead.”2
The famous socialist agitator
Crystal Eastman declared, “We
must pause and consider what
are the essential weapons in our
campaign. . . . The first thing we
need is information, complete
and accurate information about
the accidents that are happening.
It seems a tame thing to drop so
suddenly from talk of revolution
to talk of statistics, but I believe
in statistics just as firmly as I believe in revolutions. . . . And
what is more, I believe statistics
are good stuff to start a revolution with.”3
The broad labor reform movement, bringing together muckrakers, Socialists, middle-class Progressives, labor union leaders,
and enlightened capitalists in an
unlikely but at least temporarily
powerful alliance, would lead to
the creation of state bureaus of
labor statistics, the passage of
protective labor legislation, and
the appointment of factory commissioners empowered to inspect
actual working conditions and
on-site adherence to the new
labor laws.
Bethesda, Md. Theodore M. Brown is with
the Departments of History and of Community and Preventive Medicine at the
University of Rochester, Rochester, NY.
Requests for reprints should be sent to
Elizabeth Fee, PhD, Bldg 38, Room 1E21,
8600 Rockville Pike, Bethesda, MD 20894
(e-mail: [email protected]).
References
1. Rosner D, Markowitz G. The early
movement for occupational health,
1900–1917. In: Leavitt JW, Numbers
RL, eds. Sickness and Health in America: Readings in the History of Medicine
and Public Health. 2nd ed. Madison:
University of Wisconsin Press; 1985:
507–521.
2. Reeve AB. The death roll of industry. Charities and the Commons. 1907;
17:791.
About the Authors
Elizabeth Fee is with the History of Medicine Division, National Library of Medicine, National Institutes of Health,
3. Eastman C. The three essentials
for accident prevention. Annals.
1911;38:98–99.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
Confronting the Challenges in Reconnecting Urban Planning
and Public Health
Although public health and
urban planning emerged with
the common goal of preventing urban outbreaks of infectious disease, there is little
overlap between the fields
today. The separation of the
fields has contributed to uncoordinated efforts to address
the health of urban populations and a general failure to
recognize the links between,
for example, the built environment and health disparities
facing low-income populations
and people of color.
I review the historic connections and lack thereof between urban planning and public health, highlight some
challenges facing efforts to recouple the fields, and suggest
that insights from ecosocial
theory and environmental justice offer a preliminary framework for reconnecting the
fields around a social justice
agenda. (Am J Public Health.
2004;94:541–546)
| Jason Corburn, PhD, MCP
DESPITE THE COMMON
historical origins and interests of
urban planning and public health,
only minor overlaps between the
2 fields exist today. One result of
this “disconnect” is an uncoordinated approach to eliminating the
glaring health inequalities facing
the urban poor and people of
color.1–5 While public health is increasingly concentrating on biomedical factors that might contribute to different morbidity and
mortality rates between the well
off and least well off, the field is
just beginning to seriously investigate the role of land use decisions
and how the built environment
influences population health. At
the same time, urban planning
practice shows few signs of returning to one of its original missions of addressing the health of
the least well off.3,5 The result is
that work in the 2 fields is largely
disconnected, and both areas are
failing to meaningfully account
for the economic, social, and political factors that contribute to public health disparities.4 However,
the public health significance of
the disconnect between planning
and public health has not gone
unnoticed.
A series of recent reports have
emphasized the importance of reconnecting planning and public
health. For example, a 2001 Institute of Medicine report titled
Rebuilding the Unity of Health and
the Environment emphasized that
the “environment” should be understood as the interplay between ecological (biological),
physical (natural and built), social,
political, aesthetic, and economic
April 2004, Vol 94, No. 4 | American Journal of Public Health
environments.6 The National
Center for Environmental Health
of the Centers for Disease Control and Prevention, in its 2000
report Creating a Healthy Environment: The Impact of the Built Environment on Public Health, argued
for the reintegration of land use
planning and public health, explicitly linking transportation and
land use planning to public
health outcomes such as increased obesity, asthma, and
mental health.7 A 1999 report
published by the World Health
Organization, Healthy Cities and
the City Planning Process, emphasized the importance of developing a model of “healthy urban
planning” to ensure the health of
the world’s increasing urban and
poor populations.8 Finally,
Healthy People 2010 lists eliminating health disparities as 1 of
its 2 top priorities and acknowledges that only an interdisciplinary approach to health promotion will accomplish this goal.9
While these reports are important steps toward reuniting planning and public health, what is
missing is an articulation of the
challenges each field must confront in any reconnection effort
and a theory or framework articulating why and for whom the
fields should be reconnected.10
This article highlights some of
these challenges and offers a
framework by drawing on insights from ecosocial epidemiology and environmental justice. I
suggest that ecosocial epidemiology and environmental justice are
useful paradigms because the former provides an explicit frame-
work that attempts to explain
health disparities across populations and how social relations can
be pathogenic, biologically “embodied,” and expressed as health
inequalities,11–13 while the latter
outlines a democratic research
and public decisionmaking
agenda that is attentive to the distributive, procedural, and corrective justice concerns of people of
color.14–17
THE DISCONNECT
BETWEEN PLANNING
AND PUBLIC HEALTH
Public health, city planning,
and civil engineering in the
United States evolved together as
a consequence of late-19thcentury efforts to reduce the
harmful effects of rapid industrialization and urbanization, particularly infectious diseases.8,18,19
Reformers recognized that poor
housing conditions, inadequate
sanitation and ventilation, and
dangerous working conditions
helped cause devastating outbreaks of cholera and typhoid.18
Planning and public health were
regularly affiliated during this era
of miasma and contagion, and
engineering-based sanitary reforms, largely influenced by the
Chadwick report in Britain, were
instituted to limit hazardous exposures through such measures
as sewerage, garbage collection,
and rodent control.20–22 Planners
also used the power of the state
to separate out populations suspected of causing disease. Yet,
both miasma and contagion
failed to explain certain aspects
Corburn | Peer Reviewed | Reconnecting Urban Planning and Public Health | 541
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
of population health, such as why
epidemics occurred only sporadically, even with the seeming
ubiquitous filth present in many
urban areas, and how diseases
traveled.
By the end of the 19th century, the driving ideology in public health had shifted to germ theory, and this shift continued
through the first half of the 20th
century. According to germ theory, there are specific agents of
infectious disease, in particular
microbes, and these agents relate
in a one-to-one manner to specific diseases.20 This conceptual
shift was accompanied by shifts
in public health and planning
practice. Public health research
shifted from investigating ways to
improve urban infrastructure to
laboratory investigations of microbes and interventions focused
on specific immunization plans,
with physicians, not planners,
emerging as the new class of public health professionals.12,19
In urban planning, the Germaninspired “Haussman model” of
zoning began to take hold in the
United States during this same period.23 This model focused on
functionality and a hierarchical
ordering of land use that tended
to separate residential areas from
other land uses, particularly those
involving industry.24 At the core
of the Haussman model was the
idea of dividing up functions
within the economy (e.g., zoning),
isolating those functions deemed
unhealthy (e.g., industry), and
placing strict regulations on the
kind of contact occurring between people and land use functions.24 Zoning was aimed at “immunizing” urban populations
from the undesirable externalities
of the economy, such as industrial
pollution.
As clinicians increasingly implemented public health mea-
sures in the mid-to-latter half of
the 20th century, the field shifted
toward addressing the “hosts”
(e.g., individuals) of disease, because the “environment” (e.g., the
world outside of microorganisms)
was harder for physicians to influence.20 During this era, public
health largely ignored the social
dimensions of disease and emphasized modifying individual
“risk factors” reflected in one’s
lifestyle, such as diet, exercise,
and smoking.25 Planning, searching for an identity in postwar
America, turned to promoting
economic development through
large infrastructure and transportation projects.26 Planning
shifted from attempting to restrain harmful “spillovers” from
private market activities in urban
areas to promoting suburban economic development.27 Models of
economic efficiency were used in
planning new towns, regional
planning authorities were established to provide inexpensive and
reliable resources to these areas,
and an era of urban divestment
and residential segregation took
hold.26,27
By the latter half of the 20th
century, the biomedical model of
disease, which attributes morbidity and mortality to molecularlevel pathogens brought about by
individual lifestyles, behaviors,
hereditary biology, or genetics,25
was firmly entrenched as the
dominant paradigm in epidemiology. Yet, the biomedical model
was oriented toward explaining
molecular-level pathogenesis
rather than explaining the distribution of disease among populations or disease incidence or distribution at a societal level.20,25
Urban planning underwent an
analogous shift in its orientation
toward environmental health by
adopting the environmental impact assessment (EIA) process.
542 | Reconnecting Urban Planning and Public Health | Peer Reviewed | Corburn
The EIA process, institutionalized after passage of the National
Environmental Policy Act of
1969, ushered in the use of the
environmental impact statement
(EIS) for analyzing the ecological
and human health effects of plans,
projects, programs, and policies.28
The EIA process is generally accompanied by a quantitative risk
assessment in which human
health effects are considered.29
Risk assessment was institutionalized as part of the EIA process in
almost all site-specific analyses of
human health after the 1980 Supreme Court decision supporting
the use of risk assessment in the
regulation of benzene.30
Yet, both the EIS and quantitative risk assessment have been
widely criticized as methods for
assessing population health because they tend to overemphasize
carcinogenesis at the expense of
other chronic diseases,31,32 treat
all populations as similarly susceptible while ignoring the disproportionate hazardous exposures experienced by certain
populations,33 restrict analyses to
quantitative data while minimizing or ignoring other kinds of information,34 and limit the discourse and practice to experts,
which can undermine the democratic character of the process by
determining who is empowered
to frame analyses and who will
be excluded, deemed inarticulate,
irrelevant, or incompetent.29,34,35
Thus, wholesale adoption of practices such as EIS and risk assessment leads to planning becoming
disconnected from environmental
health.
CHALLENGES FACING THE
UNION OF PLANNING AND
PUBLIC HEALTH
By the late 20th century, the
fields of planning and public
health were largely disconnected
both from their original mission
of social betterment and from
working collaboratively to address the health of urban populations.8 There were some notable
exceptions in each field, such as
Alice Hamilton’s community
health work and Paul Davidoff’s
“advocacy planning” movement,28,36 both of which advocated for interventions designed
to improve the lot of the least
well off. However, such movements were the exception rather
than the rule in their respective
fields. As discussed in the sections to follow, at least 4 significant challenges for reconnecting
the fields emerge from this current disconnect.
Assessing the Health of
Places and “Place-Making”
The first challenge facing the
recoupling of planning and public
health is how to pay increased attention to the public health effects of land use and places—
often referred to as the built
environment—while simultaneously expanding our definition of
planning to include the political
processes that produce these outcomes. For instance, the fields
must develop new methods to
understand the effects of the
physical and social environment
on human health by challenging
the “geographic neutrality” assumptions of most environmental
laws. Geographic neutrality is assumed when environmental regulations control activities that cause
pollution (e.g., energy production,
agriculture, transportation).28 In
such instances, the regulations of
the Environmental Protection
Agency (EPA) involve an industry-by-industry focus or an EIS
assessment of a single facility;
there is little regard for whether
or not multiple industries or facil-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
ities are clustered in particular
communities.
Geographic neutrality is also
assumed when environmental
controls are placed on a specific
hazardous agent or pollutant (e.g.,
lead, asbestos, radon), the environmental medium, or, less frequently, the route of exposure
(e.g., drinking water, ambient air).
In the case of each of these scenarios, cumulative exposures
from multiple hazardous agents
that have effects on communities
are rarely considered.15,35 The
EPA has recognized the importance of geography in some regulatory programs, such as the state
implementation plans designated
under the Clean Air Act and watershed protection programs such
as those managing the Great
Lakes and Chesapeake Bay regions.28 However, the overall regulatory strategy remains firmly
rooted in the geographic neutrality fallacy.
While reconnecting planning
and public health will require increased attention to the health effects of plans in geographic
places, it will also demand that
the field recognize its role in the
politics of “place-making.”37,38
Planning must increasingly be understood as a profession that
manages conflicts over political
power and values that arise
when, for instance, state or private-sector objectives clash with
those of local communities. If
planning is to be reconnected
with public health, planning practice must be conceptualized as a
set of outcomes (e.g., housing,
transportation systems, urban designs) and processes that can
(1) involve the use or abuse of
power, (2) respond to or resist
market forces, (3) work to empower certain groups and disempower others, and (4) promote
multiparty consensual decision-
making discourses or simply rationalize decisions already
made.39
In other words, planning practice involves choices regarding
which information is deemed relevant, what decisionmaking processes will be used, and when, or
if, various publics will be involved
in making the plan.38 Reconnecting the fields will require increased attention to the politics of
planning practice (i.e., in terms of
shaping public agendas and attention), available evidence and
norms of inquiry, inclusive or exclusive deliberations, and responses (or lack thereof) to bias,
discrimination, inequality, and
recalcitrance.39
effectively frozen out of suburbs
by racial covenants, discriminatory mortgage practices, and racial steering since the 1950s,
Whites have benefited from access to low-cost suburban homes,
low interest rates on governmentsubsidized home mortgages, and
publicly funded transportation
projects linking their suburban
homes to employment, recreation, and commercial centers.48,49 Such housing and transportation policies promoted
segregation and continue to preclude African Americans from
enjoying the accumulation of
wealth associated with the improved health of populations.42
Addressing Health Disparities
Developing an Urban Health
Agenda
A second challenge in reconnecting the fields is developing a
coordinated, multidisciplinary approach toward eliminating health
disparities. A plethora of recent
evidence suggests that disparities
in health between people of
color and Whites have not narrowed over time, are getting
worse, and are increasingly
linked to the physical and social
environments that fall under the
traditional domain of planning,
such as housing, transportation,
streetscapes, and community or
social capital.40–47
For instance, Williams and
Collins42 noted that residential
segregation is a fundamental
cause of differences in health status between African Americans
and Whites because it shapes the
socioeconomic conditions faced
by Blacks not only at the individual and household levels but also
at the neighborhood and community levels; it also can contribute,
in residential environments, to social and physical risks that adversely affect health. While African Americans have been
In addition to addressing
health disparities, reconnecting
the fields will demand a clearly
articulated strategy to improve
the health of urban populations.
Currently, the lack of an urban
health agenda has allowed each
field to downplay the significance
of urban–suburban–rural health
disparities.1,2 Today’s absence of
an urban health agenda stems in
part from national and state
trends of divestment in cities; this
divestment has subsequently led
to a deemphasis on research
about, and deflected resources
away from, urban issues.26,48,49
With urban poverty rates approximately twice as high as suburban poverty rates (16.4% vs
8.0% in 199950), an urban
health agenda must address socioeconomic position and other
social determinants of health
unique to urban areas.2 Concentrated poverty is principally an
urban and racial phenomenon,
and people living in poor neighborhoods often face multiple simultaneous burdens that influence their health: poor schools,
April 2004, Vol 94, No. 4 | American Journal of Public Health
unemployment, psychosocial
stress, discrimination, environmental exposures, and limited access to health care.51
Democratizing Practice
Finally, reconnecting planning
and public health will require a
new conception of participatory
democracy to ensure that practices are accountable to communities that have historically been
excluded from decisionmaking
but face the greatest burden in
terms of inequalities.52 Research
and decisionmaking in both planning and public health are often
criticized for relying solely on
professional knowledge at the expense of democratic participation.52–56 Such critiques also
claim that professional “knowledge elites” tend to view the
“public” as largely ignorant of
technical and scientific issues, reflecting a professional loss of confidence in the public’s capacity to
make sense of complex problems
and disputes.
However, increasing evidence
in the natural sciences, public
health, and urban planning53–56
reveals that expert assessments
can miss important contextual information and need to be tempered by the experiences and
knowledge offered by lay
publics. Successfully reconnecting planning and public health
will require the use of expert
models, but it will also demand
that these same models be recognized as contingent and fallible.57 Democratizing practice in
both fields demands that professional knowledge not be compartmentalized from practical experience, that lay knowledge be
considered alongside expert
judgments, and that the incomplete models of the technically
literate not be mistaken for the
sum total of reality.30,35,58,59
Corburn | Peer Reviewed | Reconnecting Urban Planning and Public Health | 543
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
A RECONNECTING
FRAMEWORK: ECOSOCIAL
EPIDEMIOLOGY AND
ENVIRONMENTAL JUSTICE
Reconnecting public health
with planning will require the
fields to embrace their physical
and social dimensions, address
health disparities burdening
urban populations, and democratize research and decisionmaking
practices. Although the task is
daunting, insights from both
fields might assist in the effort. In
public health, social epidemiology, particularly ecosocial epidemiology, provides an interdisciplinary, multilevel perspective for
understanding the health status
of, and health disparities in, populations. In planning, environmental justice provides a framework
for ensuring that decisionmaking
processes and outcomes are
democratic and fair.
Ecosocial epidemiology makes
explicit the importance of an interdisciplinary understanding of
how both biology and different
forms of social organization influence the well-being of individuals
and populations and explicitly investigates social determinants of
population distributions of health,
disease, and well-being.13 Ecosocial epidemiology stresses a multidisciplinary population perspective that requires examination of
how biological, sociological, economic, and psychological phenomena influence distributions of
population health while incorporating a life-course perspective
that considers the role of early
and multiple pathogenic exposures that contribute to cumulative disadvantage.13 Through its
population and multilevel approach to health, ecosocial epidemiology recognizes that extraindividual socioeconomic factors
closely related to the physical and
social infrastructure of communities affect health above and beyond a combination of individual
“risk” factors.25
A key concept in ecosocial epidemiology is embodiment, or how
throughout our lives we literally
incorporate, biologically, the material and social world in which
we live.13 The implication for reconnecting planning and public
health is that better models are
needed to understand how our biology does or does not reflect the
physical, social, economic, and
psychosocial environments in
which we live, work, and
play.12,60–62 This insight suggests
that reconnecting public health
and planning will do more than
simply add “biology” to “social”
analyses; it will provide an understanding of health as a continual
and cumulative interplay between
exposure, susceptibility, and resistance, all of which occur at multiple levels (e.g., individual, neighborhood, national) and in multiple
domains (e.g., home, work, school,
community).13,63
Insights from environmental
justice help confront the decisionmaking challenges facing the
recoupling of planning and public health. A basic premise of environmental justice is that all
people and communities have a
right to live, work, and play in
places and communities that are
safe, healthy, and free of lifethreatening conditions.14–17
Claims of “environmental injustice” have highlighted that people
of color and poor populations
bear a disproportionate burden
of hazardous exposures, experience less stringent enforcement
of environmental regulations,
have access to fewer environmental benefits such as parks,
and have been routinely excluded from environmental decisionmaking.15,54,64 These dispro-
544 | Reconnecting Urban Planning and Public Health | Peer Reviewed | Corburn
portionate hazardous exposures
have also been shown to contribute to adverse health outcomes.17 Environmental justice
emphasizes corrective justice as
well, or the notion that polluters
should be punished and held responsible for cleanups and
should compensate or repair
communities damaged by historic pollution.15
Reconnecting the fields could
benefit from an environmental
justice decisionmaking framework that evaluates the democratic character of processes on the
basis of their openness, inclusiveness, and fairness.64 A democratic process, according to the environmental justice framework,
demands that those being asked
to bear an environmental or
health burden “speak for themselves” in the design, analysis,
and implementation stages of the
process.16 The environmental justice framework also recognizes
that improved democratic decisionmaking processes require
planners and others to work to
ensure that disadvantaged
groups receive the necessary
legal, financial, and technical resources to allow their meaningful
participation.14
A redistribution of material resources must accompany efforts
to enhance participatory democracy. Material redistribution is
necessary because, for instance,
community networks and social
capital—both of which are resources viewed as central to improving democracy and population health—cannot be built
without supporting economic capital.26,48 The conundrum is that
redistributing economic growth
alone will not guarantee the development of community networks and social organizations
that are viewed as integral to determining how the benefits of eco-
nomic growth and development
are distributed.
Ultimately, resource redistribution requires a role for the federal
government,27,49 since local governments are always constrained
by interjurisdictional competition—
that is, interstate and intrastate
(i.e., urban–suburban–rural)
competition—in formulating redistributive policies.50 Defining a
new role for the federal government in planning and public
health will be an essential part of
democratizing the reconnection
effort. The cruel irony is that
while federal policies often helped
create today’s urban–suburban
economic, social, and health disparities, policies at this same level
are necessary to revitalize urban
areas, address discriminatory programs, and help reconnect planning and public health.26,50
TOWARD HEALTHY AND
JUST URBAN PLANNING
The successful reconnection of
planning and public health will require the articulation of an explicit conceptual framework, and I
have suggested one such paradigm here. Efforts to achieve this
reconnection must confront a host
of challenges, from redefining
planning to addressing health disparities and formulating an urban
health agenda. This task will not
be easy. However, through an interdisciplinary approach that incorporates the multilevel, lifecourse, population health
perspective suggested by ecosocial epidemiology and the procedural, distributive, and corrective
justice principles advanced by environmental justice, a reconnection framework is possible.
About the Author
Jason Corburn is with the Urban Public
Health Program, Hunter College, City Uni-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
versity of New York, and the Mailman
School of Public Health, Columbia University, New York City.
Requests for reprints should be sent to
Jason Corburn, PhD, MCP, Center for Occupational and Environmental Health,
Hunter College, 425 E 25th St, Room
724w, New York, NY 10010 (e-mail: [email protected])
This article was accepted May 11,
2003.
References
1. Freudenberg N. Time for a national
agenda to improve the health of urban
populations. Am J Public Health. 2000;
90:837–840.
2. Geronimus AT. To mitigate, resist,
or undo: addressing structural influences
on the health of urban populations. Am J
Public Health. 2000;90:867–872.
3. Greenberg MF, Popper F, West B,
Krueckeberg D. Linking city planning
and public health in the United States.
J Plann Lit. 1994;8:235–239.
4. Institute of Medicine. Toward Environmental Justice: Research, Education and
Health Policy Needs. Washington, DC: National Academy Press; 1999.
5. Hancock T. Planning and creating
healthy and sustainable cities: the challenge for the 21st century. Available at:
http://www.who.dk/healthy-cities/
hcppub.htm#Our_Cities. Accessed
November 1, 2002.
6. Institute of Medicine. Rebuilding the
Unity of Health and the Environment: A
New Vision of Environmental Health for
the 21st Century. Washington, DC: National Academy Press; 2001.
7. Jackson RJ, Kochtitzky C. Creating a
healthy environment: the impact of the
built environment on public health.
Available at: htttp://www.sprawlwatch.
org. Accessed November 1, 2002.
8. Duhl LJ, Sanchez AK. Healthy cities
and the city planning process. Available
at: http://www.who.dk/document/
e67843.pdf. Accessed November 1,
2002.
9. Healthy People 2010. Washington,
DC: US Dept of Health and Human Services; 2000.
10. Krieger N, Zierler S. What explains
the public’s health? A call for epidemiologic theory. Epidemiology. 1996;7:107–
109.
11. Berkman LE, Kawachi I. Social Epidemiology. New York, NY: Oxford University Press Inc; 2001.
14. Bullard R, Johnson GS. Environmental justice: grassroots activism and
its impact on public policy decision
making. J Soc Issues. 2000;56:
555–578.
15. Lazarus R. Pursuing environmental
justice: the distributional effects of environmental protection Northwestern University Law Rev. 1993;87:787–857.
16. Di Chiro G. Environmental justice
from the grassroots. In: Faber D, ed.
The Struggle for Ecological Democracy.
New York, NY: Guilford Press; 1998:
104–136.
17. Sexton K, Adgate J. Looking at environmental justice from an environmental
health perspective. J Exp Anal Environ
Epidemiol. 1999;9:3–8.
18. Melosi MV. The Sanitary City:
Urban Infrastructure in America From
Colonial Times to the Present. Baltimore,
Md: Johns Hopkins University Press;
2000.
19. Porter D. Health, Civilization and the
State: A History of Public Health From
Ancient to Modern Times. New York, NY:
Routledge; 1999.
20. Tesh S. Hidden Arguments: Political
Ideology and Disease Prevention Policy.
New Brunswick, NJ: Rutgers University
Press; 1990.
21. Chadwick E. Report on the Sanitary
Conditions of the Laboring Population of
Great Britain. Edinburgh, Scotland: Edinburgh University Press; 1842.
30. Jasanoff S. Science at the Bar: Law,
Science, and Technology in America. Cambridge, Mass: Harvard University Press;
1995.
47. Kawachi I. Social capital and community effects on population and individual health. Ann N Y Acad Sci. 1999;
896:120–130.
31. British Medical Association. Health
and Environmental Impact Assessment: An
Integrated Approach. London, England:
Earthscan Publications Ltd; 1998.
48. Blackwell AG, Kwoh S, Pastor M Jr.
Searching for the Uncommon Common
Ground: New Dimensions on Race in
America. New York: WW Norton & Co;
2002.
32. Steinemann A. Rethinking human
health impact assessment. Environ Impact
Assess Rev. 2000;20:627–645.
33. Ozonoff D. Conceptions and misconceptions about human health impact
analysis. Environ Impact Assess Rev.
1994;14:499–516.
34. Corburn J. Combining communitybased research and local knowledge to
confront asthma and subsistence-fishing
hazards in Greenpoint/Williamsburg,
Brooklyn, NY. Environ Health Perspect.
2002;110(suppl 2):241–248.
35. Sexton K. Socioeconomic and racial
disparities in environmental health: is
risk assessment part of the problem or
part of the solution? Hum Ecological Risk
Assess. 2000;6:561–574.
36. Davidoff P. Advocacy and pluralism
in planning. J Am Inst Planners. 1965;31:
331–338.
37. Benveniste G. Mastering the Politics
of Planning. San Francisco, Calif: JosseyBass; 1989.
38. Hoch C. What Planners Do: Power,
Politics and Persuasion. Chicago, Ill:
American Planning Association; 1994.
22. Lindheim R, Syme L. Environments, people and health. Annu Rev Public Health. 1983;4:335–359.
39. Forester J. The Deliberative Practitioner. Cambridge, Mass: MIT Press;
1999.
23. Hall P. Cities of Tomorrow. Oxford,
England: Blackwell; 1996.
40. Byrd WM, Clayton LA. An American Health Dilemma: Volume 2. Race,
Medicine, and Health Care in the United
States: 1900–2000. New York, NY:
Routledge; 2002.
24. Popper FJ. The Politics of Land Use
Reform. Madison, Wis: University of Wisconsin Press; 1981.
25. Susser S, Susser E. Choosing a future for epidemiology: I. Eras and paradigms. Am J Public Health. 1996;86:
668–673.
26. Weir M. Planning, environmentalism and urban poverty. In: Fishman R,
ed. The American Planning Tradition:
Culture and Policy. Washington, DC:
Woodrow Wilson Center Press; 2000:
193–215.
27. Fishman R, ed. The American Planning Tradition: Culture and Policy. Washington, DC: Woodrow Wilson Center
Press; 2000.
12. Krieger N. Epidemiology and the
social sciences. Epidemiol Rev. 2000;22:
155–163.
28. Vig N, Kraft M. Environmental policy from the 70’s to the 90’s. In: Environmental Policy. 4th ed. Washington,
DC: Congressional Quarterly Press;
2001:1–31.
13. Krieger N. Theories of social epidemiology for the 21st century: an ecosocial perspective. Int J Epidemiol. 2001;
30:668–677.
29. Environmental Justice Guidance
Under the National Environmental Policy
Act. Washington, DC: Council on Environmental Quality; 1997.
April 2004, Vol 94, No. 4 | American Journal of Public Health
41. Smedley BD, Stith AY, Nelson AR.
Unequal Treatment: Confronting Racial
and Ethnic Disparities in Health Care.
Washington, DC: National Academy
Press; 2002.
42. Williams DR, Collins C. Racial residential segregation: a fundamental cause
of racial disparities in health. Public
Health Rep. 2001;116:404–416.
43. Bashir SA. Home is where the
harm is: inadequate housing as a public
health crisis. Am J Public Health. 2002;
92:733–738.
44. Cummins SK, Jackson RJ. The built
environment and children’s health. Pediatr Clin North Am. 2001;48:1241–
1252.
45. Frumkin H. Urban sprawl and public health. Public Health Rep. 2002;117:
201–217.
46. Fullilove MT. Promoting social cohesion to improve health. J Am Med
Womens Assoc. 1998;53:72–76.
49. Pastor M Jr, Dreier P, Grigsby E,
Garza J, Lopez-Garza M. Regions That
Work: How Cities and Suburbs Can Grow
Together. Minneapolis, Minn: University
of Minnesota Press; 2000.
50. Jargowsky PA. Poverty and Place:
Ghettos, Barrios, and the American City.
New York, NY: Russell Sage Foundation;
1997.
51. Speer MA, Lancaster B. Disease
prevention and health promotion in
urban areas: CDC’s perspective. Health
Educ Behav. 1998;25:226–233.
52. Tesh SN. Uncertain Hazards: Environmental Activists and Scientific Proof.
Ithaca, NY: Cornell University Press;
2000.
53. Fischer F. Citizens, Experts, and the
Environment: The Politics of Local Knowledge. Durham, NC: Duke University
Press; 2000.
54. Corburn J. Environmental justice,
local knowledge, and risk: the discourse
of a community-based cumulative exposure assessment. Environ Manage. 2002;
29:451–466.
55. Heiman M. Science by the people:
grassroots environmental monitoring
and the debate over scientific expertise. J Plann Educ Res. 1997;16:
291–299.
56. Epstein S. Impure Science: AIDS, Activism and the Politics of Knowledge.
Berkeley, Calif: University of California
Press; 1996.
57. Krieger N. Questioning epidemiology: objectivity, advocacy, and socially
responsible science. Am J Public Health.
1999;89:1151–1153.
58. Habermas J. Technology and science as ‘ideology.’ In: Toward a Rational
Society: Student Protest, Science and Politics. Boston, Mass: Beacon Press; 1970:
81–127.
59. Israel BA, Schulz AJ, Parker EA,
Becker AB. Review of community-based
research: assessing partnership approaches to improve public health.
Annu Rev Public Health. 1998;19:
173–202.
60. Diez Roux AV. Investigating neighborhood and area effects on health. Am J
Public Health. 2001;91:1783–1789.
61. Acevedo-Garcia D, Lochner KA,
Osypuk TL, Subramanian SV. Future directions in residential segregation and
health research: a multilevel approach.
Am J Public Health. 2003;93:215–221.
Corburn | Peer Reviewed | Reconnecting Urban Planning and Public Health | 545
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
62. Wallace R, Wallace D. Community
marginalisation and the diffusion of disease and disorder in the United States.
BMJ. 1997;314:1341–1345.
63. Krieger N. Does racism harm
health? Did child abuse exist before
1962? On explicit questions, critical science, and current controversies: an
ecosocial perspective. Am J Public Health.
2003;93:194–199.
64. Cole L, Foster S. From the Ground
Up: Environmental Racism and the Rise of
the Environmental Justice Movement. New
York, NY: New York University Press;
2001.
Ranking of Cities According to Public Health Criteria:
Pitfalls and Opportunities
Popular magazines often
rank cities in terms of various
aspects of quality of life. Such
ranking studies can motivate
people to visit or relocate to a
particular city or increase the
frequency with which they engage in healthy behaviors.
With careful consideration of
study design and data limitations, these efforts also can assist policymakers in identifying
local public health issues. We
discuss considerations in interpreting ranking studies that
use environmental measures
of a city population’s public
health related to physical activity, nutrition, and obesity.
Ranking studies such as
those commonly publicized are
constrained by statistical methodology issues and a lack of
a scientific basis in regard to
design. (Am J Public Health.
2004;94:546–549)
| Sandra A. Ham, MS, Sarah Levin, PhD, Amy I. Zlot, MPH, Richard R. Andrews, MD, and Rebecca Miles, PhD
FOR CENTURIES, PLACES TO
live have been ranked on the
basis of factors that contribute
to quality of life, such as friendliness, wealth, crime, and
health; in a 17th-century ranking, for example, areas with
more plentiful game, heavier
livestock, and lower mortality
from Indian attacks were promoted as more “livable.”1 Further, recent examples are numerous, such as the Places Rated
Almanac, a book that rates and
ranks 354 metropolitan areas in
terms of cost of living, job outlooks, transportation, education,
health care, crime, the arts,
recreation, and climate.1 Popular magazines often publish
rankings as well. For instance,
Natural Health magazine ranked
“America’s Healthiest Cities” in
2001 (in terms of 37 criteria in
the areas of amenities, physical
health, environment, and happiness),2 and Men’s Fitness magazine has ranked “America’s Fattest Cities” annually since 1999
(in terms of 16 categories including number of fitness centers and fast-food restaurants,
measures of the natural environment and climate, and number of parks and recreational
areas).3 “Best places” are also
proclaimed on the Internet, examples being Money Magazine’s
546 | Reconnecting Urban Planning and Public Health | Peer Reviewed | Ham et al.
“Best Places to Live” (factors
considered are climate, crime,
housing, education, economy,
health, arts and leisure, and
transportation)4 and Fast Forward ’s “Sperling’s Best Places”
(criteria are housing, cost of living, crime, education, economy,
health, and climate).5
Ranking studies can garner
considerable press coverage, can
influence local public health and
environmental policies, and motivate populations to work toward healthier lifestyles. In Philadelphia, after the release of
“America’s Fattest Cities 2000,”
the mayor implemented a new
public health program in which
he challenged the city’s population to lose 76 tons of weight in
76 days.6 In such ways, rankings
of cities can effectively raise
awareness of the factors influencing quality of life. In addition, local governments may use
the findings to attract new residents, businesses, or tourists. For
example, the Web site of the
Visitors Association of Portland,
Ore, touts the city as a great
place to visit and live,7 in part as
a result of the high ranking it
achieved in the “America’s Fattest Cities 2001” article.
Nevertheless, controversies
exist about whether ratings accurately reflect the “livability”
of cities and the extent to
which such reports can be misleading. A city’s ranking varies
depending on the quality of life
criteria used in a particular
study. Furthermore, these criteria typically include public
health prevalence data and environmental measures with
multiple sources of variability
that are ignored when ranking
studies are done. To date, there
has not, to our knowledge,
been a systematic analysis of
ranking studies attempting to
determine the extent to which
their findings are methodologically sound. Editors of studies
published in popular magazines
and on the Internet are not
bound by criteria imposed by
peer-reviewed journals such as
requirements regarding complete source citations and discussion of study limitations.
We provide an analysis designed to help policymakers interpret ranking studies that appear in the popular press. We
discuss considerations in developing ranking studies that use
environmental measures of a
city population’s public health
related to physical activity, nutrition, and obesity in the hopes of
stimulating greater interaction
between policymakers and those
who publish such studies.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
62. Wallace R, Wallace D. Community
marginalisation and the diffusion of disease and disorder in the United States.
BMJ. 1997;314:1341–1345.
63. Krieger N. Does racism harm
health? Did child abuse exist before
1962? On explicit questions, critical science, and current controversies: an
ecosocial perspective. Am J Public Health.
2003;93:194–199.
64. Cole L, Foster S. From the Ground
Up: Environmental Racism and the Rise of
the Environmental Justice Movement. New
York, NY: New York University Press;
2001.
Ranking of Cities According to Public Health Criteria:
Pitfalls and Opportunities
Popular magazines often
rank cities in terms of various
aspects of quality of life. Such
ranking studies can motivate
people to visit or relocate to a
particular city or increase the
frequency with which they engage in healthy behaviors.
With careful consideration of
study design and data limitations, these efforts also can assist policymakers in identifying
local public health issues. We
discuss considerations in interpreting ranking studies that
use environmental measures
of a city population’s public
health related to physical activity, nutrition, and obesity.
Ranking studies such as
those commonly publicized are
constrained by statistical methodology issues and a lack of
a scientific basis in regard to
design. (Am J Public Health.
2004;94:546–549)
| Sandra A. Ham, MS, Sarah Levin, PhD, Amy I. Zlot, MPH, Richard R. Andrews, MD, and Rebecca Miles, PhD
FOR CENTURIES, PLACES TO
live have been ranked on the
basis of factors that contribute
to quality of life, such as friendliness, wealth, crime, and
health; in a 17th-century ranking, for example, areas with
more plentiful game, heavier
livestock, and lower mortality
from Indian attacks were promoted as more “livable.”1 Further, recent examples are numerous, such as the Places Rated
Almanac, a book that rates and
ranks 354 metropolitan areas in
terms of cost of living, job outlooks, transportation, education,
health care, crime, the arts,
recreation, and climate.1 Popular magazines often publish
rankings as well. For instance,
Natural Health magazine ranked
“America’s Healthiest Cities” in
2001 (in terms of 37 criteria in
the areas of amenities, physical
health, environment, and happiness),2 and Men’s Fitness magazine has ranked “America’s Fattest Cities” annually since 1999
(in terms of 16 categories including number of fitness centers and fast-food restaurants,
measures of the natural environment and climate, and number of parks and recreational
areas).3 “Best places” are also
proclaimed on the Internet, examples being Money Magazine’s
546 | Reconnecting Urban Planning and Public Health | Peer Reviewed | Ham et al.
“Best Places to Live” (factors
considered are climate, crime,
housing, education, economy,
health, arts and leisure, and
transportation)4 and Fast Forward ’s “Sperling’s Best Places”
(criteria are housing, cost of living, crime, education, economy,
health, and climate).5
Ranking studies can garner
considerable press coverage, can
influence local public health and
environmental policies, and motivate populations to work toward healthier lifestyles. In Philadelphia, after the release of
“America’s Fattest Cities 2000,”
the mayor implemented a new
public health program in which
he challenged the city’s population to lose 76 tons of weight in
76 days.6 In such ways, rankings
of cities can effectively raise
awareness of the factors influencing quality of life. In addition, local governments may use
the findings to attract new residents, businesses, or tourists. For
example, the Web site of the
Visitors Association of Portland,
Ore, touts the city as a great
place to visit and live,7 in part as
a result of the high ranking it
achieved in the “America’s Fattest Cities 2001” article.
Nevertheless, controversies
exist about whether ratings accurately reflect the “livability”
of cities and the extent to
which such reports can be misleading. A city’s ranking varies
depending on the quality of life
criteria used in a particular
study. Furthermore, these criteria typically include public
health prevalence data and environmental measures with
multiple sources of variability
that are ignored when ranking
studies are done. To date, there
has not, to our knowledge,
been a systematic analysis of
ranking studies attempting to
determine the extent to which
their findings are methodologically sound. Editors of studies
published in popular magazines
and on the Internet are not
bound by criteria imposed by
peer-reviewed journals such as
requirements regarding complete source citations and discussion of study limitations.
We provide an analysis designed to help policymakers interpret ranking studies that appear in the popular press. We
discuss considerations in developing ranking studies that use
environmental measures of a
city population’s public health
related to physical activity, nutrition, and obesity in the hopes of
stimulating greater interaction
between policymakers and those
who publish such studies.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
THEORETICAL
FRAMEWORK
Ranking studies can compare
cities according to disease outcomes, behavior prevalence, correlates of health measures, or a
combination of these indicators.
Studies limited to behavior prevalence are the simplest, because
national health surveys report
prevalence rates at local levels.8
More commonly, studies compare
cities primarily on the basis of environmental correlates of health
measures. Ideally, the scientific
community would publish a list of
environmental and behavioral
measures derived from multilevel
ecological modeling studies, and
these measures would be
weighted in regard to their relative importance in determining
disease outcomes. Such measures
would involve the use of timely,
readily available sources of comparable data for relevant geographic units, and study designers
would construct indices with appropriate weights based on scientific theory and empirical evidence. However, this ideal
scenario does not yet exist.
Theoretical frameworks for
environment–disease relationships are still in their infancy
owing to the shift in public health
paradigms in the mid-1990s to
encompass multilevel causes of
disease.9 For example, current
frameworks for obesity-related research distinguish between physical and social environments, behaviors, and disease outcomes.10
No clear evidence exists as yet to
quantify relationships between
environment, physical activity/
nutrition, obesity, and disease,10
and ecological-level studies are
lacking that include a wide range
of readily available indicators
such as those used in city rankings. However, a recent study11
revealed that several economic
nutrition indicators (e.g., number
of full-service and fast-food
restaurants12 and average cost of
a meal prepared at home13) exhibited significant associations with
obesity prevalence rates. “Walkability” (e.g., presence of sidewalks,
enjoyable scenery)14 and number
of locations available for exercise
(e.g., walking trails, parks, indoor
gyms)15 also have been correlated
with physical activity. Future scientific research will provide additional empirical evidence on
which city rankings can be based.
Meanwhile, ranking studies are
popular and will continue to be
published in part as a result of
the plausibility of relationships
between environmental factors,
behavior, and health outcomes.
Editors are left to do their best
with limited resources, and many
appropriately choose a combination of available statistics on
health behaviors and environmental factors. Cities are complex
systems with multiple causal
pathways between environment,
population dynamics, behavior,
and health conditions. Ranking
studies may oversimplify these
complex systems. Furthermore,
combining environmental, behavioral, and disease outcome measures without clarifying the differences between them may confuse
and mislead readers. Public
health policymakers can benefit
from ranking studies while recognizing that their findings may
need to be reinterpreted once
more sound hierarchical linear
models are developed.
paucity of comparable, timely
data collected at the city level
through the use of stable and reliable procedures and representative samples of the population of
interest.
Geographical Level of
Analysis
In comparisons of cities, information is required at the city level
or the level of the metropolitan
statistical area (MSA), which comprises the central city and the suburbs and surrounding counties
economically tied to the central
city. Most cities and MSAs, however, do not routinely collect all of
the data required to create the desired indices for ranking studies.
Until recently, state-level averages
had been reported for most of the
available health data collected nationally, because these data were
derived from probability samples
and required a minimum sample
size to achieve acceptable statistical confidence.8 Yet, state-level averages may not adequately represent the health situation in any of
the state’s cities, and this is particularly the case in states covering
large geographic areas.
Another complication is that
some environmental indicators
are measured for central cities but
not the entire MSA. Faced with
this dilemma, editors may resort
to combining city-, MSA-, and
state-level data or may impute
data from the mean without documenting their decisions in published reports. Policymakers are
left to investigate the data sources
themselves.
Timeliness
DATA SOURCES
Other than lack of scientific
basis, limitations posed by available data are often the greatest
weakness associated with a ranking study. The primary issue is the
April 2004, Vol 94, No. 4 | American Journal of Public Health
Similarly, ranking studies based
on data from different years or
outdated sources should be interpreted with caution. Older
sources can be misleading when
the phenomenon under consideration is changing at a rapid pace.
For example, published summary
MSA-level health statistics on obesity trends occurring in the 1990s
were based on data gathered during the early part of the decade,
after which obesity prevalence
rates increased rapidly. Studies involving the use of older data for
indicators that tend to be more
stable over time, such as acres of
parks per 10 000 people, do not
face the same problem.
Consistency With Current
Standards
Measures used to rank cities
may not be calculated and defined according to currently accepted standards and should be
interpreted accordingly. For example, public health definitions of
“overweight,” “obesity,”16 and
“physical activity”17 have been revised in recent years, but statistics
including both current and past
definitions are widely used. Furthermore, beginning in 2001,
household, transportation, and
leisure-time physical activity were
measured in national health surveys to concur with current public
health recommendations; previous measures reported only
leisure-time exercise, one type of
all possible physical activity.8
Stability, Reliability, and
Selection Bias
Data gathered from some of
the sources used to rank cities, including federal sources (e.g., the
US Bureau of the Census [http://
www.census.gov], the US Environmental Protection Agency [http://
www.epa.gov], and the National
Oceanic and Atmospheric Association [http://www.noaa.gov]), are
collected according to scientific
methods and are well documented. The Centers for Disease
Control and Prevention’s Behavioral Risk Factor Surveillance System8 also is considered a reliable
data source that is particularly
Ham et al. | Peer Reviewed | Reconnecting Urban Planning and Public Health | 547
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
well documented and based on a
scientifically selected sample.
Other sources are more problematic and should be interpreted
with caution. For example, inhouse surveys are often derived
from convenience samples that
may systematically exclude certain groups by focusing on a nonrepresentative sample within a
city or MSA. Likewise, only about
half of all businesses are listed in
business address databases,
whereas the Census of Retail
Trade12 uses a representative
sample.
cal scores by, for example, assigning letter grades (A+, A, A–, etc.),
a method that has intuitive appeal
and commands attention in a society accustomed to being graded
or evaluated. Indices and letter
grades reduce the effects of imperfect data by giving individual
measures low weight in the overall score and by creating categorical measures. Despite its appeal,
however, the technique of creating indices via simple averaging
assumes that all data sources are
of equal quality and appropriateness, which may not be true.
CALCULATION METHODS
MESSAGE
COMMUNICATION
In ranking studies, data are
often transformed to normal distributions, although statistical
ranking methods were developed
to analyze data that are not normally distributed. Rankings based
on normal distributions identify
the best and worst cities but misrepresent the relative positions of
the many cities in between, the
reason being that most of the
scores cluster near the mean. The
highest and lowest scores are easy
to identify, but the remaining values could be statistically indistinguishable. Meaningful measures
of statistical uncertainty regarding
city scores are difficult to derive
in city ranking studies, because
such scores are sums or averages
of point estimates for differing
constructs. Therefore, the numerical scores on which these rankings are based will be more useful
to policymakers.
Rankings may be based on indices in which each construct represents a combination of related
measures (e.g., a climate index
calculated via data on temperature, precipitation, snowfall, and
sunshine) that are given numerical scores. Policymakers can
group cities with similar numeri-
Ranking studies published in
magazines and books and on
Web sites often attract media and
government attention and are
taken seriously, regardless of their
designs and limitations. When
cities refer to their rankings in
public health programs,6 on the
Internet, and in local chamber of
commerce publications,7 it would
be useful to include a brief discussion of study limitations along
with rankings and scores. As
demonstrated here, all ranking
studies involve limitations that affect data interpretation. Topics to
be touched on include, but are
not limited to, lack of a scientific
basis for linking study indicators
and lack of effective measures focusing on important aspects of the
environment, physical activity,
and nutrition; the latter issue is a
consequence of the difficulty of
measuring these factors, poor
data quality (e.g., inconsistent geographic level of analysis, outdated
statistics, poor coverage), or both.
Another limitation that should
be noted is the use of multiple geographic units of analysis for statistical data; if possible, such data
should be reanalyzed in hierarchi-
548 | Reconnecting Urban Planning and Public Health | Peer Reviewed | Ham et al.
cal linear models as they become
available. Ranking studies can be
valuable tools for the public
health field and for local governments if methodological limitations are assessed and taken into
consideration.
CONCLUSIONS
Rankings of cities can play an
important role in raising awareness of public health issues and illuminating what policymakers
can do to address these issues. In
addition, they provide policymakers and the public more information about the health challenges
they face and allow progress to
be monitored over time on a
range of indicators. If their city’s
comparisons with higher ranked
cities suggest an environmental
issue that might be health related
(e.g., poor air quality, higher
number of fast food restaurants,
lower number of recreational
areas), public health policymakers
can use published ranking studies
to justify a local investigation into
the problem. Such data also can
be used to raise awareness about
lifestyle choices among residents;
to market “healthy cities” and “active cities” as attractive places to
visit, live, and do business; and to
hold government and the private
sector accountable for doing
what is necessary to keep residents healthy.
About the Authors
At the time of the study, Sandra A. Ham,
Sarah Levin, and Amy I. Zlot were with the
Centers for Disease Control and Prevention,
Atlanta, Ga. Richard R. Andrews was with
the Centers for Disease Control and Prevention and with Johns Hopkins University
Preventive Medicine. Rebecca Miles is with
the Department of Urban and Regional
Planning, Florida State University, Tallahassee.
Requests for reprints should be sent to
Sandra A. Ham, MS, Centers for Disease
Control and Prevention, Physical Activity
and Health Branch, 4770 Buford Hwy
NE, Mail Stop K-46, Atlanta, GA 30341
(e-mail: [email protected]).
This article was accepted April 15,
2003.
Contributors
S. A. Ham planned and executed the
study, reviewed data sources, contributed to the conceptualization and
writing of the article, and revised the
article. S. Levin and A. I. Zlot reviewed
data sources and contributed to the
conceptualization and writing of the article. R. R. Andrews reviewed data
sources. R. Miles contributed to the conceptualization of the argument and to
the writing of the article.
Acknowledgment
Richard R. Andrews’ residency was
funded through a grant from the Centers
for Disease Control and Prevention.
References
1. Savageau D, D’Agostino R. Places
Rated Almanac: Millennium Edition. Foster City, Calif: Macmillan General Reference USA; 2000:7–8, 27–28.
2. Gallia K, Horn C. Second annual
America’s healthiest cities. Nat Health.
2001;123:72–81.
3. Griffiths K. America’s fattest cities
2003. Men’s Fitness. February 2003:
70–79, 148–150.
4. The best places to live. Money Magazine [serial online]. Available at: http://
money.cnn.com/best/bplive. Accessed
January 13, 2003.
5. Fast Forward Inc. Sperling’s best
places. Available at: http://www.
bestplaces.net. Accessed January 13,
2003.
6. 76 tons of fun: citywide health and
fitness challenge. Available at: http://
www.cpgi.net/76tons/Index.html. Accessed January 13, 2003.
7. Portland praises. Available at:
http://www.travelportland.com/visitors/
news_praises.html. Accessed January 13,
2003.
8. Behavioral Risk Factor Surveillance
System. Available at: http://www.cdc.
gov/brfss. Accessed January 13, 2003.
9. Susser M, Susser E. Choosing a future for epidemiology: II. From black box
to Chinese boxes to eco-epidemiology.
Am J Public Health. 1996;86:674–677.
10. Trost SG, Owen N, Bauman AE,
Sallis JF, Brown W. Correlates of adults’
participation in physical activity: review
and update. Med Sci Sports Exerc. 2002;
34:1996–2001.
11. Chou S-Y, Grossman M, Saffer H.
An economic analysis of adult obesity:
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
results from the Behavioral Risk Factor
Surveillance System. Paper presented at:
Third International Health Economics
Association Conference, July 2001, York,
England.
12. 1997 Census of Retail Trade. Washington, DC: US Bureau of the Census;
2000.
13. ACCRA Cost of Living Index. Arlington, Va: American Chamber of Commerce Researchers Association; 2001.
14. Brownson RC, Baker EA, Housemann RA, Brennan LK, Bacak SJ. Environmental and policy determinants of
physical activity in the United States. Am
J Public Health. 2001;91:1995–2003.
15. Parks SE, Housemann RA, Brownson RC. Differential correlates of physical
activity in urban and rural adults of various socioeconomic backgrounds in the
United States. J Epidemiol Community
Health. 2003;57:29–35.
16. Clinical guidelines on the identification, evaluation, and treatment of
overweight and obesity in adults: the
evidence report. Obes Res. 1998;
6(suppl 2):S54.
17. Centers for Disease Control and Prevention. Physical activity trends—United
States, 1990–1998. MMWR Morb Mortal
Wkly Rep. 2001;50:166–169.
Cost Analysis of the Built Environment:
The Case of Bike and Pedestrian Trials in Lincoln, Neb
We estimated the annual
cost of bike and pedestrian
trails in Lincoln, Neb, using
construction and maintenance
costs provided by the Department of Parks and Recreation
of Nebraska. We obtained the
number of users of 5 trails
from a 1998 census report.
The annual construction cost
of each trail was calculated by
using 3%, 5%, and 10% discount rates for a period of useful life of 10, 30, and 50 years.
The average cost per mile and
per user was calculated.
Trail length averaged 3.6
miles (range = 1.6–4.6 miles).
Annual cost in 2002 dollars
ranged from $25 762 to
$248 479 (mean = $124 927;
median = $171 064). The cost
per mile ranged from $5735
to $54 017 (mean = $35 355;
median = $37 994). The annual cost per user was $235
(range = $83–$592), whereas
per capita annual medical cost
of inactivity was $622.
Construction of trails fits a
wide range of budgets and may
be a viable health amenity for
most communities. To increase
trail cost-effectiveness, efforts
to decrease cost and increase
the number of users should
be considered. (Am J Public
Health. 2004;94:549–553)
| Guijing Wang, PhD, Caroline A. Macera, PhD, Barbara Scudder-Soucie, MEd, Tom Schmid, PhD, Michael Pratt,
MD, MPH, David Buchner, MD, MPH, and Gregory Heath, DSc, MPH
ENVIRONMENTAL FACTORS
affect the health of all people in
both developed and developing
countries. Because of industrialization and the consequent environmental pollution, environmental changes in the past several
decades have led to new challenges for public health.
Many studies have documented
links between the environment
and human health.1–7 For example, household amenities and
other environmental exposures
have been linked to children’s
health problems such as cancer
and asthma,1–3 and environmental pollution has been linked to
high morbidity and mortality in
the general population.4–7 In recent years, the worldwide increase of obesity has prompted
discussions of environmental interventions such as increasing the
availability of healthy snacks and
building environments that are
amenable to physical activity as
possible effective means to prevent and control obesity and
other costly chronic diseases.8–11
Because of suburbanization,
the transportation systems in the
United States have been designed
for automobile use. Although
April 2004, Vol 94, No. 4 | American Journal of Public Health
automobile-oriented transportation is a necessity for economic
development and people’s daily
lives, the modern transportation
system may pose a hazardous environment for public health. Recently, 42% of American adults
expressed a great deal of concern
about urban sprawl and loss of
open space,12 which can create
an environment of physical inactivity, a major risk factor for
several chronic diseases and obesity.13–25 One study has demonstrated the association between
the built environment and physical activity by showing the effects
of urban environment on walking
behavior.22 Another study
showed that environmental features such as neighborhood design appeared to affect whether
residents walked to work.24
Pedestrian-oriented urban environments may promote physical
activity,22,23 and a combination of
urban design, land use patterns,
and transportation systems that
promote walking and bicycling
may help create more livable
communities.26–28 Lieberman recently suggested that proper design and land use patterns and
policies can increase public transit
use as well as walking and bicycling.26 Efforts to increase the
pedestrian-oriented environment
through mixed-use development,
street connectivity, and good community design can enhance both
the feasibility and attractiveness
of walking and bicycling.
Participation in regular physical activity depends in part on
the availability and proximity to
such resources as community
recreation facilities and walking
and bicycling trails, so building
such environments holds much
promise in health promotion.29–31
Studies on the economic costs of
the built environment must proceed, because they may provide
critical information to policymakers regarding resource allocations. We conducted a cost
analysis of building bike and
pedestrian trails to provide some
of this information.
DATA SOURCE
We obtained the costs of construction and maintenance of 5
bike and pedestrian trails in Lincoln, Neb, from the Department
of Parks and Recreation of Nebraska, and the number of trail
Wang et al. | Peer Reviewed | Reconnecting Urban Planning and Public Health | 549
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
results from the Behavioral Risk Factor
Surveillance System. Paper presented at:
Third International Health Economics
Association Conference, July 2001, York,
England.
12. 1997 Census of Retail Trade. Washington, DC: US Bureau of the Census;
2000.
13. ACCRA Cost of Living Index. Arlington, Va: American Chamber of Commerce Researchers Association; 2001.
14. Brownson RC, Baker EA, Housemann RA, Brennan LK, Bacak SJ. Environmental and policy determinants of
physical activity in the United States. Am
J Public Health. 2001;91:1995–2003.
15. Parks SE, Housemann RA, Brownson RC. Differential correlates of physical
activity in urban and rural adults of various socioeconomic backgrounds in the
United States. J Epidemiol Community
Health. 2003;57:29–35.
16. Clinical guidelines on the identification, evaluation, and treatment of
overweight and obesity in adults: the
evidence report. Obes Res. 1998;
6(suppl 2):S54.
17. Centers for Disease Control and Prevention. Physical activity trends—United
States, 1990–1998. MMWR Morb Mortal
Wkly Rep. 2001;50:166–169.
Cost Analysis of the Built Environment:
The Case of Bike and Pedestrian Trials in Lincoln, Neb
We estimated the annual
cost of bike and pedestrian
trails in Lincoln, Neb, using
construction and maintenance
costs provided by the Department of Parks and Recreation
of Nebraska. We obtained the
number of users of 5 trails
from a 1998 census report.
The annual construction cost
of each trail was calculated by
using 3%, 5%, and 10% discount rates for a period of useful life of 10, 30, and 50 years.
The average cost per mile and
per user was calculated.
Trail length averaged 3.6
miles (range = 1.6–4.6 miles).
Annual cost in 2002 dollars
ranged from $25 762 to
$248 479 (mean = $124 927;
median = $171 064). The cost
per mile ranged from $5735
to $54 017 (mean = $35 355;
median = $37 994). The annual cost per user was $235
(range = $83–$592), whereas
per capita annual medical cost
of inactivity was $622.
Construction of trails fits a
wide range of budgets and may
be a viable health amenity for
most communities. To increase
trail cost-effectiveness, efforts
to decrease cost and increase
the number of users should
be considered. (Am J Public
Health. 2004;94:549–553)
| Guijing Wang, PhD, Caroline A. Macera, PhD, Barbara Scudder-Soucie, MEd, Tom Schmid, PhD, Michael Pratt,
MD, MPH, David Buchner, MD, MPH, and Gregory Heath, DSc, MPH
ENVIRONMENTAL FACTORS
affect the health of all people in
both developed and developing
countries. Because of industrialization and the consequent environmental pollution, environmental changes in the past several
decades have led to new challenges for public health.
Many studies have documented
links between the environment
and human health.1–7 For example, household amenities and
other environmental exposures
have been linked to children’s
health problems such as cancer
and asthma,1–3 and environmental pollution has been linked to
high morbidity and mortality in
the general population.4–7 In recent years, the worldwide increase of obesity has prompted
discussions of environmental interventions such as increasing the
availability of healthy snacks and
building environments that are
amenable to physical activity as
possible effective means to prevent and control obesity and
other costly chronic diseases.8–11
Because of suburbanization,
the transportation systems in the
United States have been designed
for automobile use. Although
April 2004, Vol 94, No. 4 | American Journal of Public Health
automobile-oriented transportation is a necessity for economic
development and people’s daily
lives, the modern transportation
system may pose a hazardous environment for public health. Recently, 42% of American adults
expressed a great deal of concern
about urban sprawl and loss of
open space,12 which can create
an environment of physical inactivity, a major risk factor for
several chronic diseases and obesity.13–25 One study has demonstrated the association between
the built environment and physical activity by showing the effects
of urban environment on walking
behavior.22 Another study
showed that environmental features such as neighborhood design appeared to affect whether
residents walked to work.24
Pedestrian-oriented urban environments may promote physical
activity,22,23 and a combination of
urban design, land use patterns,
and transportation systems that
promote walking and bicycling
may help create more livable
communities.26–28 Lieberman recently suggested that proper design and land use patterns and
policies can increase public transit
use as well as walking and bicycling.26 Efforts to increase the
pedestrian-oriented environment
through mixed-use development,
street connectivity, and good community design can enhance both
the feasibility and attractiveness
of walking and bicycling.
Participation in regular physical activity depends in part on
the availability and proximity to
such resources as community
recreation facilities and walking
and bicycling trails, so building
such environments holds much
promise in health promotion.29–31
Studies on the economic costs of
the built environment must proceed, because they may provide
critical information to policymakers regarding resource allocations. We conducted a cost
analysis of building bike and
pedestrian trails to provide some
of this information.
DATA SOURCE
We obtained the costs of construction and maintenance of 5
bike and pedestrian trails in Lincoln, Neb, from the Department
of Parks and Recreation of Nebraska, and the number of trail
Wang et al. | Peer Reviewed | Reconnecting Urban Planning and Public Health | 549
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
TABLE 1—Number of Users and Costs of Construction and Maintenance of Trails
Trail Description
1. Concrete, 2 bridges
2. Limestone chip, 0 bridges
3. Concrete, 3 bridges
4. Concrete, 0 bridges
5. Concrete, 1 bridge
Date Built
Trail
Length (mi)
Number
of Users
Construction
Cost (2002 $)
Maintenance
Cost (2002 $)
1995
1997
1996
1989
1999
4.6
4.5
4.1
3.1
1.6
1638
232
1878
238
Not available
2 366 927
90 982
1 621 994
979 600
598 863
26 183
14 980
11 828
17 196
7 040
users from the Great Plains Trails
Network (Table 1).32 In addition
to the cost and number of users,
information about surface type,
date built, and length was also obtained for each trail. The construction cost was a 1-time investment on building the trails.
Ideally, the cost would be divided
into labor cost and capital cost,
but we were able to obtain only
the total cost without further details. Maintenance cost was based
on annual upkeep expenditures.
The construction and maintenance costs were adjusted to
2002 dollars using a 5% inflation
rate based on the date each trail
was built.
The number of users was determined by the Lincoln Recreational Trails Census, which was
conducted on Sunday, July 12,
1998 (Table 1). The census
began at 7:00 AM and concluded
at 9:00 PM the same day. Census
volunteers, working in 2-hour
shifts, counted cyclists, runners,
walkers, skaters, and miscellaneous users (such as persons with
skateboards, wheelchairs, and
horses.) Ideally, this number
would be adjusted according to
weather and date of the week to
determine a representative number of users, but this information
was unavailable.
The census report used this
information for the number of
users in 1998, which is comparable to the number of users in
other years. We used this num-
ber as a snapshot of the use of
trails for a conservative estimate
of cost-effectiveness of trails.
This value is conservative because the number of users during a year should be more than
that during a day. Additionally,
we varied the number of trail
users listed in the census report
by increasing or decreasing by
50% to calculate a range for the
cost per user.
DATA ANALYSIS
The construction cost is a large
1-time capital investment, so it is
necessary to spread the investment over the useful life in years
by determining an annual value
of the capital investment. To do
this, we calculated an annuity
factor that takes into account
time preference (r, discount rate)
and length of useful life (t, number of useful years). The annuity
(A) rate [A(t,r)] for time t years at
r discount rate was derived by
using A(t,r) = 1/r [1–1/(1+r)t ].
The annual equivalent cost
(AEC) of trail construction was
calculated by AEC = C × A(t,r),
where C is the 1-time capital expenditure.
The time preference needs to
be incorporated into the cost estimate even with zero inflation
because people prefer paying
later and getting benefits earlier.33 The discount rate, r, is a
quantitative measure of time
preference.
550 | Reconnecting Urban Planning and Public Health | Peer Reviewed | Wang et al.
Different discount rates have
been used in empirical studies;
the normal range is 3% to 10%.
We used 3%, 5%, and 10% as
discount rates for cost estimation
to cover a wide range of time
preference. The higher the discount rate, the more people value
current dollars. In the case of trail
investment, a higher discount rate
was associated with a higher
AEC for construction. For the
number of years of useful life of
the trails, we used 10, 30, and
50 years to cover a wide range of
situations. The longer the useful
life of the trails, the lower the
construction AEC.
For the case of a 5% discount
rate and 30 years of useful life,
we calculated annual cost per
mile for construction, maintenance, and a total (construction
and maintenance costs combined). In addition, for the total
cost, we calculated the annual average cost per user as a measure
of cost-effectiveness. We also analyzed the composition of cost
(construction versus maintenance)
and types of users.
RESULTS
The 5 trails were built between 1989 and 1999. Their average length was 3.6 miles
(range = 1.6–4.6 miles) (Table 1).
Four trails had a concrete surface, and 1 had a limestone-chip
surface. On the day of census,
the number of users ranged from
232 persons on the limestonechip trail (trail 2) to 1878 persons on the most heavily used
concrete trail (trail 3, a concrete
surface with 3 bridges). The total
construction cost ranged from
$90 982 ($20 218 per mile) for
trail 2, the limestone-chip trail, to
$2 366 927 ($514 549 per mile)
for trail 1, a concrete surface
with 2 bridges. The annual maintenance cost ranged from $7040
($4400 per mile) for trail 5, a
concrete surface with 1 bridge,
to $26 183 ($5692 per mile) for
trail 1.
The AEC for construction of
the 5 trails under all the scenarios of time preference and period
of useful years is useful information for those deciding on resource allocations (Table 2).
Among all the scenarios, the
highest cost ($542 021) was incurred for the 4.6-mile concrete
trail 1 with its 2 bridges under
the assumption of a 10% discount rate and 10 years of useful
life. The lowest cost ($4513)
was incurred for the 4.5-mile
limestone-chip trail under the
assumption of a 3% discount
rate and 50 years of useful life.
Using a 5% discount rate and
30 years of useful life, we found
that the annual average cost per
mile for trail 4 (concrete with no
bridges) was $45 505, and the
cost for trail 3 (concrete with 3
bridges) was $37 994 per mile.
The annual total cost per user
for trail 4 was $592, whereas the
cost per user for trail 3 was $83
(Table 3). The cost ranged from
$55 to $1185 per user.
For cost composition, 85% of
the total cost was construction
cost (range = 29%–91%) under
the assumption of a 5% discount
rate and 30 years of useful life
(Figure 1). Because only 1 trail
was made of limestone chips and
it cost much less than the con-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
TABLE 2—Annual Equivalent Construction Cost of Trails (2002 $)
Annual Equivalent Construction Cost
Trail
1
2
3
4
5
Years of Useful Life
3% Discount
5% Discount
10% Discount
390 437
169 920
129 442
458 009
216 653
182 434
542 021
353 298
335 912
13 613
5924
4513
15 969
7554
6361
18 898
12 318
11 712
254 816
110 897
84 479
298 916
141 397
119 064
353 746
230 577
219 231
216 524
94 242
71 792
254 023
120 162
101 183
300 619
195 948
186 306
81 271
35 370
26 944
95 337
45 097
37 975
112 824
73 540
69 922
Concrete, 2 bridges
10
30
50
Limestone chip, 0 bridges
10
30
50
Concrete, 3 bridges
10
30
50
Concrete, 0 bridges
10
30
50
Concrete, 1 bridge
10
30
50
crete trails, the average cost
composition was very close to
the cost compositions of the concrete trails. The composition was
similar for all the concrete trails.
The majority of users were bicyclists (73%), followed by runners/walkers (20%) (Figure 2).
Because of data limitations, we
did not know how the types of
users varied with the type of
trails.
DISCUSSION
When communities decide to
build a bike or pedestrian trail, financial budgeting should be
based on trail surface type, length,
and other features such as
bridges. Both construction and
maintenance costs should be considered, because although the
construction cost of the limestonechip trail was much lower than
that of the concrete surface trails,
the maintenance cost was not
necessarily lower.
The construction AEC varied
with the discount rate and number of years of useful life. Specifically, the cost increased as the
discount rate increased, and decreased as the number of years of
useful life increased. The figures
suggest that the cost of building a
trail can vary greatly and that
trails can be developed to meet a
variety of budgets.
As an example of the variances, the construction AEC (at a
5% discount rate and 30 years of
useful life) of building a concrete
trail with 1 bridge was 6 times as
expensive as building a limestonechip trail. The total annual cost
(including both maintenance and
construction costs) for a concrete
trail was 5 times more than that
for the limestone-chip trail.
Although the cost of building
and maintaining a limestone-chip
trail was lower than the cost for a
concrete trail, the limestone-chip
trail may not be the most costeffective strategy if the number
of users is taken into account.
The cost per user for the limestone-chip trail ($111) was more
than for a concrete trail with 3
bridges ($83). Thus, both the
total cost of trails and the number
of users should be considered
TABLE 3—Annual Total Cost (2002 $) of Trails Using a 5% Discount Rate and 30 Years
of Useful Life
Construction Cost
Maintenance Cost
Total Cost
Description
Trail
Total
Per Mile
Total
Per Mile
Total
Per Mile
Per Usera
Concrete, 2 bridges
Limestone chip, 0 bridges
Concrete, 3 bridges
Concrete, 0 bridges
Concrete, 1 bridge
Average
1
2
3
4
5
216 653
7 554
141 397
120 162
45 097
106 173
47 098
1 679
34 487
38 761
28 186
30 042
31 826
18 208
14 377
20 902
8 557
18 774
6919
4046
3507
6743
5348
5312
248 479
25 762
155 774
141 064
53 654
124 947
54 017
5 725
37 994
45 505
33 534
35 355
152 (101, 303)
111 (74, 222)
83 (55, 166)
592 (395, 1185)
Not available
235 (156, 469)
a
Figures in parentheses are the cost per user calculated by increasing or decreasing the number of users listed in the census report by 50%.
April 2004, Vol 94, No. 4 | American Journal of Public Health
when decisions about trails are
made. On average across all the
trails, the cost per user was $235.
This figure is much lower than
the economic benefit of physical
activity. A conservative estimate
of direct medical cost savings
from physical activity was $330
per person in 1987.34 Using a
5% inflation rate, this savings is
about $622 in 2002, nearly 3
times as high as the trail cost.
Therefore, developing trails may
be a cost-effective means to promote physical activity.
The fact that there were more
users on concrete trails than on
the limestone-chip trail (except
trail 4, built in 1989, on which
there were similar number of
users) may suggest that concrete
trails have more desirable features and are more convenient
for cycling. Because most of
these users were bicyclists (unfortunately, we did not have detailed information about user
type on the limestone-chip trail
versus the concrete trails), building trails to fit the needs of cyclists may substantially increase
the cost-effectiveness and net
health benefits of trails.
Several limitations should be
noted to interpret the findings
properly. (1) We cannot analyze
total construction costs such as
labor and material in more detail
because of data limitations. This
lack of information restricted our
ability to examine how other
major factors (e.g., material cost,
land value, funding sources)
influenced the total cost and how
to minimize project costs. (2) The
number of trails is small, and
each trail was built in a different
year. Technology and funding
sources change over time, so the
cost of each trail may not be
fully comparable. Therefore, the
average cost across trails may be
somewhat inaccurate, although
Wang et al. | Peer Reviewed | Reconnecting Urban Planning and Public Health | 551
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
100%
9%
13%
15%
90%
15%
16%
85%
84%
Trail 4 (concrete,
0 bridges)
Trail 5 (concrete,
1 bridge)
80%
70%
Composition
71%
60%
50%
91%
87%
85%
40%
30%
20%
29%
10%
0%
Average
Trail 1 (concrete,
2 bridges)
Trail 2 (limestone Trail 3 (concrete,
chip, 0 bridges)
3 bridges)
Trail, Trail Type, and Trail Features
FIGURE 1—Cost composition (black = maintenance; white = construction) of 5 trails in Lincoln, Neb,
using a 5% discount rate and 30 years of useful life.
7%
20%
Bicyclists
Runners/walkers
Others
73%
FIGURE 2—Trail user type in Lincoln, Neb.
we adjusted all the costs to 2002
dollars. (3) The census was conducted on a Sunday in summer;
we cannot claim that it was representative of the number and
type of users on an average day.
The lack of information means
we cannot adjust the number of
users according to weather, day
of the week, purpose of using
trails, and other factors. For example, many users may commute to work. The number of
users on a Sunday would not
capture this. Therefore, if the majority of trail users used trails for
commuting, the cost per user
may be severely underestimated.
552 | Reconnecting Urban Planning and Public Health | Peer Reviewed | Wang et al.
(4) Information on various qualitative aspects of trails was lacking. We have information only
on the surface type, length, and
number of bridges for each trail.
Other attributes such as safety
and convenient access to trails
also affect the cost of construction and maintenance. Because
of these information gaps, the
cost estimates according to trail
length and surface type should
be interpreted cautiously. (5) The
trails analyzed in this study were
built as a part of community design or development planning,
not as a public health intervention project. Factors such as increased property value or a more
attractive environment may have
been major determinants of
building trails rather than health
promotion. These added values
may have significantly biased our
cost estimates because we analyzed only financial cost and did
not consider the effects of other
community features such as loca-
tion and land value. (6) We analyzed only the cost data of trails
in a local community area. The
results should not be generalized
to other areas because household
income levels, natural characteristics, and local politics influence
the development of trails.
Despite these limitations, we
derived a framework of cost analysis based on the available data,
and several strengths should be
noted. (1) We derived the costs
of construction and maintenance
for each trail and adjusted all
costs to 2002 dollars, which
should increase the comparability
of the cost across the 5 trails.
(2) We incorporated different discount rates and number of useful
years into the analysis, and therefore covered a wide range of possible cost values for trails. (3) We
used trail length and number of
users on each trail to derive the
cost per mile and cost per user.
The cost per mile is useful for
community planners who are deciding to build trails based on financial feasibility. The cost per
user is useful to demonstrate the
usability of trails as a measure of
cost-effectiveness.
For future economic research
on the built environment, detailed cost information should
be collected systematically. This
information will make analysis
much more useful in identifying
factors influencing the cost of
trails. In addition, effectiveness
of trails in changing physical activity behaviors should be incorporated into the economic analysis. To do this, data such as
consumer willingness to pay for
trail construction and use if
trails are built should be collected. When information about
trail effectiveness is available,
cost-effectiveness of trails on
health promotion can be soundly
evaluated.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RECONNECTING URBAN PLANNING AND PUBLIC HEALTH 
CONCLUSIONS
Trails can fit a wide range of
budgets depending on the needs
and resources of the community.
Our research demonstrates the
need to increase cost-effectiveness
efforts by researching ways to decrease the cost of building trails
and to increase the number of
users of trails. We have also outlined specific information that
should be gathered to more completely explore the construction
and use of trails in the future. Policymakers and community developers may use the cost information to determine their needs and
the cost-effectiveness and feasibility of built environments in their
community.
About the Authors
Guijing Wang, Caroline A. Macera, Tom
Schmid, Michael Pratt, David Buchner,
and Gregory Heath are with the Division
of Nutrition and Physical Activity, Centers for Disease Control and Prevention,
Atlanta, Ga. Barbara Scudder-Soucie is
with the Physical Activity Program, Nebraska Health and Human Services System, Lincoln.
Requests for reprints should be sent to
Guijing Wang, PhD, Division of Nutrition
and Physical Activity, Centers for Disease
Control and Prevention, 4770 Buford
Hwy, Mail Stop K-46, Atlanta, GA
30341 (e-mail: [email protected]).
This article was accepted April 18,
2003.
Contributors
All authors helped plan the study. G.
Wang and B. Scudder-Soucie obtained
the data. G. Wang also performed data
analysis and wrote the article. C. A.
Macera, T. Schmid, D. Buchner, G.
Heath, and M. Pratt revised the article.
Pratt also supervised the project.
Human Participant Protection
No protocol approval was needed for
this study.
References
1. Friedman MS, Powell KE, Hutwagner L, Graham LM, Teague WG. Impact
of changes in transportation and commuting behaviors during the 1996
Summer Olympic Games in Atlanta on
air quality and childhood asthma.
JAMA. 2001;285:897–905.
activity facilities: results from qualitative
research. Health Promot J Aust. 1997;7:
16–21.
2. Savitz DA. Environmental exposures and childhood cancer: our best
may not be good enough. Am J Public
Health. 2001;91:562–563.
16. King AC, Jeffery RW, Fridinger F,
et al. Environmental and policy approaches to cardiovascular disease prevention through physical activity: issues
and opportunities. Health Educ Q. 1995;
22:499–511.
3. McBride ML. Childhood cancer
and environmental contaminants. Can J
Public Health. 1998;89(suppl 1):S58–
S68.
4. Pope CA 3rd, Burnett RT, Thun
MJ, et al. Lung cancer, cardiopulmonary
mortality, and long-term exposure to
fine particulate air pollution. JAMA.
2002;287:1132–1141.
5. Ibald-Mulli A, Stieber J, Wichmann
HE, Koenig W, Peters A. Effects of air
pollution on blood pressure: a population-based approach. Am J Public Health.
2001;91:571–577.
6. Morgan G, Corbett S, Wlodarczyk
J. Air pollution and hospital admissions
in Sydney, Australia, 1990 to 1994. Am
J Public Health. 1998;88:1761–1766.
7. Samet JM, Dominici F, Curriero
FC, Coursac I, Zeger SL. Fine particulate air pollution and mortality in 20
US cities. N Eng J Med. 2000;343:
1742–1749.
8. Glanz K, Lankenau B, Foerster S,
Temple S, Mullis R, Schmid T. Environmental and policy approaches to cardiovascular disease prevention through nutrition: opportunities for state and local
action. Health Educ Q. 1995;22:512–
527.
17. King AC. Community intervention
for promotion of physical activity and
fitness. Exerc Sport Sci Rev. 1991;19:
211–259.
18. King AC, Blair SN, Bild DE, et al.
Determinants of physical activity and interventions in adults. Med Sci Sports
Exerc. 1992;24(suppl 6):S221–S236.
19. Linenger JM, Chesson CV 2nd,
Nice DS. Physical fitness gains following
simple environmental change. Am J Prev
Med. 1991;7:298–310.
20. Sallis JF, Johnson MF, Calfas KJ,
Caparosa S, Nichols JF. Assessing perceived physical environmental variables
that may influence physical activity. Res
Q Exerc Sport. 1997;68:345–351.
21. Sallis JF, Hovell MF, Hofstetter CR,
et al. Distance between homes and exercise facilities related to the frequency
of exercise among San Diego residents.
Public Health Rep. 1990;105:179–185.
22. Berrigan D, Troiano RP. The association between urban form and physical activity in US adults. Am J Prev Med.
2002;23(2, suppl 1):74–79.
9. Hill JO, Peters JC. Environmental
contributions to the obesity epidemic.
Science. 1998;280:1371–1374.
23. Handy SL, Boarnet MG, Ewing R,
Killingsworth RE. How the built environment affects physical activity: views
from urban planning. Am J Prev Med.
2002;23(2, suppl 1):64–73.
10. Jeffery RW, French SA, Raether C,
Baxter JE. An environmental intervention to increase fruit and salad purchases in a cafeteria. Prev Med. 1994;
23:788–792.
24. Craig CL, Brownson RC, Cragg SE,
Dunn AL. Exploring the effect of the
environment on physical activity: a
study examining walking to work. Am J
Prev Med. 2002;23(2, suppl 1):36–43.
11. French SA, Story M, Jeffery RW.
Environmental influences on eating and
physical activity. Annu Rev Public Health.
2001;22:309–335.
25. French SA, Jeffery RW, Oliphant
JA. Facility access and self-reward as
methods to promote physical activity
among healthy sedentary adults. Am J
Health Promot. 1994;8:257–262.
12. Greenberg M. Earth Day plus 30
years: public concern and support for
environmental health. Am J Public
Health. 2001;91:559–562.
13. Sallis JF, Bauman A, Pratt M. Environmental and policy interventions to
promote physical activity. Am J Prev
Med. 1998;15:379–397.
14. Brownson RC, Housemann RA,
Brown DR, et al. Promoting physical activity in rural communities: walking trail
access, use, and effects. Am J Prev Med.
2000;18:235–241.
15. Corti B, Donovan RJ, Holman CDJ.
Factors influencing the use of physical
April 2004, Vol 94, No. 4 | American Journal of Public Health
28. Prairie Chapter, Sierra Club. Defining smart growth. Available at: http://
www.sierraclub.ca/psrairie/sprawl/
defining_smart_growth.htm. Accessed
January 6, 2004.
29. Task Force on Community Preventive Services. Recommendations to increase physical activity in communities.
Am J Prev Med. 2002;22(4, suppl 1):
67–72.
30. Kahn EB, Ramsey LT, Brownson
RC, et al. The effectiveness of interventions to increase physical activity. A systematic review. Am J Prev Med. 2002;
22(4, suppl 1):73–107.
31. Owen N, Leslie E, Salmon J,
Fotheringham MJ. Environmental determinants of physical activity and sedentary behavior. Exerc Sport Sci Rev.
2000;28:153–158.
32. Lincoln recreational trails census
report. Great Plains Trails Network
Newsletter. October 1998. Available at:
http://www.bikeped.com/gptnnews5/
index.htm. Accessed January 6, 2004.
33. Shaffer PA, Haddix AC. Time preference. In: Haddix AC, Teutsch SM,
Shaffer PA, Dunet DO, eds. Prevention
Effectiveness: A Guide to Decision Analysis and Economic Evaluation. New York,
NY: Oxford University Press; 1996:
76–84.
34. Pratt M, Macera CA, Wang G.
Higher direct medical costs associated
with physical activity. Physician
Sportsmed. 2000;28(10):63–70.
26. Lieberman W. Modal alternatives
for transit-oriented communities. Presented at: Congress for the New Urbanism VI. Cities in Context: Rebuilding
Communities Within the Natural Region; April 30–May 3, 1998; Denver,
Col. Available at: http://www.cnu.org/
cnu_reports/lieberman.pdf. Accessed
January 6, 2004.
27. Frank L. Exploring land use impacts on household travel choice and
vehicle emissions in the Atlanta region.
Available at: http://www.sprawlwatch.
org/states/georgia_frankatl.html. Accessed January 6, 2004.
Wang et al. | Peer Reviewed | Reconnecting Urban Planning and Public Health | 553
 RESEARCH AND PRACTICE 
The Epidemic of Pediatric
Traffic Injuries in South
Florida: A Review of the
Problem and Initial
Results of a Prospective
Surveillance Strategy
| S. Morad Hameed, MD, MPH, Charles A.
Popkin BA, Stephen M. Cohn, MD, E. William
Johnson, MPH, and the Miami Pediatric
Traffic Injury Task Force
This study identified specific regional
risk factors for the high rate of pediatric
pedestrian trauma in Florida. Of the 29
cases studied prospectively, 3 (10%)
occurred near ice cream trucks and 13
(45%) involved “dart-outs”; mean hospital charges were $24 478 ±$43 939.
Recommendations included an engineering change for a dangerous intersection, and a population-based recommendation was to equip ice cream
trucks with extending stop signs. (Am J
Public Health. 2004;94:554–556)
Approximately 30 000 children are
struck by cars each year in the United
States.1 Florida is home to 4 of the 5 most
dangerous cities for pedestrians in this country, and the mortality rate after pedestrian
trauma (3.9 per 100 000) is higher than the
national average (2.3 per 100 000).2 Pediatric pedestrian injuries are frequently encountered at our trauma referral center in
Miami, Florida.
Efforts to reduce the rates of pedestrian injury previously centered primarily on education programs and met with little success.3
This may be partly due to an absence of data
from prospective studies. Broad demographic
trends and socioeconomic and geographic
risk factors identified in the literature are
often either region-specific or too generalized
to be useful in the creation of practical, sitespecific prevention strategies.
The purpose of this study was to outline
the distribution, determinants, and effects of
pediatric pedestrian trauma (PPT) in our community. We hypothesized that careful data
collection would uncover community-specific
PPT risk factors and suggest avenues for prevention and resource allocation.
METHODS
This study, set at the Jackson Memorial
Hospital/University of Miami Ryder Trauma
Center (the sole trauma center for approximately 3 million Miami-Dade County residents), was performed in 2 phases.
Phase 1—Retrospective Review
Medical records of pediatric pedestrians
(younger than age 16 years) who presented to
our institution between January 1994 and December 1996 were reviewed. Demographic
parameters were defined and analyzed to assess the impact of PPT in our communities.
Phase 2—Prospective Data Collection
Recommendations from a multidisciplinary
task force (including local medical, police, and
government agencies) were incorporated into
a design of a 4-month prospective cohort
study. Detailed information from hospital records, crash scene visits, patients, families, and
police interviews was compiled on consecutive cases of PPT treated at our center (July 1
through October 31, 2000). Injury scene conditions were systematically assessed and especially emphasized in the analysis.
RESULTS
Retrospective Review
A total of 235 PPT cases were evaluated.
Grade school children were most often (53%)
injured, usually in the vicinity of schools.
Boys predominated, and African American
children accounted for 60% of the cases.
High mean hospital charges ($16 553) resulted from the high incidence rates (32%) of
head injuries.
Prospective Data Collection
Population, scene, environmental, and cost
issues were explored in 29 consecutive cases
of PPT. Many children (69%) were from
single-parent homes. Although Miami is ethnically diverse, a disproportionate number of
PPT events occurred in predominantly African
554 | Research and Practice | Peer Reviewed | Hameed et al
American neighborhoods. Thirty-five percent
of children came from homes where at least 1
parent had some postsecondary education.
At most sites, intervals between marked intersections were long, allowing vehicle acceleration and predisposing random pedestrian
crossing patterns. Some intersections
(Figure 1) were observed to be poorly regulated by misplaced traffic lights and were a
source of long-standing community apprehension. Mechanisms involving obstruction of
view (“dart-outs”)4 were common (46%), although most PPT incidents (64%) occurred
in clear daylight conditions. Site visits provided insight into situational dynamics. For
example, 3 events (10%) resulted from traffic
disruption by ice cream trucks.
Hospital charges ranged from $336 to
$172 283, and at the time of the site visit
(25 ±13 days post-PPT), 44% of children
had not returned to school.
DISCUSSION
Previous studies (Table 1) have characterized region-specific risk factors for PPT,
which may not be completely generalizable to
Miami, with its unique cultural and geographic milieu. As indicated by our review,
South Florida is fertile ground for a comprehensive PPT prevention strategy. Groups such
as the North Miami Crash Traffic Safety Team
and the Safe Kids prevention programs have
taken an active role in pedestrian education,
but to date, prevention initiatives have not
been designed with specific references to objectively measured risk factors.
Available information sources, including
police reports and hospital records, lacked
sufficient detail to clarify the causes of PPT.
The second phase of this study was designed
to provide useful information for development of directed multidisciplinary prevention
policy.
All 29 cases studied during our surveillance period had implications for the design
of high-risk or population-based prevention
strategies. Miami’s uninhibited westward
growth has resulted in the creation of communities with high volumes of rapid commuter traffic and long residential streets without sidewalks. Situations such as that
summarized in Figure 1 will require innova-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Stop lights placed well
before intersection
Car
A
STOP
BUS
Car
A
Car B
Car B
NW 67 street
BUS
STOP
House
Car
A
FIGURE 1—Problematic intersection at 2nd Ave and NW 67th St, Miami, Fla. During this
site visit, numerous bystanders approached the investigators to find out when the city was
planning to modify the control of this dangerous intersection. The visit was prompted by the
injuries of a 5-year-old African American boy who had been holding his mother’s hand at
the bus stop on the corner. The driver of car A, after waiting behind a bus ahead of the
traffic lights, swerved onto the shoulder area at a high rate of speed and entered the
intersection unaware that the light had changed. The ensuing events are depicted. After the
collision with car B, the driver of car A lost control, striking the boy, his sister, and their
mother. The car then struck a fence at the corner and proceeded toward the wall of a
nearby house with the child still trapped underneath.
tive engineering approaches to eliminate highrisk scenarios. Other high-risk situations, such
as those involving ice cream trucks, will require legislation mandating the use of safety
measures such as extending stop signs on
these vehicles to help reduce the impact of
the frequently observed dart-outs. Conscientious regulation of school bus access and
pickup and drop-off practices would reduce
the incidence of injuries observed during
school hours.
Although this study was performed without
external funding, a grant from the Florida Department of Transportation will allow us to
address some of the limitations of this initial
surveillance. Information will be collected
over a school year along with an economic
evaluation, and more objective scene measurements will be made. We hope to delineate
a cost-effective surveillance-based prevention
plan that reduces the incidence of children
struck by motor vehicles.
About the Authors
S. Morad Hameed, Charles A. Popkin, Stephen M. Cohn,
and E. William Johnson are with the Divisions of Trauma
and Surgical Critical Care, Daughtry Family Department
of Surgery, University of Miami School of Medicine
Miami, Florida.
Requests for reprints should be sent to Stephen M.
Cohn, MD, Medical Director, Ryder Trauma Center,
Daughtry Family Department of Surgery, University of
Miami School of Medicine, 1800 NW 10th Ave, Suite
TABLE 1—A Summary of the Pediatric Pedestrian Traffic Injury Literature
Authors
Rivara and Barber, 19855
Brison et al., 19884
Mueller et al., 19906
Braddock et al., 19917
Roberts et al., 19953
Agran et al., 19968
Calhoun et al., 19989
Durkin et al., 199910
Miami Pediatric Traffic
Injury Task Force, 2001a
Type of Study
No. of Injuries
Location
Main Conclusion
Traffic engineering modifications are practical solution
Prevention strategies must be age-specific
Busy streets, multifamily homes are strong risk factors
High-density areas are problematic
High traffic volume in urban areas should be reduced
Parked cars and reduced speed would decrease injuries
Manageable environmental risk factors were identified;
education should be targeted toward grade school children
Community interventions (play areas, education) may be helpful
in preventing injury
Retrospective
Retrospective
Case–control
Retrospective
Case–control
Case–control
Retrospective
210
71
98
198
190
39
91
Memphis, Tenn
Washington State
King County, Washington
Hartford, Conn
Auckland, New Zealand
Orange County, California
Jefferson County, Alabama
Retrospective review of newly
implemented intervention
Harlem, New York, NY
Retrospective review
Incidence study of all injuries
(n = 981) in Harlem,
New York, NY
235
Prospective surveillance
29
Miami–Dade County, Florida
Ongoing surveillance is required for continued development of
focused prevention strategies
a
Unpublished data.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Hameed et al | Peer Reviewed | Research and Practice | 555
 RESEARCH AND PRACTICE 
227, Miami, FL 33136 (e-mail: stephe[email protected]
miami.edu).
This brief was accepted June 30, 2002.
Contributors
S. M. Hameed contributed to the study design, data collection, data analysis, and manuscript preparation. C. A.
Popkin contributed to the data collection and manuscript preparation. S. M. Cohn contributed to the study
design, data analysis, and manuscript preparation.
W. M. Johnson contributed to the study design and data
collection.
Acknowledgments
The members of the Miami Pediatric Traffic Injury Task
Force are Frank Pernas, BA, David Henderson, AICP,
Mimi Sutherland, RN, MS, Margaret Brown, MSN, J. Esteban Varela, MD, Dimeter Hristov, MD, Kimberly
Schwartz, MD, Officer Luis Taborda, BA, Julie Jackowski, RN, Tracy Byrd, BA, Gilian Hotz, PhD, Lewis
Saye, BA, and Jose Guerrier, MD.
Human Participant Protection
Institutional review board approval was obtained from
the University of Miami before this study was initiated.
References
Florida’s Motorcycle
Helmet Law Repeal and
Fatality Rates
in the state was high enough to permit a
monthly time series analysis. The Florida motorcycle helmet law change has not been
evaluated statewide.5
METHODS
| Andreas Muller, PhD
Data
On July 1 2000, the State of Florida
exempted adult motorcyclist and moped
riders from wearing helmets provided
they have medical insurance of $10000.
Monthly time series of motorcycle occupant deaths are examined from 1/1994
to 12/2001. The interrupted time series
analysis estimates a 48.6% increase in
motorcycle occupant deaths the year
after the law change. The impact estimate reduces to 38.2% and 21.3% when
trends in travel miles and motorcycle registrations are controlled. Our findings
suggest that the law’s age exemption
should be revoked. (Am J Public Health.
2004;94:556–558)
1. Accident Facts—1996 Edition. Ithasca, Ill: National
Safety Council; 1996.
2. McCann B, DeLille B. Mean Streets 2000 report.
Surface Transportation Policy Project June 2000. Available at: http://www.transact.org FCC /Reports/
ms2000/natpress.htm2000. Accessed May 15, 2002.
3. Roberts I, Norton R, Jackson R, Dunn R, Hassall I.
Effect of environmental factors on risk of injury of
child pedestrians by motor vehicles: a case-control
study. BMJ. 1995;310:91-94.
4. Brison RJ, Wicklund K, Mueller BA. Fatal pedestrian injuries to young children: a different pattern of
injury. Am J Public Health. 1988;78:793-795.
5. Rivara FP, Barber M. Demographic analysis of
childhood pedestrian injuries. Pediatrics. 1985;76:
375-381.
6. Mueller BA, Rivara FP, Lii SM, Weiss NS. Environmental factors and the risk for childhood pedestrianmotor vehicle collision occurrence. Am J Epidemiol.
1990; 132:550-560.
7. Braddock M, Lapidus G, Gregorio D, Kapp M,
Banco L. Population, income, and ecological correlates of child pedestrian injury. Pediatrics. 1991;88:
1242-1247.
8. Agran PF, Winn DG, Anderson CL, Tran C, Del
Valle CP. The role of the physical and traffic environment in child pedestrian injuries. Pediatrics. 1996;98:
1096-1103.
9. Calhoun AD, McGwin G Jr, King WD, Rousculp
MD. Pediatric pedestrian injuries: a community assessment using a hospital surveillance system. Acad Emerg
Med. 1998;5:685-690.
10. Durkin MS, Laraque D, Lubman I, Barlow B. Epidemiology and prevention of traffic injuries to urban
children and adolescents. Pediatrics. 1999;103(6):e74.
556 | Research and Practice | Peer Reviewed | Muller
Between 1997 and 2001, nationwide motorcycle rider fatalities increased by 50%
while motorcycle registrations increased by
31%.1,2 The rise in death rates may be related
to the concurrent weakening of motorcycle
helmet laws in Arkansas, Texas, Kentucky,
Louisiana, and Florida. In comparing rates the
year before (1996) and the year after (1998)
the helmet law change, Preusser et al.3 found
a 21% increase in motorcyclist deaths in
Arkansas and a 30% increase in Texas. This
analysis tries to determine the effect of weakening Florida’s motorcycle helmet law.
Since July 1, 2000, Florida statutes have
required motorcycle riders younger than 21
years of age to wear helmets. Adult motorcycle and moped riders are exempted provided
they have insurance for motorcycle accident
injuries with minimum medical benefit coverage of $10 000.4 Before July 1, 2000, Florida
had a helmet law that required all riders to
wear safety helmets.
The State of Florida is of interest because it
accounts for 9% of all motorcycle rider
deaths in the United States. Coinciding with
the helmet law change, the number of Florida’s motorcycle registrations increased substantially. The number of motorcycle deaths
Florida’s monthly motorcycle rider deaths
for the period January 1994 to December
2001 were analyzed. Motorcycle rider deaths
included operators or passengers of motorcycles, mopeds, minibikes, motorized threewheelers, off-road, other, and unknown types
of motorcycles. All-terrain vehicles were excluded. The definition is intentionally comprehensive to allow for comparisons over
time. It matches the National Highway Traffic
Safety Administration’s definition of motorcycle rider death.
The time series data were obtained from
the Fatal Accident Reporting System database.1 Yearly issues of Highway Statistics provided motorcycle registration and travel
miles for the period 1996 to 2001, and earlier years came from the 1995 summary volume.2 To obtain a more realistic representation of the Florida motorcycle registration
and travel trends, annual data were converted into a monthly series by 12-month
centered moving averages. The smoothing
operation removes 12 observations and reduces the sample size to 84 months, July
1994 to June 2001.
The following series were analyzed: motorcycle rider deaths, motorcycle rider deaths
per billion travel miles, and motorcycle rider
deaths per 10 000 registered motorcycles. Restricting the analysis to rider deaths of adult
motorcyclists (>20 years) generates 3 additional series.
Time Series Models
The method of analysis was interrupted
time series analysis using Box–Jenkins models.6,7 To approximate normal distributions
closely, all time series were converted into
natural log units. A step function for an
abrupt, permanent impact models Florida’s
motorcycle helmet law change beginning in
July 2000. The residuals of all time series follow random process properties (data available
from the author upon request).
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
227, Miami, FL 33136 (e-mail: [email protected]
miami.edu).
This brief was accepted June 30, 2002.
Contributors
S. M. Hameed contributed to the study design, data collection, data analysis, and manuscript preparation. C. A.
Popkin contributed to the data collection and manuscript preparation. S. M. Cohn contributed to the study
design, data analysis, and manuscript preparation.
W. M. Johnson contributed to the study design and data
collection.
Acknowledgments
The members of the Miami Pediatric Traffic Injury Task
Force are Frank Pernas, BA, David Henderson, AICP,
Mimi Sutherland, RN, MS, Margaret Brown, MSN, J. Esteban Varela, MD, Dimeter Hristov, MD, Kimberly
Schwartz, MD, Officer Luis Taborda, BA, Julie Jackowski, RN, Tracy Byrd, BA, Gilian Hotz, PhD, Lewis
Saye, BA, and Jose Guerrier, MD.
Human Participant Protection
Institutional review board approval was obtained from
the University of Miami before this study was initiated.
References
Florida’s Motorcycle
Helmet Law Repeal and
Fatality Rates
in the state was high enough to permit a
monthly time series analysis. The Florida motorcycle helmet law change has not been
evaluated statewide.5
METHODS
| Andreas Muller, PhD
Data
On July 1 2000, the State of Florida
exempted adult motorcyclist and moped
riders from wearing helmets provided
they have medical insurance of $10000.
Monthly time series of motorcycle occupant deaths are examined from 1/1994
to 12/2001. The interrupted time series
analysis estimates a 48.6% increase in
motorcycle occupant deaths the year
after the law change. The impact estimate reduces to 38.2% and 21.3% when
trends in travel miles and motorcycle registrations are controlled. Our findings
suggest that the law’s age exemption
should be revoked. (Am J Public Health.
2004;94:556–558)
1. Accident Facts—1996 Edition. Ithasca, Ill: National
Safety Council; 1996.
2. McCann B, DeLille B. Mean Streets 2000 report.
Surface Transportation Policy Project June 2000. Available at: http://www.transact.org FCC /Reports/
ms2000/natpress.htm2000. Accessed May 15, 2002.
3. Roberts I, Norton R, Jackson R, Dunn R, Hassall I.
Effect of environmental factors on risk of injury of
child pedestrians by motor vehicles: a case-control
study. BMJ. 1995;310:91-94.
4. Brison RJ, Wicklund K, Mueller BA. Fatal pedestrian injuries to young children: a different pattern of
injury. Am J Public Health. 1988;78:793-795.
5. Rivara FP, Barber M. Demographic analysis of
childhood pedestrian injuries. Pediatrics. 1985;76:
375-381.
6. Mueller BA, Rivara FP, Lii SM, Weiss NS. Environmental factors and the risk for childhood pedestrianmotor vehicle collision occurrence. Am J Epidemiol.
1990; 132:550-560.
7. Braddock M, Lapidus G, Gregorio D, Kapp M,
Banco L. Population, income, and ecological correlates of child pedestrian injury. Pediatrics. 1991;88:
1242-1247.
8. Agran PF, Winn DG, Anderson CL, Tran C, Del
Valle CP. The role of the physical and traffic environment in child pedestrian injuries. Pediatrics. 1996;98:
1096-1103.
9. Calhoun AD, McGwin G Jr, King WD, Rousculp
MD. Pediatric pedestrian injuries: a community assessment using a hospital surveillance system. Acad Emerg
Med. 1998;5:685-690.
10. Durkin MS, Laraque D, Lubman I, Barlow B. Epidemiology and prevention of traffic injuries to urban
children and adolescents. Pediatrics. 1999;103(6):e74.
556 | Research and Practice | Peer Reviewed | Muller
Between 1997 and 2001, nationwide motorcycle rider fatalities increased by 50%
while motorcycle registrations increased by
31%.1,2 The rise in death rates may be related
to the concurrent weakening of motorcycle
helmet laws in Arkansas, Texas, Kentucky,
Louisiana, and Florida. In comparing rates the
year before (1996) and the year after (1998)
the helmet law change, Preusser et al.3 found
a 21% increase in motorcyclist deaths in
Arkansas and a 30% increase in Texas. This
analysis tries to determine the effect of weakening Florida’s motorcycle helmet law.
Since July 1, 2000, Florida statutes have
required motorcycle riders younger than 21
years of age to wear helmets. Adult motorcycle and moped riders are exempted provided
they have insurance for motorcycle accident
injuries with minimum medical benefit coverage of $10 000.4 Before July 1, 2000, Florida
had a helmet law that required all riders to
wear safety helmets.
The State of Florida is of interest because it
accounts for 9% of all motorcycle rider
deaths in the United States. Coinciding with
the helmet law change, the number of Florida’s motorcycle registrations increased substantially. The number of motorcycle deaths
Florida’s monthly motorcycle rider deaths
for the period January 1994 to December
2001 were analyzed. Motorcycle rider deaths
included operators or passengers of motorcycles, mopeds, minibikes, motorized threewheelers, off-road, other, and unknown types
of motorcycles. All-terrain vehicles were excluded. The definition is intentionally comprehensive to allow for comparisons over
time. It matches the National Highway Traffic
Safety Administration’s definition of motorcycle rider death.
The time series data were obtained from
the Fatal Accident Reporting System database.1 Yearly issues of Highway Statistics provided motorcycle registration and travel
miles for the period 1996 to 2001, and earlier years came from the 1995 summary volume.2 To obtain a more realistic representation of the Florida motorcycle registration
and travel trends, annual data were converted into a monthly series by 12-month
centered moving averages. The smoothing
operation removes 12 observations and reduces the sample size to 84 months, July
1994 to June 2001.
The following series were analyzed: motorcycle rider deaths, motorcycle rider deaths
per billion travel miles, and motorcycle rider
deaths per 10 000 registered motorcycles. Restricting the analysis to rider deaths of adult
motorcyclists (>20 years) generates 3 additional series.
Time Series Models
The method of analysis was interrupted
time series analysis using Box–Jenkins models.6,7 To approximate normal distributions
closely, all time series were converted into
natural log units. A step function for an
abrupt, permanent impact models Florida’s
motorcycle helmet law change beginning in
July 2000. The residuals of all time series follow random process properties (data available
from the author upon request).
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
correct for the motorcycle registration trend
may understate the law’s impact. The large
increase (19.6%) in Florida motorcycle registrations in 2001 (Figure 1) suggests that
changing the law may have stimulated interest in motorcycling and increased motorcycle
registrations. Substantial increases in motorcycle registrations also occurred in Arkansas
(47%), Louisiana (13%), and Texas (12%)
the year after their helmet laws were weakened. The extent of such a law-induced effect is currently unknown. On the basis of
registration and miles traveled, it is estimated that between 46 and 82 additional
motorcyclists died in Florida the year after
the helmet law changed.
In 2001, only 53% of Florida underage
motorcyclists who died in crashes wore motorcycle helmets; for adults the figure was
39%.1 That is, the legal age restriction is
barely effective and amounts to a de facto
helmet law repeal.
350
300
250
200
150
100
50
Registrations, 1000s
Deaths
Deaths, age >20 y
Travel miles, billions
0
1994
1995
1996
1997
1998
1999
2000
2001
Source. Fatality Analysis Reporting System1 and Federal Highway Administration.2
FIGURE 1—Florida motorcycle registrations, motorcycle rider deaths, and travel miles for all
motor vehicles, 1994 to 2001.
RESULTS
Figure 1 presents annual trends in motorcyclist deaths, motorcycle registrations,
and travel miles in Florida. During the year
2000, Florida motorcyclist deaths increased by 81 (45.5%), motorcycle registrations by 19 494 (8.1%), and travel miles by
9 billion (6.3%). The upward trends in motorcycle registrations and travel miles are
noteworthy.
Figure 2 presents impact estimates based
on the analyses of 6 monthly time series.
The estimates indicate that the change in
Florida’s helmet law increased motorcycle
rider deaths. The impact on all motorcycle
rider deaths is strongest, 48.6%. Controlling for travel miles reduces the estimate to
38.2%, and correcting for the motorcycle
registration trend reduces the estimate to
only 21.3%. Restricting the analysis to
adults reduces the previous estimates only
slightly.
CONCLUSION
change depends on which “exposure” measures are controlled. Since travel increased
in Florida, the impact estimates based on the
absolute number of deaths are probably
overstated. Conversely, the estimates that
100%
90%
80%
70%
60%
50%
48.6
44.9
40%
38.2
The analysis suggests that exempting adult
motorcyclists from wearing helmets increased the number of motorcyclist fatalities
in Florida. However, the effect of the law
April 2004, Vol 94, No. 4 | American Journal of Public Health
34.6
30%
21.3
20%
18.4
10%
0%
Deaths
DISCUSSION
This study finds that the current agerestricted version of Florida’s motorcycle helmet law resulted in more motorcyclist deaths
Deaths per
billion miles
Deaths per
10 000
registrations
Deaths,
age >20
Deaths, age
>20, per
billion miles
Deaths, age
>20, per
10 000
registrations
FIGURE 2—Estimated impact of the change in Florida’s motorcycle helmet law by series,
with 95% confidence limits.
Muller | Peer Reviewed | Research and Practice | 557
 RESEARCH AND PRACTICE 
even after adjustment for concurrent increases in motorcycle registrations or miles
traveled. Exempting adult motorcycle riders
from wearing motorcycle helmets is counterproductive for motorcyclists’ health and unnecessarily increases insurance and medical
care expenses.
Femur Fractures in
Infants and Young
Children
| Desmond Brown, MD, and Elliott Fisher, MD,
MPH
About the Author
The author is with the Department of Health Services Administration, University of Arkansas at Little Rock, and
the Department of Health Policy and Management, College
of Public Health, University of Arkansas for Medical Sciences, Little Rock.
Requests for reprints should be sent to Andreas
Muller, PhD, UALR, Ross Hall 202, 2801 S University Ave, Little Rock, AR 72204 (e-mail: [email protected]
ualr.edu).
This brief was accepted June 18, 2003.
Human Participation Protection
No protocol approval was needed because no individuals are identified by the analysis.
Acknowledgments
I would like to thank the anonymous reviewers for their
helpful comments.
References
1. US Dept of Transportation, National Highway
Traffic Safety Administration, National Center for Statistics and Analysis. Fatality Analysis Reporting System
(FARS) Web-based encyclopedia. Reports: people: motorcyclists. Available at: http://www-fars.nhtsa.dot.gov.
Accessed February 18, 2003.
2. US Dept of Transportation, Federal Highway Administration. Highway statistics (multiple years). Available at: http://www.fhwa.dot.gov/policy/ohpi/hss/
hsspubs.htm. Accessed February 18, 2003.
3. Preusser DF, Hedlund JH, Ulmer RG. Evaluation
of Motorcycle Helmet Law Repeal in Arkansas and Texas.
Final Report, DTNH22–97-D-05018. Springfield Va:
National Technical Information Service; September
2000.
4. Florida Senate. The 2002 Florida Statutes, Title
XXIII, Chap 316.211. Available at: http://www.flsenate.
gov/Statutes. Accessed March 5, 2004.
5. Hotz GA, Cohn SM, Popkin C, et al. The impact
of a repealed motorcycle helmet law in Miami-Dade
County. J Trauma. 2002;52:469–474.
6. Box GEP, Jenkins GM. Time Series Analysis: Forecasting and Control. Rev ed. San Francisco, Calif:
Holden-Day; 1976.
7. Pankratz A. Forecasting With Dynamic Regression
Models. New York, NY: John Wiley & Sons; 1991.
Using an administrative database,
we determined rates of femur fracture
by year of age for children younger
than 6 years and by month of age. The
highest rate of femur fracture was in
children younger than 1 year and in 2year-olds; the greatest number of fractures occurred during the third month
of life. While femur fractures in children are often due to accidental injury, the reasons for the peak in the
first year and the subsequent decline
are not clear. (Am J Public Health.
2004;94:558–560)
The incidence of femur fractures in children is believed to have 2 peaks, one at the
age of 2 to 3 years and another during adolescence.1 This view is based, however, on
older studies from Scandinavia2–4 and a
more recent study from Maryland5 and may
not reflect the experience of the US population. Previous studies have also categorized
children by year of age, which may be insufficiently precise for the infant or young
child in whom rapid changes in size, physical ability, and behavior may affect the risk
of fracture.
Although most femur fractures in children
are caused by falls or other unintentional injuries, abuse is considered more likely in the
child aged younger than 1 year or not yet
able to walk. In this brief, we focus on this
youngest group, reporting data on hospital
discharges for femur fractures from a national
database in which children were categorized
by age in months.
resenting nearly a third of the estimated 6.7
million pediatric discharges during that year.
Using International Classification of Diseases,
Ninth Revision, Clinical Modification (ICD-9CM) codes for fracture of the proximal femur,
femoral shaft, and distal femur (diagnosis
codes 820–821.39), we identified 3308 records of children under the age of 6 discharged from a hospital with a diagnosis of
femur fracture. Fractures occurring during
childbirth were excluded.
Using population weights provided with
the database, we calculated national estimates
for the number of femur fractures in each 1year age group. We determined fracture incidence rates by dividing the number of fractures by the estimated number of children in
each age group, using population estimates
for 1997 from the US Bureau of the Census.
To examine the relationship between age and
femur fractures more closely, we identified
2753 records for which the age in months
was available. Because we lacked the population denominator to determine rates of fracture, we report the counts for this subset of
patients.
RESULTS
The rate of femur fracture was highest during the first year of life and in 2-year olds
(Table 1). One-year-olds were less likely to
sustain a fractured femur than those aged
younger than 1 year. While the ratio of boys
to girls was nearly equal in those aged younger than 1 year, all older age groups had
more boys.
In children for whom the age in months
was known, the greatest number of fractures
occurred during the third month of life (Figure 1). There were slightly fewer fractures in
children aged 4 to 11 months, and fewer still
in children aged 12 to 20 months. After the
first peak during infancy, there was a second
peak in children aged 20 to 40 months. In
children older than 40 months but younger
than 72 months, the number of fractures was
lower and relatively constant.
METHODS
DISCUSSION
The 1997 Kids’ Inpatient Database6 contains 1.9 million records of hospital discharges for children aged 18 or younger, rep-
558 | Research and Practice | Peer Reviewed | Brown and Fisher
Previous studies of femur fractures in childhood have identified a peak in incidence at
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
even after adjustment for concurrent increases in motorcycle registrations or miles
traveled. Exempting adult motorcycle riders
from wearing motorcycle helmets is counterproductive for motorcyclists’ health and unnecessarily increases insurance and medical
care expenses.
Femur Fractures in
Infants and Young
Children
| Desmond Brown, MD, and Elliott Fisher, MD,
MPH
About the Author
The author is with the Department of Health Services Administration, University of Arkansas at Little Rock, and
the Department of Health Policy and Management, College
of Public Health, University of Arkansas for Medical Sciences, Little Rock.
Requests for reprints should be sent to Andreas
Muller, PhD, UALR, Ross Hall 202, 2801 S University Ave, Little Rock, AR 72204 (e-mail: [email protected]
ualr.edu).
This brief was accepted June 18, 2003.
Human Participation Protection
No protocol approval was needed because no individuals are identified by the analysis.
Acknowledgments
I would like to thank the anonymous reviewers for their
helpful comments.
References
1. US Dept of Transportation, National Highway
Traffic Safety Administration, National Center for Statistics and Analysis. Fatality Analysis Reporting System
(FARS) Web-based encyclopedia. Reports: people: motorcyclists. Available at: http://www-fars.nhtsa.dot.gov.
Accessed February 18, 2003.
2. US Dept of Transportation, Federal Highway Administration. Highway statistics (multiple years). Available at: http://www.fhwa.dot.gov/policy/ohpi/hss/
hsspubs.htm. Accessed February 18, 2003.
3. Preusser DF, Hedlund JH, Ulmer RG. Evaluation
of Motorcycle Helmet Law Repeal in Arkansas and Texas.
Final Report, DTNH22–97-D-05018. Springfield Va:
National Technical Information Service; September
2000.
4. Florida Senate. The 2002 Florida Statutes, Title
XXIII, Chap 316.211. Available at: http://www.flsenate.
gov/Statutes. Accessed March 5, 2004.
5. Hotz GA, Cohn SM, Popkin C, et al. The impact
of a repealed motorcycle helmet law in Miami-Dade
County. J Trauma. 2002;52:469–474.
6. Box GEP, Jenkins GM. Time Series Analysis: Forecasting and Control. Rev ed. San Francisco, Calif:
Holden-Day; 1976.
7. Pankratz A. Forecasting With Dynamic Regression
Models. New York, NY: John Wiley & Sons; 1991.
Using an administrative database,
we determined rates of femur fracture
by year of age for children younger
than 6 years and by month of age. The
highest rate of femur fracture was in
children younger than 1 year and in 2year-olds; the greatest number of fractures occurred during the third month
of life. While femur fractures in children are often due to accidental injury, the reasons for the peak in the
first year and the subsequent decline
are not clear. (Am J Public Health.
2004;94:558–560)
The incidence of femur fractures in children is believed to have 2 peaks, one at the
age of 2 to 3 years and another during adolescence.1 This view is based, however, on
older studies from Scandinavia2–4 and a
more recent study from Maryland5 and may
not reflect the experience of the US population. Previous studies have also categorized
children by year of age, which may be insufficiently precise for the infant or young
child in whom rapid changes in size, physical ability, and behavior may affect the risk
of fracture.
Although most femur fractures in children
are caused by falls or other unintentional injuries, abuse is considered more likely in the
child aged younger than 1 year or not yet
able to walk. In this brief, we focus on this
youngest group, reporting data on hospital
discharges for femur fractures from a national
database in which children were categorized
by age in months.
resenting nearly a third of the estimated 6.7
million pediatric discharges during that year.
Using International Classification of Diseases,
Ninth Revision, Clinical Modification (ICD-9CM) codes for fracture of the proximal femur,
femoral shaft, and distal femur (diagnosis
codes 820–821.39), we identified 3308 records of children under the age of 6 discharged from a hospital with a diagnosis of
femur fracture. Fractures occurring during
childbirth were excluded.
Using population weights provided with
the database, we calculated national estimates
for the number of femur fractures in each 1year age group. We determined fracture incidence rates by dividing the number of fractures by the estimated number of children in
each age group, using population estimates
for 1997 from the US Bureau of the Census.
To examine the relationship between age and
femur fractures more closely, we identified
2753 records for which the age in months
was available. Because we lacked the population denominator to determine rates of fracture, we report the counts for this subset of
patients.
RESULTS
The rate of femur fracture was highest during the first year of life and in 2-year olds
(Table 1). One-year-olds were less likely to
sustain a fractured femur than those aged
younger than 1 year. While the ratio of boys
to girls was nearly equal in those aged younger than 1 year, all older age groups had
more boys.
In children for whom the age in months
was known, the greatest number of fractures
occurred during the third month of life (Figure 1). There were slightly fewer fractures in
children aged 4 to 11 months, and fewer still
in children aged 12 to 20 months. After the
first peak during infancy, there was a second
peak in children aged 20 to 40 months. In
children older than 40 months but younger
than 72 months, the number of fractures was
lower and relatively constant.
METHODS
DISCUSSION
The 1997 Kids’ Inpatient Database6 contains 1.9 million records of hospital discharges for children aged 18 or younger, rep-
558 | Research and Practice | Peer Reviewed | Brown and Fisher
Previous studies of femur fractures in childhood have identified a peak in incidence at
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
180
160
Femur Fractures
140
120
100
80
60
40
20
0
1
4
7
10 13
16 19 22 25
28 31 34 37
40 43 46 49
52 55 58 61
64 67 70
Age, months
Source. Kids’ Inpatient Database, 1997.6
FIGURE 1—Estimated number of femur fractures among children in the United States, by
month of age.
TABLE 1—US Population Estimates for Femur Fractures in Children, by Year of Age
No. of Femur Fractures
Femur Fractures/100 000 (95% CI)
Age, y
Male
Female
Total
Male
Female
Total
<1
1
2
3
4
5
849
759
1126
889
680
622
763
485
449
349
307
342
1612
1244
1575
1238
987
964
44 (39, 49)
40 (35, 44)
58 (53, 63)
45 (40, 50)
34 (29, 38)
30 (26, 34)
41 (37, 46)
26 (22, 30)
24 (21, 28)
19 (15, 22)
16 (13, 19)
17 (14, 20)
43 (39, 46)
33 (30, 36)
42 (38, 45)
32 (29, 35)
25 (23, 27)
24 (22, 26)
Note. CI = confidence interval.
Source. Kids’ Inpatient Database, 1997.6
age 2 to 3 years. By contrast, femur fractures
in children younger than 1 year of age are
thought to be less common and, when they
occur, to be highly suggestive of abuse.7,8 We
found that femur fractures were as common
in children younger than 1 year as in those
aged 2 years and older, with the greatest
number of fractures occurring during the
third month of life. There are few plausible
explanations for a femur fracture in this age
group other than intentional injury. These
data suggest that an infant has as great a
chance of sustaining a femur fracture from
physical abuse as an older child does from all
causes.
The reason for the rise in incidence at age
2 to 3 years, and the subsequent fall, is less
clear. Although most children are walking by
age 15 months, femur fractures were infrequent at this age. The 2- to 3-year-old may
be at increased risk of injury owing to
changes in gait,9 increased mobility, greater
climbing ability, and exposure to vehicular
traffic. The decline in femur fractures after
age 3 may be due to improvements in gait
and judgment, as well as to increased bone
strength. Although child abuse is thought to
be a less common cause for femur fracture in
children who are walking,10 there are widely
varying estimates of its occurrence, reflecting
April 2004, Vol 94, No. 4 | American Journal of Public Health
the difficulty of establishing the diagnosis of
abuse with certainty.11
Our study, based on an administrative
database, lacks the clinical detail of a case series. The sample size is large, however, and
the coding of femur fractures and age are
likely to be accurate.12 The rate of femur fracture in children younger than 2 years of age
was 38.0 per 100 000; this is greater than
the rate of 25.5 per 100 000 reported by
Hinton and colleagues for femoral shaft fractures in this age group in Maryland.5 We included fractures of the proximal and distal
femur, which may contribute to the higher
rate we report.
We cannot determine how often fractures
were due to abuse or neglect, but child abuse
is thought to be common in children younger
than 1 year old with femur fractures.7,8 Other
possible causes include heritable disorders of
connective tissue such as osteogenesis imperfecta13 and motor vehicle accidents. Short
falls, as occur when a child rolls off a bed or
table, are unlikely to cause a femur fracture
in an infant.14,15 The equal number of boys
and girls younger than age 1, and the predominance of boys among those older than 1
year, may signify a shift from intentional to
accidental injury.
Although not as specific for abuse as the
metaphyseal corner fracture or rib fracture, a
single long-bone fracture may be the most
common type of fracture due to abuse.16
Abuse should be suspected if caretakers provide inconsistent or implausible accounts of
how a femur fracture occurred, or if there are
additional unexplained injuries. A skeletal
survey may provide evidence of occult injuries and may support a diagnosis of abuse.
Efforts to prevent femur fractures in children
should focus on preventing physical abuse
in infants and accidental injury in the 2- and
3-year-old children at greatest risk.
About the Authors
At the time of this study, Desmond Brown was with the
Center for the Evaluative Clinical Sciences, Dartmouth
Medical School, Hanover, NH. Elliott Fisher is with the
Department of Community and Family Medicine, Dartmouth Medical School, Hanover.
Requests for reprints should be sent to Desmond Brown,
MD, Boston University School of Medicine, Department of
Orthopaedic Surgery, 850 Harrison Ave, Boston, MA
02118 (e-mail: [email protected]).
This brief was accepted June 30, 2003.
Brown and Fisher | Peer Reviewed | Research and Practice | 559
 RESEARCH AND PRACTICE 
Contributors
D. Brown conceived the study, performed the analyses,
and wrote the brief. E. Fisher assisted in the design of the
study and statistical analyses and contributed to the design of the tables and the writing of the brief.
Human Participant Protection
Institutional review board approval was not required
for this study.
References
1. Wilkins KE. The incidence of fractures in children. In: Rockwood CA, Wilkins KE, Beaty JH, eds.
Fractures in Children. 4th ed. Philadelphia, Pa: LippincottRaven; 1996:3–17.
2. Hedlund R, Lindgren U. The incidence of femoral
shaft fractures in children and adolescents. J Pediatr Orthop. 1986;6:47–50.
3. Nafei A, Teichert G, Mikkelsen SS, Hvid I.
Femoral shaft fractures in children: an epidemiological
study in a Danish urban population, 1977–86. J Pediatr Orthop. 1992;12:499–502.
4. Landin LA. Fracture patterns in children. Analysis
of 8,682 fractures with special reference to incidence,
etiology and secular changes in a Swedish urban population 1950–1979. Acta Orthop Scand Suppl. 1983;
202:1–109.
5. Hinton RY, Lincoln A, Crockett MM, Sponseller P,
Smith G. Fractures of the femoral shaft in children. Incidence, mechanisms, and sociodemographic risk factors. J Bone Joint Surg. 1999;81:500–509.
6. Agency for Healthcare Research and Quality.
1997 Kids’ Inpatient Database. Available at: http://
www.ahrq.gov/data/hcup/hcupkid.htm. Accessed
March 2, 2004.
Asthma, Wheezing, and
Allergies in Russian
Schoolchildren in Relation
to New Surface Materials
in the Home
| Jouni
J. K. Jaakkola, MD, DSc, PhD,
Helen Parise, PhD, Victor Kislitsin, MSc,
Natalia I. Lebedeva, MD, DSc, and John D.
Spengler, PhD
In a cross-sectional study of 5951
Russian 8–12-year-old schoolchildren,
risks of current asthma, wheezing, and
allergy were related to recent renovation and the installation of materials
with potential chemical emissions. New
linoleum flooring, synthetic carpeting,
particleboard, wall coverings, and furniture and recent painting were determinants of 1 or several of these 3
health outcomes. These findings warrant further attention to the type of materials used in interior design. (Am J
Public Health. 2004;94:560–562)
7. Kocher MS, Kasser JR. Orthopaedic aspects of
child abuse. J Am Acad Orthop Surg. 2000;8:10–20.
8. Nimkin K, Kleinman PK. Imaging of child abuse.
Radiol Clin North Am. 2001;39:843–864.
9. Sutherland DH, Olshen R, Cooper L, Woo SL.
The development of mature gait. J Bone Joint Surg Am.
1980;62:336–353.
10. Schwend RM, Werth C, Johnston A. Femur shaft
fractures in toddlers and young children: rarely from
child abuse. J Pediatr Orthop. 2000;20:475–481.
11. Blakemore LC, Loder RT, Hensinger RN. Role of
intentional abuse in children 1 to 5 years old with isolated femoral shaft fractures. J Pediatr Orthop. 1996;16:
585–588.
12. Fisher ES, Baron JA, Malenka DJ, Barrett J,
Bubolz TA. Overcoming potential pitfalls in the use of
Medicare data for epidemiologic research. Am J Public
Health. 1990;80:1487–1490.
13. Ablin DS, Sane SM. Non-accidental injury: confusion with temporary brittle bone disease and mild osteogenesis imperfecta. Pediatr Radiol. 1997;27:
111–113.
14. Tarantino CA, Dowd MD, Murdock TC. Short vertical falls in infants. Pediatr Emerg Care. 1999;15:5–8.
15. Nimityongskul P, Anderson LD. The likelihood of
injuries when children fall out of bed. J Pediatr Orthop.
1987;7:184–186.
16. King J, Diefendorf D, Apthorp J, Negrete VF, Carlson M. Analysis of 429 fractures in 189 battered children. J Pediatr Orthop. 1988;8:585–589.
The Soviet era has been followed by increased activity in construction and renovation of housing in the Russian Federation, as
well as an introduction of new building technology and new materials used in interior design, furniture, and textiles. Two recent studies indicated that exposure to plastic flooring
and wall materials may increase the risk of
respiratory conditions in children.1,2 As part
of a cross-sectional study of air pollution and
respiratory health in Russia in 1996 to
1997,3,4 we tested a hypothesis that the risks
of children’s asthma and allergic diseases are
related to recent renovation, especially newly
installed synthetic surface materials, furniture,
and painting.
METHODS
The study population included 5951 children in second to fifth grade (8–12 years old)
in 8 Russian cities in the Sverdlovsk Oblast re-
560 | Research and Practice | Peer Reviewed | Jaakkola et al.
gion and the city of Cherepovets in the Upper
Volga Oblast.3 The participation rate in schools
varied from 96% to 98%. The questionnaire,
modified from previous European and North
American questionnaires for the Russian conditions,5,6 inquired about the child’s personal
characteristics, health information, and socioeconomic factors. Local elementary schoolteachers were trained to conduct the interviews, and parents and guardians were invited
to meetings after the school day was finished.
After signing an informed consent form, a parent completed the questionnaire.
The current study focused on asthma,
wheezing, and allergy. Current asthma was defined as a history of doctor-diagnosed asthma
and symptoms, signs, or medication of asthma
during the past 12 months. Current wheezing
was defined as wheezing during the past 12
months. Any allergy was defined as any history of doctor-diagnosed allergy or parentalreported hay fever or pollinosis.
Exposure assessment was based on the following question: “Have you conducted any of
the following renovations in your home within
the past 12 months or earlier?” The choices
were installation of linoleum floor, painting,
particleboard, new furniture, synthetic carpet,
wall covering, and suspended ceiling.
We used the odds ratio (OR) as a measure
of effect and logistic regression analysis to adjust for age, gender, preterm birth, low birthweight, parental atopy, maternal smoking during pregnancy, exposure to environmental
tobacco smoke at home (at ages 0–1 years,
ages 2–6 years, and currently), and mother’s
and father’s education.
RESULTS
Of the children, 1.5% had current asthma,
13.4% had current wheezing, and 33.2% had
an allergy. Table 1 shows the occurrence of
the potential sources of emissions.
The risks of current wheezing (adjusted
OR = 1.36; 95% confidence interval [CI] =
1.00, 1.86) and allergy (adjusted OR = 1.31;
95% CI = 1.05, 1.65) were significantly related
to the installation of linoleum flooring during
the past 12 months (Table 2). The corresponding risk estimates were slightly lower when
focusing on exposure earlier than 12 months
ago. There was a general pattern of positive as-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Contributors
D. Brown conceived the study, performed the analyses,
and wrote the brief. E. Fisher assisted in the design of the
study and statistical analyses and contributed to the design of the tables and the writing of the brief.
Human Participant Protection
Institutional review board approval was not required
for this study.
References
1. Wilkins KE. The incidence of fractures in children. In: Rockwood CA, Wilkins KE, Beaty JH, eds.
Fractures in Children. 4th ed. Philadelphia, Pa: LippincottRaven; 1996:3–17.
2. Hedlund R, Lindgren U. The incidence of femoral
shaft fractures in children and adolescents. J Pediatr Orthop. 1986;6:47–50.
3. Nafei A, Teichert G, Mikkelsen SS, Hvid I.
Femoral shaft fractures in children: an epidemiological
study in a Danish urban population, 1977–86. J Pediatr Orthop. 1992;12:499–502.
4. Landin LA. Fracture patterns in children. Analysis
of 8,682 fractures with special reference to incidence,
etiology and secular changes in a Swedish urban population 1950–1979. Acta Orthop Scand Suppl. 1983;
202:1–109.
5. Hinton RY, Lincoln A, Crockett MM, Sponseller P,
Smith G. Fractures of the femoral shaft in children. Incidence, mechanisms, and sociodemographic risk factors. J Bone Joint Surg. 1999;81:500–509.
6. Agency for Healthcare Research and Quality.
1997 Kids’ Inpatient Database. Available at: http://
www.ahrq.gov/data/hcup/hcupkid.htm. Accessed
March 2, 2004.
Asthma, Wheezing, and
Allergies in Russian
Schoolchildren in Relation
to New Surface Materials
in the Home
| Jouni
J. K. Jaakkola, MD, DSc, PhD,
Helen Parise, PhD, Victor Kislitsin, MSc,
Natalia I. Lebedeva, MD, DSc, and John D.
Spengler, PhD
In a cross-sectional study of 5951
Russian 8–12-year-old schoolchildren,
risks of current asthma, wheezing, and
allergy were related to recent renovation and the installation of materials
with potential chemical emissions. New
linoleum flooring, synthetic carpeting,
particleboard, wall coverings, and furniture and recent painting were determinants of 1 or several of these 3
health outcomes. These findings warrant further attention to the type of materials used in interior design. (Am J
Public Health. 2004;94:560–562)
7. Kocher MS, Kasser JR. Orthopaedic aspects of
child abuse. J Am Acad Orthop Surg. 2000;8:10–20.
8. Nimkin K, Kleinman PK. Imaging of child abuse.
Radiol Clin North Am. 2001;39:843–864.
9. Sutherland DH, Olshen R, Cooper L, Woo SL.
The development of mature gait. J Bone Joint Surg Am.
1980;62:336–353.
10. Schwend RM, Werth C, Johnston A. Femur shaft
fractures in toddlers and young children: rarely from
child abuse. J Pediatr Orthop. 2000;20:475–481.
11. Blakemore LC, Loder RT, Hensinger RN. Role of
intentional abuse in children 1 to 5 years old with isolated femoral shaft fractures. J Pediatr Orthop. 1996;16:
585–588.
12. Fisher ES, Baron JA, Malenka DJ, Barrett J,
Bubolz TA. Overcoming potential pitfalls in the use of
Medicare data for epidemiologic research. Am J Public
Health. 1990;80:1487–1490.
13. Ablin DS, Sane SM. Non-accidental injury: confusion with temporary brittle bone disease and mild osteogenesis imperfecta. Pediatr Radiol. 1997;27:
111–113.
14. Tarantino CA, Dowd MD, Murdock TC. Short vertical falls in infants. Pediatr Emerg Care. 1999;15:5–8.
15. Nimityongskul P, Anderson LD. The likelihood of
injuries when children fall out of bed. J Pediatr Orthop.
1987;7:184–186.
16. King J, Diefendorf D, Apthorp J, Negrete VF, Carlson M. Analysis of 429 fractures in 189 battered children. J Pediatr Orthop. 1988;8:585–589.
The Soviet era has been followed by increased activity in construction and renovation of housing in the Russian Federation, as
well as an introduction of new building technology and new materials used in interior design, furniture, and textiles. Two recent studies indicated that exposure to plastic flooring
and wall materials may increase the risk of
respiratory conditions in children.1,2 As part
of a cross-sectional study of air pollution and
respiratory health in Russia in 1996 to
1997,3,4 we tested a hypothesis that the risks
of children’s asthma and allergic diseases are
related to recent renovation, especially newly
installed synthetic surface materials, furniture,
and painting.
METHODS
The study population included 5951 children in second to fifth grade (8–12 years old)
in 8 Russian cities in the Sverdlovsk Oblast re-
560 | Research and Practice | Peer Reviewed | Jaakkola et al.
gion and the city of Cherepovets in the Upper
Volga Oblast.3 The participation rate in schools
varied from 96% to 98%. The questionnaire,
modified from previous European and North
American questionnaires for the Russian conditions,5,6 inquired about the child’s personal
characteristics, health information, and socioeconomic factors. Local elementary schoolteachers were trained to conduct the interviews, and parents and guardians were invited
to meetings after the school day was finished.
After signing an informed consent form, a parent completed the questionnaire.
The current study focused on asthma,
wheezing, and allergy. Current asthma was defined as a history of doctor-diagnosed asthma
and symptoms, signs, or medication of asthma
during the past 12 months. Current wheezing
was defined as wheezing during the past 12
months. Any allergy was defined as any history of doctor-diagnosed allergy or parentalreported hay fever or pollinosis.
Exposure assessment was based on the following question: “Have you conducted any of
the following renovations in your home within
the past 12 months or earlier?” The choices
were installation of linoleum floor, painting,
particleboard, new furniture, synthetic carpet,
wall covering, and suspended ceiling.
We used the odds ratio (OR) as a measure
of effect and logistic regression analysis to adjust for age, gender, preterm birth, low birthweight, parental atopy, maternal smoking during pregnancy, exposure to environmental
tobacco smoke at home (at ages 0–1 years,
ages 2–6 years, and currently), and mother’s
and father’s education.
RESULTS
Of the children, 1.5% had current asthma,
13.4% had current wheezing, and 33.2% had
an allergy. Table 1 shows the occurrence of
the potential sources of emissions.
The risks of current wheezing (adjusted
OR = 1.36; 95% confidence interval [CI] =
1.00, 1.86) and allergy (adjusted OR = 1.31;
95% CI = 1.05, 1.65) were significantly related
to the installation of linoleum flooring during
the past 12 months (Table 2). The corresponding risk estimates were slightly lower when
focusing on exposure earlier than 12 months
ago. There was a general pattern of positive as-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 1—New Surface Materials,
Furniture, and Recent Painting in
Russian Homes
Emission Source
Past 12 Mo, %
Earlier, %
New linoleum flooring
New synthetic carpet
New wall covering
Recent painting
New particleboard
New furniture
9.9
6.5
35.9
32.9
4.7
12.9
34.0
22.0
38.2
39.9
20.7
39.3
sociation between installation of synthetic carpet during the past 12 months and the 3 outcomes (adjusted ORs from 1.39 to 1.84), although for asthma, the association was not
statistically significant. The effect estimates for
the past 12 months were greater than those for
earlier installation. The adjusted odds ratios for
new wall covering during the past 12 months
(from 1.20 to 1.25) and earlier (from 1.12 to
1.22) were lower. The odds ratios for recent
painting were elevated for current wheezing
and allergy. The odds ratios for new particleboard were substantially elevated for all the
studied relations except for recent installation
of particleboard and the risk of current
asthma. The adjusted odds ratios for current
asthma (1.33; 95% CI = 0.57, 3.06), current
wheezing (1.32; 95% CI = 0.99, 1.77), and
any allergy (1.43; 95% CI = 1.16, 1.75) were
increased in relation to new furniture during
the past 12 months but weaker in relation to
new furniture installed earlier.
with an unknown proportion of polyvinyl
chlorides; therefore, the exposure parameter
was rather nonspecific.
Substantial evidence indicates that in the
working-age population (13–65 y), painters
have an increased risk for developing asthma
and asthma-related and other respiratory
symptoms.7–9 Paints used in home renovation
are likely to emit similar chemical substances
as the paints used by professional painters, although exposure levels in occupational settings are much higher than in the home environments after renovation. In the homes, the
exposure levels are the highest during and
shortly after painting, but low levels of exposure may remain for several months. Wooden
furniture and also painted or varnished and
DISCUSSION
Consistent with our hypothesis, the risks of
current asthma and wheezing and allergic diseases were related to installation of materials
with potential chemical emissions.
Two previous studies provided evidence
of the role of polyvinyl chlorides and other
plastic surface materials.1,2 We asked about
installation of linoleum flooring to identify
polyvinyl chloride materials. In line with the
Norwegian study, the risks of asthma and
wheezing in the current study were related to
installation of linoleum floors. Linoleum in
colloquial Russian represents a large heterogeneous group of synthetic floor materials
April 2004, Vol 94, No. 4 | American Journal of Public Health
new furniture are likely to emit chemical substances. Also, synthetic carpets, furniture,
painting, and wall covering used as exposure
indicators constitute a heterogeneous group
of potential emitting materials.
The current findings warrant further attention to the type of materials used in interior
design.
About the Authors
Jouni J. K. Jaakkola is with the Institute of Occupational
Health, The University of Birmingham, Edgbaston, Birmingham, UK, and the Department of Public Health, University of Helsinki, Helsinki, Finland. Helen Parise is with
the Department of Mathematics and Statistics, Boston
University, Boston, Mass. Victor Kislitsin and Natalia I.
Lebedeva are with the Center for the Preparation and Implementation of International Technical Assistance Projects, Moscow, Russian Federation. John D. Spengler is
TABLE 2—Adjusted Odds Ratios (ORs) and 95% Confidence Intervals (CIs) for Current
Asthma, Current Wheezing, and Presence of Any Allergy According to Recent Installation of
Surface Materials and Furniture
Current Asthma
Current Wheezing
a
New linoleum flooring
No
Yes, past 12 mo
Yes, earlier
New synthetic carpet
No
Yes, past 12 mo
Yes, earlier
New wall covering
No
Yes, past 12 mo
Yes, earlier
Recent painting
No
Yes, past 12 mo
Yes, earlier
New particleboard
No
Yes, past 12 mo
Yes, earlier
New furniture
No
Yes, past 12 mo
Yes, earlier
Any Allergy
a
Crude OR
Adjusted OR
(95% CI)
Crude OR
Adjusted OR
(95% CI)
Crude OR
Adjusted ORa
(95% CI)
1.00
1.44
1.64
1.00
1.13 (0.44, 2.04)
1.39 (0.69, 2.77)
1.00
1.36
1.31
1.00
1.36 (1.00, 1.86)
1.25 (0.99, 1.59)
1.00
1.47
1.41
1.00
1.31 (1.05, 1.65)
1.22 (1.04, 1.45)
1.00
2.70
1.60
1.00
1.84 (0.73, 4.65)
1.26 (0.58, 2.72)
1.00
1.81
1.29
1.00
1.70 (1.21, 2.40)
1.24 (0.96, 1.61)
1.00
1.56
1.44
1.00
1.39 (1.07, 1.80)
1.22 (1.02, 1.46)
1.00
1.60
1.61
1.00
1.25 (0.63, 2.51)
1.22 (0.62, 2.43)
1.00
1.28
1.19
1.00
1.20 (0.95, 1.52)
1.12 (0.88, 1.41)
1.00
1.40
1.32
1.00
1.25 (1.06, 1.48)
1.16 (0.99, 1.37
1.00
1.45
1.58
1.00
1.09 (0.53, 2.22)
1.29 (0.65, 2.53)
1.00
1.34
1.19
1.00
1.25 (0.99, 1.58)
1.11 (0.88, 1.40)
1.00
1.40
1.29
1.00
1.25 (1.05, 1.47)
1.16 (0.99, 1.37)
1.00
1.10
1.78
1.00
0.60 (0.13, 2.77)
1.38 (0.65, 2.94)
1.00
1.44
1.50
1.00
1.33 (0.89, 2.00)
1.39 (1.07, 1.80)
1.00
1.72
1.48
1.00
1.49 (1.12, 2.00)
1.28 (1.07, 1.54)
1.00
1.65
1.57
1.00
1.33 (0.57, 3.06)
1.27 (0.64, 2.51)
1.00
1.32
1.22
1.00
1.32 (0.99, 1.77)
1.16 (0.92, 1.47)
1.00
1.59
1.39
1.00
1.43 (1.16, 1.75)
1.24 (1.05, 1.46)
a
From logistic regression, adjusted for age, gender, preterm birth, low birthweight, parental atopy, maternal smoking during
pregnancy, exposure to environmental tobacco smoke at home (at ages 0–1 years, ages 2–6 years, and currently), and
mother’s and father’s education.
Jaakkola et al. | Peer Reviewed | Research and Practice | 561
 RESEARCH AND PRACTICE 
with the Environmental Science and Engineering Program, Harvard School of Public Health, Boston, Mass.
Requests for reprints should be sent to Jouni J.K. Jaakkola,
MD, DSc, PhD, Institute of Occupational Health, The University of Birmingham, Edgbaston, Birmingham B15 2TT,
United Kingdom (e-mail: [email protected]).
This brief was accepted April 8, 2003.
Contributors
J. J. K. Jaakkola conceived the hypothesis, participated
in the planning of the study and statistical analyses and
in the interpretation of the results, and wrote the brief.
H. Parise conducted the statistical analyses and contributed to the interpretation of the results. V. Kislitsin
and N. I. Lebedeva participated in the planning of the
study and contributed to the writing of the brief. J. D.
Spengler participated in planning of the study and statistical analyses and in the interpretation of the results
and provided input on the writing of the brief.
Acknowledgments
This study was supported by a World Bank loan to the
Russian Federation and administered under the Environmental Epidemiology Component, the Center for
Preparation and Implementation of International Technical Assistance Projects.
We are indebted to our colleagues in the Ural Region Environmental Epidemiology Center who, under
the direction of Dr Sergey Kuzmin, conducted a successful comprehensive study of air pollution and children’s health in 9 Russian cities.
symptoms, bronchial hyperresponsiveness, and lung
function. Int Arch Occup Environ Health. 1994;66:
261–267.
8. Mastrangelo G, Paruzzolo P, Mapp C. Asthma due
to isocyanates: a mail survey in a 1% sample of furniture workers in the Vento region, Italy. Med Lav. 1995;
86:503–510.
9. Ucgun I, Ozdemir N, Metintas S, Erginel S, Kolsuz
M. Prevalence of occupational asthma among automobile and furniture painters in the center of Eskisehir
(Turkey): the effects of atopy and smoking habits on
occupational asthma. Allergy. 1998;53:1096–1100.
The Impact of the SARS
Epidemic on the
Utilization of Medical
Services: SARS and the
Fear of SARS
METHODS
| Hong-Jen Chang, MD, MPH, Nicole Huang, MPH,
Cheng-Hua Lee, MD, DrPH, Yea-Jen Hsu, MS,
Chi-Jeng Hsieh, MS, Yiing-Jenq Chou, MD, PhD
Human Participant Protection
Parents were informed that participation was voluntary.
Questionnaires were completed by parents and returned in sealed envelopes. No personal identifiers were
used in our data files and all questionnaires have been
destroyed. Data were not collected from children.
Using interrupted time-series analysis and National Health Insurance data
between January 2000 and August
2003, this study assessed the impacts of the severe acute respiratory
syndrome (SARS) epidemic on medical
service utilization in Taiwan. At the
peak of the SARS epidemic, significant
reductions in ambulatory care (23.9%),
inpatient care (35.2%), and dental care
(16.7%) were observed. People’s fears
of SARS appear to have had strong impacts on access to care. Adverse
health outcomes resulting from accessibility barriers posed by the fear
of SARS should not be overlooked. (Am
J Public Health. 2004;94:562–564)
References
1. Jaakkola JJK, Øie L, Nafstad P, Botten G,
Samuelsen SO, Magnus P. Interior surface materials in
the home and the development of bronchial obstruction in young children in Oslo, Norway. Am J Public
Health. 1999;89:188–192.
2. Jaakkola JJK, Verkasalo PA, Jaakkola N. Plastic wall
materials in the home and respiratory health in young
children. Am J Public Health. 2000;90:797–799.
3. Spengler JD, Jaakkola JJK, Parise H, et al. Housing
characteristics and children’s respiratory health in the
Russian Federation. Am J Public Health. 2004;94:
657–662.
4. Jaakkola JJK, Cherniack M, Spengler JD, et al. Use
of health information systems in the Russian Federation in the assessment of environmental health effects.
Environ Health Perspect. 2000;108:589–594.
5. Ferris BG. Epidemiology Standardization Project
(American Thoracic Society). Am Rev Respir Dis. 1978;
118:1–120.
6. Jaakkola JJK, Jaakkola N, Ruotsalainen R. Home
dampness and molds as determinants of respiratory
symptoms and asthma in pre-school children. J Expo
Anal Environ Epidemiol. 1993;3(suppl 1):129–142.
7. Wieslander G, Janson C, Norback D, Bjornsson E,
Staleheim G, Edling C. Occupational exposure to waterbased paints and self-reported asthma, lower airway
Since the outbreak of severe acute respiratory syndrome (SARS), its etiology, transmission routes, treatments, and outcomes have received much research attention.1–5 SARS has
low mortality and morbidity; however, the
health consequences of the SARS epidemic are
not limited to people who have been infected.6
The potentially serious impact of SARS on
people’s accessibility to medical services
562 | Research and Practice | Peer Reviewed | Chang et al.
should not be overlooked.7–10 However, no
study has systematically evaluated the impact
of the fear of SARS on the general population.
People’s fears of SARS were mainly caused
by its novel, rapid nosocomial transmission,
and the vulnerability of hospitals and health
care workers. Many wondered whether the
fears of SARS among patients and health care
workers alike deterred people from seeking
care or providers from offering services.
Therefore, a critical challenge is to determine
how public health agencies should respond to
utilization changes and possible accessibility
barriers to the general population created by
the SARS epidemic. In this study, we aimed
to assess how people’s fears of SARS influenced their utilization patterns of medical services in Taiwan.
The SARS epidemic in Taiwan started in
mid-March 2003 and lasted for almost 4
months. The epidemic was effectively contained during the initial SARS period (March
14 to April 21, 2003).11 However, multiple
clusters of hospital outbreaks among patients
and health care workers initially struck at the
end of April and extended to May and June,
dramatically exacerbating the epidemic. As a
result, overwhelming fears of SARS spread
over the entire island along with the SARS
epidemic. The situation persisted until July 5,
when Taiwan was officially removed from the
World Health Organization’s list of SARSaffected countries.11,12
We retrieved all claims made to the National Health Insurance program between January 1, 2000, and August 31, 2003, including inpatient care, Western medicine
ambulatory care, Chinese medicine services,
and dental services. An interrupted timeseries design was used. Trends for different
types of services were analyzed separately to
determine whether utilization changes were
specific to certain services. The time-series
autoregressive-moving average (ARIMA) analysis13 was applied to determine whether the
SARS epidemic was significantly associated
with changes in medical service utilization
rates. Relative differences between observed
and ARIMA-predicted values were expressed
in percentages. All analyses were performed
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
with the Environmental Science and Engineering Program, Harvard School of Public Health, Boston, Mass.
Requests for reprints should be sent to Jouni J.K. Jaakkola,
MD, DSc, PhD, Institute of Occupational Health, The University of Birmingham, Edgbaston, Birmingham B15 2TT,
United Kingdom (e-mail: [email protected]).
This brief was accepted April 8, 2003.
Contributors
J. J. K. Jaakkola conceived the hypothesis, participated
in the planning of the study and statistical analyses and
in the interpretation of the results, and wrote the brief.
H. Parise conducted the statistical analyses and contributed to the interpretation of the results. V. Kislitsin
and N. I. Lebedeva participated in the planning of the
study and contributed to the writing of the brief. J. D.
Spengler participated in planning of the study and statistical analyses and in the interpretation of the results
and provided input on the writing of the brief.
Acknowledgments
This study was supported by a World Bank loan to the
Russian Federation and administered under the Environmental Epidemiology Component, the Center for
Preparation and Implementation of International Technical Assistance Projects.
We are indebted to our colleagues in the Ural Region Environmental Epidemiology Center who, under
the direction of Dr Sergey Kuzmin, conducted a successful comprehensive study of air pollution and children’s health in 9 Russian cities.
symptoms, bronchial hyperresponsiveness, and lung
function. Int Arch Occup Environ Health. 1994;66:
261–267.
8. Mastrangelo G, Paruzzolo P, Mapp C. Asthma due
to isocyanates: a mail survey in a 1% sample of furniture workers in the Vento region, Italy. Med Lav. 1995;
86:503–510.
9. Ucgun I, Ozdemir N, Metintas S, Erginel S, Kolsuz
M. Prevalence of occupational asthma among automobile and furniture painters in the center of Eskisehir
(Turkey): the effects of atopy and smoking habits on
occupational asthma. Allergy. 1998;53:1096–1100.
The Impact of the SARS
Epidemic on the
Utilization of Medical
Services: SARS and the
Fear of SARS
METHODS
| Hong-Jen Chang, MD, MPH, Nicole Huang, MPH,
Cheng-Hua Lee, MD, DrPH, Yea-Jen Hsu, MS,
Chi-Jeng Hsieh, MS, Yiing-Jenq Chou, MD, PhD
Human Participant Protection
Parents were informed that participation was voluntary.
Questionnaires were completed by parents and returned in sealed envelopes. No personal identifiers were
used in our data files and all questionnaires have been
destroyed. Data were not collected from children.
Using interrupted time-series analysis and National Health Insurance data
between January 2000 and August
2003, this study assessed the impacts of the severe acute respiratory
syndrome (SARS) epidemic on medical
service utilization in Taiwan. At the
peak of the SARS epidemic, significant
reductions in ambulatory care (23.9%),
inpatient care (35.2%), and dental care
(16.7%) were observed. People’s fears
of SARS appear to have had strong impacts on access to care. Adverse
health outcomes resulting from accessibility barriers posed by the fear
of SARS should not be overlooked. (Am
J Public Health. 2004;94:562–564)
References
1. Jaakkola JJK, Øie L, Nafstad P, Botten G,
Samuelsen SO, Magnus P. Interior surface materials in
the home and the development of bronchial obstruction in young children in Oslo, Norway. Am J Public
Health. 1999;89:188–192.
2. Jaakkola JJK, Verkasalo PA, Jaakkola N. Plastic wall
materials in the home and respiratory health in young
children. Am J Public Health. 2000;90:797–799.
3. Spengler JD, Jaakkola JJK, Parise H, et al. Housing
characteristics and children’s respiratory health in the
Russian Federation. Am J Public Health. 2004;94:
657–662.
4. Jaakkola JJK, Cherniack M, Spengler JD, et al. Use
of health information systems in the Russian Federation in the assessment of environmental health effects.
Environ Health Perspect. 2000;108:589–594.
5. Ferris BG. Epidemiology Standardization Project
(American Thoracic Society). Am Rev Respir Dis. 1978;
118:1–120.
6. Jaakkola JJK, Jaakkola N, Ruotsalainen R. Home
dampness and molds as determinants of respiratory
symptoms and asthma in pre-school children. J Expo
Anal Environ Epidemiol. 1993;3(suppl 1):129–142.
7. Wieslander G, Janson C, Norback D, Bjornsson E,
Staleheim G, Edling C. Occupational exposure to waterbased paints and self-reported asthma, lower airway
Since the outbreak of severe acute respiratory syndrome (SARS), its etiology, transmission routes, treatments, and outcomes have received much research attention.1–5 SARS has
low mortality and morbidity; however, the
health consequences of the SARS epidemic are
not limited to people who have been infected.6
The potentially serious impact of SARS on
people’s accessibility to medical services
562 | Research and Practice | Peer Reviewed | Chang et al.
should not be overlooked.7–10 However, no
study has systematically evaluated the impact
of the fear of SARS on the general population.
People’s fears of SARS were mainly caused
by its novel, rapid nosocomial transmission,
and the vulnerability of hospitals and health
care workers. Many wondered whether the
fears of SARS among patients and health care
workers alike deterred people from seeking
care or providers from offering services.
Therefore, a critical challenge is to determine
how public health agencies should respond to
utilization changes and possible accessibility
barriers to the general population created by
the SARS epidemic. In this study, we aimed
to assess how people’s fears of SARS influenced their utilization patterns of medical services in Taiwan.
The SARS epidemic in Taiwan started in
mid-March 2003 and lasted for almost 4
months. The epidemic was effectively contained during the initial SARS period (March
14 to April 21, 2003).11 However, multiple
clusters of hospital outbreaks among patients
and health care workers initially struck at the
end of April and extended to May and June,
dramatically exacerbating the epidemic. As a
result, overwhelming fears of SARS spread
over the entire island along with the SARS
epidemic. The situation persisted until July 5,
when Taiwan was officially removed from the
World Health Organization’s list of SARSaffected countries.11,12
We retrieved all claims made to the National Health Insurance program between January 1, 2000, and August 31, 2003, including inpatient care, Western medicine
ambulatory care, Chinese medicine services,
and dental services. An interrupted timeseries design was used. Trends for different
types of services were analyzed separately to
determine whether utilization changes were
specific to certain services. The time-series
autoregressive-moving average (ARIMA) analysis13 was applied to determine whether the
SARS epidemic was significantly associated
with changes in medical service utilization
rates. Relative differences between observed
and ARIMA-predicted values were expressed
in percentages. All analyses were performed
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
21 000
Expenditure, NT$, millions
18 000
15 000
12 000
9 000
6 000
Ambulatory care ( observed )
Inpatient care ( observed )
Ambulatory care ( predicted )
Inpatient care ( predicted )
3 000
0
JAN
2002
MAR
2002
MAY
2002
JUL
2002
SEP
2002
NOV
2002
JAN
2003
MAR
2003
MAY
2003
JUL
2003
Month and Year
Note. NT$ = New Taiwan dollars.
FIGURE 1—Observed and predicted expenditures for ambulatory and inpatient care in the preepidemic, epidemic, and postepidemic periods,
January 2002 through August 2003. The date of the initial outbreak is marked with a vertical line.
using SAS for Windows, Version 8.2 (SAS Institute Inc, Cary, NC) and Stata 8.0 (Stata
Corp, College Station, Tex).
RESULTS
Figure 1 compares the observed trends in
expenditures for ambulatory and inpatient
care in Taiwan with the predicted trends estimated by the ARIMA model that assumes the
absence of the SARS epidemic. During the
epidemic, the figure shows significant reductions in observed expenditures compared
with those expected. The general patterns for
both ambulatory and inpatient services were
quite similar and corresponded to each transition period of the SARS epidemic. Correspondingly, virtually no impact was observed
before the first hospital cluster in late April,
when the epidemic was effectively contained.
A significant reduction was observed in May
and continued to expand significantly in June,
when the fears of SARS grew after the expansion of the epidemic to all of Taiwan. Finally,
the expenditures increased gradually in July
and August after the SARS epidemic was
over. Compared with ambulatory care, inpa-
tient care experienced larger reductions in expenditure at the peak period and rebounded
to levels closer to usual values toward the end
of the epidemic. This suggests that the SARS
epidemic had a stronger influence on inpatient services than on ambulatory services.
Although the responses of medical service
expenditures were similar to those of medical
service utilization, reductions in utilization
were relatively larger. Inpatient services experienced the largest reduction (35.2%), followed by dental services (23.9%) and Western medicine ambulatory services (16.7%) at
the peak of the SARS epidemic (Table 1). On
the other hand, unlike other types of medical
services, Chinese medicine services experienced an increase in utilization (1.8%) during
the SARS epidemic. One plausible explanation may be that Chinese medicine services
served as a substitute for Western medicine
ambulatory services.
on utilization reductions translated into an approximate $18.8 billion new Taiwan dollars
decrease (approximately 6% of the annual National Health Insurance expenditure) in health
care expenditure during the SARS epidemic
from April 2003 through August 2003. The
results strongly suggest that the fears of SARS
significantly influenced people’s care-seeking
behavior and that this fear seriously compromised their accessibility to quality care.
Although all the international attention is
focused on the direct causes of SARS, serious
health consequences resulting from people’s
fears of SARS should not be overlooked. The
results presented here could provide public
health agencies with a more complete picture
of overall health impacts of the SARS epidemic, so that when SARS re-emerges, it can
guide public health officials to prevent avoidable health consequences because of the fears
people have regarding SARS.
DISCUSSION
About the Authors
Over the study period, we observed significant utilization reductions at the peak of the
SARS epidemic. Overall, this short-term impact
April 2004, Vol 94, No. 4 | American Journal of Public Health
Hong-Je Chang, Cheng-Hua Lee, and Chi-Jeng Hsieh are
with the Bureau of National Health Insurance, Taipei, Taiwan. Nicole Huang is with the Department of Health Policy
and Management, Johns Hopkins Bloomberg School of
Chang et al. | Peer Reviewed | Research and Practice | 563
 RESEARCH AND PRACTICE 
TABLE 1—Observed and Predicted Monthly Medical Expenditures and Utilizations, by Type of
Medical Service: Taiwan, January 2003 through August 2003
Pre-SARS and Initial SARS Period
Jan–Mar, Avg $/No. April, $/No.
Peak SARS Period
Difference, %
May, $/No. Difference, %
Post-SARS Period
June, $/No. Difference, %
July, $/No. Difference, % August, $/No. Difference, %
Expenditures
Inpatient care
Observed
Expected
Ambulatory care
Observed
Expected
Dental care
Observed
Expected
Chinese medicine
Observed
Expected
10 143
10 269
10 845
10 839
0.1
8808
11 278
–21.9
7888
10 729
–26.5
10 228
11 533
–11.3
10 334
11 272
–8.3
17 131
16 977
17 161
17 601
–2.5
15 726
18 427
–14.7
13 922
17 006
–18.1
16 503
18 467
–10.6
16 226
17 623
–7.9
2297
2323
2339
2572
–9.1
1991
2603
–23.5
2083
2488
–16.3
2538
2736
–7.2
2495
2568
–2.8
1343
1362
1458
1446
0.8
1418
1546
–8.3
1331
1351
–1.4
1454
1525
–4.7
1450
1421
2.0
Utilization
Inpatient care
Observed
Expected
Ambulatory care
Observed
Expected
Dental care
Observed
Expected
Chinese medicine
Observed
Expected
241
245
245
262
–6.3
180
266
–32.4
167
258
–35.2
227
272
–16.8
228
271
–15.7
23 117
22 323
22 525
22 245
1.3
18 665
23 979
–22.2
15 744
20 692
–23.9
18 668
22 783
–18.1
19 043
21 418
–11.1
2040
2073
2055
2275
–9.7
1717
2299
–25.3
1833
2201
–16.7
2269
2404
–5.6
2231
2262
–1.4
2454
2430
2654
2519
5.4
2558
2715
–5.8
2359
2319
1.8
2575
2612
–1.4
2606
2431
7.2
Note. Avg = monthly average; $ = new Taiwan dollars in millions; No. = number of visits per admissions in thousands; Difference = [(observed value – predicted value)/predicted value] 100. The
official exchange rate for 2003 published by the Central Bank of China is 1 US$ = 34.24 New Taiwan $. Available at: http://www.cbc.gov.tw. Accessed February 26, 2004.
Public Health, Baltimore, Md. Yea-Jen Hsu and Yiing-Jenq
Chou are with the Department of Social Medicine, School of
Medicine, National Yang Ming University, Taipei, Taiwan.
Requests for reprints should be sent to Hong-Jen Chang,
140, Sec 3, Hsin-Yi Rd, Taipei, Taiwan 106 (e-mail:
[email protected]).
This brief was accepted November 30, 2003.
Contributors
H. J. Chang planned the study and supervised all aspects of its implementation. N. Huang assisted with the
study and led the writing. C. H. Lee synthesized analyses and contributed to the writing of the article. C. J.
Hsieh assisted with the data management and the
study. Y. J. Hsu assisted with the study and analyses.
Y. J. Chou planned the study, completed the statistical
analysis, and supervised the study implementation. All
authors helped to conceptualize ideas, interpret findings, and review drafts of the brief.
Acknowledgments
We thank Roger Haesevoets for editing the brief.
Human Participant Protection
6. Emanuel EJ. The lessons of SARS. Ann Intern
Med. 2003;139:589–591.
No protocol approval was needed for this study.
References
1. Guan Y, Zheng BJ, He YQ, et al. Isolation and
characterization of viruses related to the SARS coronavirus from animals in southern China. Science. 2003;
302:276-278.
2. Cinatl J, Morgenstern B, Bauer G, et al. Treatment
of SARS with human interferons. Lancet. 2003;362:
293–294.
3. Kuiken T, Fouchier RA, Schutten M, et al. Newly
discovered coronavirus as the primary cause of severe
acute respiratory syndrome. Lancet. 2003;362:263–270.
4. Donnelly CA, Ghani AC, Leung GM, et al. Epidemiological determinants of spread of causal agent of
severe acute respiratory syndrome in Hong Kong.
Lancet. 2003;361:1761–1766.
5. Lipsitch M, Cohen T, Cooper B, et al. Transmission dynamics and control of severe acute respiratory
syndrome. Science. 2003;300:1966–1970.
564 | Research and Practice | Peer Reviewed | Chang et al.
7. Haines CJ, Chu YW, Chung TK. The effect of severe acute respiratory syndrome on a hospital obstetrics
and gynaecology service. BJOG. 2003;110:643–645.
8. Clark J. Fear of SARS thwarts medical education
in Toronto. BMJ. 2003;326:784.
9. Yeoh SC, Lee E, Lee BW, et al. Severe acute respiratory syndrome: private hospital in Singapore took
effective control measures. BMJ. 2003;326:1394.
10. Maunder R, Hunter J, Vincent L, et al. The immediate psychological and occupational impact of the
2003 SARS outbreak in a teaching hospital. CMAJ.
2003;168:1245–1251.
11. Centers for Disease Control and Prevention. Severe acute respiratory syndrome—Taiwan, 2003.
JAMA. 2003;289:2930–2932.
12. Chien LC, Yeh WB, Chang HT. Lessons from Taiwan. CMAJ. 2003;169:277.
13. Box GEP, Jenkins GM. Time Series Analysis Forecasting and Control. San Francisco, Calif: Holden-Day; 1976.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Health Effects
Associated With
Recreational Coastal
Water Use: Urban Versus
Rural California
14 000
12 000
Mean Total Coliform Count
North Orange County
| Ryan H. Dwight, PhD, Dean B. Baker, MD,
MPH, Jan C. Semenza, PhD, MPH, and
Betty H. Olson, PhD
10 000
Santa Cruz County
8 000
6 000
4 000
2 000
We compared rates of reported
health symptoms among surfers in
urban North Orange County (NOC) and
rural Santa Cruz County (SCC), California, during 2 winters (1998 and
1999) to determine whether symptoms
were associated with exposure to
urban runoff. NOC participants reported almost twice as many symptoms as SCC participants during the
1998 winter. In both study years, risk
increased across symptom categories
by an average of 10% for each 2.5
hours of weekly water exposure. Our
findings suggest that discharging untreated urban runoff onto public
beaches can pose health risks. (Am J
Public Health. 2004;94:565–567)
Coastal waters along public beaches can
be polluted by urban runoff, which is water
that carries non–point-source pollution via
surface waterways to the ocean.1 A variety of
illnesses have been associated with exposure
to polluted recreational coastal waters.2–4 In
this study, which involved 2 geographic watersheds differing in terms of urbanization,
we measured reported health effects on individuals with high levels of exposure to coastal
waters.
METHODS
North Orange County (NOC), California,
was the “urban” site because its watershed is
1 of the most developed areas in the world
and generates highly polluted runoff waters.5–9 We selected Santa Cruz County (SCC),
California, as the comparison “rural” site be-
0
Jan
Feb
Mar
Apr
Jan
Feb
Mar
Apr
Winter Study Months
FIGURE 1—Mean monthly total coliform counts (per 100 mL) during 1998 El Niño and
1999 La Niña winters: North Orange County and Santz Cruz County coastal waters (data
provided by Orange County Health Care Agency and Santa Cruz Health Agency).
cause of its coastal water quality indicators
(Figure 1) and watershed characteristics.
We conducted 2 cross-sectional surveys of
surfers from NOC and SCC, 1 in April 1998
and 1 in April 1999, and gathered data on
reported health symptoms (e.g., vomiting, diarrhea, sore throat) experienced during the
previous 3 months. The 1998 El Niño winter
had led to record high precipitation throughout California, while the 1999 La Niña winter
had led to record low precipitation in NOC.
NOC had lower total rainfall than SCC in
both years, yet the former had higher coastal
water coliform (a water quality measure of
pollution) levels (Figure 1).
Surfers were selected as the study population because of their regular exposure to
coastal waters. Interviewers at surfing
beaches recruited participants by approaching
all individuals who had wetsuits and surfboards. Those who reported surfing at least
once a week and were 18 years or older were
eligible to be interviewed. Demographic information was collected, as well as information
on symptoms experienced during the previous 3-month period. Participants also reported the amount of time they were exposed
to coastal waters.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Multiple reports of 1 symptom were combined, allowing only 1 symptom report per
participant, equivalent to a 3-month period
prevalence. Logistic regression analysis was
used to estimate adjusted odds ratios (ORs)
comparing symptom reporting rates between
the 2 counties, stratified by year. The final logistic model included the following variables:
county, water exposure, gender, age, occupation, educational level, annual income, political outlook, and level of concern about water
quality. The latter 2 variables were included
to control for potential reporting bias associated with perspectives about the potential
health effects of environmental pollution and
water quality.
RESULTS
In 1998, investigators interviewed 479
participants in NOC and 374 in SCC. In
1999, investigators interviewed 662 participants in NOC and 358 in SCC. At each site,
response rates were above 80% in both
1998 and 1999. The mean age of the participants was 30 years, and 93% were male.
The urban versus rural analysis showed
that NOC participants reported almost twice
Dwight et al. | Peer Reviewed | Research and Practice | 565
 RESEARCH AND PRACTICE 
during both years and used comparable
groups of surfers with relatively similar social
characteristics. The high participation rates
(above 80%) lowered potential bias due to selective participation. The study was cross sectional, which represents a limitation in terms
of assessment of symptoms over a 3-month
period, but any recall bias was likely to be
nondifferential and toward the null. To reduce
potential differential reporting bias, we adjusted for participants’ level of concern about
coastal water quality. Another limitation is
that we did not measure water quality at the
sites, so we were unable to determine the specific nature of the pollutants associated with
symptoms.
In summary, this study suggests that discharging untreated urban runoff onto public
beaches can pose health risks. These potential
health risks warrant greater public health surveillance, as well as greater efforts to reduce
pollutants discharged onto public beaches.
Large-scale prospective investigations are
needed to further characterize the health
risks of people exposed to untreated urban
runoff in coastal waters.
TABLE 1—Odds Ratios for Reported Symptoms: North Orange County and Santa Cruz
County, 1998 and 1999
1998 (El Niño Winter)
Any symptom
SRD
HCGI
Fever
Nausea
Stomach pain
Vomiting
Diarrhea
Sinus problems
Cough
Phlegm
Sore throat
Eye redness
Ear pain
Skin infection
1999 (La Niña Winter)
OR
95% CI
OR
95% CI
1.85
1.29
2.32
1.63
1.18
2.51
2.13
2.10
1.41
1.36
1.33
1.96
2.44
1.36
1.93
1.36, 2.52
0.91, 1.82
1.27, 4.25
1.08, 2.44
0.74, 1.90
1.45, 4.32
0.95, 4.78
1.33, 3.31
1.05, 1.91
0.96, 1.91
0.92, 1.92
1.42, 2.70
1.20, 4.93
0.89, 2.09
1.12, 3.33
1.17
0.75
0.97
0.89
0.89
0.90
0.84
1.06
1.25
1.10
0.52
1.55
1.42
1.55
0.71
0.87, 1.57
0.53, 1.05
0.62, 1.51
0.61, 1.28
0.58, 1.36
0.60, 1.37
0.46, 1.53
0.69, 1.63
0.93, 1.68
0.80, 1.51
0.35, 0.76
1.13, 2.14
0.60, 3.33
0.98, 2.46
0.42, 1.21
Note. Odds ratios (ORs) were adjusted for water exposure, gender, age, occupation, education, income, political outlook, and
level of concern about coastal water quality. CI = confidence interval; SRD = significant respiratory disease (fever and sinus
problems, fever and sore throat, or cough and phlegm); HCGI = highly credible gastrointestinal illness (vomiting, diarrhea and
fever, or stomach pain and fever).
as many symptoms overall as SCC participants (OR = 1.85; 95% confidence interval
[CI] = 1.4, 2.5) during the 1998 El Niño winter (Table 1). In that year, NOC participants
reported higher rates of every symptom.
During the 1999 La Niña winter, NOC participants reported only slightly more symptoms than SCC participants (OR = 1.17; 95%
CI = 0.9, 1.6) and reported slightly higher
frequencies in regard to 6 of the 12 symptoms. Odds ratios decreased consistently
across all symptoms between the 2 winters.
In both study years, risk increased across almost every symptom category by an average
of about 10% (OR = 1.1) for each additional
2.5 hours of water exposure per week.
DISCUSSION
Results from this investigation and other
studies9,10 suggest that discharging untreated
urban runoff onto public beaches can pose
health risks. This conclusion is supported by
the higher reporting rates of symptoms
among urban NOC participants during the
rainy 1998 El Niño winter, after controling
for possible confounding (due to demographic
characteristics) and reporting bias (due to
concern about coastal water quality). The
exposure–response relationship demonstrated
for most of the symptoms further supports
this conclusion. Direct associations have been
reported between pollution levels in runoff
waters and urban land use, population levels,
and amount of impervious surface area in the
watershed.11–15
Research on the health consequences of
urban runoff represents a relatively new area
of investigation, despite decades of urban
runoff contaminating coastal waters.7,12 Most
previous epidemiological studies focused on
waters contaminated with domestic sewage,
and the majority found associations between
water pollution levels and incidence levels of
symptoms.2–4,16–21 Most epidemiological studies of recreational water use have focused on
single exposure events rather than exposure
over time.2,3 Our study assessed 3-month
prevalence rates of symptoms and demonstrated that average symptom prevalence
was associated with different levels of water
pollution.
To reduce the potential for confounding,
we sampled from the same source population
566 | Research and Practice | Peer Reviewed | Dwight et al.
About the Authors
Ryan H. Dwight and Betty H. Olson are with the Environmental Health Science and Policy Program, Department of
Environmental Health, Science, and Policy, University of
California, Irvine. Dean B. Baker is with the Department
of Medicine and Center for Occupational and Environmental Health, University of California, Irvine. Jan C. Semenza is with the School of Community Health, Portland
State University, Portland, Ore.
Requests for reprints should be sent to Dean B. Baker,
MD, MPH, UC Irvine Center for Occupational and Environmental Health, 5201 California Ave, Suite 100, Irvine,
CA 92612 (e-mail: [email protected]).
This brief was accepted April 25, 2003.
Contributors
All authors contributed to the design, writing, and final
approval of the article.
Acknowledgments
We are grateful to Dr Harvey Molotch, former director
of the Ocean and Coastal Policy Center at the University of California, Santa Barbara, and the University of
California Toxic Substances Research and Teaching Program for their early financial support. We also thank Dr
JoAnne Prause, Dr Dele Ogunseitan, the interviewers,
and all of the participants for taking part.
Human Participant Protection
The use of human participants in this study was approved by the institutional review board of the Univer-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
sity of California at Irvine. The participants provided
verbal informed consent.
References
1. Field R, O’Shea M, Brown MP. The detection and
disinfection of pathogens in storm-generated flows.
Water Sci Technol. 1993;28:311–315.
2. Saliba LJ, Helmer R. Health risks associated with
pollution of coastal bathing waters. World Health Stat Q.
1990;43:177–184.
3. Pruss A. Review of epidemiological studies on
health effects from exposure to recreational water. Int J
Epidemiol. 1998;27:1–9.
19. Seyfried PL, Tobin RS, Brown NE, Ness PF. A prospective study of swimming-related illness: I. Swimmingassociated health risk. Am J Public Health. 1985;75:
1068–1070.
20. Corbett SJ, Rubin GL, Curry GK, Kleinbaum DG.
The health effects of swimming at Sydney beaches. Am
J Public Health. 1993;83:1701–1706.
21. Fleisher JM, Kay D, Salmon RL, Jones F, Wyer
MD, Godfree AF. Marine water contaminated with domestic sewage: non-enteric illnesses associated with
bather exposure in the United Kingdom. Am J Public
Health. 1996;86:1228–1234.
4. Cabelli VJ. Swimming-associated illness and recreational water quality criteria. Water Sci Technol. 1989;
21:13–21.
5. Bay SM, Greenstein DJ. Toxicity of dry weather
flow from the Santa Monica Bay watershed. Bull South
Calif Acad Sci. 1996;95:33–45.
6. Gold M, Bartlett M, Dorsey J, McGee C. Storm
Drains as a Source of Surf Zone Bacterial Indicators and
Human Enteric Viruses to Santa Monica Bay. Santa
Monica, Calif: Santa Monica Bay Restoration Project;
1991.
7. Cross J, Schiff K, Schafer H. Surface runoff to the
Southern California Bight. In: Annual Report. Long
Beach, Calif: Southern California Coastal Water Research Project; 1992:19–28.
8. Schafer H, Gossett R. Storm Runoff in Los Angeles
and Ventura Counties. Long Beach, Calif: Southern California Coastal Water Research Project; 1988. Final
report.
9. Dwight RH, Semenza JC, Baker DB, Olson BH.
Association of urban runoff with coastal water quality
in Orange County, California. Water Environment Res.
2002;74:82–90.
10. Haile RW, Witte JS, Gold M, et al. The health effects of swimming in ocean water contaminated by
storm drain runoff. Epidemiology. 1999;10:355–363.
11. Arnold C, Gibbons J. Impervious surface coverage:
the emergence of a key environmental indicator. J Am
Plann Assoc. 1996;62:243–258.
12. California’s Ocean Resources: An Agenda for the Future. Sacramento, Calif: California Resources Agency;
1997.
13. Young KD, Thackston EL. Housing density and
bacterial loading in urban streams. J Environ Eng.
1999;125:1177–1180.
14. Mallin MA, Williams KE, Esham EC, Lowe RP.
Effect of human development on bacteriological water
quality in coastal watersheds. Ecological Applications.
2000;10:1047–1056.
15. Schueler TR. The importance of imperviousness.
Watershed Protection Techniques. 1994;1:100–111.
16. Cabelli VJ, Dufour AP, McCabe LJ, Levin MA.
Swimming-associated gastroenteritis and water quality.
Am J Epidemiol. 1982;115:606–616.
17. Kay D, Fleisher JM, Salmon RL, et al. Predicting
likelihood of gastroenteritis from sea bathing: results
from randomized exposure. Lancet. 1994;344:
905–909.
18. Fleisher JM, Jones F, Kay D, et al. Water and nonwater related risk factors for gastroenteritis among
bathers exposed to sewage-contaminated marine waters. Int J Epidemiol. 1993;22:698–708.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Dwight et al. | Peer Reviewed | Research and Practice | 567
 RESEARCH AND PRACTICE 
Raised Speed Limits, Speed Spillover, Case-Fatality Rates,
and Road Deaths in Israel: A 5-Year Follow-Up
| Elihu D. Richter MD, MPH, Paul Barach MD, MPH, Lee Friedman, BA, Samuel Krikler, MS, Abraham Israeli, MD, MBA
On November 1, 1993, the government of
Israel increased the enforced speed limit for
all vehicles, including trucks, from 90 to 100
kilometers per hour (55.9 to 62.1 mph) on
segments (115 km, or 71.4 miles) of 3 major
interurban highways connecting its 4 major
cities: Tel Aviv, Jerusalem, Haifa, and Beersheba. The government made major improvements on these highways and many other
roads and declared the increased speed limit
a 1-year “experiment.”1 Simultaneously, it
mandated the use of rear seat belts and daytime running lights.
Lower travel speeds and fewer deaths
usually follow lowered speed limits.2,3
Higher travel speeds and more deaths follow
increased speed limits.4–10 Recent data demonstrate a 17% increase in deaths after a
4% increase in speeds on US interstate
highways.11 High-speed driving on highways
induces speed adaptation (a situation in
which vehicle speed is influenced by the
speed and duration of recent travel in the
vehicle) on connecting interurban roads, and
even urban roads. This so-called spillover effect may persist for 5 to 6 years.12–15 Yet
there is still worldwide controversy over the
impact of increased speed limits.16–18 One
view holds that increased speed limits not
only shorten travel time but also are protective when increased vehicle mileage is used
to correct for increases in death tolls.16 The
US Centers for Disease Control and Prevention, despite 41 967 road deaths in 1997,
has not cited higher speed limits as contributing to the high number of deaths.19
The British government, by contrast, is committed to strategies to reduce speeds.20
Israel, with a size of 21 501 km2, provides
an ideal setting for observing the effects of
speed limits. Israel has a fairly modern car
fleet and roads and relatively low drunkdriving rates and is isolated from traffic from
neighboring states.21 All its highways are interurban. The 3 highways on which the en-
Objectives. We assessed the 5-year, nationwide impact on road deaths of the raise in the
speed limit (November 1, 1993) on 3 major interurban highways in Israel from 90 to 100 kph.
Methods. We compared before–after trends in deaths as well as case fatality—an outcome
independent of exposure (defined as vehicle-kilometers traveled).
Results. After the raise, speeds rose by 4.5%–9.1%. Over 5 years, there was a sustained
increase in deaths (15%) and case fatality rates (38%) on all interurban roads. Corresponding
increases in deaths (13%) and case fatality (24%) on urban roads indicated “speed spillover.”
Conclusions. Immediate increases in case fatality predicted and tracked the sustained
increase in deaths from increased speeds of impact. Newtonian fourth power models predicted the effects of “small” increases in speed on large rises in case fatality rates. Countermeasures and congestion reduced the impact on deaths and case-fatality rates by more
than half. (Am J Public Health. 2004;94:568–574)
forced speed limit was raised serve as the
major conduits of Israel’s interurban traffic.
We examined the suddenness, size, distribution, and persistence of nationwide
changes in death and injury tolls after the increase in the speed limit on these highways,
with specific attention to speed spillover and
its nationwide effect on road deaths. We examined the utility of 2 empirically derived
models that demonstrate the relations between speed and fatality risks; the models are
based on Newtonian physics. The first model
demonstrates that case-fatality rates (CFRs)
vary to the fourth power of the velocity at
vehicular impact with both unbelted22 and
belted23 drivers; the second model demonstrates that the number of crashes, injuries,
and deaths varies with the first, second, and
fourth power, respectively, of increases in average traffic speeds.24,25 We also examined
whether an increase in the CFR, a crashphase outcome (those variables which influence survival in the event of a crash, such as
speed of impact, seat belt use, and trauma
care) independent of exposure (billion
vehicle-kilometers of travel [bvkmt]),26 predicted and tracked sustained trends in
increased road death tolls. Finally, we assessed the degree to which protective countermeasures and increased traffic congestion
offset and conceal the full impact of increases
in travel speeds on road deaths.
568 | Research and Practice | Peer Reviewed | Richter et al.
DATA SOURCES AND METHODS
Speed Trends
Data on speed trends on the 3 high-speed
highways (Tel Aviv to Jerusalem [highway 1];
Tel Aviv to Haifa [highway 2]; Tel Aviv to
Beersheba [highway 4]) came from sporadic
roadside daytime monitoring in the years
1971–1994 using roadside radar and laser
cameras.27–30
Road Deaths and Injuries
We collected data on road deaths (up to
30 days after crash injury), serious injuries
(hospitalized more than 24 hours), and light
injuries (not hospitalized, or hospitalized less
than 24 hours), and exposure—as measured
by billion vehicle-kilometers of travel—from
the Central Bureau of Statistics.27,29 We also
used a surrogate measure for the CFR—the
proportion killed among all seriously injured
(hereafter CFRS, for “CFR surrogate”)—to
avoid biases from transient underreporting of
light injuries, which the Central Bureau of
Statistics estimated to be of the order of 10%.
Impacts Within Subgroups
We carried out a 1-year comparison of
deaths and case-fatality rates (CFRs and
CFRSs), before and after the increase to
100 kph on November 1, 1993, for highspeed roads, other interurban roads, and all
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
urban roads, and for major crash types and
driver subgroups.
RESULTS
Speed Trends: 1 Year
Sustained Impact of 100 kph
We analyzed changes in death rates and
CFRs between 3 years before and 5 years
after the increase in the speed limit with the
Student t test and cumulative summing. This
method involves subtracting the differences
between monthly totals for deaths from the
overall mean of the 3-year control period
derived from a baseline of monthly average
death totals and testing these differences
with simple t tests.31 We then estimated the
specific effect of increased speed limits without countermeasures and congestion by
comparing the observed change in the number of deaths per year with that attributable
(K(ATTRIB)) specifically to the change in CFRs.
We made these estimates using the formula
(1)
K(ATTRIB) = {K(B)(CFRS(A)/CFRS(B))} − K(B),
where K(B) represents persons killed per year
3 years before, and CFRS(B) and CFRS(A) are
the proportion of those killed among all
those seriously injured before and after the
speed limit change, respectively.19 Using models developed by Evans,3 we also estimated
the average change in speeds (V) of travel
and crash impact, as well as the change in
number of deaths per each 1% change in
speed, using algebraic fourth-root models in
which
(2)
K(A)/K (B) = V(A)4/V(B)4
where K(A) and K(B) are persons killed per
year 5 years after and before the increase, respectively, and V(A) and V(B) are average
speeds on roads after and before the speed
increase, respectively.
(A separate autoregressive integrated
moving averages analysis of observed–expected ratios for deaths, deaths per billion
vehicle-kilometers of travel, and CFRs on interurban and urban roads of the first year
after the increased speed limit is available
from the authors on request. This analysis
used monthly totals for these parameters
going back 13 years as a baseline for predicting expected results during the first 6
and subsequent 8 months after the increase
in the speed limit.)
Sporadic monitoring from 1971 to 1994
(Table 1) indicated that right after the increase
in the speed limit, travel speeds on high-speed
roads increased by 4.5% on the slow lane of
the Tel Aviv–Haifa road (highway 2), by 9.1%
on the fast lane of the Tel Aviv–Jerusalem
road (highway 1), and even more on a newly
widened stretch of a major connecting road
off the Tel Aviv–Jerusalem road (Table 1,
highway 40)—all compared with the year before. Other data showed that speeds rose on
all 3 highways after the speed limits were
raised, and that the mean estimated increase
in speeds on the high-speed roads later fell
back to a net increase of approximately 4%
(range: −4% to 13%) in 1995.27–30
Modifiers and Confounders
Deaths and Case-Fatality Rates:
Immediate Effects
A sudden increase in monthly nationwide
death tolls and CFRSs followed the increase
in the speed limit on November 1, 1993. The
first month after speed limits were increased,
deaths (n = 61) increased 32.6% from October (n = 46). Interurban deaths (n = 38) and
the CFRS (26.4%) were the highest since November 1990.
First-Year Trends: Subgroups
In the first year after the increase in the
speed limit, deaths increased by 24%, from
257 to 319, and CFRSs increased by 29.5%
on all interurban roads combined, compared
with corresponding increases of 3%—from
230 to 236—and less than 10%, respectively, on urban roads (Table 2). On newly
widened segments of the 3 high-speed highways (and extensions) with 100-kph limits
(roads 1, 2, and 3), deaths increased by
67%, from 21 to 35, a reversal of downward trends from 1990, and CFRSs increased by 50%. Even so, 48 (77%) of the
62 added deaths on interurban roads in the
12 months after the increase in the speed
limit occurred not on the 3 high-speed roads
but on other interurban connecting roads. A
separate autoregressive integrated moving
averages analysis29 verified that the abrupt,
large jump in deaths in the first 6 months
after the increase in the speed limit was es-
April 2004, Vol 94, No. 4 | American Journal of Public Health
pecially marked on interurban roads and
was directly attributable to increases in CFRs
and offset the long-term drops in deaths per
billion vehicle-kilometers of travel. Before
and after the increase in the speed limit,
90% or more of those killed in truck crashes
were occupants of passenger cars.32 After
the increase in the speed limit, much—60%—
of the increase in the nationwide road death
toll came from large increases in deaths from
truck crashes, mainly on interurban roads.
Table 2 also shows increased CFRs and
deaths in 1-vehicle and motorcycle crashes
nationwide and decreased deaths among
pedestrians and cyclists. Despite retention of
the 90-kph limit by the military, reported
deaths involving soldiers—both drivers and
occupants—increased 106% (from 15 to 31),
and reported CFRs increased 30%.
Exposure. In the first year following the increase in the speed limit to 100 kph, interurban traffic increased 5% from 11.4 to 12.0
bvkmt, whereas urban traffic increased much
more—from 13.2 to 15.5 bvkmt (17%). The
number of road deaths per year was weakly
correlated (r = 0.15) with billion vehiclekilometers of travel annually from 1963 to
1995, but negatively correlated with the number of licensed drivers (r = −0.46) from 1970
to 1995. These results rule out more vehicle
traffic and drivers as plausible explanations for
the sudden large increase in deaths.
Precrash and crash countermeasures. The
government introduced several countermeasures, including laws requiring rear seat belts
and daytime running lights (both mandated
on November 1, 1993), more capital investment in upgrading old roads, building of new
roads, midline concrete barriers and flyovers,
and nighttime lighting. Hospital trauma services increased from 1 to 5, and police enforcement, measured by issuance of speeding
tickets, increased approximately fourfold in
the years 1994 to 1998 (Cdr E. Efrat, Traffic
Police Division, written communication, May
2001).
There were no changes in before–after ratios of billion vehicle-kilometers of travel for
trucks to all vehicles (27.3%:27.5%), drivers
aged 19 to 24 years–all drivers (17.2%:17.5%),
or fuel costs or alcohol sales.
Richter et al. | Peer Reviewed | Research and Practice | 569
 RESEARCH AND PRACTICE 
TABLE 1—Measured Speeds on 3 Main Highways and Other Roads Before and After the Speed Limit
Increase (90 kph to 100 kph) of November 1993: Israel, 1971–1994
Highway
Date
No. of Vehicles
Weighted Mean
Speed Traveled, kpha
Range of SDs
(by Siting)
Weighted 90th
Percentileb
3 533
90.7
9.9–12.7
108.7
3 132
94.8
12.8–13.4
109.9
11 974
8 076
8 340
10 000
94.4
98.6
96
98.9
7.7–13.9
12.5–22.9
13.6–20.0
11.0–14.7
NA
114.6
111.7
114
11
2 762
Single-lane connecting road
107.9
8.6–17.3
126.6
No. of Sitings
Right lanec
Before November 1993
Highway 2 (Tel Aviv–Haifa)
After November 1993
Highway 2 (Tel Aviv–Haifa)
2/7/93–9/14/93
5
1/11/94–3/23/94
3
Left lanec
Before November 1993
Highway 2 (Tel Aviv–Haifa)
Highway 2 (Tel Aviv–Haifa)
Highway 4 (Tel Aviv–Beersheba)
Highway 1 (Tel Aviv–Jerusalem)
After November 1993
Highway 1 (Tel Aviv–Jerusalem)
Before November 1993
Highway 40 (off Tel Aviv–Jerusalem)
After November 1993
Highway 40
1971–1990d
9/7/93–9/9/93
8/23/93–8/25/93
8/24/93–9/1/93
4/27/94–5/30/94
10
7
6
8
3/92
3
1 461
72.2
11–15
84.8
7/1/94–7/4/94
8
1 901
90.7
11.2–15.3
113.6
Note. NA = not available.
a
Weighted mean speed traveled: (sum of number of vehicles per siting multiplied by mean speed for individual siting) divided by total number of vehicles.
b
Weighted 90th percentile for speed traveled: (sum of number of vehicles per siting multiplied by 90th percentile speed for individual siting) divided by total number of vehicles.
c
Right lane is slow lane; left lane is fast lane.
d
From 1971 to 1990, mean speeds (and number of sitings) were as follows: 1971: 92.3 (1537); 1975: 97.4 (1338); 1976: 97.3 (1446); 1977: 94.0 (1758); 1980: 88.3 (1860); 1981: 87.6
(1866), 1983: 95.9 (1974); 1988: 103.8 (NA); 1990: 102.8 (NA).
5-Year Trends in Deaths and
Case-Fatality Rates
During the entire 5-year period following
the increase in the speed limit (November 1,
1993, to October 31, 1998), there were substantial drops in the number of persons reported with serious injuries (Table 3). In July
1995, the monthly death toll (n = 62) peaked.
In the third year after the increase in the
speed limit, the death toll on interurban roads
began to fall from a peak in 1994–1995
(n = 327), corresponding to indications of decreases in average interurban speeds
(Table 1) but continued to increase on urban
roads. During the entire 5-year period, there
were mean increases of 39.2 (15%) and 27.2
(13%) deaths per year on interurban and
urban roads, respectively. The corresponding
increases in CFRs, which tracked trends in
speeds of impact, were much greater: 38%
(from 12.5% to 17.3%), and 24% (from 7.9%
to 9.8%) (Table 3). Using validated Newton-
ian models,3 we estimated that without countermeasures and congestion, increased speeds
of travel and impact would have resulted in
increases of 100.2, and 51.1 deaths per year
on interurban and urban roads, respectively.
Countermeasures and congestion would have
resulted, if not for the increased speed limits,
in corresponding reductions of 61 and 23.9
deaths per year. The 5-year nationwide increase in deaths per year (n = 151.3) expected
from increased speeds of impact greatly exceeded the observed increase (n = 66.4) in
total deaths per year (Table 4).
creases in deaths from increases in travel
speeds on the 3 major highways and spillover
of these effects to other urban and interurban
roads. More than three quarters of the firstyear increase in deaths (n = 62) on interurban
roads occurred from a systemwide spillover
effect from high-speed roads on which the
speed limits were legally increased to other
interurban roads. All these findings state the
case for systemwide increases in real travel
speeds.
DISCUSSION
Our observations state the case for a direct
cause-and-effect relation between the increase
in the speed limit and the increase in the
death toll. First, the step function increase in
deaths coincided with the increase in the
speed limit. Second, the increase in deaths is
attributable specifically to the increase in
CFRs—in all vehicle and crash types—a find-
After the increase in the speed limit from
90 to 100 kph, sporadic data suggested that
travel speeds increased on Israel’s 3 major
highways and other roads, later falling back
somewhat. In the first year after the increase
in the speed limit, there were abrupt in-
570 | Research and Practice | Peer Reviewed | Richter et al.
The Case for a Cause-and-Effect
Relation
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 2—Deaths and Case-Fatality Rates in Israel Before and After the Speed Limit Increase (90 kph to 100 kph)
of November 1993: Israel, November 1992–October 1993 vs November 1993–October 1994
Interurban Roads
High Speeda
Other b
Allb
Urban roadsb
Trucksc
Interurban
Urban
Single vehicled
Soldierse
Off-duty
On-duty
Motorcyclesb
Pedestriansb
Bicyclesb
No. Killed
1992–1993
No. Killed
1993–1994
Absolute
Change, No.
Ratio of Nos.
CFR
(1993–1994/1992–1993) 1992–1993, %
CFR
1993–1994, %
CFR
Ratio
CFRS
1992–1993, %
CFRS
1993–1994, %
CFRS
Ratio
21
236
257
230
35
284
319
236
14
48
62
6
1.7
1.2
1.2
1.0
NA
2.5f
2.1
0.9
NA
3.0f
2.7
0.9
NA
1.2
1.2
1.2
14.0
12.8
12.9
8.7
21.0
16.1
16.7
9.3
1.5
1.3
1.3
1.1
48
25
85
74
37
108
26
12
23
1.5
1.5
1.3
3.6
1.9
1.80
5.5
3.6
2.7
1.5
1.9
1.5
NA
NA
8.8
NA
NA
11.9
NA
NA
1.4
14
1
22
199
12
20
11
39
189
10
6
10
17
–10
–2
1.4
11.0
1.8
1.0
0.8
10.6
0.40
0.8
3.7
1.2
7.5
2.9
1.3
3.8
1.1
0.7
8.0
1.6
1.0
0.9
19.4
3.6
5.9
13.7
6.6
26.0
25.6
9.1
14.2
6.5
1.3
7.2
1.5
1.0
1.0
Note. CFR = standard case-fatality rate: killed/(killed + seriously injured + lightly injured), expressed as percentage; CFRS = modified case-fatality rate: killed/(killed + seriously injured only),
expressed as percentage; NA = not available.
a
Data received from the Israel’s Police National Headquarters; data comprising only killed and seriously injured on 100-kph and 90-kph sections of highways 1 (Tel Aviv–Jerusalem), 2 (Tel
Aviv–Haifa), and 4 (Tel Aviv–Ashdod) from 1990 to 1994.
b
Data from Israel Central Bureau of Statistics 1992–1994: Road Accidents with Casualties: Part I.
c
Truck data received by Israel’s Police National Headquarters; data comprising killed and injured (serious + light) for 11 months of 1993 and 11 months of 1994 (November through September).
d
Single-vehicle crash data from Israel Central Bureau of Statistics 1992–1994 Road Accidents with Casualties: Part I. Single-vehicle crashes include categories skidding, overturning, running of the
way, and collision with fixed object.
e
Soldier data received by Israel Defense Forces. Data consists of 8-month periods (November 1992 to July 1993; November 1993 to July 1994). For the CFRS, we combine moderate and serious
injuries (Israel Defense Forces have a different classification: slight, moderate, serious, and killed).
f
CFR ratios for 1993–1994 were corrected using the correction factor provided by the Central Bureau of Statistics to correct for underreporting in 1994.
ing that suggests increased speeds of impact.
Third, the increases in deaths and CFRs on
the high-speed roads were proportionately
much greater than on other interurban roads
and on urban roads. Fourth, the degree of increase in CFRs and deaths matched that expected from the reported increases in travel
speeds based on the validated models. Fifth,
time trends in all the modifiers and confounders (enforcement, seat belts, trauma care)
should have resulted in reductions, not increases, in death tolls.
Predictive Models for the Size
of the Relation
We confirmed the utility and validity of
predictive Newtonian models in which deaths
and CFRs increase in proportion to the fourth
power of increases in speeds of travel of all
vehicles and impact speeds of crashing vehicles, respectively. In the first year after the
100-kph speed limit was implemented, the
observed increase in average travel speeds of
4% to 4.5%, based on sporadic measurements, accords with the increase of 5.5% in
average travel speeds predicted from the observed increase of 24% in deaths on all
roads.3,23–25 Increases in CFRs of 50% on
highways and 26% on other interurban roads
in the first year imply that average increases
in speeds of impact on these roads were of
the order of 11% and 5.6%, respectively.
During the entire 5-year period after implementation of the 100-kph speed limit, similar
calculations suggest that impact speeds of
crashing vehicles increased on interurban and
urban roads by some 8.3% and 5.5%, respectively. These increases exceeded the estimated increases of 3.6% and 3.1% in average speeds of travel for all vehicles (Table 4).
These calculations imply that each 1% increase in average speeds of travel and impact
April 2004, Vol 94, No. 4 | American Journal of Public Health
resulted in increases of approximately 11
deaths per year on interurban roads, and 9
deaths per year on urban roads.
Subgroups at Risk for the Effect
of the Increased Speed Limit
The large increases in the number of
deaths and CFRs were seen in crashes involving trucks, motorcycles, single vehicles, and
soldiers (both on and off-duty) but not pedestrians and bicyclists. Other evidence suggested
that the fatal crash risk of truck drivers from
higher speeds increased with longer hours, irregular shifts, and incentive premiums.32,33
The fact that increases in deaths from increases in CFRs occurred in all subgroups except pedestrians and bicyclists rules out
changes in the case mix of crash types as the
reason for the increase in total CFRs of all
crash groups combined. Soldiers—both drivers
and occupants—were at increased risk, despite
Richter et al. | Peer Reviewed | Research and Practice | 571
 RESEARCH AND PRACTICE 
TABLE 3—Deaths and Case-Fatality Rates in Israel Before and After the Speed Limit
Increase (90 kph to 100 kph) of November 1993: All Roads, Interurban, and Urban, Israel,
1990–1997
Yeara
Killed, No.
Seriously Injured, No.
CFRS, %
Range
Interurbanb
Before increase
1990–1991
1991–1992
1992–1993
Mean
219
307
257
261
1718
2078
1735
1843.7
11.8
12.9
12.9
12.5
6.99–17.73
10.04–18.07
7.50–19.33
1993–1994
1994–1995
1995–1996
1996–1997
1997–1998
Mean
319
327
293
278
284*
300.2
1596
1596
1407
1259
1344
1440.4
16.7
17.0
17.2
18.1
17.4**
17.3
11.36–26.39
11.58–21.09
12.99–23.53
12.82–24.37
12.10–24.42
After increase
Urbanb
Before increase
1990–1991
1991–1992
1992–1993
Mean
211
195
230
212
2448
2592
2410
2483.3
7.94
7.00
8.71
7.9
5.49–12.28
3.46–9.65
4.74–13.86
1993–1994
1994–1995
1995–1996
1996–1997
1997–1998
Mean
236
223
235
249
253*
239.2
2303
2376
2231
2156
2008
2214.8
9.3
8.6
9.5
10.4
11.2**
9.8
7.39–11.27
6.15–11.77
6.35–12.76
5.47–14.04
7.11–14.42
After increase
Note. CFR = case-fatality rate: the standard rate killed/all casualties; CFRS = modified case-fatality rate: killed/(killed +
seriously injured). Student’s t test compared monthly death and case-fatality rates of November 1990 through October 1993
versus November 1993 through October 1998 for all roads combined, interurban alone, and urban alone. Slightly Injured
were corrected for underreporting in 1993 by 1% and in 1994 by 9% as recommended by the Central Bureau of Statistics.
Change in 1996 reporting of slightly injured resulted in an increase in reported light injuries by more than 8000 during
1995–1996. Change in reporting (“attendance only” requirement) probably reduced number of seriously injured as well.
a
Calendar year in this study is from November through October. The speed limit was raised officially on November 1, 1993.
b
All data obtained from Israel Central Bureau of Statistics Road Accidents with Casualties: Part I for years 1990 to 1997.
* P < .01; ** P < .001.
the military’s retention of the 90-kph limit during working hours. For pedestrians and bicyclists, the drop in death tolls may be explained
by a trend seen in the United Kingdom, where
there has been a reported decrease in walking
and cycling on interurban roads.34 The fact that
CFRs from 1-vehicle crashes—which are not influenced by vehicle–vehicle interactions—were
not less than those from other crash types undermines the claim that increased speed variance35 and not increased speed is the real
cause for the increase in deaths.
Speed Versus Congestion and
Countermeasures
The observed increase in deaths per year
following the increase in the speed limit to
100 kph substantially underestimated the increase in deaths directly attributable to the
increase in CFRs. The exponential effect of
“small” increases in speed and speed spillover
on nationwide increases in CFRs over the
next 5 years more than offset the decreases
in death risks per vehicle-traveled from protective countermeasures, as well as conges-
572 | Research and Practice | Peer Reviewed | Richter et al.
tion in urban areas. The persistence of high
CFRs indicates that increased speeds of impact were negating the protective effects of
newly widened roads, improved lighting,
cloverleafs, air bags, rear seat belt laws, more
speed enforcement and changes in trauma
care, and other countermeasures, as well as
increased congestion. Without protective
countermeasures and increased traffic congestion, there would have been many more
deaths, and without the increase in the speed
limit, there would have been many fewer
deaths (Table 4).
In Israel, increases in traffic congestion and
road safety countermeasures have produced a
long-term strong inverse relation between
deaths per billion vehicle-kilometers of travel
and both the number of vehicles (r = −0.88)
and vehicles per population (r = −0.85).19,27 In
the United States, car occupant deaths dropped
11% between 1975 and 1997, despite a
ninefold increase in cars.36 But from 1992 to
1997, by which time US states were raising
speed limits, deaths—and even deaths per billion vehicle-kilometers of travel—did not
drop.19 By contrast, in the years 1991 to
1998, road deaths per year fell by 25.2% in
the United Kingdom with correspondingly
larger decreases in deaths per billion vehiclekilometers of travel.37
Risks for deaths per billion vehiclekilometers of travel have always decreased
with increases in billion vehicle-kilometers
of travel38 (“the soccer field is tilted downwards”19); therefore, before and after differences following increased speed limits will underestimate the increases in death tolls from
increased speed limits over many years. Studies of the long-term impact of increased speed
limits that “correct” for increases in billion vehicle-kilometers of travel may underestimate
the full direct impact of increased speed limits
and travel speeds on road deaths. Long-term
follow-up in Washington State, for example,
showed that when corrected for exposure, a
25% increase in deaths on interstates was reduced to a 10% increase.39 These studies ignore the role of countermeasures and increased congestion in producing the falling
trends in deaths per billion vehicle-kilometers
of travel in urban areas. Congestion from increased billion vehicle-kilometers of travel
during the so-called rush hours offsets the ef-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 4—Effect of Increase in Speed Limit With and Without Countermeasures and Congestion:
Deaths, Estimated Change in Speeds, and Estimated Deaths per Year per 1% Increase in Speed
Mean Deaths per
Year November
1990 to October
1993 (Baseline),
No.
Observedd
Expected
Speed limit raised; without
increase in countermeasures and
congestion (expected scenario 1)e
Speed limit not raised; with
increase in countermeasures and
congestion (expected scenario 2)e
Mean Deaths per
Year November
1993 to October
1998 (Follow-up),
No.
Ratio Change
(Baseline to
Follow-up)
Estimated Change
in Speeds (Nilssona
and Jokschb
Formulas), %
Mean Change in
Deaths per Year
for Each 1%
Increase in
Speed, No.c
39.2
1.15
3.6a
10.9
Absolute Change
(Baseline to
Follow-up), No.
261
Interurban roads
300.2
261
361.2
100.2
1.38
8.3b
12.1
261
200
–61
0.77
–6.4a
–9.53
212
239.2
27.2
1.13
3.1a
8.8
212
263.3
51.3
1.24
5.5b
9.3
212
188.1
–23.9
0.89
–2.9a
–8.24
Urban roads
Observedd
Expected
Speed limit raised; without
increase in countermeasures and
congestion (expected scenario 1)e
Speed limit not raised; with
increase in countermeasures and
congestion (expected scenario 2)f
a
Change in travel speeds derived from fourth root of ratio of number of deaths after to deaths before increase in speed limit, using equation K(a)/K(b) = V(a)4/V(b)4 to solve for V(a), where K(a) and K(b)
and V(a) and V(b) are deaths and average vehicular speeds, respectively, on roads after and before (Evans3,23).
b
Change in impact speeds calculated as fourth root of ratio of case-fatality rate after increase in speed limit to that before increase in speed limit (see references 23 and 24).
c
Derived by dividing absolute change in deaths by estimated percentage change in speeds.
d
See the means for deaths per year before and after the increase in speed limit in Table 3.
e
Derived from increase in mean case-fatality rate alone from 12.5% to 17.3% on interurban roads and from 7.9% to 9.8% on urban roads after the increase in the speed limit from 90 to 100 kph
on November 1, 1993. We used the formula K(ATTRIB) = {K(B) × (CFRS(A)/ CFRS(B))} – K(B), where K(ATTRIB) is persons killed per year attributable to the increase in speed 5 years after the speed limit
increase, K(B) is persons killed per year 3 years before the increase in the speed limit, and CFRS(B) and CFRS(A) are the proportion of those killed among all those seriously injured before and after
the speed limit increase. We estimated average changes in speed of travel and crash impact and number of deaths per each 1% change in speed, using fourth-root models.
f
The difference in observed deaths in the 5-year follow-up period and the absolute change (baseline to follow-up) in scenario 1 gives us expected deaths for scenario 2.
fects of higher speeds during other hours, notably during nighttime.
Evans has shown how relations between increases (and decreases) in speed and the increases (and decreases) in death tolls are direct, obey algebraically defined laws derived
from Newtonian physics, and are reversible.23
Based on these premises, we suggest that a reduction by 10% in the average speeds of impact during the period we studied would have
prevented 121 (or 40%) of the 300 interurban deaths per year, and 85 (or 35.4%) of the
240 urban deaths per year (Table 4). Support
for this inference comes from the observation
in the United Kingdom that there are even bigger decreases in death tolls—up to 70%—with
reductions in speeds from massive use of road-
side speed cameras.40 Substantial reductions in
deaths, injuries, and crashes from the use of
speed cameras state the case for their use.41
We suggest, however, that achieving sustainable major reductions in road death tolls
requires not only lower speed limits and increased detection and deterrence of high
speeds, but also lower design speeds for cars,
and a downward shift in speed distributions,
in keeping with the principle of treating sick
populations, not just sick individuals.42 The
public health stakes involved in applying this
principle are enormous, given that globally,
there are now more than 1 170 000 road
deaths per year43 and more than 40 000
deaths per year in the United States alone. We
predicted increases in death tolls from new
April 2004, Vol 94, No. 4 | American Journal of Public Health
highways and spillover roads with even higher
design speeds and speed limits, a trend now
seen in many rapidly motorizing countries.44
Elsewhere, we have suggested using the CFR
to track the direct long-term impact of increased travel speeds on death tolls in the
United States.19 The CFR, the outcome of concern, is a parameter based on a universe; because it is extremely sensitive to small changes
in speed well within the range of sampling
and measurement errors, it paradoxically may
be a more valid indicator of speed trends than
sporadic speed measurements themselves.
Elsewhere, we have addressed the ethical
and scientific lapses underlying the decision
to increase the speed limit.45,46 In retrospect,
the sentinel increases in travel speeds on the
Richter et al. | Peer Reviewed | Research and Practice | 573
 RESEARCH AND PRACTICE 
study’s highways, CFRs, and deaths in the
very first months of the 100-kph “experiment”
predicted its subsequent 5-year nationwide
impact and stated the case for its cancellation.
In 2002, despite an increase in deaths to
540 from 476 the year before, there were renewed pressures to increase the speed limit
still further, to 110 kph–120 kph.
About the Authors
Elihu D. Richter and Lee Friedman are with Hebrew
University-Hadassah School of Community Medicine and
Public Health, Unit of Occupational and Environmental
Medicine and Injury Prevention Center, Jerusalem, Israel.
Paul Barach was with the Center for Patient Safety, Department of Anesthesia and Critical Care, Pritzker School
of Medicine, University of Chicago, Ill. Samuel Krikler is
with the Department of Statistics and Abraham Israeli was
with the Department of Medical Ecology and Health Services, Hebrew University-Hadassah, Jerusalem.
Requests for reprints should be sent to Elihu D Richter
MD, MPH, Unit of Occupational and Environmental
Medicine, Hebrew University-Hadassah Medical School,
Jerusalem 91120, Israel ([email protected]).
This brief was accepted September 10, 2003.
Contributors
E. D. Richter collected preliminary data, conceived the
study, and wrote the brief. P. Barach collected much of
the data on the trends in the first year and reviewed
the literature. L. Friedman did the data collection and
statistical analysis for the 5-year follow-up. S. Krikler
oversaw time series analyses connected with the shortterm effects. A. Israeli served as co-tutor and adviser to
P. Barach.
Acknowledgments
Partial support for this work came from university
scholarship grants.
We thank Zvi Weinberger of Jerusalem College of
Technology, Tali Tal of the Central Bureau of Statistics
in Israel, Dr Orly Manor of the Department of Medical
Ecology of Hebrew University-Hadassah School of Public Health and Community Medicine, and Professor
Gerald Ben-David, Dr Jacob Adler, and Zelda Harris of
Metuna for advice and helpful suggestions.
Human Participant Protection
No protocol approval was needed for this study.
References
1. Fishman RH. Anti-speeding Israeli doctors. Lancet.
1995;345:1040.
2. Transportation Research Board. 55: A Decade of
Experience. Washington, DC: National Research Council. 1984. Special Report No. 204.
3. Evans L. Traffic Safety and the Driver. New York,
NY: Van Nostrand Reinhold Co; 1990.
4. Robertson L. Injury Epidemiology. New York, NY:
Oxford University Press; 1998.
5. Freedman M, Williams AF. Speed associated with
55 mph and 65 mph speed limits in northeastern
states. Instit Transportation Engineers J. 1992;62:7–21.
6. Wagenaar AC, Streff FM, Schultz RH. Effects of
the 65 mph speed limit on injury morbidity and mortality. Accid Anal Prev. 1990;22:571–585.
7. Garber S, Graham JD. The effect of the new 65
mph speed limit on rural highway fatalities: a state-bystate analysis. Accid Anal Prev. 1990;22:137–149.
8. Baum HM, Wells JK, Lund AK. The fatality consequences of the 65 mph speed limit. J Safety Res. 1999;
22:171–177.
9. Rock SM. Impact of the 65 mph speed limit on
accidents, deaths and injuries in Illinois. Accid Anal
Prev. 1995;27:207–224.
10. Insurance Institute For Highway Safety (USA).
Deaths go up on interstate highways where higher
speed limits are posted. Status Report. 1999;34:4–5.
11. Farmer CM, Retting RA, Lund AK. Changes in
motor vehicle occupant fatalities after repeal of the national maximum speed limit. Accid Anal Prev. 1999;31:
537–543.
12. Schmidt DF, Tiffin J. Distortion of drivers’ estimate of automobile speed as a function of speed adaptation. J Appl Psychol. 1969;53:536–539.
13. Mathews ML. A field study of the effects of drivers’ adaptation to automobile velocity. Hum Factors.
1978;20:709–776.
14. Casey SM, Lund AK. Changes in Speed and Speed
Adaptation Following Increase in National Maximum
Speed Limit. J Safety Res. 1992;23(3):135–146.
15. Casey SM, Lund AK. Changes in speed and speed
adaptation following increase in national maximum
speed limit. J Safety Res. 1992;23:35–46.
16. Lave L, Elias P. Did the 65 mph speed limit save
lives? Accid Anal Prev. 1994;26:49–62.
17. Beenstock M, Gafni D. Globalization in road safety:
explaining the downward trend in road accident rates in a
single country (Israel). Accid Anal Prev. 2000;32:71–84.
18. BBC News Talking Point: Should there be more
speed cameras? August 2, 2001. Available at: http://
www.bbcnews.com. Accessed August 3, 2001.
19. Richter ED, Barach P, Ben-Michael E, Berman T.
Death and injury from motor vehicle crashes: a public
health failure, not an achievement. Inj Prev. 2001;7:
176–178.
20. Tomorrow’s Roads—Safer for Everyone: The Government’s Road Safety Strategy and Casualty Reduction Targets For 2010. London: DETR Free Literature; March
2000.
21. Richter ED, Meltzer U, Bendov Tyger G, Bloch B.
Alcohol levels in drivers and pedestrians killed in road
accidents in Israel. Int J Epidemiol. 1986;5:272–273.
22. Joksch H. Velocity change and fatality risk in a
crash: a rule of thumb. Accid Anal Prev. 1985;7:55–70.
23. Evans L. Driver injury and fatality risk in two-car
crashes versus mass ratio inferred using Newtonian
mechanics. Accid Anal Prev. 1994;26:609–616.
24. Nilsson G. The effect of speed limits on traffic accidents in Sweden. Linkoping, Sweden: National Road
and Traffic Research Institute; 1982. VTI Report 68.
S-58101 1–10, 1982.26. 27.
25. Rumar K. Speed—a sensitive matter for drivers.
Nordic Road Transport Res. 1999;11:20–22.
26. Van Beeck EF, Mackenback JP, Looman CWN,
Kunst A. Determinants of traffic accident mortality in
the Netherlands: a geographical analysis. Int J Epidemiol. 1999;20:698–706.
574 | Research and Practice | Peer Reviewed | Richter et al.
27. Road Accidents with Casualties Annual Reports, and
Transport Statistics Quarterly Reports. Jerusalem, Israel:
Israel Central Bureau of Statistics; 1990, 1991, 1992,
1993, 1994, 1995, 1996, 1997, 1998.
28. Caspit N, Mehalel D, Livneh M. Influence of
Change in Speed Limit on Road Safety [in Hebrew].
Haifa, Israel: Technion-Israel Institute of Technology;
1997. Research Report 97–248.
29. Barach P. Effects of Raised Speed Limit on Road
Deaths In Israel [MPH thesis; in Hebrew with English
summary]. Jerusalem, Israel: Hebrew University; 1998.
30. Hocherman I, Cohen A, Dubeh E. Trends in
speed and road accidents on fast roads in Israel [in Hebrew]. Traffic Transportation. November 1996:33–38.
31. Chiu WK. The economic design of CUSUM charts
for controlling normal means. Appl Stat. 1974;
23,420–433.
32. Ben-David G, Neeman V, Apter J, Richter ED.
Trends in deaths and injury from truck crashes: Israel.
Paper presented at: 4th International Conference on
Traffic Safety. 1997; Tel Aviv.
33. Barach P, Ben-David G, Richter ED. Truck injuries and fatigue. N Engl J Med. 1998;338:390.
34. Allsop RE. British experience with a national road
casualty reduction target. Paper presented at: 4th International Conference on Safety in the Environment in the
21st Century; November 23–27, 1997; Tel Aviv, Israel.
35. Rodriguez R. Speed, speed dispersion and the
highway fatality rate. South Econ J. 1990;57:349–356.
36. Insurance Institute For Highway Safety (USA).
Motor vehicle deaths unchanged overall but important
differences still apparent. Status Report. 1998:33;4–6.
37. International Accident Facts. 2nd ed. National
Safety Council, Israel Engineering Association: Tel Aviv.
1999.
38. Smeed RJ. Variation in the pattern of accident
rates of different countries and their causes. Traffic Engine Control. 1968;11:364–371.
39. Osiander E, Cummings P. Freeway speed limits
and traffic fatalities in Washington State. Accid Anal
Prev. 2002;341:13–18.
40. West R. The effect of speed cameras on injuries
from road accidents. BMJ. 1998;36:5–6.
41. Richter ED, Reingold S. Injury and environmental
epidemiology: a converging agenda. Global Change
Hum Health. 2002;3:2–12
42. Rose G. Sick individuals and sick populations. Int
J Epidemiol. 1985;4:32–38.
43. World Health Organization. Leading causes of
death (world). 1998. Available at: http://www3.who.
int/whosis/menu.cfm. Accessed February 2, 2000.
44. Ginsberg G, Fletcher E, Ben-Michael E, Richter ED.
How many shall live, how many shall die? Deaths resulting from the Trans-Israel Highway and alternatives: a risk
assessment. World Transport Policy Practice. 1997;3:4–10.
45. Richter ED, Barach P, Ben-Michael E, Weinberger
Z. Junk ethics and junk science in transport risk assessment. Presented at: the Proceedings of the 7th International Conference, Israel Social Ecology and Environmental Health Sciences; June 13–18, 1999; Jerusalem.
46. Richter ED, Barach P, Berman T, Ben-David G,
Weinberger Z. Extending the boundaries of the Declaration of Helsinki: a case study of an unethical experiment in a non-medical setting. J Med Ethics. 2000:27:
126–129.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Knee Pain and Driving Duration: A Secondary Analysis
of the Taxi Drivers’ Health Study
| Jiu-Chiaun Chen, MD, MPH, ScD, Jack T. Dennerlein, PhD, Tung-Sheng Shih, ScD, Chiou-Jong Chen, PhD, Yawen Cheng, ScD, Wushou P. Chang, MD,
ScD, Louise M. Ryan, PhD, and David C. Christiani, MD, MPH, MS
Knee pain is a common health problem
worldwide. Data from the First National
Health and Nutrition Examination Survey
(NHANES I) suggest that in the 1970s, it was
the second most common musculoskeletal
symptom, affecting 13.3% of people aged 25
to 74 years.1 Results of NHANES III
(1988–1994) revealed that 18.1% of US
men and 23.5% of US women aged 60 years
or older suffered from significant knee pain.2
During the same period surveyed by
NHANES III, the estimated 1-year prevalence
of persistent knee pain in England was 25%
among those aged 55 years and older.3 Similar statistics showing that knee pain is a prevailing public health problem can be derived
from studies conducted in Europe.4–7 Other
research findings demonstrate that people
who live in the nonindustrialized world are
not exempt from this endemic problem, because estimates of knee pain prevalence from
nonindustrialized countries either were comparable to those in industrialized countries8–11
or were even higher,12 partially because of
the greater prevalence of heavy physical activities in nonindustrialized countries.
Knee pain is very likely a health problem
with tremendous health care costs, despite the
lack of direct cost estimates. In 1996–1997,
more than 6 million Americans sought medical care for knee problems,13 about 5 million
of whom visited offices of orthopedic surgeons and 1.4 million of whom went to a hospital emergency room. A survey of US orthopedic surgeons conducted in 1997 found that
the knee was the most often treated anatomic
site, accounting for 26% of all orthopedic visits.13 Pain relief remains one of the major reasons for joint replacement.14 In 1999,
311 106 inpatient hospital stays involving
total knee replacement in the United States
accrued a “national bill” of more than $6.5
billion.15 The annual rate at which patients request total knee replacements to ameliorate
Objectives. We explored a postulated association between daily driving time and
knee pain.
Methods. We used data from the Taxi Drivers’ Health Study to estimate 1-year prevalence of knee pain as assessed by the Nordic musculoskeletal questionnaire.
Results. Among 1242 drivers, the prevalence of knee pain, stratified by duration of
daily driving (≤ 6, > 6 through 8, > 8 through 10, and > 10 hours), was 11%, 17%, 19%,
and 22%, respectively. Compared with driving 6 or fewer hours per day, the odds ratio
of knee pain prevalence for driving more than 6 hours per day was 2.52 (95% confidence
interval = 1.36, 4.65) after we adjusted for socioeconomic, work-related, and personal
factors in the multiple logistic regression.
Conclusions. The dose-related association between driving duration and knee pain
raises concerns about work-related knee joint disorders among professional drivers. (Am
J Public Health. 2004;94:575–581)
knee pain and restore mobility has increased
since the early 1990s.13 A similar trend also
has been reported in Europe. After examining
data from the Swedish Knee Arthroplasty
Registry, researchers found that the number
of knee arthroplasties per year between the
periods 1976–1980 and 1996–1997 increased more than fivefold.16 On the basis of
the 1996 and 1997 data, it was projected
that, from 2000 through 2030, in the absence of an effective preventive treatment, the
number of knee arthroplasties per year will
increase by at least one third.
Moreover, knee pain imposes a significant
disability burden on modern societies.3,17–20
Both cross-sectional and prospective studies
have consistently shown that knee pain,
rather than radiographically detectable abnormalities, is the major determinant of knee
osteoarthritis–related physical disability.6,21–26
Longitudinal studies have demonstrated that
previous knee pain is associated with both
the development of disease27 and the progression of radiographically evident knee osteoarthritis.28,29 In the NHANES Epidemiologic Follow-Up Study30 on the relative risk
of experiencing difficulty in ambulation and
transfer (as from a chair to a standing position), the estimated relative risk for knee osteoarthritis patients (4.42 and 4.08, respec-
April 2004, Vol 94, No. 4 | American Journal of Public Health
tively) were twice those for heart disease patients (2.27 and 2.13). Framingham Osteoarthritis Study31 researchers estimated
that approximately 15% of the risk for the
overall population of experiencing difficulty
in walking—the highest attributable proportion for any single medical comorbidity—was
attributable to knee osteoarthritis. Knee pain
also may lead to accidental falls,32–35 which,
together with arthritis, account for more than
30% of all restricted-activity days among
older US adults.36 As the baby boom generation ages, the knee pain–related disability
burden will become even more substantial;
therefore, studying the multifaceted problem
of knee pain is a public health task of fundamental importance.
Researchers should seek a better understanding of the mechanisms and the impacts
on health of knee pain.2,17,37 Because most
musculoskeletal pain is chronic and recurrent,38 studies of knee pain with onset at a
younger age, such as knee pain precipitated by
work-related injury or strain, and the contribution of knee pain to later disability will provide
us with better information about the natural
history of knee osteoarthritis. Such knowledge
will help us to develop effective prevention
strategies and management modalities tailored
to different stages of the disease. A similar re-
Chen et al. | Peer Reviewed | Research and Practice | 575
 RESEARCH AND PRACTICE 
search direction has been adopted in studies of
other types of musculoskeletal pain.39–43
Descriptive results of 2 previous reports directed our attention to work-related knee
pain among professional drivers. Anderson
and Raanaas44 conducted a survey of musculoskeletal complaints of taxi drivers in Norway. They used the Nordic musculoskeletal
questionnaire45 and found that the 1-year
prevalence of knee pain among 703 full-time
taxi drivers was higher than that among the
reference group from the local community
(29% vs 25%, respectively). A nationwide occupational health survey in Taiwan46,47 that
used a modified version of the Nordic musculoskeletal questionnaire also found that employed professional drivers had a knee pain
prevalence slightly higher than the national
average (11% vs 8.6%). However, no further
data were available to explain the higher
prevalence of knee pain among professional
drivers observed in these 2 studies.
In 2000, the Taxi Drivers’ Health Study
(TDHS)48—an occupational, epidemiological
study of cardiovascular disease risk, job stress,
and low back pain—was launched in Taipei,
Taiwan. The TDHS baseline data allowed us
to test the hypothesis that prolonged driving
is associated with increased knee pain prevalence among taxi drivers.
METHODS
The TDHS is integral to a medical-monitoring program sponsored by the Taipei city
government that provides taxi drivers with
free physical examinations each year.48,49
From January 31 to May 31, 2000, 3295
taxi drivers participated in this program. From
the 5 hospitals designated to provide free
physical examinations (each hospital had a
maximum number of taxi drivers it could
serve), we selected the one with the largest
assigned service volume as our study base for
the TDHS. For drivers to be eligible for enrollment in our study, they had to (1) have
been registered taxi drivers in Taipei for at
least 1 year, (2) be voluntary participants, and
(3) be able to read.
A standardized, self-administered questionnaire was delivered to each participant in the
selected hospital. Its feasibility was tested
among a volunteer sample of taxi drivers,
who were recruited from cab companies, cooperative practices, local unions, and resting
areas (a large parking area wher drivers can
take a break, wash their cars, etc.), before the
study began. In addition to questions about
demographics and health behaviors, the questionnaire contained items regarding driver
profiles (professional seniority in years, average number of driving days per month, and
duration of daily driving in hours) and average frequency of physical activities (lifting
and bending/twisting) during both work and
leisure time. Previous studies50,51 have shown
that self-reporting is a relatively reliable and
valid method to assess time spent driving a
motor vehicle. In a small subset of baseline
data from drivers who also participated in an
exposure assessment study,52 we found that
97% of self-reported daily driving times
(grouped by periodic categories) agreed with
data we retrieved from diary records and
structured interviews. Although self-reported
daily driving estimates exceeded actual measurements by an average of 0.9 hour, this
measurement error was independent of knee
pain (P = .73). The modified Nordic musculoskeletal questionnaire, the same questionnaire used in a previous nationwide survey,47
presented a graph of 9 body parts and asked
subjects to mark the anatomic sites at which
they had experienced any pain in the past 12
months. (The Nordic musculoskeletal questionnaire has been demonstrated to possess
acceptable validity and reliability.45,53) The
modified questionnaire also included a job
dissatisfaction subscale from the Job Content
Questionnaire (Chinese version) and 5 questions about mental health from the Taiwanese
version of the 36-item Medical Outcomes
Study short form (SF-36).54,55 Anthropometric and laboratory data were retrieved from
annual free physical examination records.
We used multiple logistic regression analysis to estimate the odds ratio of knee pain
prevalence associated with a change in duration of driving time. We grouped drivers by 4
categories according to duration of daily
driving (≤ 6, > 6 through 8, > 8 through 10,
and > 10 hours) and calculated the crude
odds ratio for knee pain prevalence in each
group. Drivers who had driving times of 6 or
fewer hours composed the reference group.
We wanted to make a statistical inference
576 | Research and Practice | Peer Reviewed | Chen et al.
about the effect of daily driving time on knee
pain prevalence that controlled for biomechanically or biologically plausible risk factors for knee pain and osteoarthritis. We
searched for these potential predictors before
we examined the relationship between any
covariate and knee pain prevalence in the
univariate analysis. This process identified
age, body mass index (BMI), education,
smoking, lifting, bending/twisting, and psychosocial variables as predictors retained in
the final model. We then fit the univariate
model, driving time only (base model). All
other variables had to cause at least a 10%
change in the estimate of the odds ratio of
knee pain prevalence associated with duration of daily driving in the base model to be
included in the final logistic model, or they
had to be significant in the univariate analysis (P = .25). We assumed no interactions
among the potential predictors and included
only subjects with complete data in the final
analyses. The Hosmer–Lemeshow test56 was
used to assess the goodness of fit. Finally, we
performed the jackknife dispersion test57 to
obtain an unbiased adjusted odds ratio of
knee pain prevalence associated with a
change in duration of daily driving. All of
these statistical analyses were conducted with
Stata 7.0 statistical software (Stata Corp, College Station, Tex).
RESULTS
Of the 1355 drivers who received medical
examinations in the selected hospital, 1242
(92%) completed the 2 sets of questionnaires.
The study population’s mean age ± SD was
44.5 ± 8.7 years, drivers drove an average of
9.8 hours per day and 26 days per month,
and 234 (19%) drivers had experienced knee
pain in the past 12 months. Personal characteristics and occupational factors are shown in
Table 1. We also tabulated the population reference statistics58 and the demographic and
other characteristics of the other 1940 drivers who were not enrolled in the TDHS but
who had received physical examinations in
other hospitals during the study period. With
respect to the distribution of age, gender, professional seniority, daily driving duration,
BMI, marital status, and registration type, the
TDHS–enrolled drivers were not significantly
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 1—Demographic and Occupational Characteristics of Participants in the Taxi Drivers’
Health Study (TDHS) and Other Driversa: Taipei, Taiwan, 2000
TDHS Participants (N = 1242)
Other Drivers (N = 1940)
n1
Mean ± SD or %
n2
Mean ± SD or %
Reference Groupb
Age, y
1242
44.5 ± 8.7
1403
46.6 ± 8.7
43.9
Professional seniority, y
1234
11.4 ± 7.8
1890
11.0 ± 7.5
9.2
Total driving per month, days
1239
26.2 ± 2.6
1780
25.2 ± 3.6
26.8
Total driving per day, h
1238
9.8 ± 2.8
1889
9.9 ± 2.5
Body mass index, kg/m2
1242
24.9 ± 3.6
1780
25.2 ± 3.6
...
1193
96%
1854
96%
97%
49
4%
82
4%
3%
Characteristics
10
Gender
Male
Female
Education
Less than high school
405
33%
770
40%
...
High school
782
63%
1067
56%
...
53
4%
69
4%
...
College or more
Marital status
Single
201
16%
257
14%
...
Married
960
75%
1469
77%
...
Separated/divorced/widowed
116
9%
178
10%
...
Individual
497
40%
808
43%
...
Cooperative
395
32%
606
33%
...
Affiliated with taxicab company
341
28%
447
24%
...
Never/rare/seldom
604
49%
...
...
...
Often/sometimes
508
41%
...
...
...
Very frequently
122
10%
...
...
...
Registration type
Lifting activities
Bending/twisting
Never/rare/seldom
643
52%
...
...
...
Often/sometimes
482
39%
...
...
...
Very frequently
111
9%
...
...
...
Leisure-time physical exertion
Never/rare/seldom
602
49%
...
...
...
Often/sometimes
506
41%
...
...
...
Very frequently
126
10%
...
...
...
Perceived job stress
None
282
23%
...
...
...
Mild
639
52%
...
...
...
311
25%
...
...
...
Mental health score (0–100)
Moderate to severe
1218
63.1 ± 16.8
...
...
...
Job dissatisfaction index (0.01–1.00)
1225
0.61 ± 0.17
...
...
...
Low back pain in past 12 months
628 (1241)
51%
988 (1798)
55%
...
Knee pain in past 12 months
234 (1241)
19%
395 (1798)
22%
...
Note. n1 = number of subjects in TDHS group; n2 = number of subjects not in study base. The total number summed up across
each category varies slightly because of missing data.
a
Other drivers received medical examinations at hospitals outside the study.
b
Data from Dept of Statistics, Ministry of Transportation and Communication, Taiwan.58
April 2004, Vol 94, No. 4 | American Journal of Public Health
different from drivers who were not enrolled,
although they had a slightly lower prevalence
of both knee pain and low back pain. We also
noted that the demographic features of these
2 groups of drivers were comparable to the
reference statistics.
Crude estimates of the 1-year prevalence
of knee pain—stratified by duration of daily
driving ( ≤ 6, 6–8, 8–10, and > 10 hours)—
were 11%, 17%, 19%, and 22%, respectively.
Compared with drivers who drove 6 or fewer
hours per day, the crude odds ratio of knee
pain prevalence for drivers who drove more
than 6 hours per day was 2.06 (95% confidence interval [CI] = 1.23, 3.43). Univariate
analyses indicated that high frequency of
bending/twisting activities during both work
and leisure time, moderate to severe self-perceived job stress, a low mental health score,
and high job dissatisfaction were significantly
and positively associated with knee pain prevalence (P < .05).
The results of the multiple logistic regression analyses are shown in Table 2. After we
adjusted for age, gender, BMI, income, education, marital status, smoking habit, frequency
of regular exercise, mental health score, selfperceived job stress, job dissatisfaction index
score, physical exertion during both work and
leisure time, and professional seniority, taxi
drivers with long driving times ( > 6 hours/
day) had a significantly higher prevalence of
knee pain than drivers with short driving
times (≤ 6 hours/day): an adjusted odds ratio
of 2.52 (95% CI = 1.36, 4.65). In contrast to
the case for crude analyses, this increase in
odds ratio estimate resulted mainly from the
joint negative confounding by high physical
exertion during leisure time, low income, and
registration as an individual driver (as opposed to being in a cooperative practice or affiliated with a taxicab company). Those drivers with any 1 of these 3 characteristics
tended to drive less than their counterparts.
The result of the Hosmer–Lemeshow test
(P = .74) supported the goodness of fit of the
multiple logistic model. The jackknifed odds
ratio associated with long driving times ( > 6
hours/day) was 2.40 (95% CI = 1.24, 4.63).
All of the jackknife estimates of the odds ratios for knee pain prevalence associated with
each category of daily driving time were similar to estimates provided by all observations,
Chen et al. | Peer Reviewed | Research and Practice | 577
 RESEARCH AND PRACTICE 
TABLE 2—Odds Ratios (ORs) and 95% Confidence Intervals (CIs) for Prevalence of Knee
Pain in Past 12 Months (n = 1115): TDHS, Taipei, Taiwan, 2000
Characteristic
Total driving per day, h
≤6
6–8
8–10
> 10
Bending/twisting
Never/rare/seldom
Often/sometimes
Very frequently
Leisure-time physical exertionb
Never/rare/seldom
Often/sometimes
Very frequently
Perceived job stress
None
Mild
Moderate to severe
Low mental health scorec
No
Yes
High job dissatisfactiond
No
Yes
Registration type
Affiliated with taxicab company or cooperative practice
Individual practice
Crude OR (95% CI)
Adjusteda OR (95% CI)
1.00
1.70 (0.93, 3.11)
1.95 (1.12, 3.40)*
2.30 (1.35, 3.93)**
1.00
1.99 (1.00, 3.98)
2.55 (1.32, 4.94)**
3.14 (1.62, 6.08)**
1.00
1.25 (0.92, 1.29)
1.75 (1.09, 2.80)*
1.00
1.08 (0.75, 1.55)
1.56 (0.88, 2.75)
1.00
1.48 (1.09, 2.01)*
1.78 (1.12, 2.82)*
1.00
1.35 (0.94, 1.93)
1.94 (1.12, 3.34)*
1.00
1.58 (1.05, 2.38)*
2.49 (1.61, 3.84)**
1.00
1.36 (0.85, 2.15)
1.78 (1.06, 2.99)*
1.00
2.12 (1.57, 2.88)**
1.00
1.77 (1.26, 2.50)**
1.00
1.50 (1.07, 2.11)*
1.00
1.31 (0.90, 1.91)
1.00
1.22 (0.91, 1.62)
1.00
1.60 (1.09, 2.35)*
a
Adjusted for age, gender, education level, body mass index, marital status, income, smoking, professional seniority in years,
days of driving per month, full-time status, frequency of heavy lifting activities, regular exercise, and all the other covariates in
the table.
b
The frequency of bending/twisting and/or heavy lifting when not at work.
c
Low mental health score is defined as standardized mental health score lower than the first quartile as measured by the
Taiwanese version of the 36-item Medical Outcomes Study short form (SF-36).
d
High job dissatisfaction is defined as those whose job dissatisfaction index are in the highest quartile as measured by the
job dissatisfaction subscale in the Job Content Questionnaire.
*P < .05; **P < .01.
suggesting that no observations were overly
influential.
DISCUSSION
To our knowledge, analytic studies that
show the association between knee pain and
long driving times have not been reported in
the literature in English. Our study indicated a
likely association between long driving times
and increased knee pain prevalence, both in
the crude analysis and after adjustment for a
large set of potential confounders and risk factors for knee pain and knee osteoarthritis.
Few previous studies have examined this
interesting association. In a nationwide survey
of musculoskeletal symptoms among post office pensioners in England,59 Sobti et al.
found that having driven more than 4 hours
per day in previous occupations was common
(15%) among post office employees. About
43% of pensioners reported experiencing
knee pain or stiffness in the past month, but a
significant association between driving and
578 | Research and Practice | Peer Reviewed | Chen et al.
knee pain was not reported. In a survey of
musculoskeletal pain in 12 groups of newly
hired young workers (median age = 23 years),
Nahit et al.60 examined whether the 1-month
prevalence of knee pain (22%) was related to
daily driving duration. The odds ratio associated with driving 15 minutes or more per day
was found to be nonsignificant (odds ratio
[OR] = 1.0; 95% CI = 0.6, 1.7). Because both
study populations in Sobit et al. and Nahit et
al. consisted of members of occupation
groups with different background risks for
musculoskeletal disorders, the limited variability in driving duration may not have provided the investigators with sufficient power
to detect a significant association between
long driving times and knee pain.
Our finding of a significant association between daily driving duration and knee pain
was in accord with the results of previous
studies. In an earlier study by Jajic et al.,61
a significant increased concentration of
99 m Tc-polyphosphate in bone scans of knee
joints (indicating increased bone rebuild, an
early sign of degenerative changes referred
to as a “preosteoarthrotic condition” in the
report) was found among professional drivers. A recent study by Coggon et al.62
showed an association (OR = 2.3; 95% CI =
1.4, 4.0) between long driving times ( ≥ 4
hours/day) and knee cartilage injuries in a
community-based case–control study. Several survey results63–65 have indicated that
the knee is one of the joints most frequently
injured in motor vehicle accidents. Studies of
musculoskeletal injuries among bus drivers
also showed that injuries to the lower extremities, including the knees, were the most
common musculoskeletal injuries.66 In a subset of 893 drivers who had information on
previous motor vehicle accident–related
knee injuries, we found previous knee injury
to be strongly associated with knee pain
prevalence (adjusted OR = 6.54; 95% CI =
1.62, 26.4). However, after we adjusted for
previous motor vehicle accident–related
knee injuries, long driving times were still
significantly associated with increased knee
pain prevalence (adjusted OR = 2.30; 95%
CI = 1.18, 4.47).
We further examined the associations between knee pain and vehicle characteristics to
provide some mechanic implications of our
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
findings. Our analyses yielded no consistent
association between knee pain and vehicle
manufacturers or engine sizes. Interestingly,
the crude knee pain prevalence among drivers who operated vehicles made in 1990 or
earlier was 25%, but only 18% among those
who operated vehicles made after 1990. This
association was marginally significant (adjusted OR = 1.63; P = .07) after we controlled
for all variables retained in the final multiple
logistic model (Table 2). In a previous exposure assessment study on back disorders,52
we found that 37% of the taxicabs in Taipei
had manual transmissions. Presumably, most
taxicabs made before 1990 had manual
transmissions; more repetitive motion in the
lower extremities is required when driving
such vehicles. Nevertheless, it is noteworthy
that the association between duration of daily
driving and knee pain remained statistically
significant (OR = 2.63; 95% CI = 1.42, 4.88).
Regardless of the potential measurement errors of this rough classification, our analyses
imply that in addition to repetitive motions of
lower extremities, the contribution of other
physical factors associated with prolonged
driving (e.g., strenuous knee postures, relative
immobilization of the left knee when using an
automatic transmission) should be investigated in future studies.
Other physical and psychosocial factors associated with knee pain in our study conform
to previous observations. Physical activity
during both work and leisure time has been
found to be a risk factor for developing knee
osteoarthritis.67,68 Many studies have identified psychosocial variables, such as selfperceived job stress, job dissatisfaction, and
mental health (all included in our study), that
are important determinants of knee pain in
both occupational and community settings.4,19,69–71 Another interesting finding,
which is probably related to psychosocial
context as well, was that independent drivers
had slightly higher knee pain prevalence
(21%) than did drivers in a cooperative practice (18%) or those affiliated with taxicab
companies (17%). This difference was statistically significant in the multiple logistic regression, which suggests that factors other than
the physical and psychosocial variables retained in our model may be more common
among independent drivers and may account
for their higher knee pain prevalence. We
posited that the social-network function (e.g.,
social support) could partially explain this observation, because independent taxi drivers
may be more isolated than other taxi drivers.
Detailed analysis of data from the Job Content Questionnaire is needed to support this
speculation.
We wanted to be cautious about the observed association between driving and knee
pain. Therefore, we took the following steps
to rule out plausible alternative explanations
of our finding. Because a few studies had
found that clustering of musculoskeletal
symptoms is very common,59,70,72 we first examined whether the reported knee pain was
merely a co-symptom of other more frequent
musculoskeletal complaints, such as pain in
the low back (51%), neck (50%), and shoulder (30%) in this group of taxi drivers. After
adding these 3 variables into the final multiple logistic regression, we found that our
data did support the clustering of musculoskeletal symptoms. Taxi drivers who reported musculoskeletal pain in these 3 sites
had significantly higher knee pain prevalence, with a corresponding adjusted odds
ratio of 1.88 (95% CI = 1.35, 2.65) for
those who had low back pain, 1.90 (95%
CI = 1.35, 2.71) for those who had neck
pain, and 1.71 (95% CI = 1.22, 2.41) for
those who had shoulder pain. However,
even after we adjusted for the clustering of
musculoskeletal symptoms, the association
between long driving times and knee pain
remained statistically significant (adjusted
OR = 2.41; 95% CI = 1.28, 4.50).
Our sensitivity analysis73 (data not shown)
was intended to examine the likelihood that
our analyses had missed an important confounder not provided by the TDHS data. The
sensitivity analysis was conducted to determine how severe an unmeasured confounder
would have to be to affect our results. For a
presumably confounded odds ratio to be depressed from 2.52, for example, to 1.50, we
would have had to miss an unmeasured confounder. However, such a confounder either
must be related to long driving times ( > 6
hours/day) with an odds ratio greater than 3
and associated with knee pain with an odds
ratio of 4 or greater or must be related to
long driving times with an odds ratio of 2 and
April 2004, Vol 94, No. 4 | American Journal of Public Health
associated with knee pain with an odds ratio
greater than 5. Because no such strong factors have ever been documented, and because our association had been adjusted for a
large set of variables retained in the multiple
logistic model, we considered the odds of
having missed such important factors to be
small.
As a secondary analysis of existing data,
our study had several limitations. First, the
TDHS baseline data depended on subjective
reporting to estimate the frequency of musculoskeletal disorders. No further objective information was available on the nature of the
reported knee pain, such as the sidedness of
knee complaints and the clinical significance
of the observed association between driving
and knee pain. Future studies need to include
these distinctions, especially when investigating knee pain in relation to early knee osteoarthritis and the resultant disability among
professional drivers.
Second, our study may have been limited
by shortcomings of the cross-sectional design.
Although we employed a widely used occupational study questionnaire to measure the
prevalence of knee pain, the Nordic musculoskeletal questionnaire does not include detailed items that assess severity of musculoskeletal symptoms. In a small subset of 319
drivers who were administered questionnaire
items on severity of musculoskeletal complaints, 61% of those who had knee pain recalled that they had lost at least 1 day of
work in the past year because of knee pain.
The average number of lost workdays likely
related to knee pain was 4.4 days (range:
1–30 days). Because of the study’s crosssectional design, it is therefore arguable that
the TDHS baseline data may overrepresent
cases of knee pain with relatively longer
symptomatic duration (and probably with less
severe underlying knee joint disorders). Counteracting this length-biased sampling is the
healthy worker effect, which may either have
excluded former drivers who had more severe knee pain (and therefore were forced to
retire or quit) from the TDHS baseline data
or led to changes of driving duration among
symptomatic drivers who remained in the
taxicab business. A prospective study should
provide a more appropriate design to address
these complexities.
Chen et al. | Peer Reviewed | Research and Practice | 579
 RESEARCH AND PRACTICE 
CONCLUSIONS
Our exploratory analyses of the TDHS
baseline data revealed a strong and robust association between long driving times and
knee pain. The public health impact of workrelated knee pain among professional drivers
could be substantial. For this reason, findings
from our cross-sectional study need to be
replicated in longitudinal studies and in biomechanical studies that examine the nature
and the mechanisms of knee pain and its relationship with early osteoarthritis.
About the Authors
Jiu-Chiuan Chen, Jack T. Dennerlein, and David C. Christiani are with the Occupational Health Program, Department of Environmental Health, Harvard School of Public
Health, Boston, Mass. Tung-Sheng Shi and Chiou-Jong
Chen are with the Institute of Occupational Safety and
Health, Council of Labor Affairs, Taipei, Taiwan. Yawen
Cheng is with the Department of Public Health, College of
Medicine, National Cheng-Kung University, Tainan, Taiwan. Wushou P. Chang is with the Institute of Environmental Health Science, National Yang-Ming University,
Taipei. Louis M. Ryan is with the Department of Biostatistics, Harvard School of Public Health.
Requests for reprints should be sent to David C. Christiani, Occupational Health Program, Dept of Environmental Health, Harvard School of Public Health, Bldg I, Rm
1402, 665 Huntington Ave, Boston, MA 02115 (e-mail:
[email protected]).
This article was accepted May 23, 2003.
Contributors
All of the authors conceptualized the study and interpreted the results. J. C. Chen, W. P. Chang, and Y. Cheng
developed the survey instrument. J. C. Chen performed
the analysis and led the writing of the article. L. M.
Ryan supervised the data analysis. W. P. Chang and C. J.
Chen were the principal investigators of the Taxi Drivers’ Health Study.
Acknowledgments
The first phase of the Taxi Drivers’ Health Study was
jointly funded by the Institute of Occupational Safety
and Health, Council of Labor Affairs, Taiwan, and the
Harvard–Liberty Mutual Program.
The authors thank Mei-Shu Wang, Michelle Yen, and
Yu-Ping Wu for their assistance in both data collection
and research administration. The authors are grateful
for Queenie E. Lee, Chi-Chia Liang, and Zai-Jung
Huang for their contribution to data management.
References
1. Cunningham LS, Kelsey JL. Epidemiology of musculoskeletal impairments and associated disability. Am J
Public Health. 1984;74:574–579.
2. Andersen RE, Crespo CJ, Ling SM, Bathon JM,
Bartlett SJ. Prevalence of significant knee pain among
older Americans: results from the Third National
Health and Nutrition Examination Survey. J Am Geriatr
Soc. 1999;47:1435–1438.
3. Peat G, McCarney R, Croft P. Knee pain and osteoarthritis in older adults: a review of community burden and current use of primary health care. Ann
Rheum Dis. 2001;60:91–97.
4. Bergenudd H, Nilsson B, Lindgarde F. Knee pain
in middle age and its relationship to occupational work
load and psychosocial factors. Clin Orthop. 1989;(245):
210–215.
5. Allander E. Prevalence, incidence, and remission
rates of some common rheumatic diseases or syndromes. Scand J Rheumatol. 1974;3:145–153.
6. Odding E, Valkenburg HA, Algra D,
Vandenouweland FA, Grobbee DE, Hofman A. Associations of radiological osteoarthritis of the hip and
knee with locomotor disability in the Rotterdam
Study. Ann Rheum Dis. 1998;57:203–208.
16. Robertsson O, Dunbar MJ, Knutson K, Lidgren L.
Past incidence and future demand for knee arthroplasty in Sweden: a report from the Swedish Knee
Arthroplasty Register regarding the effect of past and
future population changes on the number of arthroplasties performed. Acta Orthop Scand. 2000;71:
376–380.
17. Symmons DP. Knee pain in older adults: the latest
musculoskeletal “epidemic.” Ann Rheum Dis. 2001;60:
89–90.
18. Urwin M, Symmons D, Allison T, et al. Estimating
the burden of musculoskeletal disorders in the community: the comparative prevalence of symptoms at different anatomical sites, and the relation to social deprivation. Ann Rheum Dis. 1998;57:649–655.
19. O’Reilly SC, Muir KR, Doherty M. Knee pain and
disability in the Nottingham community: association
with poor health status and psychological distress. Br J
Rheumatol. 1998;37:870–873.
20. Tennant A, Fear J, Pickering A, Hillman M, Cutts
A, Chamberlain MA. Prevalence of knee problems in
the population aged 55 years and over: identifying the
need for knee arthroplasty. BMJ. 1995;310:
1291–1293.
7. Claessens AA, Schouten JS, van den Ouweland
FA, Valkenburg HA. Do clinical findings associate with
radiographic osteoarthritis of the knee? Ann Rheum Dis.
1990;49:771–774.
21. Miller ME, Rejeski WJ, Messier SP, Loeser RF.
Modifiers of change in physical functioning in older
adults with knee pain: the Observational Arthritis
Study in Seniors (OASIS). Arthritis Rheum. 2001;45:
331–339.
8. Zhang N, Shi Q, Zhang X. An epidemiological
study of knee osteoarthritis [in Chinese]. Chin J Intern
Med [Zhonghua Nei Ke Za Zhi]. 1995;34:84–87.
22. McAlindon TE, Cooper C, Kirwan JR, Dieppe PA.
Determinants of disability in osteoarthritis of the knee.
Ann Rheum Dis. 1993;52:258–262.
9. Pountain G. Musculoskeletal pain in Omanis, and
the relationship to joint mobility and body mass index.
Br J Rheumatol. 1992;31:81–85.
23. McAlindon TE, Cooper C, Kirwan JR, Dieppe PA.
Knee pain and disability in the community. Br J
Rheumatol. 1992;31:189–192.
10. Chaiamnuay P, Darmawan J, Muirden KD, Assawatanabodee P. Epidemiology of rheumatic disease
in rural Thailand: a WHO-ILAR COPCORD study.
Community Oriented Programme for the Control of
Rheumatic Disease. J Rheumatol. 1998;25:1382–1387.
24. Davis MA, Ettinger WH, Neuhaus JM, Mallon KP.
Knee osteoarthritis and physical functioning: evidence
from the NHANES I Epidemiologic Followup Study.
J Rheumatol. 1991;18:591–598.
11. Darmawan J, Valkenburg HA, Muirden KD,
Wigley RD. Epidemiology of rheumatic diseases in
rural and urban populations in Indonesia: a World
Health Organisation International League Against
Rheumatism COPCORD study, stage I, phase 2. Ann
Rheum Dis. 1992;51:525–528.
12. Zhang Y, Xu L, Nevitt MC, et al. Comparison of
the prevalence of knee osteoarthritis between the elderly Chinese population in Beijing and whites in the
United States: the Beijing Osteoarthritis Study. Arthritis
Rheum. 2001;44:2065–2071.
13. American Academy of Orthopaedic Surgeons. 6
million a year seek medical care for knees [press release]. Rosemont, Ill: American Academy of Orthopaedic Surgeons, Dept of Research and Scientific Affairs;
March 1997.
Human Participant Protection
14. Recommendations for the medical management of
osteoarthritis of the hip and knee: 2000 update. American College of Rheumatology Subcommittee on Osteoarthritis Guidelines. Arthritis Rheum. 2000;43:
1905–1915.
The study protocol was approved by the human subjects committee of the Harvard School of Public
Health and by the institutional review board of the
Taipei Veterans General Hospital, Taipei, Taiwan.
15. HCUPnet. Healthcare Cost and Utilization Project.
Rockville, Md: Agency for Healthcare Research and
Quality. Available at: http://www.ahrq.gov/data/hcup/
hcupnet.htm. Accessed February 23, 2004.
580 | Research and Practice | Peer Reviewed | Chen et al.
25. Jordan J, Luta G, Renner J, Dragomir A, Hochberg
M, Fryer J. Knee pain and knee osteoarthritis severity
in self-reported task specific disability: the Johnston
County Osteoarthritis Project. J Rheumatol. 1997;24:
1344–1349.
26. Jordan JM, Luta G, Renner JB, et al. Self-reported
functional status in osteoarthritis of the knee in a rural
southern community: the role of sociodemographic factors, obesity, and knee pain. Arthritis Care Res. 1996;9:
273–278.
27. Hart DJ, Doyle DV, Spector TD. Incidence and
risk factors for radiographic knee osteoarthritis in middle-aged women: the Chingford Study. Arthritis Rheum.
1999;42:17–24.
28. Cooper C, Snow S, McAlindon TE, et al. Risk factors for the incidence and progression of radiographic
knee osteoarthritis. Arthritis Rheum. 2000;43:
995–1000.
29. Spector TD, Dacre JE, Harris PA, Huskisson EC.
Radiological progression of osteoarthritis: an 11-year
follow-up study of the knee. Ann Rheum Dis. 1992;51:
1107–1110.
30. Ettinger WH, Davis MA, Neuhaus JM, Mallon KP.
Long-term physical functioning in persons with knee
osteoarthritis from NHANES, I: effects of comorbid
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
medical conditions. J Clin Epidemiol. 1994;47:
809–815.
31. Guccione AA, Felson DT, Anderson JJ, et al. The
effects of specific medical conditions on the functional
limitations of elders in the Framingham Study. Am J
Public Health. 1994;84:351–358.
32. O’Mahony D, Foote C. Prospective evaluation
of unexplained syncope, dizziness, and falls among
community-dwelling elderly adults. J Gerontol Series A
Biol Sci Med Sci. 1998;53:M435–M440.
33. Northridge ME, Nevitt MC, Kelsey JL. Nonsyncopal falls in the elderly in relation to home environments. Osteoporosis Int. 1996;6:249–255.
34. Lau EM, Woo J, Lam D. Neuromuscular impairment: a major cause of non-syncopal falls in elderly
Chinese. Public Health. 1991;105:369–372.
35. Arden NK, Nevitt MC, Lane NE, et al. Osteoarthritis and risk of falls, rates of bone loss, and osteoporotic
fractures. Study of Osteoporotic Fractures Research
Group. Arthritis Rheum. 1999;42:1378–1385.
36. Kosorok MR, Omenn GS, Diehr P, Koepsell TD,
Patrick DL. Restricted activity days among older adults.
Am J Public Health. 1992;82:1263–1267.
37. Hadler NM. Knee pain is the malady—not osteoarthritis. Ann Intern Med. 1992;116:598–599.
38. Berg M, Sanden A, Torell G, Jarvholm B. Persistence of musculoskeletal symptoms: a longitudinal
study. Ergonomics. 1988;31:1281–1285.
39. Zitting P, Vanharanta H. Why do we need more
information about the risk factors of the musculoskeletal pain disorders in childhood and adolescence? Int J
Circumpolar Health. 1998;57:148–155.
40. Von Korff M. Studying the natural history of back
pain. Spine. 1994;19(18 suppl):2041S–2046S.
41. Croft PR, Lewis M, Papageorgiou AC, et al. Risk
factors for neck pain: a longitudinal study in the general population. Pain. 2001;93:317–325.
42. Burton AK, Clarke RD, McClune TD,
Tillotson KM. The natural history of low back pain in
adolescents. Spine. 1996;21:2323–2328.
43. Macfarlane GJ, Hunt IM, Silman AJ. Predictors of
chronic shoulder pain: a population based prospective
study. J Rheumatol. 1998;25:1612–1615.
44. Anderson D, Raanaas R. Psychosocial and physical factors and musculoskeletal illness in taxi drivers.
In: McCabe PT, Hanson MA, Robertson SA, eds. Contemporary Ergonomics 2000. London, England: Taylor
& Francis; 2000:322–327.
45. Kuorinka I, Jonsson B, Kilbom A, et al. Standardized Nordic questionnaires for the analysis of musculoskeletal symptoms. Appl Ergonomics. 1987;18:
233–237.
46. Taiwan Institute of Occupational Safety and
Health (IOSH). Professional drivers are suffering from
significant job stress and musculoskeletal pain [press release]. Taipei, Taiwan: Council of Labor Affairs; 1999.
47. Taiwan IOSH. Survey of Employees’ Perception of
Safety and Health in the Work Environment in 1998 Taiwan. Taipei, Taiwan: Taiwan IOSH, Council of Labor
Affairs; 1999.
48. Taiwan IOSH. Occupational Safety Evaluation of
Taxi Drivers in Taipei City. Taipei, Taiwan: Taiwan
IOSH, Council of Labor Affairs; 2000.
49. Chang WP. The Periodic Medical Examination Program for Taipei Taxi Drivers. Taipei, Taiwan: Bureau of
Transportation, Taipei City Government; 2000.
50. Palmer KT, Haward B, Griffin MJ, Bendall H, Coggon D. Validity of self-reported occupational exposures
to hand-transmitted and whole-body vibration. Occup
Environ Med. 2000;57:237–241.
51. Wiktorin C, Vingard E, Mortimer M, et al. Interview versus questionnaire for assessing physical loads
in the population-based MUSIC–Norrtalje Study. Am J
Ind Med. 1999;35:441–455.
52. Chen JC, Chang WR, Shih TS, et al. Predictors of
whole-body vibration among urban taxi drivers. Ergonomics. 2003;46:1075–1090.
53. Baron S, Hales T, Hurrell J. Evaluation of symptom surveys for occupational musculoskeletal disorders.
Am J Ind Med. 1996;29:609–617.
54. Cheng Y, Luh WM, Guo YL. Reliability and validity of the Chinese version of the Job Content Questionnaire in Taiwanese workers. Int J Behav Med. 2003;10:
15–30.
55. Fuh JL, Wang SJ, Lu SR, Juang KD, Lee SJ. Psychometric evaluation of a Chinese (Taiwanese) version
of the SF-36 health survey amongst middle-aged
women from a rural community. Qual Life Res. 2000;9:
675–683.
56. Lemeshow S, Hosmer DW Jr. A review of
goodness-of-fit statistics for use in the development
of logistic regression models. Am J Epidemiol. 1982;
115:92–106.
hazardous environmental conditions. Ergonomics.
1987;30:335–342.
67. Imeokparia RL, Barrett JP, Arrieta MI, et al. Physical activity as a risk factor for osteoarthritis of the
knee. Ann Epidemiol. 1994;4:221–230.
68. McAlindon TE, Wilson PW, Aliabadi P, Weissman
B, Felson DT. Level of physical activity and the risk of
radiographic and symptomatic knee osteoarthritis in
the elderly: the Framingham study. Am J Med. 1999;
106:151–157.
69. Creamer P, Lethbridge-Cejku M, Costa P, Tobin
JD, Herbst JH, Hochberg MC. The relationship of anxiety and depression with self-reported knee pain in the
community: data from the Baltimore Longitudinal
Study of Aging. Arthritis Care Res. 1999;12:3–7.
70. Nahit ES, Pritchard CM, Cherry NM, Silman AJ,
Macfarlane GJ. The influence of work-related psychosocial factors and psychological distress on regional musculoskeletal pain: a study of newly employed workers.
J Rheumatol. 2001;28:1378–1384.
71. Bergenudd H, Nilsson B. The prevalence of locomotor complaints in middle age and their relationship
to health and socioeconomic factors. Clin Orthop.
1994;(308):264–270.
72. Magnusson ML, Pope MH, Wilder DG, Areskoug
B. Are occupational drivers at an increased risk for developing musculoskeletal disorders? Spine. 1996;21:
710–177.
73. Greenland S. Basic methods for sensitivity analysis
of biases. Int J Epidemiol. 1996;25:1107–1116.
57. Shao J, Tu D. The Jackknife and Bootstrap. New
York, NY: Springer Verlag; 1995.
58. Taiwan Ministry of Transportation and Communications (MOTC). Survey Report on Taxi Service Business.
Taipei, Taiwan: Taiwan MOTC; 2000.
59. Sobti A, Cooper C, Inskip H, Searle S, Coggon D.
Occupational physical activity and long-term risk of
musculoskeletal symptoms: a national survey of post
office pensioners. Am J Ind Med. 1997;32:76–83.
60. Nahit ES, Macfarlane GJ, Pritchard CM,
Cherry NM, Silman AJ. Short-term influence of mechanical factors on regional musculoskeletal pain: a
study of new workers from 12 occupational groups.
Occup Environ Med. 2001;58:374–381.
61. Jajic I, Jelcic I, Schwarzwald M, Delimar N. Detection of preosteoarthrotic condition of the knee joint in
professional drivers with 99mTc-polyphosphate. Acta
Med Iugosl. 1976;30:295–305.
62. Coggon D, Baker P, Reading I, Barrett D,
McLaren M, Copper C. Knee cartilage injury and occupational activities. Paper presented at: The Fourth International Scientific Conference on Prevention of
Work-Related Musculoskeletal Disorders; Amsterdam,
The Netherlands; September 30–October 4, 2001.
63. Schelp L, Ekman R. Road traffic accidents in a
Swedish municipality. Public Health. 1990;104:55–64.
64. Nagel DA, Burton DS, Manning J. The dashboard
knee injury. Clin Orthop. 1977;(126):203–208.
65. Atkinson T, Atkinson P. Knee injuries in motor
vehicle collisions: a study of the National Accident
Sampling System database for the years 1979–1995.
Accid Anal Prev. 2000;32:779–786.
66. Barak D, Djerassi L. Musculoskeletal injuries
among bus drivers due to motor vehicle accidents and
April 2004, Vol 94, No. 4 | American Journal of Public Health
Chen et al. | Peer Reviewed | Research and Practice | 581
 RESEARCH AND PRACTICE 
Estimating Capacity Requirements for Mental Health Services
After a Disaster Has Occurred: A Call for New Data
| Carole E. Siegel, PhD, Eugene Laska, PhD, and Morris Meisner, PhD
In the chaotic aftermath of a disaster, authorities are faced with the need to provide an extensive array of services to the affected population. Such a situation occurred after the
terrorist attacks of September 11, 2001, when
mental health and other related support systems mobilized to deliver services to persons
who were psychologically or psychiatrically
affected by the events.
Planning efforts required estimates of both
anticipated mental health needs and the capacity required to respond to these needs.
The New York State Office of Mental Health
(NYSOMH), in conjunction with researchers
from Columbia University’s Joseph P. Mailman
School of Public Health, conducted a mental
health needs assessment. Their report focused
on persons suffering from posttraumatic stress
disorder (PTSD). For this group, they estimated the breadth of the need, the likely
number of services required, and sources of
payment for care.1–3 Data were presented regarding the current capacity of the New York
State mental health specialty sector, and a
general formula to estimate the service capacity that would be required after a disaster appears in an appendix to that report.
The rationale for the formula and its formulation are presented in this article. By envisioning the formula being applied to cover the
largest population likely to seek help, information that is currently available to numerically
calculate the value of the formula was identified, as were gaps that limit the ability to provide realistic estimates. An examination of
these gaps has led to recommendations for
local and national data collection that would
enhance the potential for appropriate capacity
planning following disasters. The formula,
when applied in limited scope, has immediate
utility for estimating the service requirements
of priority populations. An example of this use
is given for persons living below 110th Street
in Manhattan and who experienced PTSD
after the September 11 disaster.
Objectives. We sought to estimate the extended mental health service capacity requirements of persons affected by the September 11, 2001, terrorist attacks.
Methods. We developed a formula to estimate the extended mental health service
capacity requirements following disaster situations and assessed availability of the information required by the formula.
Results. Sparse data exist on current services and supports used by people with
mental health problems outside of the formal mental health specialty sector. There
also are few systematically collected data on mental health sequelae of disasters.
Conclusions. We recommend research-based surveys to understand service usage in
non–mental health settings and suggest that federal guidelines be established to promote uniform data collection of a core set of items in studies carried out after disasters. (Am J Public Health. 2004;94:582–585)
METHODS
Basic Formula
Extended capacity after a disaster has occurred is defined as the service capacity above
the usual service delivery levels required; extended capacity is defined in terms of units of
service. These units may be converted to monetary or staff requirements. Extended capacity
may be required to provide services to disaster
victims who experience emotional distress that
is severe enough to require a mental health intervention. Victims could include both persons
not currently receiving mental health services
(new, or incidence, cases) and persons already
receiving services whose problems have been
exacerbated by the disaster (old, or prevalence,
cases). New cases will require services at some
rate to be agreed upon, whereas old cases may
require services in addition to those they currently receive as a result of exposure to the disaster.4 The extended capacity requirement for
each group is simply the product of the number
of persons in the group and the number of anticipated services required as a result of the disaster. The total extended capacity requirement
is the sum of the requirements of the 2 groups.
More formally, extended capacity, ∆C, is
based on the number of new cases requiring
services postdisaster, Nnew; the number of old
cases requiring additional services postdisaster,
Nold; the average number of services per per-
582 | Research and Practice | Peer Reviewed | Siegel et al.
son required by new cases, Rnew; and the average number of additional services required per
person for old cases, R+old. The equation is as
follows:
(1)
∆C=Nnew Rnew +Nold R+old.
A conservative estimate of extended capacity
assumes that old cases will not require any additional services and that new cases will receive
services at current or lower-than-current levels.
That is, Rnew ≤Rold, where Rold is the current service delivery rate, R+old =0, and ∆C=Rnew Nnew.
At the other extreme, old cases may require
new services, and new cases may require services at a rate that is higher than the exacerbated rate of old cases. That is, Rnew >Rold +
R+old, and R+old >0. All other cases are intermediate to these 2 cases.
Range of Possibilities
Time frame. Service requirements will differ
in the acute and postacute phases in the aftermath of a disaster. New cases may emerge
over time, whereas distress may abate in some
persons.
Population groups. Different population
groups will have different diagnostic and careseeking patterns, leading to different service requirements. Population groups can be defined
in terms of geographical areas or exposure to
the disaster (e.g., “first responders,” adults living
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
closest to the scene of the disaster). They could
be further classified demographically (e.g., by
age group, by racial/ethnic categories).
Diagnoses/disorders. Victims of disasters are
at risk for experiencing a gamut of mental
health disorders ranging in type and severity.
For new cases, the disorders expected to occur
after a disaster include, but are not limited to,
acute stress disorder, PTSD, depression, anxiety, panic disorder, and traumatic grief. Diagnoses for existing cases cover all their current
diagnoses and possibly new ones similar to
those of the new cases.
Services. The types of services that will be required are diagnostic specific and are likely to
include assessments, crisis counseling, psychoeducation, psychotherapy, and pharmacotherapy. The number of services required to treat
cases will change over time, with higher rates
of treatment expected in the initial phases following the disaster.
Sector/service venue. It is well known that
even under normal times, persons experiencing
mental distress may seek help in service venues
other than the organized mental health sector.5,6 In particular, persons who experience
nonpsychotic disorders are quite likely to seek
services first from non–mental health specialists. For example, persons with depressive disorders will often turn to primary care physicians,7
and persons from particular cultural groups
may first seek services from traditional healers.8
Many who experience distress at subclinical
threshold levels seek services from the clergy or
self-help groups. Regier5 first noted this collection of providers from whom persons experiencing mental distress seek help and labeled
them as members of a de facto mental health
system. More than likely, these same providers
will be approached by disaster victims for help.
To model the extended mental health service
requirement following a disaster, data are required on how many of those from different
population and clinical groups will seek and receive services from these de facto venues.
Comprehensively, the sectors where persons
are expected to seek services include mental
health services provided by programs funded,
certified, or operated by state offices of mental
health; the Department of Veterans Affairs (VA);
general hospital emergency rooms; and other
non–mental health sectors such as schools, social service agencies, and family agencies (re-
ferred to hereafter as “non–mental health sectors”). Services may also be provided by individuals who are mental health specialists in private
practices that are not part of a state office of
mental health (“mental health specialists”); primary care physicians in private practices, clergy,
and self-help groups (“other specialists”); and
others in nonformal settings.
Persons with existing severe disorders or persons who experience distress that reaches clinical diagnostic thresholds are likely to be served
by a state office of mental health, the VA, or
mental health individuals. Persons with disorders that do not reach threshold levels are
likely to use the remaining sectors. They may
well account for the bulk of the new service
needs, especially in the acute phase after a disaster has occurred. The provider list could be
expanded to include sectors that serve persons
with alcohol and substance abuse disorders, depending on the scope and purview of the capacity assessment.
General Formula
The general formula sums the basic formula
in equation 1 over the range of possibilities to
provide an estimate of total extended capacity
requirements, ∆C. It relates to a length of time
after the disaster, T; service sectors, S; population groups that use these sectors, g(S); disorders of the groups that seek services in the sector, d(g(S)); and units of service type u required
for a disorder, u(d(g(S))). For a fixed T, total extended capacity requirements are expressed as
(2)
∆C(T)=Σ u Σ d Σ g Σ S ∆C (T, S, g, d, u),
where we have suppressed the notation indicating the sequential dependencies of groups on
sectors, diagnoses on groups, and services on
diagnoses.
To avoid double counting, the assumption is
made that population groups do not overlap
and that service requirements are distinct
across diagnoses and sectors. (Note, however,
that when the formula is used in limited scope,
any population group can be singled out and its
extended capacity requirement estimated.) The
formula also assumes that no 2 diagnoses are
associated with the same service requirement.
Although a particular service requirement of a
person with comorbidities might be counted
with respect to each diagnosis, this will not hap-
April 2004, Vol 94, No. 4 | American Journal of Public Health
pen if the range of service requirements is restricted to those closely connected to the diagnosis. If this is not possible, the comorbid condition itself could be introduced as a diagnosis
and the service requirement could be attributed to the comorbid diagnosis.
It is also possible that service requirements
cannot be distinctly ascribed to a sector. Use
of multiple sectors for similar needs has been
documented for veterans. It has been observed that a small percentage ( < 5%) of veterans use similar type services from both VA
and non–federally funded providers (C. Siegel,
PhD, S. Lin, PhD, E. Laska, PhD, unpublished
data, 2003). If estimates of the usage of multiple venues for similar services were available,
then a model-based adjustment to the total estimate could be made.9
Although the estimate of extended capacity
does not depend on current service capacities,
the ability to provide the services required
clearly does. An important step is to estimate
whether extended capacity requirements can
be met, the projected shortfalls, and the concomitant budgetary and staffing requirements.10
Limited Versions
Most likely, however, estimates of extended
capacity will be desired for high-priority situations that limit the range of possibilities covered in the general formula. For example, an
estimate may be required of the extended capacity needs of a specific population group
for special types of disorders with their service requirements, delivered in the service
sectors in which these services are apt to be
delivered. To obtain estimates limited in
scope, ∆C (T, S, g, d, u) is summed over specific subsets that delineate the coverage of the
capacity estimate. For example, if a sector S*
requires an estimate of its total extended capacity requirement for T months after the disaster, it is Σ u Σ d Σ g ∆C (T, S*, g, d, u). An estimate of high and immediate priority might
consider the extended capacity required to
treat, in the formal mental health sector S*,
within the first 6 months postdisaster, first responders (population group g*) experiencing
PTSD (say disorder d*), where they would require within 6 months an amount u* of specialized treatment u*. The estimate formula is
∆C (6, S*, g*, d*, u*). Estimates restricted to
geographical areas most directly affected by
Siegel et al. | Peer Reviewed | Research and Practice | 583
 RESEARCH AND PRACTICE 
Little of the information required to estimate
extended capacity is available or can be extrapolated from studies on disasters before September 11. Recent studies conducted after September 1111–13 do provide some new data that are
useful for budget justifications and planning for
increased staffing requirements.
Nold may be obtained directly for some sectors from utilization data related to recent time
frames before the disaster. These data are available in New York State for the NYSOMH and
for the VA. Some sector usage data can be extrapolated from the Epidemiologic Catchment
Area Study5 and the National Comorbidity Survey.6 Both studies provide an estimate of the
proportion of persons with mental disorders
who seek services in these other sectors, and
the latter study provides some limited data on
actual utilization.
Studies of other disasters and studies
mounted soon after a disaster has occurred do
provide information on the risk of a disease,
given exposure to the new disaster. The risk
times the population size is Nnew.
There are few estimates of R+old and Rnew
available from studies conducted of other disasters. R+old could be informed by the current service delivery rate=Cold/Nold, where Cold =current capacity. Local providers and other key
informants can be asked to estimate exacerbation rates, but they might find it difficult to
make guesses specific to diagnostic groups or
service types. Consensus approaches would
need to be used to avoid overinflated estimates.
who experienced PTSD. Galea et al.11 estimated the percentage of new cases with PTSD
among this group. They randomly sampled
adult persons from the area 5 to 8 weeks after
September 11 and administered a telephone interview to assess their psychiatric symptoms.
They found that 7.5% of the sample had symptoms severe enough to classify them as having
PTSD related to the September 11 attacks.
Persons with this serious diagnosis would
most likely require services within the formal
mental health specialty sector, but prior studies
suggest that not all of these persons will seek
services. Boscarino et al.12 reported that 19.4%
of those interviewed in the Galea sample had
mental health visits, but estimates for utilization
specific to PTSD and specific to symptoms related to the disaster were not provided. Kessler
et al.14 reported that 28% of persons with
PTSD sought services, and we used this higher
estimate in our calculation.
The population size of adult persons living in
Manhattan below 110th street is 919000. Assuming no exacerbation of symptoms in old
cases, a conservative estimate of extended capacity requirements is based on the number of
new cases, Nnew, that will emerge and their rate,
Rnew, of service usage. The calculation of Nnew is
.075×.28×919000=19299. Jack and Glied3
concluded that in a 6-month period, treatment
among those with a diagnosis of PTSD should
consist on average of 7 outpatient visits and 6
monthly medication visits at a cost in New York
City of approximately $1500 per person.
Using these data for Rnew in the conservative
version of the formula provides an estimate of
an extended capacity requirement of the formal mental health sector for this population/
diagnostic group in the 6-month period after
September 11 of Rnew Nnew =135093 PTSD
visits and 115794 medication visits, with a
total cost of $28948500.
A Valuation of the Formula
CONCLUSIONS
One of the most likely estimates to be required immediately after a disaster is the capacity requirements for populations that are
close to the disaster and that experience severe
emotional distress. Some data have appeared
since September 11 that enable an estimate to
be made of extended capacity requirements for
the New York City adult population living
below 110th Street (close to the disaster site)
Information that is currently available includes epidemiological estimates of mental disorders and mental health service utilization
after disaster incidents,11,15,16 epidemiological information on the incidence of mental disorders
in the general population and the naturalistic
use of the various service sectors predisaster,5,6
and sector-specific administrative service uti-
the attacks would also be of high priority. In
this case, g is held fixed to represent the geographic area, and all else in equation (2) is
summed.
RESULTS
Obtaining Data to Valuate the Formula
584 | Research and Practice | Peer Reviewed | Siegel et al.
lization data sets on pre- and postdisaster utilization (e.g., state data sets, VA data sets,
county-level data sets).
These data, however, neither adequately
cover the scope nor provide the comprehensive
information required to accurately estimate the
full range of extended capacity requirements.
The information available from other disaster
incidents does not span all manifestations nor
all venues in which persons seek help. Further,
these data may be only partially applicable to
those affected by a new disaster because of differences in population demographics and service system characteristics.
Sector-specific data sets on utilization are limited to the organized specialty mental health
services and are unavailable for the sectors
most likely to be used after a disaster has occurred (e.g., general hospital emergency rooms,
non–mental health sectors, mental health individuals, private primary care physicians, other
non–mental health individuals). Data that are
reported on these latter sectors in epidemiological studies conducted to date5,6 provide information on the likelihood of using the sector and
offer only limited data on actual utilization, with
estimates based on small sample sizes. Other
administrative data sets on utilization that are
available are payer specific (e.g., Medicaid data
sets, behavioral health care data sets). Although
useful for examining the disaster impact on payers, they are less useful for planning for services
in the locations where they are needed.
Finally, although there are resource data on
the number of persons in a given profession
who are capable of providing mental health services, these numbers are not readily convertible
to estimates of current mental health capacity.
Persons in these professions (e.g., social workers) may already be included in other sector
counts or may not provide mental health services. If they do provide such services, the
amount provided is unknown.
There are 2 classes of information that
would be useful for estimating service requirements related to disasters. The first is current
service usage of the various sectors of the de
facto mental health system. This information
would enable natural pathways to care to be
identified, current capacities to be documented
better, and multiple use of sectors to be understood better. The second class of information is
data that are collected more systematically
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
about future disaster incidents and their mental
health sequelae. These 2 data sets would enable better modeling of needs that would
emerge after a disaster occurs.
One effective way to collect sector usage data
would be through a 2-part survey, a provider
inventory and a survey of usage of a provider
by persons with mental health problems. The
first part would inventory, within a sector, individuals and agencies that are capable of providing both formal and informal mental health services, producing in effect a resource directory
for that sector. The coverage area of the survey
should coincide with geographical areas that
have been designated as service areas for disaster response. Mounting such a survey in a large
urban area such as New York City would be a
daunting task, but once computerized mechanisms are in place, the resource could serve as
an invaluable management tool if a disaster
were to occur (as well as in normal times). It
would increase the ability to coordinate services
and also provide a basis for estimating training
and recruiting requirements, should enhanced
capacity be required.
The inventory is needed to conduct the second part of the survey in which data are collected to enable estimation of the number and
types of persons with mental health problems
who are seen by each provider.
The NYSOMH Patient Characteristics Survey
provides one approach that could be followed.
Providers of mental health services that are
funded, contracted, or operated by New York
State are surveyed on the characteristics and
services used by persons seen in a representative week. If data are collected during periods
of normalcy, the 1-week counts can be annualized or inflated to other time periods using a
statistical method currently employed by the
NYSOMH.17,18 Analogously, for a particular sector, all providers in the sector could be surveyed using a 1-week time frame to ascertain
the number of persons using their services for
mental problems, their characteristics, and their
service use. For greater precision, screening of
persons seen by the provider for mental disorders could also be part of the survey.
Because full census surveys might not be
feasible, sampling strategies could be employed,
especially if details on the types of mental distress that manifest themselves are to be collected. During times of normalcy, these data
can be adjusted to establish base capacity rates.
If the survey is repeated after a disaster, perhaps at several different time points, more
could be learned about manifesting problems
and the new capacities that have emerged to
deal with them. Data collected in this manner
would facilitate the parsing of government
budget allocations for disaster situations to the
sectors in proportion to the assistance they provide to the population in need.
Other information needs to be extrapolated
from data of studies of disaster situations. Currently, investigators collect data according to
their own protocols, resulting in studies having
limited commonality in data elements. Guidelines are needed for data collection of at least a
core set of items. This could include specific details of the disaster, specified time frames, delineated population groups, specification of problems, specific treatment system variables, time
frames to report duration of service needs, specific outcomes, and bases for cost estimates.
Having such data would facilitate synthesis and
extrapolation to other disaster incidents.
Developing such guidelines will require a
federally sponsored effort and mandates to establish a core set of items to be uniformly collected. With such data, when new disasters
occur, needs assessment models could be used
to relate the nature of the disaster to the nature
and extent of the problems that would be expected to arise and the capacity required to
deal with them.
About the Authors
All of the authors are with the Nathan S. Kline Institute for
Psychiatric Research, Orangeburg, NY.
Requests for reprints should be sent to Carole E. Siegel,
PhD, Epidemiology and Health Services Research Laboratory, Nathan S. Kline Institute for Psychiatric Research,
140 Old Orangeburg Rd, Orangeburg, NY 10962 (e-mail:
[email protected]).
This article was accepted May 15, 2003.
Contributors
C. E. Siegel identified sources for the data required for
this article. All authors contributed to the mathematical
formulation.
Acknowledgments
This project was supported in part by a National Institute of Mental Health grant (3P50 MH51359) to the
Center for the Study of Issues in Public Mental Health
of the Nathan S. Kline Institute for Psychiatric Research,
and in part by the NYSOMH.
The views expressed in this article are strictly those of
the authors and do not represent agency endorsements.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Human Participant Protection
No protocol approval was needed for this study.
References
1. Herman D, Felton C, Susser E. New York State:
Mental Health Needs Assessment Related to Terrorist Attacks in the US. Albany, NY: New York State Office of
Mental Health; 2001.
2. Herman D, Felton C, Susser E. Mental health
needs in New York state following the September 11th
attacks. J Urban Health. 2002;79(3):322–331.
3. Jack K, Glied S. The public costs of mental health response: lessons from the New York City post-9/11 needs
assessment. J Urban Health. 2002;79(3):332–339.
4. Franklin CL, Young D, Zimmerman M. Psychiatric
patients’ vulnerability in the wake of the September 11th
terrorist attacks. J Nerv Ment Dis. 2002;190:833–838.
5. Regier DA, Narrow WE, Rae DS, Manderscheid RW,
Locke BZ, Goodwin FK. The de facto US mental and addictive disorders service system. Epidemiologic catchment
area prospective 1-year prevalence rates of disorders and
services. Arch Gen Psychiatry. 1993;50:85–94.
6. Kessler RC, McGonagle KA, Zhao S, et al. Lifetime
and 12-month prevalence of DSM-III-R psychiatric disorders in the United States. Results from the National
Comorbidity Study. Arch Gen Psychiatry. 1994;51:8–19.
7. Wells KB, Sturm R, Sherbourne CD, Meredith LS,
eds. Caring for Depression. Cambridge, Mass: Harvard
University Press; 1996.
8. Rogler LH, Cortes DE. Help-seeking pathways: a
unifying concept in mental health care. Am J Psychiatry.
1993;150:554–561.
9. Laska EM, Meisner M, Wanderling J, Siegel C. Estimating population size and duplication rates when records cannot be linked. Stat Med. 2003;22:3403–3417.
10. Siegel C, Wanderling J, Laska EM. Coping with disasters: estimation of additional capacity of the mental
health system to meet extended service demands [abstract]. J Ment Health Policy Econ. 2003;6(suppl 1):S46.
11. Galea S, Ahern J, Resnick H, et al. Psychological
sequelae of the September 11 terrorist attacks in New
York City. N Eng J Med. 2002;346:982–987.
12. Boscarino J, Galea S, Ahern J, Resnick H, Vlahov
D. Utilization of mental health services following the
September 11th terrorist attacks in Manhattan, New
York City. Int J Emerg Ment Health. 2002;4(3):143–155.
13. DeLisi LE, Maurizio A, Yost M, et al. A survey of
New Yorkers after the Sept. 11, 2001, terrorist attacks.
Am J Psychiatry. 2003;160:780–783.
14. Kessler RC, Zhao S, Katz SJ, et al. Past-year use of
outpatient services for psychiatric problems in the National
Comorbidity Study. Am J Psychiatry. 1999;156:115–123.
15. North CS, Nixon SJ, Shariat S, et al. Psychiatric
disorders among survivors of the Oklahoma City
bombing. JAMA. 1999;282:755–762.
16. Call JA, Pfefferbaum B. Lessons from the first two
years of Project Heartland, Oklahoma’s mental health
response to the 1995 bombing. Psychiatr Serv. 1999;
50:953–955.
17. Laska E, Meisner M, Siegel C. Estimating the size
of a population from a single sample. Biometrics. 1988;
44:461–472.
18. Laska E, Meisner M, Siegel C. Estimating population size when duplicates are present. Stat Med. 1996;
15:1635–1646.
Siegel et al. | Peer Reviewed | Research and Practice | 585
 RESEARCH AND PRACTICE 
The Economic Burden of Hospitalizations Associated
With Child Abuse and Neglect
| Sue Rovi, PhD, Ping-Hsin Chen, PhD, and Mark S. Johnson, MD, MPH
In 1999, an estimated 826 000 children
were abused or neglected nationally.1 This
represents a victimization rate of 11.8 per
1000 children. It was further estimated that
1100 children died as a result of abuse and
neglect. Such estimates are surely conservative because victims of abuse and neglect are
often not identified, and even if suspected,
abuse and neglect are underreported.2–7 Although the personal costs in pain and suffering to victims and their families cannot be calculated, in this study we assess the burden of
hospitalizations associated with child abuse
and neglect. By demonstrating the substantial
costs of tertiary care for victims of child abuse
and neglect, we can justify increased support
for primary prevention.8,9
Several studies have demonstrated that
hospitalized children who are identified as
abused or neglected have longer hospital
stays, more severe injuries, worse medical
outcomes, and higher hospital charges, and
such children are more likely to die during
the current hospitalization compared with
other hospitalized children.8–12 Most of this
research is based on reviews of medical records conducted in pediatric tertiary care
hospitals with trauma centers. One study
found significant differences of more than
$2000 in daily hospital charges for child
abuse patients compared with other children admitted to a pediatric intensive care
unit, with mean charges for hospitalized victims of $30 684.8 Another study conducted
at a regional pediatric trauma center reported an average hospital charge of
$20 359 for victims of child abuse, and this
was significantly higher than the hospital
charges for other injured children except
those for burn victims.12 To better understand the scope of the problem, we should
look at nationally representative data. However, we know of only 2 studies that relied
on national data, and neither reported on
the costs of hospitalization.2,10
Objectives. This study assessed the economic burden of child abuse–related hospitalizations.
Methods. We compared inpatient stays coded with a diagnosis of child abuse or neglect with stays of other hospitalized children using the 1999 National Inpatient Sample
of the Healthcare Costs and Utilization Project.
Results. Children whose hospital stays were coded with a diagnosis of abuse or neglect were significantly more likely to have died during hospitalization (4.0% vs 0.5%), have
longer stays (8.2 vs 4.0 days), twice the number of diagnoses (6.3 vs 2.8), and double
the total charges ($19266 vs $9513) than were other hospitalized children. Furthermore,
the primary payer was typically Medicaid (66.5% vs 37.0%).
Conclusion. Earlier identification of children at risk for child abuse and neglect might
reduce the individual, medical, and societal costs. (Am J Public Health. 2004;94:586–590)
Our research objective was to determine 1
aspect of the economic burden of child abuse
and neglect on the health care system using a
national probability sample of US community
hospitals. Specifically, we compared children
hospitalized with a diagnostic code of abuse
or neglect with other hospitalized children in
terms of mean hospital charges, length of hospital stay, and the numbers of diagnoses and
deaths during hospitalizations. Although research has demonstrated that diagnostic
codes for child and adult abuse are likely underutilized,13,14 analyses based on these codes
can provide valuable information on the medical response to victims of abuse and neglect.
Thus, for hospitalizations coded with a diagnosis of abuse or neglect, we expect higher
hospital charges, longer hospital stays, and
more comorbidities compared with other hospitalized children, as well as more deaths during hospitalization.
METHODS
We conducted secondary data analyses of
the 1999 Nationwide Inpatient Sample of
the Healthcare Costs and Utilization Project
(NIS-HCUP).15 The sampling design for the
NIS-HCUP was a stratified random sample of
hospitals with all discharges included from
each selected hospital. The 1999 NIS-HCUP
586 | Research and Practice | Peer Reviewed | Rovi et al.
provides data on 7 198 929 hospital inpatient
stays at 984 hospitals in 24 states, thereby
approximating a 20% stratified sample of US
community hospitals. Sample weights are provided in the NIS-HCUP to enable data users
to produce national estimates.
Identification of Cases
All inpatient hospital stays of children
aged 18 years and younger were selected
(n = 1 371 835). Of these, two thirds (64.8%)
had neonatal/maternal diagnoses/procedures.
Because these patients were less likely to
have diagnoses of abuse or neglect, all new
mothers and neonates aged less than 1 day
were omitted, thereby resulting in 636 802
inpatient hospital stays for these analyses.
Among the excluded cases, there were 19
coded with child abuse or neglect: 15 were
neonatal/maternal, and 4 were 0 days old.
To identify cases of child abuse and neglect,
we used diagnostic codes from the International Classification of Diseases, Ninth Revision,
Clinical Modification (ICD-9-CM).16 In 1996,
the single code for child maltreatment syndrome was expanded to better specify the
forms of abuse: physical, sexual, emotional/
psychological, neglect, and shaken infant syndrome (see the list of codes presented in
Table 1).17 The NIS-HCUP provides a primary
diagnosis and up to 14 secondary diagnoses
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
RESULTS
TABLE 1—Estimated 1999 US Hospital Inpatient Stays Coded With a Diagnosis of Child
Abuse or Neglect
Types of Abuse
Percentage With Diagnosis of Abuse
Weighted na
Child abuse, unspecified
Child emotional/psychological abuse
Child neglect (nutritional)
Child sexual abuse
Child physical abuse
Shaken infant syndrome
Other child abuse and neglect
Two or more types of abuse coded
4.6
0.5
16.0
8.9
38.8
21.1
6.5
3.6
217
25
761
426
1849
1008
311
173
ICD-9 Code
99550
99551
99552
99553
99554
99555
99559
Totals
100
4771
Note. ICD-9 = The International Classification of Diseases, Ninth Revision, Clinical Modification.
a
Represents population estimates based on weighted data for n = 966.
for each inpatient stay. We selected inpatient
stays with any of the ICD-9-CM diagnostic
codes of child abuse or neglect as the subset
of cases of abuse-related hospitalizations. Five
cases coded as adult abuse but involving patients younger than 18 years were recoded to
child abuse. Remaining hospitalizations constituted our comparison group.
dures, and total charges are compared and reported for both groups.
Hospital Variables
The location of the hospital (i.e., urban or
rural), whether or not it was a teaching hospital, and region of the country are also reported for both groups.
Analysis
Demographic Variables
Age at the time of admission, gender, race,
and income are presented for both groups
(see Table 2 for breakdown of variables). Age
was provided in years, with zero indicating infants younger than 1 year. For race, Hispanic,
Asian/Pacific Islander, Native American, and
other were collapsed into 1 category called
Other, and because race was not provided for
nearly 20% of patients, we created a separate
category labeled Unknown for inclusion in
analyses. Income is based on 4 categories
representing the median income for the patient’s zip code. Despite caveats about the use
of zip codes as proxies for income,18 this variable is the best approximation available in
these data. Another proxy for income is the
expected primary payer for the hospitalization, which is also reported.
Medical Discharge Variables
Admission source (e.g., emergency vs routine), whether or not the patient died during
hospitalization, the length of hospital stay in
days, the number of diagnoses and proce-
Using sample weights provided in the
NIS-HCUP data, we produced national estimates of hospitalizations coded with a diagnosis of abuse or neglect. Analyses were done
with SAS (SAS Institute Inc, Cary, NC), and
SUDAAN (Research Triangle Institute, Research Triangle Park, NC) was used for calculating variance and assessing statistical significance that takes into account the sampling
design. Unless otherwise specified, only
weighted data are reported. We estimated the
overall percentage of US hospitalizations
coded with a diagnosis of child abuse or neglect, along with a breakdown of the types of
abuse. Statistical comparisons between child
abuse–related and other hospitalizations are
presented; we used χ2 tests for categorical
variables and t tests for means. Significance
tests of total charges and length of stay were
based on analyses that used log transformations to adjust for skew. Odds ratios with confidence intervals are reported when appropriate. One-way analysis of variance was used to
compare total charges between groups and
among the types of abuse.
April 2004, Vol 94, No. 4 | American Journal of Public Health
In the 1999 NIS-HCUP, there were 966
cases of children hospitalized with 1 of the
diagnostic codes for abuse or neglect, providing a national estimate of 4771, or 0.15% of
US hospitalizations of children aged 18 years
and younger, after neonatal or maternal
diagnoses/procedures and neonates younger
than 1 day were omitted. Physical abuse was
coded most often (38.8%), followed by
shaken infant syndrome (21.1%), child neglect (16.0%), sexual abuse (8.9%), child
abuse unspecified (4.6%), and emotional/
psychological abuse (0.5%) (Table 1). Two or
more diagnoses of abuse were coded in
3.6% of cases. Abuse or neglect was the primary diagnosis 40.2% of the time.
Overall, we found significant differences
between hospital stays coded with a diagnosis
of abuse or neglect and other hospital stays in
the demographics of the child/parent, medical
utilization, and hospitals sampled.
Demographically, children whose hospital
stays were coded with a diagnosis of abuse or
neglect tended to be younger on average (2.7
vs 5.2 years; P < .0001) (Table 2). In analyses
not shown, 49.2% of those coded with abuse
or neglect were younger than 1 year compared with 40.8% of those not coded as such.
Nearly one half of hospitalized children were
White, but they represented less than one
third of the group coded as abused or neglected (P < .0001). Hospitalizations of Black
children and those hospitalized without a racial classification were proportionally more
likely to be coded as abused or neglected.
The median income based on the patient’s zip
code indicates that those coded as abused or
neglected were significantly more likely to be
in the lower income categories (P = .0027).
Race continued to discriminate between the
abused and not-abused groups even after we
controlled for income categories. Medicaid
was the expected primary payer for two
thirds (66.5%) of the hospitalizations of the
abused or neglected group compared with almost one third (37.0%) of the hospitalizations
of other groups, and the opposite tendency
was observed for private payers (24.0% vs
54.5%; P < .0001). The gender of the hospitalized children did not differ between the 2
groups.
Rovi et al. | Peer Reviewed | Research and Practice | 587
 RESEARCH AND PRACTICE 
more procedures (1.3 vs 0.8), and had double
the total charges ($19 266 vs $9513).
A higher percentage of hospitalizations
with a diagnosis of child abuse and neglect
were coded in the Midwest (28.7% vs
17.2%) and a lower percentage in the South
(33.9% vs 46.7%). Medical staff at hospitals
located in urban areas were significantly
more likely to have coded abuse or neglect
compared with hospitals located in rural areas
(P = .0029); and teaching hospitals were
nearly 3 times more likely to have coded
abuse or neglect compared with nonteaching
hospitals (OR = 2.94, 95% CI = 2.33, 3.71).
Comparisons of the mean total charges for
each type of abuse or neglect and when no
abuse was coded are presented in Table 3
and show that the abused or neglected children, regardless of type of abuse, had significantly higher average charges. Compared
with the mean total charges for hospital stays
in which no abuse or neglect was coded
($9513), the highest mean charges were for
shaken infant syndrome ($30 311), followed
by children who had experienced multiple
types of abuse or neglect ($22 070), and then
“other child abuse and neglect,” which includes multiple forms ($20 267). Because
costs at teaching hospitals are known to be
higher,19 we reanalyzed the mean total
charges for abuse-related hospitalizations
compared with those without abuse or neglect while controlling for teaching status, and
the significant differences in average total
charges, regardless of type of abuse, remained.
TABLE 2—Estimated 1999 US Hospital Inpatient Stays Coded With a Diagnosis of Abuse or
Neglect Compared With Those Not Coded With Abuse or Neglect
Variables
Abuse/Neglect Coded
Patient/Family Demographic Characteristics
Mean age, y (SE)
2.7 (0.18)
Gender (% male)
54.5
Race, %
Black
20.1
White
30.9
Other (Hispanic, Asian/Pacific Islander, Native American,
14.3
and other)
Unknown
33.9
Median income for patient’s zip code, %
$1–$24 999
10.6
$25 000–$34 999
38.3
$35 000–$44 999
30.5
≥ $45 000
20.6
Expected primary payer, %
Medicaid
66.5
Private including HMO
24.0
Other
9.5
Medical/Hospital Stay Characteristics
Admission source, %
Emergency room
58.5
Routine/birth/other
27.3
Another hospital or facility
14.2
Court/law enforcement
0.11
In-hospital deaths (% died) (OR = 8.82, 95% CI = 6.19, 12.60)
4.0
Mean length of stay, d (SE)
8.2 (0.59)
Mean number of diagnoses (SE)
6.3 (0.19)
Mean number of procedures (SE)
1.3 (0.092)
Mean total charges, $ (SE)
19 266 (1646)
Hospital Characteristics
Region, %
Northeast
17.5
Midwest
28.7
South
33.9
West
19.9
Location of hospital (% urban) (OR = 1.63; 95% CI = 1.13, 2.34) 91.2
Status of hospital (% teaching) (OR = 2.94; 95% CI = 2.33, 3.71) 81.8
No Abuse/Neglect Coded
5.2 (0.13)
54.4
P Value
< .0001
.9375
< .0001
16.0
46.6
17.6
19.9
.0027
8.4
32.9
30.9
27.9
< .0001
37.0
54.5
8.5
< .0001
37.6
56.9
5.3
0.22
0.47
4.0 (0.08)
2.8 (0.05)
0.8 (0.04)
9513 (458)
< .0001
< .0001
< .0001
< .0001
< .0001
.0004
18.9
17.2
46.7
17.4
86.4
60.5
DISCUSSION
.0029
< .0001
Notes. HMO = health maintenance organization; OR = odds ratio; CI = confidence interval. Analyses are based on weighted
data: population estimate = 3 123 626.
Hospitalized children whose stays were
coded with a diagnosis of abuse or neglect
were significantly more likely to be admitted
through the emergency room than routinely
(P < .0001) compared with children not coded
as abused or neglected; and they were nearly
9 times more likely to die during hospitaliza-
tion (odds ratio [OR] = 8.82, 95% confidence
interval [CI] = 6.19, 12.60), resulting in an estimated 190 deaths. On average, those coded
with abuse or neglect compared with those
not coded as abused or neglected spent twice
the number of days (8.2 vs 4.0), had twice
the number of diagnoses (6.3 vs 2.8), had
588 | Research and Practice | Peer Reviewed | Rovi et al.
Our analyses demonstrate that the financial
costs for children hospitalized with a diagnosis of child abuse or neglect are considerable
compared with those for other hospitalized
children. The average total charges were
nearly $10 000 more per hospitalization for
the abused or neglected group, with an estimated total 1999 cost of nearly $92 million
for fewer than 5000 children. In addition to
the diagnosis of abuse or neglect, they had
twice the number of diagnoses/comorbidities.
Sadly, these children were also nearly 9 times
more likely to die during hospitalization.
Possible explanations for higher charges for
children hospitalized with a diagnosis of abuse
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 3—Mean Total Charges for 1999 US Hospitalizations Coded With an Abuse/Neglect
Diagnosis Compared With Those Not Coded With Abuse or Neglect
Types of Abuse
Mean Total Charges, $
SE
n
Not coded as abused: comparison group
99550: Child abuse unspecified
99551 Child emotional/psychological abuse
99552: Child neglect
99553: Child sexual abuse
99554: Child physical abuse
99555: Shaken infant syndrome
99559: Other child abuse and neglect
≥ 2 types of abuse coded
9 513
12 163
9 875
14 292
11 285
17 593
30 311
20 267
22 070
458
1774
2509
2664
2420
2003
2928
4747
4743
3 024 581
207
25
729
421
1 719
979
301
163
Note. Analysis of variance between each type of abuse or neglect and the comparison group is significant at P < .05
(F = 59.56, df = 8).
or neglect include the following: (1) Those so
coded may represent the most severe cases,
and certain types of abuse may get coded
more than other types (e.g., physical abuse
and shaken infant syndrome).2,6 (2) Children
with preexisting disabilities may be overrepresented among the abused and neglected20,21
and also more likely to have higher charges
associated with their hospitalizations. (3) Children identified as abused may need to stay in
the hospital longer for additional diagnostic
evaluations/investigations of the circumstances of the injuries or neglect.8 None of
these explanations, however, negate the fact
that child abuse and neglect are costly for
both victims and society. Moreover, by excluding hospital stays with neonatal/maternal
diagnoses or procedures and neonates aged
less than 1 day, which we argued would be
less likely to be coded with a diagnosis of
abuse or neglect, a bias may have been introduced. But in fact, when these cases were included in the analyses, the difference in mean
total charges between the 2 groups was even
greater, lowering the mean total charges for
the group with no abuse or neglect diagnostic
codes to $6879.
We also found children hospitalized with a
diagnosis of child abuse or neglect to be more
likely to be younger, to be Black, to live in
lower-income areas, and to be insured by
Medicaid. Other studies support findings of differences in the recognition and identification of
abuse according to racial/socioeconomic status.2–4,22,23 It remains unclear whether some
social groups may be at greater risk for abuse
or neglect or whether they may be more
likely identified/coded as abused because of
social conceptions about abuse and about reporting it. National incidence studies of child
abuse and neglect have found no racial differences in the incidence of maltreatment, but
lower incomes are related to higher incidence
rates.24 Actual demographics of abuse and
neglect require additional study to assist in
identification of at-risk children.
The limitations of these analyses include
the fact that these data reflect only hospitalizations coded with a diagnosis of child abuse
or neglect, and therefore these analyses most
likely underestimate the numbers of hospitalized children who experienced abuse or neglect. Unfortunately, it is very unlikely that
fewer than 5000 children nationally were admitted to community hospitals for medical
care associated with child maltreatment.2,24
These data only provide the total charges incurred for the hospitalization and do not
include other costs related to services, medical and otherwise, that victims or their families may incur after discharge—and as such
reflect an underestimate of the overall financial burden. The charges also do not include
physician services. One study estimates an additional 25% for inpatient physician services,25 thus increasing the total estimated
charges for these hospitalizations in 1999 to
nearly $115 million. Additionally, we know
that abused and neglected children often experience a lifetime of poorer mental and
April 2004, Vol 94, No. 4 | American Journal of Public Health
physical health, requiring more medical and
social services.8,10–11 Therefore, our estimate
of the charges for 1 hospitalized victim of
abuse or neglect does not reflect the lifetime
of health care costs that can result in such
cases. Despite these limitations, our research
demonstrates the substantial medical bill for
the maltreatment of children and sheds light
on the social costs of such maltreatment. Furthermore, these analyses provide unique descriptions of the current use of the diagnostic
codes for child maltreatment and may provide important information for future efforts
to develop guidelines for the appropriate use
of these codes.26
Child abuse and neglect are underidentified, underdiagnosed, and undercoded. Targeting interventions for those already being
abused or neglected as well as medical education for health care providers and interventions for children at risk of abuse and neglect
can reduce the individual, medical, and social
costs. Notably, teaching hospitals were thrice
as likely to code abuse, thereby confirming
the importance of medical training for addressing the problem of child abuse and neglect. Future analyses of the diagnoses or comorbidities associated with inpatient stays
coded with a diagnosis of abuse or neglect
may help prevent future harm to children by
providing potential indicators or red flags of
present abuse. Our findings provide the economic rationale for policies and programs to
prevent child abuse and neglect.
About the Authors
The authors are with the Department of Family Medicine,
University of Medicine and Dentistry of New Jersey, New
Jersey Medical School, Newark.
Requests for reprints should be sent to Sue L. D. Rovi,
185 S Orange Ave, MSB-B646, Newark, NJ, 07103
(e-mail: [email protected]).
This article was accepted May 31, 2003.
Contributors
S. Rovi and M. S. Johnson conceived of the study. S. Rovi
and P.-H. Chen conducted data analyses. All authors interpreted findings and reviewed drafts of the article.
Acknowledgments
This study was funded by the Department of Family
Medicine at New Jersey Medical School, University of
Medicine and Dentistry of New Jersey in Newark, NJ.
The authors thank the departmental writing group,
Charles Mouton, MD, Mindy Smith, MD, and Cathy
Widom, PhD, for their comments on earlier drafts of
this article.
Rovi et al. | Peer Reviewed | Research and Practice | 589
 RESEARCH AND PRACTICE 
Human Participant Protection
The University of Medicine and Dentistry of New Jersey’s institutional review board approved this research
as exempt from review.
US census-defined geographic areas—the public health
disparities geocoding project. Am J Public Health. 2002;
92:1100–1102.
References
19. Mechanic R, Coleman K, Dobson A. Teaching hospital costs: implications for academic missions in a
competitive market. JAMA. 1998;280:1015–1019.
1. Child Maltreatment 1999. Washington, DC: US
Department of Health and Human Services, Administration on Children, Youth and Families; 2001.
20. Sullivan PM, Knutson JF. Maltreatment and disabilities: a population-based epidemiological study.
Child Abuse Neglect. 2000;24:1257–1273.
2. Hampton RL, Newberger EH. Child abuse incidence and reporting by hospitals: significance of
severity, class, and race. Am J Public Health. 1985;75:
56–60.
21. Crosse SB, Kaye E, Ratnofsky AC. A report on the
maltreatment of children with disabilities. Washington,
DC: US Department of Health and Human Services.
3. Herman-Giddens ME, Brown G, Verbiest S, et al.
Underascertainment of child abuse mortality in the
United States. JAMA. 1999;282:463–467.
22. Lane WG, Rubin DM, Monteith R, Christian CW.
Racial differences in the evaluation of pediatric fractures for physical abuse. JAMA. 2002;288:1603–
1609.
4. Jenny C, Hymel KP, Ritzen A, Reinert SE, Hay TC.
Analysis of missed cases of abusive head trauma.
JAMA. 1999;291:621–626.
5. Kempe CH, Silverman FN, Steele BF, Droegemueller W, Silver HK. The battered-child syndrome.
JAMA. 1962;181:105–112.
6. Saulsbury FT, Campbell RE. Evaluation of child
abuse reporting by physicians. Am J Dis Child. 1985;
139:393–395.
7. Warner JE, Hansen DJ. The identification and reporting of physical abuse by physicians: a review and
implications for research. Child Abuse Negl. 1994;18:
11–25.
8. Irazuzta JE, McJunkin JE, Danadian K, Arnold F,
Zhang J. Outcome and cost of child abuse. Child Abuse
Negl. 1997;21:751–757.
23. Flaherty EG, Sege R, Mattson CL, Binns HJ. Assessment of suspicion of abuse in the primary care setting. Ambulatory Pediatr. 2002; 2:120–126.
24. Sedlak AJ, Broadhurst DD. Executive summary of
the third national incidence study of child abuse and
neglect. US Department of Health and Human Services, Administration for Children and Families. Available at: http://www.calib.com/nccanch/pubs/statinfo/
nis3.cfm. Accessed November 4, 2002.
25. Quinlan KP, Sacks JJ. Hospitalizations for dog bite
injuries. JAMA. 1999;281:232–233.
26. Rovi S, Johnson MS. More harm than good? The
use of diagnostic codes for child and adult abuse. Violence Victims. 2003; 18:491–502.
9. Wright MS, Litaker D. Childhood victims of violence: hospital utilization by children with intentional
injuries. Arch Pediatr Adolesc Med. 1996;150:415–420.
10. DiScala C, Sege R, Li G, Reece RM. Child abuse
and unintentional injuries: a 10-year retrospective. Arch
Pediatr Adolesc Med. 2000;154:16–22.
11. Rivera FP, Kamitsuka MD, Quan L. Injuries to
children younger than 1 year of age. Pediatrics. 1988;
81:93–97.
12. Peclet MH, Newman KD, Eichelberger MR, et al.
Patterns of injury in children. J Pediatr Surg. 1990;25:
85–91.
13. Rovi S, Johnson MS. Physician use of diagnostic
codes for child/adult abuse. J Am Med Womens Assoc.
1999;54:211–214.
Terrorism and
Public Health
Edited by Barry S. Levy and Victor W. Sidel
Published by Oxford Univ. Press and APHA
P
ublic health is a critical element in responding to terrorist incidents and in reducing or preventing threats of future terrorism. This important book addresses terrorism
and public health and presents a balanced
approach to strengthening systems and protecting people. It is designed to assist public
health professionals and their organizations
by providing up-to-date, science-based expert
information on, and a systematic practical approach to, a wide range of relevant public
health issues as they relate to terrorism.
The book is based, in part, on scientific
sessions presented at the 2001 APHA Annual
Meeting. It is designed to complement War
and Public Health.
ISBN 0-19-515834-2
2002 ❚ hardcover ❚ 368 pages
$41.95 APHA Members
$49.95 Nonmember
plus shipping and handling
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
TER04J3
14. Runyan WJ, Davey D. Identifying domestic violence within inpatient hospital admissions using medical records. Women Health. 2000;30:1–14.
15. Healthcare Cost and Utilization Project. 1999 National Inpatient Sample. Rockville, Md: Agency for
Healthcare Research and Quality; 2001.
16. Public Health Service and Health Care Financing
Administration. The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM).
Los Angeles, Calif: Practice Management Information
Corp; 1996.
17. American Hospital Association. Child/adult abuse.
Coding Clinic ICD-9-CM. 1996;13(4):38–45.
18. Krieger N, Waterman P, Chen JT, Soobader M,
Subramanian SV, Carson R. Zip code caveat: bias due
to spatiotemporal mismatches between zip codes and
590 | Research and Practice | Peer Reviewed | Rovi et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Somali and Oromo Refugees: Correlates of Torture
and Trauma History
| James M. Jaranson, MD, MPH, James Butcher, PhD, Linda Halcon, PhD, MPH, RN, David Robert Johnson, MD, MPH, Cheryl Robertson, PhD, MPH, RN,
Kay Savik, MS, Marline Spring, PhD, Joseph Westermeyer, MD, PhD, MPH
Historically, refugees and asylum seekers
have had a high probability of experiencing
politically motivated torture.1 The United
States has resettled many groups of refugees,
including these study populations from Somalia and Ethiopia. Having experienced civil war
and a lack of formal government for more
than a decade,2 Somalis have often suffered
traumatic events. Oromos claim ongoing political oppression since their territory was incorporated into the country of Ethiopia at the
end of the 19th century.3–5
Estimating the prevalence of torture in
community samples of refugees is extremely
difficult, often impossible, and rarely attempted. The political and emotional sensitivity of torture makes it difficult to study, and
refugees are challenging groups for research.6
Between 5% and 35% of refugees have been
tortured, according to the most frequently
cited review.7 Existing studies of torture and
associated factors typically conducted in refugee clinics and in other treatment settings,1,6,8–16 report that posttraumatic stress,
anxiety, depression, and somatization are
common.17,18 However, these studies have
rarely included control groups, generally have
had small samples, and cannot address the
prevalence of torture survival in communities.
Any consequences specifically associated with
torture, compared with other traumatic events
that refugees commonly experience, still need
to be identified and the effects quantified.19,20
Only a few studies with large samples
(n > 500) have examined torture prevalence
and posttraumatic stress disorder (PTSD)
rates, the focus of our article. From national
samples, De Jong et al.21 studied postconflict
populations in Algeria, Cambodia, Gaza, and
Ethiopia, finding rates of PTSD ranging from
16% to 37%. The prevalence of PTSD in the
1200 Ethiopians surveyed was 16%, higher
for torture survivors (P < .001) than for those
not tortured. Modvig et al.22 randomly sur-
Objectives. This cross-sectional, community-based, epidemiological study characterized Somali and Ethiopian (Oromo) refugees in Minnesota to determine torture prevalence and associated problems.
Methods. A comprehensive questionnaire was developed, then administered by trained
ethnic interviewers to a nonprobability sample of 1134. Measures assessed torture techniques; traumatic events; and social, physical, and psychological problems, including posttraumatic stress symptoms.
Results. Torture prevalence ranged from 25% to 69% by ethnicity and gender, higher
than usually reported. Unexpectedly, women were tortured as often as men. Torture
survivors had more health problems, including posttraumatic stress.
Conclusions. This study highlights the need to recognize torture in African refugees,
especially women, identify indicators of posttraumatic stress in torture survivors, and
provide additional resources to care for tortured refugees. (Am J Public Health. 2004;
94:591–598)
veyed 1033 household representatives in
East Timor and found a torture prevalence
rate of 30%.
Symptom levels tend to be higher in refugee camps than in resettlement populations.23–26 Mollica et al.,27,28 studying 993
Cambodians in a Thai refugee camp, found
that a third had PTSD. Comparing 526
Bhutanese torture survivors in a Nepalese
refugee camp with matched controls,
Shrestha et al.29 found higher posttraumatic
stress and anxiety. Van Ommeren et al.30
subsequently randomly sampled 810 (418
tortured and 392 nontortured) Bhutanese
refugees from the same frame (the general
source from which the sample population is
selected), finding that torture survivors had
more occurences of PTSD (43% vs 4%).
Among refugees living in nearby countries, Iacopino et al.31 found 4% torture
prevalence among 1180 randomly sampled
households of Kosovars living in Macedonia
and Albania. Smaller samples included
Senegalese refugees in Gambian camps
(16% torture prevalence),32 Tibetan nuns
and lay students tortured in Tibet but living
in India (54% anxiety vs 29% in controls),33 and Burmese political dissidents in
Thailand (23% PTSD).34
April 2004, Vol 94, No. 4 | American Journal of Public Health
In smaller prisoner populations in Turkey,
Paker et al.35 estimated that tortured Turkish
prisoners had significantly more PTSD compared with other prisoners, and Basoglu et
al.36 showed higher rates of lifetime (33%)
and current (18%) PTSD than controls.
In Western resettlement populations, Thonneau et al.37 found 8% torture prevalence
among 1194 refugee applicants to Canada.
Smaller samples have shown PTSD rates
among Cambodians of 12%,38 50%,39 and
even 86%.40
Our 5-year, multiphased, community-based
epidemiological study aimed to identify demographic characteristics, pre- and postmigration
factors, torture prevalence, and the association
of torture survival with health and social problems in 2 resettled refugee communities. The
findings of a quantitative survey of 1134 adult
participants are presented here. Two subsequent surveys, using subsets from this sample,
will compare torture survivors with nontortured
refugees using (1) structured instruments to assess symptoms, disability, coping, social support,
and family function and (2) a brief neurological
screen to identify soft signs of impairment and
the Schedules for Clinical Assessment in Neuropsychiatry41 to make Diagnostic and Statistical
Manual of Mental Disorders, Fourth Edition,42
Jaranson et al. | Peer Reviewed | Research and Practice | 591
 RESEARCH AND PRACTICE 
and International Classification of Diseases, 10th
Revision43 psychiatric diagnoses.
METHODS
Questionnaire Development and
Interview Administration
Questionnaires were administered by 8 Somali and Oromo staff with experience in
health care or interviewing and backgrounds
in medicine, law, engineering, biology, or social services.
Investigators provided training in survey
methods and research protocols. To prevent
participants from associating the study with
torture, we used the terms health problems
and physical or psychological abuse instead of
the word torture in the informational documents and consent forms used. Three authors
(M. S., C. R., D. J.) held biweekly sessions with
staff, providing debriefing and ongoing supervision during data collection.
Because using structured assessment instruments for the entire study was too expensive
and impractical for a large sample, we chose a
multiphased design. We developed a comprehensive closed-ended questionnaire for this
first interview phase. Although questionnaires
to assess trauma and torture are described in
the literature,44–47 their validity for East African refugees has not been documented. Our
questionnaire consisted of 188 questions with
537 response items adapted from published
studies and the authors’ clinical and research
experiences. The questionnaire elicited biographical information; current and earlier life circumstances; health status; and experiences of
violence, deprivation, and physical or psychological trauma and torture. Questions were
asked in a variety of formats, including yes/
no, Likert scale, and checklist. The most sensitive trauma and torture questions were asked
at the end of the questionnaire, after staff had
the opportunity to gain participants’ trust.
Checklists, an accepted assessment method for
torture techniques,46 were used, and scales
were developed to estimate physical, psychological, and social problems.
Because PTSD is one of the most common
and controversial diagnoses associated with
torture, the PTSD Checklist (PCL-C) was administered in this interview phase and, for
comparison, in the next 2 phases. The PCL-C,
a self-report Likert scale with 17 items, has
shown high internal consistency and reliability and strong correlation with PTSD diagnosis using the Clinician Administered PTSD
Scale.48,49 Translated versions of the PCL-C
into Oromo and Somali produced high reliability with Cronbach’s α (.93).
Staff members were matched with participants by gender and ethnicity. The questionnaire was self-administered for literate participants (50.4%) and interviewer administered
for those who were illiterate (49.6%). These
percentages were not significantly different
across the groups exposed and unexposed to
torture (χ2 = 0.36, df = 1, P = .55). Staff remained with all participants during questionnaire administration to ensure that the participants understood the items and to observe
for signs of distress. The questionnaire was
translated from English into the Somali and
Oromo languages and back-translated using
standard techniques.50–60
Sampling and Classification
We estimated that a sample size of 1200
would allow us to detect a doubling of the
rate of PTSD in the tortured group from a
base prevalence of 10% (power = 80%, α =
.05, 2-tailed).
Oromos or Somalis living in the Minneapolis/
St. Paul metropolitan area and who were at
least 18 years old were eligible. Exclusion criteria included psychological inability to participate or residence in a household where a
relative had already been interviewed. Unrelated persons living in the same household
were eligible.
Census data were not available for our
study because most Oromo and Somali refugees arrived after the 1990 census, and interviewing for this sample (which took place between July 14, 1999, and September 3,
2001) began before the Year 2000 census.
Even if census data had been available, the
census has historically undercounted refugees
and minorities.61 State data were kept only
for initial resettlement of refugees, and community organizations and agencies lacked
complete data.
Because a random sampling frame was unavailable, a combination of nonrandom sampling approaches62–64 and lengthy recruitment (more than 25 months) was used.
592 | Research and Practice | Peer Reviewed | Jaranson et al.
Sampling approaches included targeting persons associated with community organizations
and geographic locations (62%) and sampling
by linkage (38%). The participation rate was
97.1% of all invited.
Lacking the ability to assess true representativeness of the sample, we compared the
sample demographics with newly available
outside data in order to identify any differences between our sample and the underlying populations.65 The outside data were used
to estimate community size, geographic distribution, and demographic feaures and included public school enrollment reports, birth
records, and state refugee resettlement data.
No large differences were found between the
sample and the underlying populations, suggesting that the sample was representative.
The outside data allowed us to estimate the
Minneapolis/St. Paul populations as 641 Oromos and 6538 Somalis aged 18 years or
older. Consequently, the Oromo sample
(n = 512) may have represented 80% of potential participants, and the Somali sample
(n = 622) may have represented only 8%.
The United Nation’s (UN) definition of torture66,67 formed the basis for classifying participants as tortured. Key components of this
definition include physical or psychological
pain and suffering, intentionally inflicted for
any reason, based on discrimination and perpetrated by persons acting officially. Although
torture can occur in many nonpolitical settings, such as domestic violence or satanic
cults, “official” perpetration differentiates the
UN definition from others. During administration of the questionnaire, staff clarified the
context of traumatic events to ensure compatibility with the UN definition. For Somalis,
who have had no formal government since
1991, “official” perpetrators included opposing clans who had taken power.
Participants were classified as torture survivors if they (1) responded in the positive to
any of 3 items directly asking whether they
had been tortured (Have you been tortured in
prison? [Y/N]; Was tortured [marked off on a
checklist]; Were you tortured in prison or jail?
[Y/N]) and reported experiencing at least 1
identified torture technique item (details
available from authors) or (2) reported experiencing 1 of the subsets of torture techniques
that investigators considered could be used
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
only during torture sessions (details available
from authors), even if participants responded
in the negative to all of the questions about
torture exposure.
Data Collection, Management,
and Analysis
Staff explained the purpose, procedures,
risks, and benefits of the study to potential
participants and their rights to refuse participation at any time throughout the interview process. Literate persons read the consent form in
the language of their choice, and interviewers
read it to those who were illiterate. Staff obtained signed or oral informed consent from
subjects before they began the interviews.
Two strategies were undertaken to appraise
the level of cooperation and credibility of participants. (1) Staff rated participants after
completion of interviews using a 1 to 5 Likert
scale, with 1 being the most positive. The median rating was 2 (very cooperative or credible). (2) The questionnaire contained 4 internal validity measures identifying participants
with response inconsistency on item pairs,68
and, for more than 95% of the sample, more
missing items,69 a greater number of unusually virtuous responses,70(i.e., responses designed to assess if respondents presented
themselves in an overly favorable manner)
and a greater number of extreme responses.71
Those participants identified according to
these 4 validity measures were classified as
“suspect.” On analysis, the suspect group of
participants was found to be associated with
torture exposure, Oromo ethnicity, and younger age. Results for all participants were compared with a subset that excluded the suspect
records. Because subjective ratings of cooperation and credibility were not different between the suspect and nonsuspect groups, results were presented for the entire sample.
Data were double entered, and periodic error
checks were made. Summary data for each
variable were extensively checked, and outof-range responses were compared with the
original paper questionnaires.
The trauma count for each participant was
controlled in order to assess any additional effect of torture beyond other types of trauma.
Several specific items to measure social, psychological, and physical problems were incorporated into the questionnaire, combined into
summative scales, and refined using an item
analysis. These scales, meant to provide only
a preliminary indication of problem areas, are
presented with the Cronbach αs in Table 1.
Descriptive statistical analysis revealed the
distribution of variables and the appropriateness of statistical tests for the data. Stepwise
logistic regression was used to compare suspect and nonsuspect groups. Bivariate com-
parisons between torture groups were conducted using χ2 and t tests. Major outcomes
of interest included social, physical, and psychological problems, and total PCL-C scores.
For the interval measures, multivariable analysis was accomplished using stepwise multiple
linear regression. Regression diagnostics were
performed for each model to assess any possible violations of assumptions. Analysis was
performed using SPSS version 8 (SPSS Inc,
Chicago, Ill) and SAS version 8 (SAS Institute
Inc, Cary, NC).
RESULTS
The final sample of 1134 included 622 Somalis and 512 Oromos, 605 men and 529
women. Table 2 describes characteristics of
the ethnic/gender subgroups. On average,
participants reported experiencing 21 of 61
possible nontorture traumatic events. All but
6 participants reported experiencing traumatic experiences.
Problem Scales and PCL-C
Table 3 displays results of the analyses assessing variables associated with the problem
scales and PCL-C. Trauma count and exposure to torture were significant at P < .0001
for all these measures. In addition, the number of social problems was greater among
TABLE 1—Problem Scales and Cronbach αs: Minneapolis, Minn, July 14, 1999–September 3, 2001
Social Problems
Psychological Problems/Source of Problems
Does not speak English easily
Does not read English easily
Does not have job
Has less than high school education
Is separated from family
Has less money to spend than in homeland
Does not anticipate opportunities for work
Does not have good friends
Has difficulty caring for monthly expenses
Feels alone
Has problems getting job
Family has stayed behind
Has problems learning English
Has hard time understanding American life
Likes home food better than American food
α = .77
Had child die
Had death in family in last 6 months
Has no good work opportunities in United States
Feels stress living in United States
Bothered by things done in home country
Has trouble sleeping
Has loss of appetite
Hears voices
Has thoughts of killing self
Has been to doctor for mental health problem
Has intense memories of torture
Has frequent headaches
Has had changes in appetite
April 2004, Vol 94, No. 4 | American Journal of Public Health
α = .63
Physical Problems
Starvation before leaving homeland
Currently taking medication
On journey from homeland:
Life-threatening illness
Life-threatening lack of food
Life-threatening lack of water
Life-threatening serious injury
Life-threatening physical assault
Head injuries from torture
Physical problems resulting from torture
Still has physical reactions from torture
Has faintness or dizziness
α = .77
Jaranson et al. | Peer Reviewed | Research and Practice | 593
 RESEARCH AND PRACTICE 
TABLE 2—Characteristics of the Sample by Ethnic/Gender Group
Age, y, mean (SD)
Age at time of leaving home, mean (SD)
Number of traumatic events endorsed, mean (SD)
Years between leaving home and arriving in United States, mean (SD)
Years in United States, mean (SD)
Marital status, no. (%)
Married, living with partner
Separated by immigration
Single
Separated, divorced, widowed
Education, no. (%)
No formal education
Less than high school
High school diploma
Post high school
Speaks English, no. (%)
Receives government aid, no. (%)
Employed, no. (%)
Owns home, no. (%)
Has no permanent address, no. (%)
Muslim, no. (%)
Christian, no. (%)
Total
(n = 1134)
Oromo Men
(n = 282)
Oromo Women
(n = 230)
Somali Men
(n = 323)
Somali Women
(n = 299)
F Statistic
P Value
35.1 (13.9)
27.5 (14.5)
20.8 (11)
4.1 (3.2)
3.4 (3.6)
31 (12.4)
24.6 (12.1)
25.3 (10.4)
3.3 (3.0)
3.1 (4.2)
34.8 (13.9)
27.8 (14.4)
25.2 (8.2)
3.1 (2.8)
3.9 (4.4)
37.4 (15.9)
28.9 (16.0)
14.1 (8.5)
4.9 (3.2)
3.4 (3.2)
36.8 (14.7)
28.4 (14.7)
20.2 (11.8)
4.8 (3.3)
3.2 (2.8)
11.63,1123
5.03,1123
83.53,1130
25.913,1100
1.93,1108
<.001
.002
<.001
<.001
.12
319 (29)
264 (24)
342 (31)
180 (16)
68 (25)
78 (29)
122 (45)
2 (1)
93 (41)
34 (15)
58 (25)
44 (19)
167 (15)
418 (37)
292 (26)
245 (22)
633 (56)
83 (7)
577 (51)
64 (6)
35 (3)
990 (87)
105 (9)
14 (5)
104 (38)
58 (21)
99 (36)
198 (70)
3 (1)
198 (72)
25 (9)
5 (2)
204 (72)
62 (22)
62 (27)
103 (45)
31 (14)
33 (14)
101 (44)
28 (12)
103 (45)
26 (11)
8 (4)
185 (80)
43 (19)
69 (22)
86 (27)
115 (37)
43 (14)
14 (4)
117 (36.5)
115 (36)
75 (23.5)
224 (69)
12 (4)
188 (58)
6 (2)
9 (3)
313 (97)
0
89 (30)
66 (23)
47 (16)
91 (31)
156.39
<.001
77 (26)
94 (31)
88 (30)
38 (13)
110 (37)
40 (13)
88 (30)
7 (2)
13 (4)
288 (96)
0
167.29
104.83
46.63
114.13
34.23
3.43
115.53
142.23
<.001
<.001
<.001
<.001
<.001
.33
<.001
<.001
TABLE 3—Multivariable Analysis of Characteristics and Problems
Social Problems
β (SE)
Gender: female
Ethnicity: Oromo
Age
Married or living with partner
High school graduate
Speaks English
Has job
Owns home
Age at which left home country
Years between leaving home and arriving in United States
Years in United States
Religious practices increased
Religious practices decreased
Trauma count
Exposure to torture
Adjusted R 2
1.54 (0.13)
(0.71 (0.14)
(0.01 (0.03)
(0.39 (0.13)
(1.94 (0.13)
(2.01 (0.14)
(1.27 (0.13)
(0.84 (0.26)
0.04 (0.03)
0.06 (0.03)
0.01 (0.03)
(0.21 (0.18)
0.53 (0.24)
0.08 (0.01)
(0.59 (0.14)
0.7170
P Value
<.0001
<.0001
.7023
.0036
<.0001
<.0001
<.0001
.0013
.0979
.0472
.6347
.2570
.0293
<.0001
<.0001
Physical Problems
β (SE)
(1.0 (0.13)
(0.78 (0.14)
(0.02 (0.03)
(0.40 (0.14)
(0.33 (0.14)
(0.22 (0.15)
(0.44 (0.14)
(0.07 (0.27)
0.02 (0.03)
0.02 (0.03)
0.03 (0.03)
0.18 (0.19)
0.25 (0.25)
0.15 (0.01)
0.60 (0.15)
0.5091
Psychological Problems
P Value
<.0001
<.0001
.5347
.0035
.0159
.1289
.001
.8027
.4239
.5977
.2778
.3144
.3398
<.0001
<.0001
β (SE)
P Value
0.16 (0.10)
.0899
0.02 (0.10)
.8650
(0.02 (0.02)
.3209
(0.18 (0.10)
.0764
0.01 (0.10)
.8877
(0.04 (0.11)
.6848
(0.18 (0.10)
.0709
(0.29 (0.19)
.1381
0.05 (0.02)
.0189
0.02 (0.02)
.3423
0.06 (0.23)
.0163
0.08 (0.14)
.5740
0.76 (0.18)
<.0001
0.09 (0.01)
<.0001
0.61 (0.11)
<.0001
0.4962
Total PCL-C Score
β (SE)
P Value
(3.35 (0.88)
.0001
2.29 (0.94)
.0149
0.04 (0.19)
.8336
(1.66 (0.91)
.0685
(0.85 (0.93)
.3615
–0.57 (0.98)
.5606
(0.55 (0.90)
.5436
0.95 (1.85)
.6048
(0.01 (0.19)
.9752
(0.35 (0.22)
.1046
(0.20 (0.23)
.3795
3.19 (1.28)
.0131
4.67 (1.67)
.0051
0.52 (0.05)
<.0001
7.18 (0.98)
<.0001
0.4122
Note. PCL-C = Posttraumatic Stress Syndrome Checklist.
594 | Research and Practice | Peer Reviewed | Jaranson et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
women, Somalis, and those who had decreased their religious practices since immigrating. English language fluency, employment, high school graduation, marriage, home
ownership, and longer residency in the
United States were associated with fewer social problems. Male gender and Somali ethnicity were associated with more physical
problems, whereas employment, marriage,
and high school graduation were associated
with fewer physical problems.
None of the assessed factors were associated with fewer psychological problems. More
psychological problems were associated with
leaving home at an older age, longer residency in the United States, and decreased religious practices since immigrating. Female gender was inversely associated with PCL-C
score, whereas Oromo ethnicity and any
change in religious practices since immigrating
were associated with increased PCL-C scores.
Associations with fewer problems on at least
2 problem scales/PCL-C included employment, high school graduation, marriage, and
continued religious practices, whereas more
problems were associated with decreased religious practices, high trauma count, and torture
exposure. Despite high scores on the scales,
less than 1% of the participants requested or
accepted referral to mental health services.
Prevalence of Torture and
Characteristics of Survivors
Prevalence rates for each torture category
were calculated (Table 4). Of the 1134 participants, 44% met criteria for torture exposure and 56% did not. Criteria for classification are described in the Methods section.
Only 15% (n = 92) met criteria solely by reporting that they experienced 1 or more of
the techniques occurring only during torture
(false negatives). Conversely, 40 participants
were excluded from the torture group because, although they gave a positive answer
to a torture question, they did not report experiencing any torture techniques (false positives).72 Had we accepted, as the only criterion, an affirmative response when asking
TABLE 4—Demographic Characteristics of the Sample by Torture Classification
Exposed to Torture
Entire sample (n = 1134)
Ethnicity
Somali, no. (%) (n = 622)
Oromo, no. (%) (n = 512)
Gender
Male, no. (%) (n = 605)
Female, no. (%) (n = 529)
Married or living with partner, no. (%)
Graduated from high school, no. (%)
Employed, no. (%)
Owns home, no. (%)
Has no permanent address, no. (%)
Current age, mean ( SD)
Age on leaving home, mean ( SD)
Years between leaving home and arriving in
United States, mean ( SD)
Years in United States, mean ( SD)
Number of traumas endorsed, mean ( SD)
Number of social problems, mean ( SD)
Number of psychological problems, mean ( SD)
Number of physical problems, mean ( SD)
PCL-C scores, mean ( SD)
Unexposed to Torture
χ2
P Value
42.61
<.0001
502 (44)
632 (56)
221 (36)
261 (55)
401 (64)
231 (45)
272 (45)
228 (43)
120 (25)
230 (47)
243 (49)
30 (06)
19 (04)
36.78 (14)
29.50 (14)
4.04 (03)
333 (55)
302 (57)
199 (32)
307 (49)
334 (53)
34 (05)
16 (03)
33.78 (15)
25.90 (15)
4.12 (03)
7.791
0.601
1.551
0.191
1.471
3.451125
4.191109
0.661102
.0052
.4387
.2128
.6656
.2255
.0006
<.0001
.5119
3.04 (04)
28.07 (10)
7.09 (04)
3.67 (02)
4.32 (03)
42.53 (15)
3.66 (04)
14.94 (08)
6.43 (03)
1.74 (01)
1.72 (02)
27.17 (10)
2.821110
25.021132
3.261132
19.141132
18.881132
18.53919
.0048
<.0001
.0012
<.0001
<.0001
<.0001
0.741
Note. PCL-C = Posttraumatic Stress Syndrome Checklist.
April 2004, Vol 94, No. 4 | American Journal of Public Health
.3900
participants whether they had been tortured,
132 (11.6%) would possibly have been misclassified according to our definition. Fifty
torture survivors (10%) reported experiencing only physical or only psychological techniques, whereas the other 90% reported experiencing both.
Torture history varied by gender and ethnicity: men (n = 272, 45% of the men) and
women (n = 228, 43% of the women) had
approximately equal exposure to torture,
and more Oromos were exposed to torture
(n = 286, 55%) than Somalis (n = 224, 36%).
Among ethnic/gender groups, those most
often exposed to torture were Oromo men
(n = 194, 69%) and Somali women (n = 141,
47%), followed by Oromo women (n = 85,
37%) and Somali men (n = 81, 25%). Gender
differences were not statistically significant
(Table 3), but differences by ethnicity (χ2 =
42.6, df = 1, P < .001) and by ethnic/gender
group (χ2 = 126.0, df = 3, P < .001) were
significant.
Participants in the tortured and nontortured
groups differed in several characteristics and
in the degree of reported problems (Table 3).
Torture survivors were less often married and
were older when they left their home countries. No significant between-group differences
were found for men or for those employed,
more educated, with permanent addresses, or
with longer times between leaving their home
countries and arriving in the United States.
For both groups, a higher number of traumatic events correlated positively with scores
on all problem scales and total PCL-C. The
tortured group averaged 13 more traumatic
events than the nontortured group (28 vs 15),
adding an average of 1 social and 1 psychological problem, 2 physical problems, and 7
points to the PCL-C scale for those exposed
to torture. Beyond the general trauma, torture
exposure added small but significant increases
to the number of psychological and physical
problems but not to the number of social
problems. For social problems, the estimated
adjusted means were 6.46 (SE = 0.11) for
those exposed and 7.03 (SE = 0.09) for those
unexposed to torture, showing no significant
statistical difference.
The most striking correlate of torture exposure was the increase in total PCL-C score.
Those exposed to torture averaged an addi-
Jaranson et al. | Peer Reviewed | Research and Practice | 595
 RESEARCH AND PRACTICE 
tional 7-point (> 10%) increase in PCL-C
score beyond the increase due to trauma
alone. Using a cutoff PCL-C score of greater
than 50 (range 17 to 85) to indicate suspected PTSD, 123 (25%) of those exposed to
torture met this criterion, but only 23 (4%) of
the unexposed did.
DISCUSSION
Our study interviewed 1134 East Africans,
the largest refugee community sample conducted in a resettlement country. As expected, social problems were common
throughout the sample, because the early
stages of resettlement for all refugees are
characterized by many problems in daily living and adjustment.9,73,74 Factors associated
with fewer social problems included speaking
English and graduating from high school,
which presumably increased the chances for
employment and home ownership. Marriage
can provide social stability, and longer residency in the United States can allow time for
adjustment. High school education, employment, and marriage were associated with
fewer physical problems, perhaps because
those with fewer physical problems could adjust more successfully. However, none of
these factors were associated with fewer psychological problems.
Low scores on at least 2 of the problem
scales/PCL-C were associated with employment, high school graduation, marriage, and
continuing religious practices since immigration. These factors indicate the importance of
education, economic stability, social support,
and religion for the well-being of refugees.
High scores for trauma and torture, as expected, were associated with more problems.
Torture rehabilitation centers have historically claimed that the more educated men
were most often targeted as leaders who
would serve as examples to their communities.1 Our study, in contrast, showed that men
and those with higher education levels were
no more likely to be tortured than women or
those with less education. Although more
time in transition to the United States, usually
in refugee camps, might be expected to increase the risk of trauma, our study did not
find any increase for those experiencing
longer transitions.
Except for Somali men (25% of whom
were tortured), the prevalence of torture exposure in this community sample was significantly higher than the 5% to 35% often reported.7 This high prevalence was found
despite the expectation that participants
would underreport.8 If these results can be
replicated in other refugee communities,
400 00075 is likely an underestimation of the
number of torture survivors living in the
United States.
Women were as likely to experience torture as men, an exposure rate not previously
reported. Because civilians are increasingly
affected by modern warfare and terrorism, it
is not surprising that women frequently experience torture.76,77
Significant differences in torture exposure
by ethnicity and ethnic/gender group were
found. Oromos were tortured more often
than Somalis, whereas Oromo men and Somali women were the ethnic/gender groups
most often tortured. A number of possible
explanations can be posited. The very high
rates in the Oromo community may reflect
long-standing interethnic conflicts. Somali
women were more often tortured than Somali men. Anecdotally, Somali men were
either killed in their home country or able to
escape unharmed, whereas women and children had a more difficult time leaving the
country.
Our results suggest that torture’s effect may
be additive to other forms of trauma. Torture
survivors were more likely than other refugees to experience physical and psychological
problems, even after we analyzed the group
differences using nontorture trauma as a covariate. The most striking correlate of torture
exposure in our study was the increase in
total PTSD symptoms. Of those tortured,
25% had suspected PTSD compared with
only 4% of those not tortured. These rates
are comparable to those cited in earlier studies. In addition, PTSD also showed the greatest additive effect of torture. In another study,
Silove et al.19 found an additive effect of
PTSD in Tamil torture survivors in Australia
after they accounted for other traumatic
events. Although the published literature is
contradictory, our findings also support a possible dose–response relationship between torture and PTSD.
596 | Research and Practice | Peer Reviewed | Jaranson et al.
Several limitations were inherent in the
study design. Because we could not use true
random sampling methods, our prevalence results are estimates. However, our analyses indicate that an estimated 80% of the Oromo
community, nearly the entire Oromo “village,”
may have been sampled. With such a comprehensive sample of Oromo, the extremely high
torture prevalence of Oromo men (69%) is
plausible. The study variables were measured
at the same time, a limitation shared by all
cross-sectional studies. Therefore, inferring
causality between torture history and problems was not possible. Although the present
analyses strongly support associations between torture history and key problem areas,
our 2 follow-up studies will further elucidate
these findings. The final clinician-administered
diagnostic phase of our research will also assess the duration and frequency of torture experiences in greater depth.
Our study of 2 communities represents
only a small proportion of refugees and displaced persons but contributes to the enormous gap in understanding of the effects of
torture and trauma on refugee populations
worldwide. The capacity to care for survivors
falls far short of the need. This study suggests that the shortfall is even greater than
previously thought. The 31 torture rehabilitation centers in the United States and the
more than 200 worldwide78 cannot treat all
torture survivors.
From a public health perspective, our findings warrant screening refugees for a history
of torture at least among East African and
women refugees. However, less than 1% of
our highly traumatized sample either requested or followed up a referral to Western
mental health services. This highlights the
need to investigate the reasons for underutilization and to adequately address the needs
of torture survivors living not only in Minnesota but throughout the world.
About the Authors
At the time of the study, James M. Jaranson was with the
HealthPartners Division of Behavioral Health and the departments of Epidemiology and Psychiatry at the University of Minnesota, Minneapolis. James Butcher was with
the Psychology Department, University of Minnesota.
David Robert Johnson and Joseph Westermeyer are with
the Veterans Administration Medical Center and the Psychiatry Department, University of Minnesota. David Rob-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
ert Johnson is also with the Center for Victims of Torture,
Minneapolis. Linda Halcon, Cheryl Robertson, and Kay
Savik are with the School of Nursing, and Marline Spring
is with the Division of Epidemiology, School of Public
Health, University of Minnesota.
Requests for reprints should be sent to Marline Spring,
Ph.D., Division of Epidemiology, School of Public Health,
University of Minnesota, 1300 S Second St, Suite 300,
Minneapolis, MN 55454 (e-mail: [email protected]).
This article was accepted July 16, 2003.
Contributors
All authors contributed to the conceptualization of
ideas, interpretation of findings, and review of the article. J. Jaranson (principal investigator) conceived the
study, drafted the article, and supervised operations
and personnel. J. Butcher contributed to design and developed the questionnaire. L. Halcon contributed the
epidemiological perspective for the research. D. R. Johnson contributed to design and provided insights from
his work at the Center for Victims of Torture. C. Robertson contributed to design from the public health nursing perspective. K. Savik supervised the analyses of the
data and the research assistants. M. Spring was project
coordinator, supervised the interview staff, ensured the
quality of the data, and provided a cross-cultural perspective. J. Westermeyer contributed to design with his
extensive background in cross-cultural psychiatry.
Acknowledgments
This research was funded by the National Institutes of
Health (National Institute of Mental Health) grants
5R01-MH59579 and 1R01-MH59570 (project officer
Farris Tuma, ScD.)
The authors thank Dr Alan Lifson for his contributions to the initial development of this study and Dr
Russell Luepker for his insights into management of
the research project. The authors acknowledge the valuable contributions in advising, translating, recruiting,
and interviewing by our Somali community coordinators, Nadifa Osman, Dr Abdullahi Gas, Dr Osman
Ahmed, Dr Abdulqadir Omar, and Dr Mohamed Wadi,
and by our Oromo community coordinators, Israel
Gobena, Johara Mohammed, and Elizabeth Namarra.
We are grateful to our research assistants, Michelle Peterson, Katherine Grimm, MPH, and Dr Sheng Zhong
for data management.
Human Participant Protection
The University of Minnesota institutional review board
and human subjects protection program and the
HealthPartners Foundation institutional review board
approved the study before interviewing began and subsequently approved protocol changes, consent forms,
and annual progress reports.
References
1. Jaranson JM. The science and politics of rehabilitating torture survivors: an overview. In: Jaranson
JM, Popkin, MK, eds. Caring for Victims of Torture.
Washington, DC: American Psychiatric Press; 1998:
15–40.
2. Putman DB, Noor MC. The Somalis: Their History
and Culture. Washington, DC: The Refugee Service
Center, Center for Applied Linguistics; 1993. CAL Refugee Fact Sheet 9.
3.
Trueman T. Genocide against the Oromo people of
Ethiopia? Western influence. Malvern, UK: Oromia Support Group. Occasional paper: June 7, 2001.
4. Country Profile: Ethiopia, Eritrea, Somalia, Djibouti.
London, England: Economist Intelligence Unit;
1994–1995.
5. Annual Human Rights Reports, 1997. Washington
DC: US State Department: Bureau of Democracy,
Human Rights and Labor; 1998.
6. Basoglu M, Jaranson JM, Mollica R, Kastrup M.
Torture and mental health: a research overview. In:
Gerrity E, Keane TM, Tuma F, eds. The Mental Health
Consequences of Torture. New York, NY: Kluwer Academic/
Plenum Publishers; 2001:35–62.
7. Baker R. Psychological consequences for tortured
refugees seeking asylum and refugee status in Europe.
In: Basoglu M, ed. Torture and Its Consequences: Current
Treatment Approaches. Cambridge, England: Cambridge
University Press; 1992:83–101.
8. Jaranson JM, Kinzie JD, Friedman M, et al. Assessment, diagnosis, and intervention. In: Gerrity E, Keane
TM, Tuma F, eds. The Mental Health Consequences of
Torture. New York, NY: Kluwer Academic/Plenum Publishers; 2001:249–275.
9. Jaranson JM, Martin SF, Ekblad S. Refugee mental
health: issues for the new millennium. In: Manderscheid RW, Henderson MJ, eds. Mental Health, United
States, 2000. Rockville, Md: US Department of Health
and Human Services; 2001:120–133.
10. Jaranson JM, Popkin MK, eds. Caring for Victims of
Torture. Washington, DC: American Psychiatric Press;
1998.
11. Basoglu M, ed. Torture and Its Consequences: Current Treatment Approaches. Cambridge, England: Cambridge University Press; 1992.
12. Lavik NJ, Hauff E, Skrondal A, et al. Mental disorder among refugees and the impact of persecution and
exile: some findings from an out-patient population. Br
J Psychiatry. 1996;169:726–732.
13. Moore LJ, Boehlein JK. Posttraumatic stress disorder, depression, and somatic symptoms in US Mien patients. J Nerv Ment Dis. 1991;179:728–733.
14. Boehnlein JK, Kinzie JD. Refugee trauma. Transcultural Psychiatr Res Rev. 1995;32:223–252.
21. De Jong JTVM, Komproe IH, Van Ommeren M, et al.
Lifetime events and posttraumatic stress disorder in 4
postconflict settings. JAMA. 2001;286:555–562.
22. Modvig J, Pagaduan-Lopez J, Rodenburg J,
Salud CM, Cabigon RV, Panelo CI. Torture and trauma
in post-conflict East-Timor. Lancet. 2000;356:1763.
23. Hinton WL, Chen YCJ, Du N. DSM-III-R disorders
in Vietnamese refugees: prevalence and correlates.
J Nerv Ment Dis. 1993;181(2):113–122.
24. Carlson EB, Rosser-Hogan R. Cross-cultural response to trauma: a study of traumatic experiences
and posttraumatic symptoms in Cambodia refugees.
J Trauma Stress. 1994;7(1):43–58.
25. Harding RK, Looney JG. Problems of Southeast
Asian children in a refugee camp. Am J Psychiatry.
1977;134:407–411.
26. Sughandabhirom B. Experiences in a first asylum
country: Thailand. In: Williams CL, Westermeyer J,
eds. Refugee Mental Health in Resettlement Countries.
New York, NY: Hemisphere Publishing Corp; 1986:
81–96.
27. Mollica RF, Donelan K, Tor S, et al. The effect of
trauma and confinement on functional health and mental health status of Cambodians living in ThailandCambodia border camps. JAMA. 1993;270:581–586.
28. Mollica RF, Poole C, Tor S. Symptoms, functioning, and health problems in massively traumatized
populations: the legacy of Cambodian tragedy. In:
Dohrenwend BP, ed. Adversity, Stress, and Psychopathology. New York, NY: Oxford University Press; 1998:
34–51.
29. Shrestha NM, Sharma B, Van Ommeren M, et al.
Impact of torture on refugees displaced within the developing world: symptomatology among Bhutanese refugees in Nepal. JAMA. 1998;280:443–448.
30. Van Ommeren M, de Jong JT, Sharma B, et al.
Psychiatric disorders among tortured Bhutanese refugees in Nepal. Arch Gen Psychiatry. 2001;58:
475–482.
31. Iacopino V, Frank, MW, Bauer HM, et al. A
population-based assessment of human rights abuses
committed against ethnic Albanian refugees from
Kosovo. Am J Public Health. 2001;91:2013–2018.
15. Mollica RF, Wyshak G, Lavelle J. The psychosocial impact of war trauma and torture on Southeast
Asian refugees. Am J Psychiatry. 1987;144:
1567–1572.
32. Tang SM, Fox SH. Traumatic experiences and the
mental health of Senegalese refugees. J Nerv Ment Dis.
2001;189:507–512.
16. Ekblad S, Roth G. Diagnosing posttraumatic stress
disorder in multicultural patients in a Stockholm psychiatric clinic. J Nerv Ment Dis. 1997;185:102–107.
33. Holtz TH. Refugee trauma versus torture trauma:
a retrospective controlled cohort study of Tibetan refugees. J Nerv Ment Dis. 1998;186:24–34.
17. Westermeyer J, Bouafuely M, Neider J, et al. Somatization among refugees: an epidemiological study.
Psychosomatics. 1989;30:34–43.
34. Allden K, Poole C, Chantavanich S, et al. Burmese
political dissidents in Thailand: trauma and survival
among young adults in exile. Am J Public Health. 1996;
86:1561–1569.
18. Beiser M, Fleming JA. Measuring psychiatric disorder among Southeast Asian refugees. Psychol Med.
1986;16:627–639.
19. Silove D, Steel Z, McGorry P, et al. The impact of
torture on post-traumatic stress symptoms in war-affected
Tamil refugees and immigrants. Compr Psychiatry.
2002;43:49–55.
35. Paker M, Paker O, Yuksel S. Psychological effects of
torture: an empirical study of tortured and non-tortured
non-political prisoners. In: Basoglu M, ed. Torture and
Its Consequences: Current Treatment Approaches. Cambridge, England: Cambridge University Press; 1992:
72–82.
20. Steel Z, Silove D, Bird K, et al. Pathways from war
trauma to posttraumatic stress symptoms among Tamil
asylum seekers, refugees, and immigrants. J Trauma
Stress. 1999;12:421–435.
36. Basoglu M, Paker M, Paker O, et al. Psychological
effects of torture: a comparison of tortured with nontortured political activists in Turkey. Am J Psychiatry.
1994;151:76–81.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Jaranson et al. | Peer Reviewed | Research and Practice | 597
 RESEARCH AND PRACTICE 
37. Thonneau P, Gratton J, Desrosiers G. Health profile of applicants for refugee status (admitted into Quebec between August 1985 and April 1986). Can J Public Health. 1990;81:182–186.
38. Chung RC, Kagawa-Singer M. Predictors of psychological distress among Southeast Asian refugees. Soc
Sci Med. 1993;36:631–639.
39. Kinzie JD. The psychiatric effects of massive
trauma on Cambodian refugees. In: Wilson JP, Harel Z,
Kahana B, eds. Human Adaptation of Extreme Stress:
From the Holocaust to Vietnam. New York, NY: Plenum
Press; 1988:305–317.
40. Carlson EB, Rosser-Hogan R. Trauma experiences,
posttraumatic stress, dissociation, and depression in
Cambodian refugees. Am J Psychiatry. 1991;148:
1548–1551.
41. Wing JK, Babor T, Brugha T, et al. SCAN: Schedules for Clinical Assessment in Neuropsychiatry. Arch
Gen Psych. 1990;47:589–593.
42. Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition. Washington, DC; American Psychiatric Association; 1994.
56. Robins LN, Wing J, Wittchen HU, et al. The Composite International Diagnostic Interview: an epidemiologic instrument suitable for use in conjunction with
different diagnostic systems and in different cultures.
Arch Gen Psychiatry. 1988;45:1069–1077.
57. Sartorius N. Making of a common language for
psychiatry: development of the classification of mental,
behavioral and developmental disorders in the 10th revision of the ICD. WPA Bull. 1989;1(1):3–6.
58. Bravo M, Woodbury-Farina FM, Canino GJ, et al.
The Spanish translation and cultural adaptation of the
Diagnostic Interview Schedule (DISC) in Puerto Rico.
Cult Med Psychiatry. 1993;17:329–344.
59. Westermeyer J, Janca A. Language, culture, and
psychopathology. Transcultural Psychiatry. 1997;34:
291–311.
44. Marsella AJ. Measurement issues. In: Gerrity E,
Keane TM, Tuma F, eds. The Mental Health Consequences of Torture. New York, NY: Kluwer Academic/
Plenum Publishers; 2001:277–290.
63. Lepkowski JM. Sampling the Difficult-to-Sample.
Nutritional Assessment and Intervention: Interface of Science and Policy. Washington, DC: Federation of American Societies for Experimental Biology; 1991.
45. Mollica RF, Caspi-Yavin Y. Measuring torture and
torture-related symptoms. Psychol Assess. 1991;3:
581–587.
64. Frank O, Snijders T. Estimating the size of hidden
populations using snowball sampling. J Official Statistics. 1994;10(1):53–67.
46. Willis GB, Gonzalez A. Methodological issues in
the use of survey questionnaires to assess the health
effects of torture. J Nerv Ment Dis. 1998;186:
283–289.
65. Spring M, Westermeyer J, Halcon L, et al. Sampling in difficult to access refugee and immigrant communities. J Nerv Ment Dis. 2003:191: 813–819.
49. Blanchard EB, Jones-Alexander J, Buckley TC, et
al. Psychometric properties of the PTSD checklist
(PCL). Behav Res Ther. 1996;34:669–673.
50. Hall ET. The Silent Language. Garden City, NY:
Doubleday; 1959.
51. Brislin R. Back-translation for cross-cultural research. J Cross-Cultural Psychol. 1970;1:185–216.
52. Butcher JN, Garcia R. Cross-national application of
psychological tests. Personnel Guidance. 1978;56:
472–475.
53. Hulin CL. A psychometric theory of evaluations
of item and scale translations: fidelity across languages.
J Cross-Cultural Psychol. 1987;18:115–142.
54. Sabin JE. Translating despair. Am J Psychiatry.
1975;132:197–199.
55. Kinzie JD, Manson SM, Vinh DT, et al. Development and validation of a Vietnamese language depression rating scale. Am J Psychiatry. 1982;139:
1276–1281.
76. Voices. New York, NY: Womens’ Commission for
Women and Children; 1993.
77. Robertson C. Patterns of survival for Bosnian rural
mothers during the war and post-war period. Minneapolis, Minn: School of Nursing, University of Minnesota;
2000.
78. Rehabilitation of Torture Victims: Centres and Programmes Worldwide. Copenhagen, Denmark: International Rehabilitation Council for Torture Victims;
2000.
61. El Nasser H. Cities, minorities lose in census undercount. USA Today. March 28, 2001: A1, A6.
62. Waters JK, Biernaski P. Targeted sampling: options
for the study of hidden populations. Soc Probl. 1989;
36:416–430.
48. Weathers FW, Litz BT, Herman DS, et al. The
PTSD Checklist (PCL): reliability, validity, and diagnostic utility. Paper presented at: 9th Annual Meeting of
the International Society for Traumatic Stress Studies;
October 24–27, 1993; San Antonio, Tex.
75. Jaranson JM. Government-sanctioned torture: status of the rehabilitation movement. Transcultural Psychiatr Res Rev. 1995;32:253–286.
60. Ebigo PO. Development of a culture-specific
(Nigeria) screening scale of somatic complaints indicating psychiatric disturbance. Cult Med Psychiatry. 1982;
6:29–43.
43. International Classification of Diseases, 10th Revision. Geneva, Switzerland: World Health Organization;
1980.
47. Hollifield M, Warner TD, Lian N, et al. Measuring
trauma and health status in refugees: a critical review.
JAMA. 2002;288:611–621.
Mental Health Consequences of Torture. New York, NY:
Kluwer Academic/Plenum Publishers; 2001:111–120.
66. Convention Against Torture and Other Cruel, Inhuman, or Degrading Treatment or Punishment. New York,
NY: Office of Public Information, United Nations;
1984.
67. UN Convention Against Torture and Other Cruel,
Inhuman, or Degrading Treatment or Punishment: Methods of Combating Torture. Geneva, Switzerland: United
Nations Centre for Human Rights; 1989:17.
68. Tellegen A. The analysis of consistency in personality assessment. J Pers. 1988;56:621–663.
69. Berry DTR, Adams JJ, Smith GT, et al. MMPI-2
clinical scales and 2-point code types: impact of varying levels of omitted items. Psychol Assess. 1997;9:
158–160.
70. Hathaway SR, McKinley JC. The Minnesota Multiphasic Personality Schedule. Minneapolis, Minn: University of Minnesota Press; 1943.
71. Arbisi P, Ben-Porath YS. An MMPI-2 infrequency
scale for use with psychopathological populations: The
Infrequency-Psychopathology Scale, F (p). Psychol Assess. 1995;7:424–431.
72. Montgomery E, Foldspang A. Criterion-related validity of screening for exposure to torture. Dan Med
Bull. 1994;41:500–591.
73. Kunz E. Exile and resettlement: refugee theory.
Int Migr Rev. 1981;15:42–51.
74. Kinzie JD, Jaranson JM. Refugees and asylumseekers. In: Gerrity E, Keane TM, Tuma F, eds. The
598 | Research and Practice | Peer Reviewed | Jaranson et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Gender Differences in Long-Term Health Consequences of Physical Abuse
of Children: Data From a Nationally Representative Survey
| Martie P. Thompson, PhD, J. B. Kingree, PhD, and Sujata Desai, PhD
The public health significance of physical
abuse in childhood is manifested by its incidence and prevalence1,2 and also its long-term
psychological and physical health consequences. In its most recent report, the Administration on Children, Youth, and Families estimated physical abuse incidence rates for boys
to be 2.1 per 1000 and for girls to be 2.2 per
1000.1 In terms of prevalence, national data
reveal that approximately 1.5 million children
have experienced physical abuse.2 Psychological and behavioral problems that have been
found to be associated with physical abuse in
childhood include poorer academic and intellectual outcomes, posttraumatic stress disorder, depression, substance abuse, personality
disorders, suicidal behavior, and aggression.3–15 Physical health problems associated
with physical abuse in childhood include gastrointestinal problems, greater physical functional disability, more physical health symptoms, and more hospitalizations.16–20
According to the National Research Council’s Panel on Child Abuse and Neglect, there
is little research on gender differences in the
consequences of child abuse. The panel recommended that research be conducted to determine whether there are differential consequences of child abuse for boys and girls.21
The lack of research on gender differences is
likely because most studies on the consequences of child maltreatment have focused
on females.22 Studies that have included
males have typically examined the consequences of maltreatment for males and females separately and have not tested for the
interaction between gender and maltreatment, or compared magnitudes of associations across gender. Further, most of these
studies have not focused on the effects of
physical abuse per se, but rather on maltreatment in general.10,15,23–25
Studies that have examined gender differences in the long-term consequences of child
abuse have produced mixed findings. Some
Objectives. This study investigated the effects of physical abuse in childhood on health
problems in adulthood and assessed gender differences in these associations.
Methods. We used data from 8000 men and 8000 women who were interviewed in
the National Violence Against Women Survey. We used multivariate logistic regression
to test for main and interactive effects and conducted post hoc probing of significant
moderational effects.
Results. Men were more likely than women to have experienced physical abuse during childhood. Whereas abuse had negative consequences for both boys and girls, it was
generally more detrimental for girls.
Conclusions. Findings suggest the need to consider gender differences and long-term
adverse health consequences in the development of intervention strategies to address
physical abuse in childhood. (Am J Public Health. 2004;94:599–604)
of these studies have focused on sexual
abuse,26 some on physical abuse,27 and some
on maltreatment in general.15,24 Although
findings have been somewhat inconsistent,
by and large, the results suggest that females
are more affected by child abuse than males.
For example, although both men and women
who had experienced physical abuse in childhood were more likely to have higher lifetime prevalence rates of anxiety disorders
and alcohol abuse/dependence than their
nonabused counterparts, only female victims
were at increased risk for a major depressive
disorder or drug abuse/dependence.27 Similarly, in a sample of adult prisoners, child
maltreatment (combined sexual, physical, and
emotional abuse) was more strongly related
to depression, suicidal behavior, and substance abuse in women than in men.24 Consistent with this, in one of the most rigorous
longitudinal studies of child maltreatment to
date, Widom and White15 found that abused
and neglected females, but not males, were
at significantly higher risk for substance
abuse/dependence than their nonabused
counterparts.
The purpose of the current study was to
test for gender differences in the associations
of physical abuse in childhood with health
problems in adulthood. Data are from a large,
nationally representative survey conducted in
the United States with 8000 men and 8000
April 2004, Vol 94, No. 4 | American Journal of Public Health
women. This study extends the literature on
child abuse by including a large sample of
men and by specifically testing the differential
effects of physical abuse on health problems
for males and females.
METHODS
Sample and Procedures
Data for this study came from the National
Violence Against Women Survey. Respondents were told that the survey was about
personal safety. The study was conducted by
the Center for Policy Research and was jointly
funded by the National Institute of Justice and
the Centers for Disease Control and Prevention. Data were collected between November
1995 and May 1996. The sample of 8000
men and 8000 women was derived using
random-digit dialing among households with
telephones in all 50 states and the District of
Columbia. The participation rates for women
and men were 72% and 69%, respectively.
Before the interview, respondents were informed that their participation was voluntary
and that their answers would be kept confidential. Computer-assisted telephone interviewing was used. A Spanish-language version of the survey was used for Spanishspeaking respondents. For more information
on the sampling procedures, see Tjaden and
Thoennes.28
Thompson et al. | Peer Reviewed | Research and Practice | 599
 RESEARCH AND PRACTICE 
Measures
Predictor Variable: Physical Abuse in Childhood. Physical abuse in childhood was assessed using 12 questions from the Conflict
Tactics Scales.29 Respondents were asked if
they had experienced as a child (child was not
defined) any of 12 violent behaviors perpetrated by a parent, stepparent, or guardian.
The violent behaviors included having something thrown at [them]; being pushed,
grabbed, or shoved; having hair pulled; being
slapped or hit; being kicked or bitten; being
choked or experiencing attempted drowning;
being hit with some object; being beaten;
being threatened with a gun; being threatened with a weapon other than a gun; having
a gun used on [them]; and having another
type of weapon used on [them]. Respondents
were classified as victims (1 = yes; 47%) or
nonvictims (0 = no; 53%) based on whether
they had experienced any of these violent behaviors as a child.
Dependent Variables: Health Problems in
Adulthood. We included in the analyses only
those health problems for which we were
able to determine the onset and thus could
create the correct temporal sequence (i.e., occurrence of childhood abuse preceded onset
or occurrence of health problems). Physical
injury assessed whether after the age of 17
years respondents had sustained a serious injury (e.g., head injury) that was disabling or
interfered with their normal activities (8%
said yes). Chronic physical health condition assessed whether after the age of 17 years respondents had acquired a chronic physical
health problem (e.g., high blood pressure) that
was disabling or interfered with their normal
activities (11% said yes). Chronic mental health
condition assessed whether after the age of 17
years respondents had acquired a chronic
mental health problem (e.g., depression) that
was disabling or interfered with their normal
activities (2% said yes). Alcohol use assessed
whether respondents drank alcohol every day
or nearly every day during the past 12
months (6% said yes). Drug use assessed respondents’ past-month use of tranquilizers
(5% said yes), prescription pain killers (10%
said yes), antidepressants (4% said yes), and
illegal drugs such as marijuana, crack, heroin,
or angel dust (3% said yes). Current perceptions of unfavorable physical health assessed
whether respondents perceived their health to
be fair or poor (12% said yes) rather than
good, very good, or excellent.
Demographics and Other Violence Experienced. We controlled for 6 demographic variables: gender, marital status, employment status, education, race, and age. We also
controlled for sexual abuse occurring in childhood. This variable assessed whether respondents had experienced completed or attempted forced vaginal, oral, or anal
intercourse before the age of 18 years.
Interaction Term. We computed an interaction term to represent the cross-product of
gender and physical abuse in childhood.
Statistical Analysis
First, we examined the demographic composition of the total sample, as well as for
males and females separately. Second, we examined differences between males and females on the reported prevalence of physical
abuse in childhood. Third, we conducted bivariate logistic regression analyses to test the
main effects of physical abuse in childhood
on health problems in adulthood. Fourth, we
conducted multivariate logistic regression
analyses that tested for the main effects of
physical abuse in childhood on health problems in adulthood while controlling for the
demographic variables and child sexual
abuse. In this way, we could assess the
unique contribution of physical abuse in
childhood on health problems in adulthood
while holding constant other factors (e.g., age)
that have also been shown to be related to
health problems. Fifth, to determine whether
physical abuse in childhood had differential
effects on the health measures based on the
respondent’s gender, we conducted moderational analyses using Baron and Kenny’s30 criteria for testing moderation. Specifically, the
interaction term between the predictor (physical abuse in childhood) and the hypothesized
moderator (gender) must be significantly related to the dependent variables (health measures) after we controlled for the main effects
of both the predictor and hypothesized moderator variable. We conducted the moderator
analyses for all of the outcome variables. Last,
we conducted post hoc probing of significant
moderational effects using Holmbeck’s31 recommended procedures.
600 | Research and Practice | Peer Reviewed | Thompson et al.
RESULTS
Descriptive Statistics
The sample comprised 50% men aged 18
years and older and 50% women aged 18 years
and older. Approximately two thirds (64.9%)
of the sample were married (men = 66.9%,
women = 62.9%), 68.4% were employed
(men = 78.2%, women = 59.0%), 89.6% had
at least a high school education (men = 89.9%,
women = 89.3%), 82.5% were White (men =
82.8%, women = 82.2%), 9.2% were Black
(men = 8.5%, women = 9.9%), and 8.3%
(men = 8.7%, women = 7.9%) were of other
races (e.g., Native American/Alaska Native;
Asian/Pacific Islander). The mean age of
the sample was 43.33 years (SD = 15.76)
(men = 42.47 (SD = 15.33), women = 44.19
(SD = 16.13)).
Prevalence of Physical Abuse
in Childhood by Gender
Table 1 presents descriptive data for males
and females on each of the abuse items (these
data also appear in Tjaden and Thoennes28(p38)).
Men were significantly more likely than
women to have experienced 7 of the 12 violent behaviors perpetrated by a parent, stepparent, or guardian. Specifically, men were
more likely than women to have had something thrown at them that could hurt; to have
been pushed, grabbed, or shoved; to have
been slapped or hit; to have been kicked or
bitten; to have been beaten up; to have been
hit with some object; and to have been threatened with a weapon other than a gun during
their childhood. No gender differences were
found on the other items.
Bivariate Associations Between
Physical Abuse in Childhood
and Health Problems in Adulthood
Results from bivariate analyses are presented in Table 2 and indicated that respondents who had experienced physical abuse in
childhood were significantly more likely than
their nonabused counterparts to have sustained a serious injury in adulthood (crude
odds ratio [COR] = 1.73, 95% confidence interval [CI] = 1.54, 1.95), acquired a mental
health condition in adulthood (COR = 2.05,
95% CI = 1.57, 2.67), used alcohol daily in
the past year (COR = 1.51, 95% CI = 1.32,
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 1—Descriptive Data on the Reported Prevalence of Physical Abuse in Childhood for
Men and Women Interviewed in the National Violence Against Women Survey: 1995–1996a
Violent Act
Males, % (n = 8000)
Females, % (n = 8000)
Had something thrown at youb
Were pushed, grabbed, or shovedb
Had hair pulled
Were slapped or hitb
Were kicked or bittenb
Were choked or experienced drowning
Were beat upb
Were hit with objectb
Were threatened with gun
Were threatened with another weaponb
Had gun used on you
Had another weapon used on you
Experienced any violenceb
8.4
25.1
12.2
43.8
3.8
1.2
6.3
25.9
0.8
2.1
0.4
1.3
53.8
6.1
15.7
11.4
33.3
3.0
1.5
5.5
16.9
0.9
1.4
0.4
1.1
40.0
a
These data also appear in Tjaden and Thoennes.28(p38)
Differences between males and females are statistically significant.
b
1.73), and used the following substances in
the past month: tranquilizers (COR = 1.74,
95% CI = 1.51, 2.01), pain killers (COR = 1.45,
95% CI = 1.31, 1.61), antidepressants (COR =
1.84, 95% CI = 1.55, 2.17), and illegal drugs
(COR = 2.78, 95% CI = 2.25, 3.42). There
were no significant differences between
abused and nonabused respondents in their
likelihood of acquiring a physical health condition in adulthood or reporting perceptions
of unfavorable health.
Multivariate Results for Effects of
Gender and Physical Abuse in Childhood
on Health Problems in Adulthood
Findings from the multivariate analyses
are also presented in Table 2. Results for the
demographic statistical controls are presented although we only describe the main
effects for gender and physical abuse in
childhood.
Gender. When we controlled for the other
demographic variables, childhood sexual
abuse, and childhood physical abuse, men
were significantly more likely than women to
have sustained a serious injury in adulthood
(adjusted odds ratio [AOR] = 1.33, 95% CI =
1.17, 1.52), used alcohol daily in the past
year (AOR = 2.68, 95% CI = 2.29, 3.13), and
used illegal drugs in the past month (AOR =
2.80, 95% CI = 2.21, 3.56). Women were sig-
nificantly more likely than men to have acquired a physical health condition in adulthood (AOR = 0.79, 95% CI = 0.71, 0.89), acquired a mental health condition in
adulthood (AOR = 0.69, 95% CI = 0.51,
0.94), used tranquilizers in the past month
(AOR = 0.61, 95% CI = 0.51, 0.71), and used
antidepressants in the past month (AOR =
0.38, 95% CI = 0.31, 0.46).
Physical Abuse in Childhood. When we
controlled for demographic variables and
childhood sexual abuse, childhood physical
abuse was significantly associated with all
of the assessed health problems in adulthood. Specifically, respondents who had
experienced physical abuse in childhood
were significantly more likely than their
nonabused counterparts to have sustained a
serious injury in adulthood (AOR = 1.74,
95% CI = 1.53, 1.97); acquired a physical
health condition in adulthood (AOR = 1.23,
95% CI = 1.10, 1.38); acquired a mental
health condition in adulthood (AOR = 2.36,
95% CI = 1.75, 3.18); used alcohol daily in
the past year (AOR = 1.42, 95% CI = 1.24,
1.63); used tranquilizers (AOR = 2.10, 95%
CI = 1.80, 2.46), painkillers (AOR = 1.58,
95% CI = 1.42, 1.76), antidepressants
(AOR = 2.13, 95% CI = 1.78, 2.54), and illegal drugs (AOR = 2.38, 95% CI = 1.91,
2.96) in the past month; and reported per-
April 2004, Vol 94, No. 4 | American Journal of Public Health
ceptions of unfavorable health (AOR = 1.25,
95% CI = 1.13, 1.39). The largest effects
were found for mental health problems,
tranquilizer use, antidepressant use, and illegal drug use.
Test of Gender × Childhood Physical Abuse
Interaction Terms. The interaction term was
statistically significant for acquiring a
chronic mental health condition in adulthood (Wald = 6.48, P < .01) and reporting
perceptions of unfavorable health (Wald =
4.02, P < .05). These significant interaction
terms indicate that the effects of physical
abuse in childhood on health problems in
adulthood differed for men and women.
However, they did not tell us in what way
the effects differed.
To interpret the nature of the interaction
terms (i.e., for which group—men, women, or
both—physical abuse in childhood was significantly associated with health problems in
adulthood), we conducted post hoc probing
of the moderational effects. This entailed
computing 2 conditional moderator variables and then running 2 more regression
models, 1 for each conditional moderator
variable. In the first model, men were assigned a score of zero, and in the second
model, women were assigned a score of
zero. On the basis of these post hoc analyses,
we derived adjusted odds ratios, 95% confidence intervals, and t values for the associations between physical abuse in childhood
and the health measures for both men and
women. In this way, we were able to determine whether physical abuse in childhood
was significantly associated with acquiring a
mental health condition in adulthood and reporting perceptions of unfavorable health for
men only, for women only, or for both genders but in differing magnitudes.
Results from these analyses indicated that
physical abuse in childhood was significantly
related to acquiring a mental health condition in adulthood for women (AOR = 3.10;
95% CI = 2.13, 4.50; t = 5.92), but not for
men (AOR = 1.40; 95% CI = 0.86, 2.28; t =
1.36). Physical abuse in childhood also was
significantly related to reporting current perceptions of unfavorable health for women
(AOR = 1.38; 95% CI = 1.20, 1.59; t = 4.44),
but not for men (AOR = 1.12; 95% CI = 0.96,
1.30; t = 1.38).
Thompson et al. | Peer Reviewed | Research and Practice | 601
0.97
1.37
1.45
2.06
1.16
1.06
0.79
1.44
1.23
0.86, 1.09
1.20, 1.56b
1.25, 1.70b
1.72, 2.45b
0.92, 1.46
1.05, 1.06b
0.71, 0.89b
1.25, 1.83b
1.10, 1.38b
1.95
2.69
1.35
1.22
1.05
1.01
0.69
2.68
2.36
1.46, 2.60b
1.97, 3.66b
0.93, 1.95
0.79, 1.90
0.64, 1.74
1.01, 1.02b
0.51, 0.94b
1.79, 4.00b
1.75, 3.18b
1.01
0.94
0.56
0.55
0.88
1.00
2.68
1.00
1.42
0.87, 1.17
0.79, 1.12
0.43, 0.74b
0.40, 0.77b
0.66, 1.16
0.70, 1.45b
2.29, 3.13b
0.70, 1.45
1.24, 1.63b
1.36
1.76
1.29
0.59
1.30
1.02
0.61
1.92
2.10
1.17, 1.59b
1.49, 2.09b
1.04, 1.59b
0.42, 0.82b
1.00, 1.69
1.02, 1.03
0.51, 0.71b
1.48, 2.48b
1.80, 2.46b
1.14
1.58
1.47
1.48
1.35
1.01
0.96
1.46
1.58
1.02, 1.28b
1.40, 1.78b
1.26, 1.72b
1.24, 1.75b
1.16, 1.62b
1.01, 1.02b
0.86, 1.07
1.89, 1.78b
1.42, 1.76b
1.42
1.39
0.99
0.46
0.72
1.01
0.38
1.81
2.13
1.19, 1.69b
1.15, 1.68
0.75, 1.31
0.31, 0.68b
0.51, 1.03
1.01, 1.02b
0.31, 0.46b
1.38, 2.38b
1.78, 2.54b
2.52
1.29
1.24
1.04
0.92
0.93
2.80
2.21
2.38
1.99, 3.18b
1.02, 1.62b
0.91, 1.68
0.77, 1.42
0.66, 1.26
0.92, 0.94b
2.21, 3.56b
1.57, 3.13b
1.91, 2.96b
1.27
2.18
2.56
1.49
1.70
1.03
0.99
1.63
1.25
1.14, 1.41b
1.95, 2.45b
2.24, 2.91b
1.26, 1.77b
1.43, 2.02b
1.02, 1.03b
0.88, 1.10
1.33, 2.00b
1.13, 1.39b
DISCUSSION
602 | Research and Practice | Peer Reviewed | Thompson et al.
Note. OR = odds ratio; CI = confidence interval.
a
Odds ratios presented are crude odds ratios.
b
The 95% confidence interval does not include 1.
c
Odds ratios presented are adjusted odds ratios.
1.03, 1.33b
1.21, 1.60b
1.08, 1.55b
1.08, 1.62b
1.08, 1.66b
1.01, 1.02b
1.17, 1.52b
1.27, 2.05b
1.53, 1.97b
1.17
1.39
1.30
1.32
1.34
1.01
1.33
1.61
1.74
2.25, 3.42b 1.09 1.00, 1.20
1.55, 2.17b 2.78
2.05 1.57, 2.67b 1.51 1.32, 1.73b 1.74 1.51, 2.01b 1.45 1.31, 1.61b 1.84
1.73 1.54, 1.95b 0.97 0.88, 1.08
Bivariate resultsa
Physical abuse in childhood
Multivariate resultsc
Unmarried vs married
Unemployed vs employed
Education < high school vs ≥ high school
Black vs White
Other vs White
Age
Male vs female
Sexual abuse in childhood
Physical abuse in childhood
95% CI
OR
95% CI
Illegal Drugs
OR
95% CI
Antidepressants
Painkillers
OR
95% CI
OR
95% CI
Alcohol Use
OR
95% CI
Chronic Mental
OR
95% CI
OR
Chronic Physical
Injury
95% CI
OR
Predictors
TABLE 2—Odds Ratios and 95% Confidence Intervals for Predicting Health Problems in Adulthood:
National Violence Against Women Survey, November 1995 to May 1996
Tranquilizers
OR
95% CI
Perceived Health
 RESEARCH AND PRACTICE 
Using data from a large, nationally representative sample, we found that physical
abuse in childhood was more prevalent
among men than women. We also found that
physical abuse in childhood was related to
health problems in adulthood for the sample
as a whole and adversely affected the mental
health and general perceptions of health of
women more than men.
The inclusion of a large sample of men and
women allowed testing how the association
between physical abuse in childhood and
health problems in adulthood might vary by
gender. Our results are consistent with other
studies that have examined gender differences in the effects of child abuse. That is,
child abuse is generally detrimental for both
males and females. However, female abuse
victims appear to be at greater risk for some
health problems than their male counterparts.
As with prior studies, we found this to be the
case with mental health problems.24,27 However, unlike prior studies,15,24,27 we did not
find gender differences in the effects of child
abuse on alcohol or drug use.
Although this study had several strengths,
including its large national sample and the
statistical test of the interaction between
physical abuse in childhood and gender,
there were some limitations. First, even
though most of the examined relations were
statistically significant, the magnitudes of the
odds ratios were small. This is not surprising
given the amount of time that elapsed between the occurrence of physical abuse in
childhood and the occurrence of health problems in adulthood. The elapsed time allowed
a large and varied number of experiences
that could have affected the associations between child abuse and health problems in
adulthood. Second, although we took efforts
to ensure that the temporal order between
physical abuse in childhood and health problems in adulthood was correct, our findings
do not provide firm conclusions that child
abuse is causally related to health problems
in adulthood. Third, data were retrospective,
so recall bias was possible. For 2 reasons, recall bias may be particularly problematic
when the stressor under study is child victimization. First, because the amount of time that
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
elapsed since the childhood victimization allows a great and varied number of experiences, recall bias may be more likely as people perceive past events in light of their later
and current experiences.32 Second, research
suggests that emotional trauma might cause
memory impairment.33
Although our data suggest that physical
abuse in childhood is related to adverse
health outcomes in adulthood, they do not
address why childhood physical abuse would
lead to later health problems, or why some of
these associations differed by gender. Future
research should address variables that might
explain or mediate the child abuse—adult
health problems relation. Physical abuse in
childhood has been found to be related to
several potential mediators of the child
abuse—adult health problem association, including insecure attachment patterns and
more aggression (interpersonal problems),
deficits in receptive and expressive language
and poor academic achievement (cognitive
problems), and an increased likelihood of
risky sexual behavior, physical inactivity, and
smoking (risky health behaviors).7,34,35 These
potential mediators, in turn, have been found
to be associated with health problems.
Our data also do not address why physical
abuse in childhood was associated with poor
mental health and perceived physical health
in adulthood for women but not for men.
Some researchers have speculated that child
abuse may be a marker for other negative
childhood experiences that are more common for girls than for boys.27 Although we
were able to control for childhood sexual
abuse, we were not able to control for other
early childhood experiences that may account for the observed gender differences. In
1 study, abused and neglected girls were at
increased risk of substance abuse and arrests
for violent crimes compared with their counterparts even after researchers controlled for
such family background characteristics as parental substance abuse, parental arrest, and
family welfare status. For boys, however, previously significant bivariate relations between
abuse and these outcomes were reduced to
nonsignificance after researchers controlled
for these family background variables.15
Some researchers have speculated that females may be more likely than males to en-
gage in self-blame after child abuse, and that
this accounts for females’ increased risk for
mental health problems in adulthood.36,37
Others have speculated that males and females react to stress differently, with females
being more likely to internalize stress symptoms (e.g., depression) and males being more
likely to externalize stress reactions (e.g., aggressive behaviors).15,38 Because our study
did not include measures of externalizing behaviors, we could not test this hypothesis. It
has also been suggested that females may be
more likely than males to evidence problems
after abuse because the abuse was more persistent or severe.36 Data from the current
study indicate that males actually experience
more forms of physical abuse in childhood
than females (men had higher mean score on
sum of Conflict Tactics Scale items than
women did), so differences in the magnitude
of abuse did not account for our finding of
gender differences. However, Widom and
White15 suggest that abuse may be tolerated
more for boys than girls, and hence nonvictim comparison groups for boys may have
more false negatives than nonvictim comparison groups for girls. More research is needed
to determine in what ways and why males
and females differ in the consequences of
physical abuse in childhood.
Although this study did not directly address why there may be gender differences
in the association between physical abuse in
childhood and health problems in adulthood,
the findings suggest the importance of considering potential long-term adverse health
consequences in the development of intervention strategies to address physical abuse
in childhood. Health care providers should
be aware that physical abuse in childhood
might be associated with health problems in
adulthood, especially among females. Intervening at an early stage may reduce a child’s
likelihood of developing long-term health sequelae and also reduce the public health
burden of child abuse by preventing future
health problems. Attention also should be
paid to the primary prevention of physical
abuse of children. Some efforts to prevent
the initial occurrence of child abuse have
shown promising results.39–41
In sum, little research has focused on differences between males and females in the
April 2004, Vol 94, No. 4 | American Journal of Public Health
consequences of physical abuse in childhood.
This study helps to address this research gap
by examining the moderating role of gender
in the associations between physical abuse in
childhood and health problems in adulthood.
We found that childhood physical abuse was
more prevalent among males, and although it
was related to adverse health outcomes for
both genders, the effect was generally greater
for females. These findings can help inform
intervention strategies by alerting public
health and medical practitioners of the potential for physical abuse in childhood to be related to health problems in adulthood.
About the Authors
Martie P. Thomson and J. B. Kingree are with the Department of Public Health Sciences, Clemson University,
Clemson, SC. Sujata Desai is with the Family Violence
Unit, Texas Health and Human Services Commission,
Austin, Tex.
Requests for reprints should be sent to Martie P.
Thompson, PhD, Department of Public Health Sciences,
Clemson University, 511 Edwards Hall, Clemson, SC
29634–0745 (e-mail: [email protected]).
This article was accepted July 18, 2003.
Contributors
M. P. Thompson took the lead role in conceptualizing
the study, conducting the analyses, and writing the article. J. B. Kingree and S. Desai assisted with study conceptualization and manuscript preparation.
Human Participant Protection
No protocol approval was needed for this study.
References
1. Child Maltreatment 2000: Reports From the States
to the National Child Abuse and Neglect Data System.
Washington, DC: US Department of Health and
Human Services, Administration on Children, Youth,
and Families; 2002.
2. Straus MA, Gelles RJ. Physical Violence in American Families: Risk Factors and Adaptations to Violence in
8145 Families. New Brunswick, NJ: Transaction; 1990.
3. Brown G, Anderson B. Psychiatric morbidity in
adult inpatients with childhood histories of sexual and
physical abuse. Am J Psychiatry. 1991;148:55–61.
4. Brown J, Cohen P, Johnson J, Smailes EM. Childhood abuse and neglect: specificity of effects on adolescent and young adult depression and suicidality.
J Am Acad Child Adolesc Psychiatry. 1999;38:1490–
1496.
5. Bryant SL, Range LM. Suicidality in college
women who were sexually and physically abused and
physically punished by parents. Violence Vict. 1995;10:
195–201.
6. Fox KM, Gilbert BO. The interpersonal and psychological functioning of women who experienced
childhood physical abuse, incest, and parental alcoholism. Child Abuse Negl. 1994;18:849–858.
Thompson et al. | Peer Reviewed | Research and Practice | 603
 RESEARCH AND PRACTICE 
7. Kaplan SJ, Pelcovitz D, Labruna V. Child and adolescent abuse and neglect research: a review of the
past 10 years, I: Physical and emotional abuse and
neglect. J Am Acad Child Adolesc Psychiatry. 1999;38:
1214–1222.
8. McCauley J, Kern DE, Kolodner K, et al. Clinical
characteristics of women with a history of childhood
abuse. JAMA. 1997; 277:1362–1368.
9. Perez CM, Widom CS. Childhood victimization
and long-term intellectual and academic outcomes.
Child Abuse Negl. 1994;18:617–633.
10. Silverman A, Reinherz H, Giaconia R. The longterm sequela of child and adolescent abuse: a longitudinal community study. Child Abuse Negl. 1996;20:
709–723.
11. Thompson MP, Kaslow NJ, Lane DB, Kingree JB.
Childhood maltreatment, PTSD, and suicidal behavior
among African American females. J Interpersonal Violence. 2000;15:3–15.
12. Weiler BL, Widom CS. Psychopathy and violent
behavior in abused and neglected young adults. Criminal Behav Mental Health. 1996;6:253–271.
13. Widom CS. Posttraumatic stress disorder in
abused and neglected children grown up. Am J Psychiatry. 1999;156:1223–1229.
14. Widom CS. Childhood victimization and the development of personality disorders: unanswered questions remain. Arch Gen Psychiatry. 1999;56:607–608.
15. Widom CS, White HR. Problem behaviors in
abused and neglected children grown-up: prevalence
and co-occurrence of substance abuse, crime, and violence. Criminal Behav Mental Health. 1997;7:287–310.
16. Drossman DA, Leserman J, Nachman G, et al.
Sexual and physical abuse in women with functional or
organic gastrointestinal disorders. Ann Intern Med.
1990;113:828–833.
17. Leserman J, Drossman DA, Li ZM, Toomey TC,
Nachman G, Glogau L. Sexual and physical abuse history in gastroenterology practice: how types of abuse
impact health status. Psychosom Med. 1996;58:4–15.
18. Moeller TP, Bachmann GA, Moeller JR. The combined effects of physical, sexual, and emotional abuse
during childhood: long-term health consequences for
women. Child Abuse Negl. 1993;17:623–640.
19. Walker EA, Gelfand A, Katon WJ, et al. Adult
health status of women with histories of childhood
abuse and neglect. Am J Med. 1999;107:332–339.
20. Walker EA, Unutzer J, Rutter C, et al. Costs of
health care use by women HMO members with a history of childhood abuse and neglect. Arch Gen Psychiatry. 1999;56:609–613.
male and female prisoners. Criminal Justice Behav.
1997;24:455–477.
26. Beitchman JH, Zucker KJ, Hood JE, DaCosta GA,
Akman D, Cassavia E. A review of the long-term effects of child sexual abuse. Child Abuse Negl. 1992;16:
101–118.
27. MacMillan HL, Fleming JE, Streiner DL, et al.
Childhood abuse and lifetime psychopathology in a community sample. Am J Psychiatry. 2001;158:1878–1883.
28. Tjaden P, Thoennes N. Full Report of the Prevalence, Incidence, and Consequences of Violence Against
Women: Findings from the National Violence Against
Women Survey. Washington, DC: US Department of
Justice; 2000. Report NCJ 183781.
29. Straus M. Measuring intrafamily conflict and violence: The Conflict Tactics (CT) scales. J Marriage Fam.
1979;41:75–88.
30. Baron RM, Kenny DA. The moderator-mediator
variable distinction in social psychological research:
conceptual, strategic, and statistical considerations.
J Pers Soc Psychol. 1986;51:1173–1182.
31. Holmbeck GN. Post-hoc probing of significant
moderational and mediational effects in studies of pediatric populations. J Pediatr Psychol. 2002;27:87–96.
32. Widom CS. Does violence beget violence? A critical examination of the literature. Psychol Bull. 1989;
106:3–28.
33. Figley C. Trauma and its wake: the study and
treatment of posttraumatic stress disorder. New York,
NY: Bruner/Mazel; 1885.
34. Felitti VJ, Anda RF, Nordenberg D, et al. Relationship of childhood abuse and household dysfunction to
many of the leading causes of death in adults. Am J
Prev Med. 1998;14:245–258.
35. Widom CS, Kuhns JB. Childhood victimization
and subsequent risk for promiscuity, prostitution, and
teenage pregnancy: a prospective study. Am J Public
Health. 1996;86:1607–1612.
36. Cutler SE, Nolen-Hoeksema S. Accounting for sex
differences in depression through female victimization:
childhood sexual abuse. Sex Roles. 1991;24:425–438.
37. Nolen-Hoeksema S. Sex Differences in Depression.
Stanford, Calif: Stanford University Press; 1990.
Confronting
Violence
George A. Gellert, MD
With a foreword by Frank Keating,
Governor of Oklahoma
T
his book discusses interpersonal violence, including child and elder
abuse, sexual assault, murder, suicide,
stranger violence, and youth violence.
It is written in a series of easy-to-reference questions and answers, and provides tips for avoiding high-risk situations. Confronting Violence includes lists
of organizations and public agencies
that provide help.
The 2nd Edition includes a new preface by APHA Executive Director
Mohammad N. Akhter as well as new
statistics and references to recent
events, such as the Columbine High
School massacre and the child sex
abuse scandal in the Catholic Church.
ISBN 0-87553-001-X
2002 ❚ 384 pages ❚ softcover
$19.95 APHA Members
$24.95 Non-members
Plus shipping and handling
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
CV2D07J9
FAX: (301) 843-0159
38. Dohrenwend BP, Dohrenwend BS. Sex differences in
psychiatric disorders. Am J Sociol. 1976;81:1447–1459.
21. National Research Council. Panel on Research on
Child Abuse and Neglect. Understanding Child Abuse
and Neglect. Washington, DC: National Academy Press;
1993.
39. Olds DL, Eckenrode J, Henderson CR, et al. Longterm effects of home visitation on maternal life course
and child abuse and neglect: fifteen-year follow up of a
randomized trial. JAMA. 1997;278:637–643.
22. Haskett ME, Marziano B, Dover ER. Absence of
males in maltreatment research: a survey of recent literature. Child Abuse Negl. 1996;20:1175–1182.
40. Chalk R, King PA. Violence in Families: Assessing
Prevention and Treatment Programs. Washington, DC:
National Academy Press; 1998.
23. Rosen LN, Martin L. Impact of childhood abuse
history on psychological symptoms among male and female soldiers in the US Army. Child Abuse Negl. 1996;
20:1149–1160.
41. Luzker JR, Bigelow KM, Doctor RM, Gershater
RM, Greene BF. An ecobehavioral model for the prevention and treatment of child abuse and neglect: history and applications. In: Lutker J, ed. Handbook of
Child Abuse Research and Treatment. New York, NY:
Plenum Press; 1998: 239–266.
24. McClellan DS, Farabee D, Crouch BM. Early victimization, drug use, and criminality: a comparison of
Second
Edition
25. Widom CS, Weiler BL, Cottler LB. Childhood victimization and drug abuse: a comparison of prospective
and retrospective findings. J Consult Clin Psychol. 1999;
67:867–880.
604 | Research and Practice | Peer Reviewed | Thompson et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Prevalence and 3-Year Incidence of Abuse
Among Postmenopausal Women
| Charles P. Mouton, MD, MS, Rebecca J. Rodabough, MS, Susan L. D. Rovi, PhD, Julie L. Hunt, PhD, Melissa A. Talamantes, MA, Robert G. Brzyski,
MD, PhD, and Sandra K. Burge, PhD
Abuse, including physical, sexual, financial,
or psychological mistreatment, is a serious
problem for adults aged 65 years and older.1
According to the National Elder Abuse Incidence Study, approximately 450 000 older
adults in domestic settings were abused, neglected, or both during 1996.2 This number
increases to approximately 551 000 when
older adults who experienced self-neglect are
included. In a population-based survey of
metropolitan Boston, Pillemer and Finkelhor
found a rate of elder abuse of 3.2%.3 In the
long-term care setting, 23% of older adults
either have been or still are victims of
abuse.4–6
The public health implications of abuse
are its associations with premature mortality
and morbidity.7–13 Lachs and colleagues
found that among older adults who were victims of abuse, only 9% were alive 2 years
later compared with 40% of older adults
who had not been abused.11 Other studies
have found a risk of death for older abuse
victims that is 3 times higher than for nonvictims.12,13 The direct medical costs associated
with these violent injuries are estimated to
add over $5.3 billion to the nation’s annual
health expenditures (K. Fullin et al., unpublished data, 1994).
Gender is an important factor in abuse exposure. Worldwide, between 10% and 50%
of women report being physically assaulted
at some point in their adult lives; 14% to
25% of women seen at ambulatory medical
clinics and 20% of women seen in emergency departments have been physically
abused.7–10 Older, postmenopausal women
(65 years or older) are more likely than older
men to be the victims of all forms of abuse,
except for abandonment, even when taking
into account the fact that they make up a
larger proportion of the aging population.3,4,14,15 While females made up about
57.6% of the total national population aged
Objectives. We examined prevalence, 3-year incidence, and predictors of physical and
verbal abuse among postmenopausal women.
Methods. We used a cohort of 91 749 women aged 50 to 79 years from the Women’s
Health Initiative. Outcomes included self-reported physical abuse and verbal abuse.
Results. At baseline, 11.1% reported abuse sometime during the prior year, with 2.1%
reporting physical abuse only, 89.1% reporting verbal abuse only, and 8.8% reporting
both physical and verbal abuse. Baseline prevalence was associated with service occupations, having lower incomes, and living alone. At 3-year follow-up, 5.0% of women
reported new abuse, with 2.8% reporting physical abuse only, 92.6% reporting verbal
abuse only, and 4.7% reporting both physical and verbal abuse.
Conclusions. Postmenopausal women are exposed to abuse at similar rates to younger women; this abuse poses a serious threat to their health. (Am J Public Health. 2004;
94:605–612)
65 years and older in 2000, women were
the victims in 76.3% of reports of emotional
or psychological abuse, 71.4% of physical
abuse, 63.0% of financial or material exploitation, and 60.0% of neglect.2 Women in
the early postmenopausal ages (aged 50–65
years) are exposed to abuse by intimate partners at a rate of 0.5 per 1000 and account
for 30% of homicides committed by an intimate partner.16 Cognitive or physical impairment, or both, is an additional factor in
abuse exposure. In a study of mortality due
to mistreatment of elders, over 85% of victims of elder abuse had some impairment of
their activities of daily living.2,11
Unfortunately, most studies examining the
associations with abuse exposure have focused on younger women in their childbearing years or on frail, functionally dependent
older adults. To date, no study has examined
the associations with physical and verbal
abuse in functionally independent, cognitively
intact, older women. We conducted this study
to (1) describe the 1-year baseline prevalence
and 3-year incidence of physical and verbal
abuse in a cohort of functionally independent
older women and (2) examine the sociodemographic factors and health behaviors associated with this prevalence and incidence of
abuse.
April 2004, Vol 94, No. 4 | American Journal of Public Health
METHODS
Subjects
We analyzed survey responses from
93 205 women enrolled in the observational
study arm of the Women’s Health Initiative
(WHI). The study design of the WHI and its
observational study arm has been described
in detail previously.17 In brief, the WHI is a
large, multicenter study with 2 components,
an observation study and a clinical trial. Postmenopausal women, aged 50 to 79 years old
at baseline, were recruited through targeted
mass mailings to voter registration lists, vehicle registration lists, and driver’s license lists
and invited to participate in the clinical trial.
Subjects who were eligible and interested enrolled in 1 or more of the 3 WHI clinical trials: (1) hormone replacement therapy to prevent cardiovascular disease, (2) a low-fat,
high-fiber diet to prevent breast and colorectal cancer, and (3) calcium and vitamin D to
prevent osteoporosis-related fractures.
Subjects who were ineligible or unwilling
to participate in the clinical trials were invited to participate in the observational
study, a longitudinal study of health outcomes. In general, women were ineligible
for any clinical trial if they had a medical
condition with a predicted survival of less
Mouton et al. | Peer Reviewed | Research and Practice | 605
 RESEARCH AND PRACTICE 
than 3 years, cancer within the last 10
years, or dementia rendering them unable
to answer study questions. Women were excluded from the hormone replacement therapy clinical trial study if they were taking
hormone replacement therapy and were unwilling to stop use. Women were ineligible
for the low-fat diet clinical trial study if they
had a baseline body mass index of less than
18 kg/m2 or if they consumed more than
6000 kcal per day. Women were ineligible
for the vitamin D/calcium clinical trial study
if they had a history of an osteoporosisrelated fracture or medical contraindications
to taking study medication. All observational study participants completed several
study questionnaires at the time of enrollment, including questions about abuse in
the previous year. Three years after enrollment, participants were scheduled for a
follow-up clinic visit and administered the
same study questionnaires.
To determine the occurrence of physical
abuse at baseline, the following question was
asked: “Over the past year, were you physically abused by being hit, slapped, pushed,
shoved, punched or threatened with a
weapon by a family member or close friend?”
Subjects could choose from the following responses: (1) no, (2) yes, and it upset me not
too much, (3) yes, and it upset me moderately
(medium), or (4) yes, and it upset me very
much. We classified women who answered
yes (responses 2–4) as having been exposed
to physical abuse.
To determine the occurrence of verbal
abuse at baseline, the following question was
asked: “Over the past year, were you verbally abused by being made fun of, severely
criticized, told you were a stupid or worthless person, or threatened with harm to
yourself, your possessions, or your pets, by a
family member or close friend?” Subjects
could chose from the following responses:
(1) no, (2) yes, and it upset me not too
much, (3) yes, and it upset me moderately
(medium), or (4) yes, and it upset me very
much. We classified women who answered
yes (responses 2–4) as having been exposed
to verbal abuse. Women who fell into either
the physical or verbal abuse categories at
baseline determined the exposure group for
our abuse prevalence estimates.
Using these questions, women were
screened for physical and verbal abuse again
3 years after enrollment. Women who responded no at baseline but who answered yes
3 years after enrollment determined our 3year incidence estimates of abuse. Any
woman who screened positive for physical or
verbal abuse at baseline or follow-up was
given information about the Domestic Violence Hotline, self-help information about domestic violence, and information about the
nearest battered women’s shelter. They were
also urged to seek help from adult protective
services and receive psychological counseling
for domestic violence.
Responses to these abuse questions determined 3 mutually exclusive variables: physical abuse only, verbal abuse only, and physical and verbal abuse. These 3 variables
became our main outcomes of interest. Our
baseline predictor variables included age,
race/ethnicity, occupation, marital status, income, education, smoking, alcohol intake, and
living arrangement. These predictor variables
were chosen on the basis of previous literature suggesting an association of sociodemographics (age, race/ethnicity, education, occupation, and income) and health behaviors
(smoking and alcohol use) with elder abuse
and intimate partner violence.18–20
Data Analysis
We first examined the descriptive statistics
of the predictor variables and the abuse variables (at baseline and year 3): no abuse, physical abuse only, verbal abuse only, and combined physical and verbal abuse. Chi-square
tests were then performed to examine the bivariate association of the various variables
with reports of physical, verbal, and combined physical and verbal abuse vs no abuse.
The bivariate analyses examined the association of each variable without adjusting for
other factors.
We considered abuse to be the outcome
variable and our sociodemographic and
health behavior variables to be covariates.
Two sets of multivariate regression models
were developed for both baseline abuse prevalence data and 3-year abuse incidence data.
Complete case analysis was used for all modeling and all explanatory variables were kept
in each model, regardless of statistical signifi-
606 | Research and Practice | Peer Reviewed | Mouton et al.
cance. Thus, estimates of odds ratios for each
predictor variable were adjusted for all other
variables in the model. Continuous variables
were included as linear covariates and categorical variables as indicator levels. Logistic
regression models were developed to examine the association of study covariates with
each level of abuse status versus no abuse
(i.e., a separate model for each level of abuse
vs no abuse). All analyses were performed
with the SAS System, Version 8 (SAS Institute
Inc, Cary, NC).
RESULTS
Of the 91 749 subjects responding to survey questions on abuse at baseline, 10 199
(11.1%) reported exposure to abuse within
the preceding 12 months. Most women in our
sample were non-Hispanic White (82.9%),
well educated (40.3% had at least a college
degree), and married (64.9%) (Table 1).
While most women in our sample were not
currently employed, those who were employed tended to work in managerial or professional occupations. Of those women who
were married, most reported that their
spouse was not currently employed. Most
women reported drinking less than 1 alcoholic beverage per week and were not currently smokers.
Of the 10 199 women exposed to abuse,
218 women (2.1%) were exposed to physical
abuse only, 9083 (89.1%) to verbal abuse
only, and 898 (8.8%) to physical and verbal
abuse sometime during the year before the
baseline interview. Exposure to abuse was associated with being in the younger age cohort (<58 years), being of non-White race/
ethnicity, having less than a high school education, having a family income of less than
$20 000, being divorced or separated, being
a past or current smoker, and drinking more
than 1 drink per week (all P values < .01)
(Table 1).
The associations with exposure to physical abuse at baseline only, after control for
other covariates, are shown in Table 2.
Black women were 2.84 times more likely
(95% confidence interval [CI] = 1.89, 4.26)
to report exposure to physical abuse only at
baseline than non-Hispanic White women.
Other ethnic minority subgroups were also
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 1—Baseline Abuse Prevalence in Cohort of Postmenopausal Women, by Subjects’
Characteristics (N = 91 749)
Any Abuse (n = 10 199)
Characteristic
Overall
Age, y
< 58
59–64
65–69
70–74
> 74
Ethnicity
American Indian/Alaska Native
Asian/Pacific Islander
African American
Hispanic/Latino
White
Education
0–8 y
Some HS/HS diploma/GED
School after high school
College graduate or higher
Family income, $
< 20 000
20 000–34 999
35 000–49 999
50 000–74 999
> 75 000
Occupation
Managerial/professional
Technical/sales/administrative
Service/labor
Homemaker only
Currently employed (yes)
Marital status
Never married
Divorced/separated
Widowed
Presently married
Partner’s main job
Homemaker
Managerial/professional
Technical/sales/administrative
Service/labor
Other
Partner currently employed (yes)
Smoking
Never smoked
Past smoker
Current smoker
No Abuse, No. (%)
(n = 81 550)
Physical Abuse
Only, No. (%)
P
218 (2.1)
Verbal Abuse
Only, No. (%)
9 083 (89.1)
.19
22 136 (27.1)
20 620 (25.3)
18 367 (22.5)
14 052 (17.2)
6 375 (7.8)
73 (33.5)
57 (26.1)
43 (19.7)
29 (13.3)
16 (7.3)
339 (0.4)
2 393 (2.9)
6 682 (8.2)
2 950 (3.6)
69 186 (84.8)
4 (1.8)
8 (3.7)
55 (25.2)
19 (8.7)
132 (60.6)
1 284 (1.6)
16 141 (20.0)
29 242 (36.1)
34 232 (42.3)
13 (6.0)
62 (28.8)
77 (35.8)
63 (29.3)
11 730 (15.5)
17 567 (23.2)
15 287 (20.2)
15 378 (20.3)
15 655 (20.7)
65 (33.0)
39 (19.8)
40 (20.3)
29 (14.7)
24 (12.2)
33 991 (43.7)
22 155 (28.5)
13 151 (16.9)
8 489 (10.9)
28 018 (35.4)
66 (33.0)
48 (24.0)
65 (32.5)
21 (10.5)
69 (32.9)
3 940 (4.9)
12 379 (15.3)
14 717 (18.1)
50 133 (61.8)
8 (3.7)
59 (27.4)
38 (17.7)
110 (51.2)
152 (0.3)
26 926 (56.3)
6 697 (14.0)
8 371 (17.5)
5 715 (11.9)
18 446 (38.0)
2 (1.9)
37 (35.2)
15 (14.3)
33 (31.4)
18 (17.1)
36 (34.6)
41 115 (51.0)
34 568 (42.9)
4 865 (6.0)
100 (46.3)
96 (44.4)
20 (9.3)
P
<.001
<.001
<.001
14 (1.6)
35 (3.9)
137 (15.3)
103 (11.5)
609 (67.8)
<.001
150 (1.7)
1 541 (17.1)
3 527 (39.2)
3 784 (42.0)
<.001
<.001
47 (5.3)
193 (21.7)
381 (42.9)
267 (30.1)
<.001
1 496 (17.8)
2 002 (23.8)
1 676 (19.9)
1 653 (19.7)
1 580 (18.8)
<.001
<.001
283 (34.5)
204 (24.9)
118 (14.4)
114 (13.9)
101 (12.3)
<.001
3 655 (42.4)
2 526 (29.3)
1 596 (18.5)
852 (9.9)
3 340 (38.1)
<.001
<.001
335 (3.7)
1 665 (18.4)
999 (11.1)
6 032 (66.8)
<.001
.003
.09
<.001
384 (42.8)
210 (23.4)
162 (18.0)
102 (11.4)
40 (4.5)
56 (0.6)
219 (2.4)
639 (7.0)
458 (5.0)
7 711 (84.9)
<.001
<.001
260 (31.1)
254 (30.4)
221 (26.4)
101 (12.1)
330 (38.5)
4 410 (49.3)
3 844 (43.0)
694 (7.8)
.06
<.001
12 (1.3)
279 (31.2)
127 (14.2)
475 (53.2)
<.001
22 (0.4)
2 986 (51.9)
815 (14.2)
1 219 (21.2)
716 (12.4)
2 228 (38.4)
P
898 (8.8)
<.001
3 229 (35.5)
2 408 (26.5)
1 821 (20.0)
1 122 (12.4)
503 (5.5)
.44
<.001
Physical and Verbal
Abuse, No. (%)
<.001
<.001
<.001
4 (0.9)
168 (37.5)
67 (15.0)
140 (31.3)
69 (15.4)
186 (41.1)
<.001
<.001
420 (47.7)
354 (40.2)
107 (12.1)
Continued
April 2004, Vol 94, No. 4 | American Journal of Public Health
Mouton et al. | Peer Reviewed | Research and Practice | 607
 RESEARCH AND PRACTICE 
TABLE 1—Continued
Alcohol intake
Nondrinker
Past drinker
< 1 drink/wk
1–6 drinks/wk
≥ 7 drinks/wk
Living alone (yes)
<.001
9 139 (11.3)
14 879 (18.3)
25 516 (31.5)
21 122 (26.0)
10 430 (12.9)
21 940 (27.1)
24 (11.3)
65 (30.5)
56 (26.3)
43 (20.2)
25 (11.7)
52 (24.0)
.31
<.001
881 (9.8)
1 975 (21.9)
3 034 (33.6)
2 145 (23.7)
999 (11.1)
1 884 (20.9)
<.001
<.001
138 (15.5)
242 (27.2)
266 (29.9)
156 (17.5)
89 (10.0)
233 (26.2)
.56
Note. HS =high school; GED = general equivalency diploma.
TABLE 2—Multivariate Associations With Baseline Reports of Abuse vs No Abuse Among
Postmenopausal Women
Age, y (vs 50–58 y)
59–64
65–69
70–79
Race (vs non-Hispanic White)
American Indian/Alaska Native
Asian/Pacific Islander
African American
Hispanic American
Education (vs college graduate)
≤ HS diploma
Some college/technical school
Income, $ (vs > $75 000)
< 20 000
20 000–34 999
35 000–49 999
50 000–75 000
Employment (vs managerial)
Technical
Service
Homemaker
Marital status (vs married)
Never married
Divorced
Widowed
Smoking status (vs never smoked)
Past smoker
Current smoker
Alcohol use (vs past/never drank)
< 1 drink/wk
≥ 1 drink/wk
Living alone (vs no)
Physical Abuse Only
OR (95% CI)
Verbal Abuse Only
OR (95% CI)
Physical and Verbal Abuse
OR (95% CI)
0.81 (0.55, 1.19)
0.76 (0.49, 1.17)
0.67 (0.42, 1.06)
0.79 (0.75, 0.84)
0.68 (0.64, 0.73)
0.57 (0.53, 0.62)
0.56 (0.47, 0.68)
0.49 (0.39, 0.60)
0.38 (0.30, 0.48)
2.54 (0.62, 10.45)
2.04 (0.98, 4.24)
2.84 (1.89, 4.26)
1.74 (0.93, 3.26)
1.34 (0.98, 1.83)
0.79 (0.68, 0.92)
0.73 (0.66, 0.80)
1.08 (0.96, 1.22)
3.10 (1.73, 5.54)
1.52 (1.04, 2.24)
1.26 (0.99, 1.59)
1.95 (1.49, 2.54)
1.45 (0.90, 2.33)
1.10 (0.72, 1.66)
0.70 (0.65, 0.76)
0.98 (0.93, 1.05)
0.82 (0.64, 1.04)
1.14 (0.94, 1.38)
2.72 (1.43, 5.18)
1.64 (0.93, 2.89)
1.18 (0.73, 1.90)
1.42 (0.91, 2.22)
2.12 (1.86, 2.42)
1.72 (1.56, 1.88)
1.43 (1.33, 1.53)
1.22 (1.14, 1.30)
5.15 (3.75, 7.06)
3.14 (2.40, 4.11)
1.94 (1.54, 2.44)
1.29 (1.01, 1.64)
0.95 (0.61, 1.49)
1.68 (1.08, 2.62)
1.03 (0.57, 1.86)
1.03 (0.97, 1.10)
1.08 (1.00, 1.17)
0.96 (0.87, 1.05)
1.16 (0.94, 1.43)
1.40 (1.12, 1.75)
1.04 (0.78, 1.40)
0.83 (0.35, 1.99)
1.55 (0.97, 2.49)
1.06 (0.63, 1.78)
0.71 (0.62, 0.82)
1.05 (0.96, 1.14)
0.64 (0.58, 0.71)
0.28 (0.15, 0.52)
1.42 (1.13, 1.79)
0.75 (0.57, 0.99)
1.34 (0.97, 1.84)
1.30 (0.74, 2.26)
1.06 (1.01, 1.12)
1.30 (1.18, 1.43)
1.07 (0.91, 1.26)
1.69 (1.33, 2.16)
0.79 (0.54, 1.17)
1.02 (0.70, 1.50)
0.61 (0.39, 0.95)
0.97 (0.92, 1.03)
0.80 (0.76, 0.86)
0.75 (0.69, 0.81)
0.86 (0.72, 1.03)
0.73 (0.60, 0.89)
0.76 (0.61, 0.95)
Note. OR = odds ratio; CI = confidence interval; HS =high school.
608 | Research and Practice | Peer Reviewed | Mouton et al.
more likely to report physical abuse exposure than non-Hispanic White women, although these associations did not reach statistical significance. When other variables
are controlled for, women who had incomes
of less than $20 000 (odds ratio [OR] =
2.72; 95% CI = 1.43, 5.18) and who worked
in service-type occupations (OR = 1.68;
95% CI = 1.08, 2.62) were more likely to
report exposure to physical abuse. Women
who were living alone were nearly half as
likely to report exposure to physical abuse
at baseline.
Table 2 also demonstrates the multivariate
associations with exposure to verbal abuse
only at baseline. When other variables are
controlled, women in the 3 older age categories were less likely than women aged 50
to 58 years to report verbal abuse only at
baseline. Black and Asian/Pacific Islander
women were less likely to report verbal
abuse only at baseline than non-Hispanic
White women (OR = 0.73 and 0.79, respectively), as were women who were never
married/widowed, drank less than 1 drink
per week, or who lived alone. Women who
had incomes of less than $75 000 annually
or who were current smokers were more
likely to report verbal abuse only.
For women reporting both physical and
verbal abuse, those in the older age categories were less likely to report abuse at baseline than women aged 50 to 58 years, as
were women who were never married, were
widowed, or lived alone. Ethnic minority
women, those with incomes of less than
$75 000, those employed in service-type jobs,
and those who were current smokers were
more likely to report both physical and verbal
abuse.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Of the 48 522 women with follow-up data
at year 3 and who reported no exposure to
domestic violence at baseline, 2431 women
(5.01%) reported exposure to abuse at their
follow-up visit 3 years later. Of these 2431
women, 67 (2.8%) reported physical abuse
only, 2250 (92.6%) verbal abuse only, and
114 (4.7%) both physical and verbal abuse
(Table 3). Ethnicity was associated with all 3
abuse categories, while education and income
were associated with both physical abuse only
and verbal abuse only. Age and marital status
were associated with verbal abuse only and
the combined abuse category.
The associations with 3-year incident exposure to physical and verbal abuse, after control for other covariates, are demonstrated in
Table 4. Women in the 2 older age categories
were less likely to have been exposed to either physical or verbal abuse at the 3-year
follow-up visit than women aged 50 to 58
years. Non-White women were more likely to
report exposure to either physical or verbal
abuse at the 3-year follow-up visit than nonHispanic White women, as were women who
had lower annual household incomes (i.e., incomes of less than $75 000 annually).
Women who were past or current smokers
were more likely to report 3-year incident exposure to verbal abuse only. Women who
were living alone were less likely to report incident exposure to verbal abuse only.
DISCUSSION
In this study, we found that many functionally independent, older women are exposed
to physical and verbal abuse. Our finding that
1.2% of the women in our self-selected, postmenopausal cohort were physically abused is
similar to the prevalence estimates reported
in other population-based surveys.21–23 However, our finding that 10% of women reported verbal abuse is 3 to 10 times higher
than population-based results showing a 1.1%
to 3.2% prevalence of verbal abuse.21–24
These findings suggest that even for nondependent older women, physical and verbal
abuse is occurring at rates similar to, or
higher than, those for younger women. Perhaps more importantly, we found that 3.7 per
1000 older women reported new exposure to
physical abuse and 46 per 1000 older
women reported becoming new victims of
verbal abuse. This result compares with population estimates that show the annual incidence of abuse ranging from 735 000 to 2
million out of an estimated 31 million older
women.24 To our knowledge, our findings are
the first estimate of incidence of physical and
verbal abuse in a large sample of postmenopausal women.
Exposure to abuse among these postmenopausal women is associated with younger
age and lower income. These findings are
comparable to data in intimate partner
abuse research but contrast with elder abuse
data. Studies demonstrate that victims of intimate partner abuse are more likely to be
younger than 35 years old, not to be college
educated, and to have lower socioeconomic
status.11,18,19,25–28 Studies on abuse among
older adults, however, show that advanced
age (>75 years old), functional dependency,
shared living arrangement, social isolation,
depression, personality disorder, cognitive
impairment, and excessive use of drugs or
alcohol place an older adult at risk for
abuse.20,28
The discrepancies between our findings
and previous research with regard to age and
living situation may be related to the fact that
all the women in our sample were functionally independent. Given the high level of
physical functioning in our sample, it is unlikely that abuse by caregivers, neglect, or
self-neglect was a predominate cause of abuse
in our study. By focusing on the frail elderly,
most of the previous research on the abuse of
older adults was influenced by issues of caregiver abuse and neglect. These findings suggest that there is a transition in abuse risk factors for women as they age. If a woman
remains functionally independent, the risk
factors for abuse mirror those for intimate
partner violence. If she becomes dependent
functionally, and perhaps more vulnerable,
the risk factors for abuse mirror those of caregiver abuse and neglect.
One interesting finding was the relationship
between race/ethnicity and abuse. NonHispanic White women reported more exposure to verbal abuse than their minority counterparts, while African American women
reported more exposure to physical abuse.
Our 3-year incidence results show a similar
April 2004, Vol 94, No. 4 | American Journal of Public Health
pattern for African American women, with
less verbal abuse in this group, although the
results did not reach statistical significance.
The 3-year incidence results, however, show
a stronger association of all 3 types of abuse
exposure among Hispanic women.
These results are in contrast to the findings
on elder abuse and abuse in younger women
that show non-Whites as being more likely to
be victimized by all types of abuse. Previous
research demonstrates a 4-fold influence of
ethnicity on reports of abuse.19 There has not
been any distinction demonstrated in the
types of abuse experienced across racial subgroups. Since intimate relationships have
strong culturally specific meanings, the interpretation of what constitutes abuse across cultures may influence the association of racial/
ethnic group with certain types of abuse. Perhaps race/ethnicity is a factor for abuse exposure that has more specific targets in older,
functionally independent women as contrasted with more broad categories of race/
ethnicity in more frail older women. Thus,
despite their older age, functionally independent victims of abuse in our study seem to be
similar to younger victims of intimate partner
violence.
In addition to race/ethnicity, other lifestyle
factors are associated with abuse exposure.
Current smoking seems to be associated with
greater exposure to abuse, particularly for
verbal abuse. However, alcohol use seems to
be less likely among those who were exposed
to abuse, particularly verbal abuse. The associations with verbal abuse are consistent for
both our prevalence and 3-year incidence results. While previous research has not examined smoking behaviors in women exposed to
violence, our findings regarding alcohol use
are in contrast with most previous research.
Research on intimate partner violence and
elder abuse suggests that abuse victims in
both groups have a higher rate of alcohol and
substance use.20,29 Our results may reflect the
fact that the functionally independent older
women in our study did not perceive a need
to “escape” an abusive relationship through
alcohol use. Another possibility may be that
these women perceived alcohol use as increasing their vulnerability and thus escalating their potential of being victimized by
greater violence.
Mouton et al. | Peer Reviewed | Research and Practice | 609
 RESEARCH AND PRACTICE 
TABLE 3—Three-Year Abuse Incidence in Cohort of Postmenopausal Women, by Subjects’
Characteristics (N = 48 522)
Any Abuse (n = 2 431)
Characteristic
Overall
Age, y
< 58
59–64
65–69
70–74
> 74
Ethnicity
American Indian/Alaska Native
Asian/Pacific Islander
African American
Hispanic/Latino
White
Education
0–8 y
Some HS/HS diploma/GED
School after high school
College graduate or higher
Family income, $
< 20 000
20 000–34 999
35 000–49 999
50 000–74 999
≥ 75 000
Occupation
Managerial/professional
Technical/sales/administrative
Service/labor
Homemaker only
Currently employed (yes)
Marital status
Never married
Divorced/separated
Widowed
Presently married
Partner’s main job
Homemaker
Managerial/professional
Technical/sales/administrative
Service/labor
Other
Partner currently employed (yes)
Smoking
Never smoked
Past smoker
Current smoker
No Abuse
(n = 46 091)
Physical Abuse
Only, No. (%)
P
67 (2.8)
Verbal Abuse
Only, No. (%)
2 250 (92.6)
.12
14 272 (31)
10 903 (23.7)
9 860 (21.4)
7 700 (16.7)
3 356 (7.3)
27 (40.3)
20 (29.9)
8 (11.9)
7 (10.4)
5 (7.5)
156 (0.3)
1 320 (2.9)
2 831 (6.1)
1 119 (2.4)
40 665 (88.2)
0 (0.0)
4 (6.0)
11 (16.4)
8 (11.9)
44 (65.7)
509 (1.1)
8 611 (18.8)
16 228 (35.5)
20 405 (44.6)
3 (4.5)
14 (21.2)
25 (37.9)
24 (36.4)
6 073 (14.1)
9 989 (23.2)
8 770 (20.4)
8 949 (20.8)
9 264 (21.5)
18 (29.5)
12 (19.7)
12 (19.7)
10 (16.4)
9 (14.8)
19 732 (45.1)
12 425 (28.4)
7 028 (16.1)
4 550 (10.4)
16 675 (37.6)
20 (31.7)
18 (28.6)
16 (25.4)
9 (14.3)
19 (29.2)
2 220 (4.8)
6 804 (14.8)
7 750 (16.9)
29 155 (63.5)
4 (6.0)
14 (20.9)
11 (16.4)
38 (56.7)
67 (0.2)
16 261 (58)
3 934 (14.0)
4 540 (16.2)
3 232 (11.5)
11 242 (40.2)
1 (2.9)
12 (34.3)
4 (11.4)
13 (37.1)
5 (14.3)
16 (43.2)
23 332 (51.2)
19 710 (43.2)
2 556 (5.6)
35 (52.2)
28 (41.8)
4 (6.0)
P
<.001
<.001
<.001
2 (1.8)
2 (1.8)
14 (12.3)
14 (12.3)
82 (71.9)
<.001
28 (1.3)
369 (16.6)
879 (39.4)
953 (42.8)
.01
.31
2 (1.8)
23 (20.7)
46 (41.4)
40 (36.0)
.007
356 (16.9)
469 (22.3)
432 (20.5)
410 (19.5)
436 (20.7)
.08
.06
22 (20.6)
32 (29.9)
17 (15.9)
21 (19.6)
15 (14.0)
.06
967 (45.7)
579 (27.4)
375 (17.7)
193 (9.1)
929 (43.6)
<.001
<.001
99 (4.4)
417 (18.6)
267 (11.9)
1 457(65.0
<.001
.052
.97
<.001
53 (46.5)
32 (28.1)
14 (12.3)
10 (8.8)
5 (4.4)
8 (0.4)
81 (3.6)
139 (6.2)
109 (4.8)
1 913 (85)
.04
.36
45 (41.7)
35 (32.4)
21 (19.4)
7 (6.5)
46 (41.1)
1 070 (48.2)
995 (44.8)
155 (7.0)
.45
.02
4 (3.6)
27 (24.1)
12 (10.7)
69 (61.6)
<.001
5 (0.4)
733 (52.8)
200 (14.4)
279 (20.1)
171 (12.3)
604 (43.6)
P
114 (4.7)
<.001
940 (41.8)
533 (23.7)
420 (18.7)
260 (11.6)
97 (4.3)
.16
.51
Physical and Verbal
Abuse, No. (%)
.02
.003
.08
0 (0.0)
25 (39.1)
12 (18.8)
16 (25.0)
11 (17.2)
37 (53.6)
.08
.14
56 (50.5)
44 (39.6)
11 (9.9)
Continued
610 | Research and Practice | Peer Reviewed | Mouton et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 3—Continued
Alcohol intake
Nondrinker
Past drinker
< 1 drink/wk
1–6 drinks/wk
≥ 7 drinks/wk
Living alone (yes)
.11
4 698 (10.2)
7 849 (17.1)
14 511 (31.6)
12 525 (27.3)
6 291 (13.7)
12 087 (26.3)
9 (13.4)
17 (25.4)
21 (31.3)
17 (25.4)
3 (4.5)
17 (25.4)
.86
.04
226 (10.1)
431 (19.2)
717 (32.0)
597 (26.6)
271 (12.1)
520 (23.3)
<.001
<.001
17 (15.2)
35 (31.3)
29 (25.9)
18 (16.1)
13 (11.6)
26 (23.2)
.45
Note. HS =high school; GED = general equivalency diploma.
TABLE 4—Multivariate Associations With 3-Year Incidence of Abuse vs No Abuse Among
Postmenopausal Women
Age, y (vs 50–58 y)
59–64
65–69
70–79
Race (vs non-Hispanic White)
American Indian
Asian/Pacific Islander
African American
Hispanic American
Education (vs college graduate)
≤ HS diploma
Some college/technical school
Income, $ (vs > $75 000)
< 20 000
20 000–34 999
35 000–49 999
50 000–75 000
Employment (vs managerial)
Technical
Service
Homemaker
Marital status (vs married)
Never married
Divorced
Widowed
Smoking status (vs never smoked)
Past smoker
Current smoker
Alcohol use (vs past/never drank)
< 1 drink/wk
≥ 1 drink/wk
Living alone (vs no)
Physical Abuse Only
OR (95% CI)
Verbal Abuse Only
OR (95% CI)
Physical and Verbal Abuse
OR (95% CI)
1.17 (0.62, 2.19)
0.44 (0.18, 1.06)
0.46 (0.20, 1.09)
0.75 (0.67, 0.85)
0.61 (0.53, 0.70)
0.50 (0.43, 0.57)
0.67 (0.41, 1.10)
0.33 (0.17, 0.64)
0.33 (0.17, 0.64)
…
2.11 (0.63, 7.00)
1.66 (0.66, 4.13)
4.50 (1.90, 10.66)
1.03 (0.48, 2.21)
1.45 (1.14, 1.84)
0.87 (0.71, 1.06)
1.65 (1.31, 2.08)
5.40 (1.29, 22.65)
0.70 (0.17, 2.91)
1.39 (0.71, 2.73)
3.56 (1.77, 7.15)
0.52 (0.22, 1.22)
0.86 (0.44, 1.67)
0.87 (0.75, 1.02)
1.13 (1.01, 1.27)
1.21 (0.64, 2.30)
1.13 (0.67, 1.90)
2.18 (0.58, 8.18)
2.74 (1.08, 6.95)
1.24 (0.53, 2.89)
1.48 (0.69, 3.19)
2.05 (1.57, 2.66)
1.65 (1.38, 1.98)
1.34 (1.17, 1.54)
1.25 (1.10, 1.42)
1.37 (0.42, 4.52)
2.46 (1.18, 5.15)
2.02 (1.12, 3.62)
1.11 (0.59, 2.09)
1.47 (0.68, 3.14)
2.18 (1.00, 4.77)
2.23 (0.88, 5.63)
0.89 (0.78, 1.00)
0.95 (0.82, 1.10)
0.90 (0.75, 1.08)
0.90 (0.53, 1.52)
0.67 (0.35, 1.28)
0.47 (0.19, 1.20)
0.87 (0.19, 4.05)
1.34 (0.56, 3.25)
1.45 (0.58, 3.62)
0.85 (0.66, 1.10)
1.09 (0.92, 1.29)
0.79 (0.65, 0.96)
0.63 (0.18, 2.19)
1.39 (0.73, 2.65)
1.08 (0.51, 2.26)
0.88 (0.50, 1.54)
0.72 (0.22, 2.39)
1.16 (1.05, 1.28)
1.22 (1.01, 1.48)
0.98 (0.64, 1.51)
1.58 (0.78, 3.17)
0.84 (0.43, 1.63)
0.85 (0.43, 1.68)
0.76 (0.34, 1.72)
0.96 (0.85, 1.08)
0.90 (0.79, 1.01)
0.83 (0.70, 0.97)
0.55 (0.34, 0.90)
0.50 (0.30, 0.83)
0.73 (0.39, 1.36)
Note. HS =high school.
April 2004, Vol 94, No. 4 | American Journal of Public Health
This study has important limitations. The
detection of exposure to physical and verbal
abuse relies on the self-report of the victims.
Subjects may have been reluctant to admit to
abuse, resulting in an underestimate of the
prevalence and 3-year incidence. This underestimate may also diminish the differences
found in the association of abuse with our
predictor variables. Also, the subjects recruited for the WHI are drawn from a volunteer sample of older healthier women. These
women may differ from other women of their
age in exposure to abuse and its effects on
their health status.
Despite these limitations, our finding that
11.1% of women aged 50 to 79 years reported exposure to abuse in the past year,
and that an additional 5% in this age group
reported exposure to abuse over a 3-year
interval, reveals an important problem for
older women. While it is unclear if this
abuse is a continuation of a lifelong cycle of
violence or the result of late-life onset of violence, these results suggest that abuse is
occurring at rates too great to ignore. If
abuse of older women yields the same
untoward morbidity and mortality seen in
younger women and fragile elders, there is a
great threat to public health. Although a recent article by Ramsay et al. challenges the
effectiveness of screening for domestic violence,30 screening these postmenopausal
women may trigger an investigation by
agencies like Adult Protective Services that
can provide help to abuse victims. Our results suggest that additional investigations
regarding the impact of abuse in this population and the impact of screening for abuse
in postmenopausal women should be
encouraged.
Mouton et al. | Peer Reviewed | Research and Practice | 611
 RESEARCH AND PRACTICE 
About the Authors
Charles P. Mouton, Melissa A. Talamantes, and Sandra K.
Burge are with the Department of Family and Community
Medicine and Robert G. Brzyski is with the Department of
Obstetrics and Gynecology, the University of Texas Health
Science Center at San Antonio. Rebecca J. Rodabough and
Julie L. Hunt are with the Fred Hutchinson Cancer Center,
the University of Washington School of Medicine, Seattle.
Susan L. D. Rovi is with the Department of Family Medicine, the University of Medicine and Dentistry of New
Jersey–New Jersey Medical School, Newark.
Requests for reprints should be sent to Charles P. Mouton, MD, MS, Department of Family and Community
Medicine, UTHSCSA, 7703 Floyd Curl Dr, San Antonio,
TX 78229-7795 (e-mail: [email protected]).
This article was accepted May 14, 2003.
Contributors
C. P. Mouton conceived of the study, developed the
study design, and supervised the data acquisition and
analysis, and drafting of the manuscript. R. J. Rodabough
retrieved study data and completed data analysis.
S. L. D. Rovi assisted in the study design. S. K. Burge assisted in defining the categories of abuse. All authors
assisted in the interpretation of the data analysis and
drafting of the manuscript.
Acknowledgments
This study was supported by National Institutes of
Health grants KO8AG00822, HL 63293, and HL
07575.
We acknowledge the editorial support of E. Mikaila
Adams and the University of Texas Health Science Center at San Antonio writing group.
Human Participants Protection
Protocol and consent forms were approved by the institutional review boards of all the Women’s Health Initiative participating institutions, including the University of
Texas Health Science Center at San Antonio. All
women provided written informed consent.
References
1. Khan FI, Welch TL, Zillmer EA. MMPI-2 profiles
of battered women in transition. J Pers Assess. 1993;
60:100–111.
2. National Center on Elder Abuse. Reporting on
elder abuse. Available at: http://www.elderabusecenter.
org. Accessed June 2003.
3. Pillemer K, Finkelhor D. The prevalence of elder
abuse: a random sample survey. Gerontologist. 1988;
28:51–57.
4. US Dept of Health and Human Services. Abuse of
the elderly. Elder Abuse. 1980;23:24–32.
5. Hajjar I, Duthie E Jr. Prevalence of elder abuse in
the United States: a comparative report between the
national and Wisconsin data. WMJ. 2001;100(6):
22–26. [erratum: WMJ, 2001;100(8):4]
6. Reay AM, Browne KD. Risk factor characteristics
in carers who physically abuse or neglect their elderly
dependants. Aging Ment Health. 2001;5:56–62.
7. Hamburger L, Saunders D, Hover M. Prevalence
of domestic violence in community practice and rate of
physician inquiry. Fam Med. 1992;24:283–287.
8. McCauley J, Kern DE, Kolodner K, et al. The “battering syndrome”: prevalence and clinical characteris-
tics of domestic violence in primary care internal medicine practices. Ann Intern Med. 1995;123:737–746.
abuse or neglect. J Am Geriatr Soc. 2000;48:
205–208.
9. Tilden VP, Schmidt TA, Limandri BJ, Chiodo GT,
Garland MJ, Loveless PA. Factors that influence clinicians’ assessment and management of family violence.
Am J Public Health. 1994;84:628–633.
29. Coker AL, Davis KE, Arias I, et al. Physical and
mental health effects of intimate partner violence for
men and women. Am J Prev Med. 2002;23:260–268.
10. Watts C, Zimmerman C. Violence against women:
global scope and magnitude. Lancet. 2002;359:
1232–1237.
30. Ramsay J, Richardson J, Carter YH, Davidson LL,
Feder G. Should health professionals screen women for
domestic violence? Systematic review. BMJ. 2002;325:
314–318.
11. Lachs MS, Williams CS, O’Brien S, Pillemer KA,
Charlson ME. The mortality of elder mistreatment.
JAMA. 1998;280:428–432.
12. American Medical Association white paper on elderly health. Report of the Council on Scientific Affairs.
Arch Intern Med. 1990;150:2459–2472.
13. Diagnostic and Treatment Guidelines on Domestic
Violence. Vol 1. Chicago, Ill: American Medical Association; 1993.
14. Barrier PA. Domestic violence. Mayo Clin Proc.
1998;73:271–274.
15. Lay T. The flourishing problem of elder abuse in
our society. AACN Clin Issues Crit Care Nurs. 1994;5:
507–515.
16. Rennison CM. Intimate Partner Violence and Age of
Victim, 1993–1999. Washington, DC: Bureau of Justice Statistics, US Dept of Justice; October 2001.
17. Gore MJ. The Women’s Health Initiative: studying
interventions over the long term. Clin Lab Sci. 1995;8:
311–316.
18. Jones JS, Holstege C, Holstege H. Elder abuse and
neglect: understanding the causes and potential risk
factors. Am J Emerg Med. 1997;15:579–583.
19. Lachs MS, Williams C, O’Brien S, Hurst L, Horwitz R. Older adults. An 11-year longitudinal study of
adult protective service use. Arch Intern Med. 1996;
156:449–453.
20. Lachs MS, Williams C, O’Brien S, Hurst L, Horwitz R. Risk factors for reported elder abuse and neglect: a nine-year observational cohort study. Gerontologist. 1997;37:469–474.
Communicating Public
Health Information
Effectively:
A Guide for Practitioners
Edited by David E. Nelson, MD, MPH; Ross C.
Brownson, PhD; Patrick L. Remington, MD, MPH;
and Claudia Parvanta, PhD
A
24. McCreadie C, Bennett G, Gilthorpe MS, Houghton
G, Tinker A. Elder abuse: do general practitioners
know or care? J R Soc Med. 2000;93:67–71.
s the first of its kind, this book provides a comprehensive approach to
help public health practitioners improve
their ability to communicate with different audiences. Covering all modes of
communication, each chapter provides
practical, real-world recommendations
and examples of how to communicate
public health information to nonscientific
audiences more effectively. The knowledge and skills gleaned from this book
will assist with planning and executing
communication activities commonly done
by public health practitioners.
25. Fulmer T, McMahon DJ, Baer-Hines M, Forget B.
Abuse, neglect, abandonment, violence, and exploitation: an analysis of all elderly patients seen in one
emergency department during a six- month period.
J Emerg Nurs. 1992;18:505–510.
ISBN 0-87553-027-3
2002 ❚ 240 pages ❚ softcover
$25.95 APHA Members
$33.95 Non-members
21. Comijs HC, Pot AM, Smit JH, Bouter LM, Jonker
C. Elder abuse in the community: prevalence and consequences. J Am Geriatr Soc. 1998;46:885–888.
22. Kurrle SE, Sadler PM, Lockwood K, Cameron ID.
Elder abuse: prevalence, intervention and outcomes in
patients referred to four aged care assessment teams.
Med J Aust. 1997;166:119–122.
23. McCreadie C, Tinker A. Review: abuse of elderly
people in the domestic setting: a UK perspective. Age
Ageing. 1993;22:65–69.
26. Paveza GJ, Cohen D, Eisdorfer C, et al. Severe
family violence and Alzheimer’s disease: prevalence
and risk factors. Gerontologist. 1992;32:493–497.
American Public Health Association
27. Bosker G. Elderly abuse: patterns, detection, and
management. Resid Staff Physician. 1990;36(3):39–44.
28. Dyer CB, Pavlik VN, Murphy KP, Hyman DJ. The
high prevalence of depression and dementia in elder
612 | Research and Practice | Peer Reviewed | Mouton et al.
Plus shipping and handling
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
PHIn12J1
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Protection Orders and Intimate Partner Violence: An
18-Month Study of 150 Black, Hispanic, and White Women
| Judith McFarlane, DrPH, Ann Malecha, PhD, Julia Gist, PhD, Kathy Watson, MS, Elizabeth Batten, BA, Iva Hall, PhD, and Sheila Smith, PhD
Abused women use a variety of methods in
seeking assistance to halt violence inflicted
upon them, including court orders of protection. Such orders restrict the access of 1 person (e.g., a male abuser) to another person
(e.g., an abused woman) for a specified time.
(The synonym “restraining order” is used in
some jurisdictions.) Protection orders, both
temporary and permanent, represent public
documentation that abuse has occurred, and if
the order is violated, the assailant is subject to
prosecution. A protection order offers the victim legal action when the victim does not want
the abuser charged criminally or jailed for an
offense. However, choice of this action does
not preclude other civil or criminal action.
Results of research on the effectiveness of
protection orders are inconsistent. We identified 8 recent longitudinal studies that measured additional intimate partner violence
committed against women after a protection
order had been filed. Six of the studies reported positive results,1–6 meaning that the respondents felt the protection order helped to
end or reduce the violence. The remaining 2
studies reported high reassault rates after filing
of the protection order.7,8 In the case of most
of these studies, low response rates, short followup periods, and lack of comparison groups do
not allow generalizations to be made.
In addition, we did not identify any studies
that included non–English-speaking women or
measures of worksite harassment. To test the
effectiveness of protection orders, we entered
into a partnership with a local district attorney’s
office in a large urban city in an attempt to determine whether women who are granted a 2year protection order experience lower levels
of violence than women who apply and qualify
for such an order but are not granted one.
METHODS
Our study was conducted from January
2001 to June 2002 at a special family violence unit of the Houston, Tex, district attor-
Objectives. We compared types and frequencies of intimate partner violence experienced by women before and after receipt of a 2-year protection order.
Methods. Participants were 150 urban English- and Spanish-speaking Black, Hispanic,
and White women who qualified for a 2-year protection order against an intimate partner.
Results. One woman committed suicide 6 weeks into the study. The remaining 149
women completed all interviews. Results showed significant reductions in threats of assault, physical assault, stalking, and worksite harassment over time among all women,
regardless of receipt or nonreceipt of a protection order.
Conclusions. Abused women who apply and qualify for a 2-year protection order, irrespective of whether or not they are granted the order, report significantly lower levels
of violence during the subsequent 18 months. (Am J Public Health. 2004;94:613–618)
ney’s office that serves an ethnically diverse
population of 3 million citizens. The primary
service of the family violence unit is processing of protection orders. During the 12
months preceding this study, 2932 women
applied to the unit for a protection order;
1980 (68%) met qualifying criteria, and 962
(49%) were granted the protection order.
Qualification criteria for protection orders are
set by state law and include applicants providing evidence (i.e., police or witness report, visible injury) that the respondent (e.g., abuser)
has been violent with them and is likely to
continue this violence toward them. In addition, the applicant must have previously lived
with the abuser in the same household, or
they must be the biological parents of the
same child.9
If the applicant’s case is accepted, the attorneys file the case with the family law court
and ask for a court date to be set for a hearing. After the case has been filed, the court issues a temporary protection order. A copy of
this order is sent to the applicant by mail, and
a copy is served to the abuser in person. The
temporary protection order is similar to the
final 2-year protection order in that it informs
the abuser that he or she must stay 200 ft
(60 m) away from the applicant’s home and
workplace and prohibits the respondent from
assaulting the applicant, from threatening the
applicant directly or through another person,
and from harassing or stalking the applicant.
April 2004, Vol 94, No. 4 | American Journal of Public Health
However, the temporary protection order
differs from the final 2-year protection order
in that a violation of the temporary order cannot be charged as a criminal offense; it can be
filed only as a civil contempt of court. Furthermore, the temporary protection order is
valid for only 20 days. The court date is set
within those 20 days, and the order expires
whether or not the abuser is served or the
hearing takes place. However, the temporary
protection order may be extended if the
abuser is not served by the hearing date.
The applicant is not responsible for any
fees in association with the protection order.
The order is granted for 2 years and can result in both criminal and civil penalties if violated. Applicants are informed at the time of
application as to whether they do or do not
qualify to receive the order. All qualifying applicants are assigned to a case worker who
provides them with educational information
about violence and safety planning as well as
information regarding community resources
(e.g., emergency shelters, counseling, legal and
medical assistance). Applicants are encouraged to contact the case worker for further
questions about the protection order process.
All women who presented to the special
family violence unit at the district attorney’s
office to apply for a protection order, completed the application process, qualified for the
protection order, and met our inclusion criteria
(e.g., female, 18 years or older, English or
McFarlane et al. | Peer Reviewed | Research and Practice | 613
 RESEARCH AND PRACTICE 
Spanish speaker) were invited into the study
by 1 of the 6 investigators until 150 women
agreed to participate and were entered into the
study. Four women refused to participate. One
woman committed suicide 6 weeks into the
study. All of the remaining 149 women completed the 3-month, 6-month, 12-month, and
18-month follow-up interviews, resulting in a
retention rate of 99%.
Instruments
Demographic data form. This form was used
to document information on participants’ age,
education, income, self-identified race/ethnicity,
employment status, relationship to the abuser,
and primary language.
Severity of Violence Against Women Scales
(SVAWS). This 46-item instrument is designed to measure threats of physical violence (19 items) and physical assault (27
items).10 Examples of behaviors that threaten
physical violence are threats to destroy property, do bodily harm, or harm other family
members. Examples of behaviors that represent physical violence are kicking, choking,
beating up, and engaging in forced sex. For
each item, respondents use a 4-point scale to
indicate how often the behavior occurred
(1 = never, 2 = once, 3 = 2–3 times, 4 = 4 or
more times). Possible score ranges were 19 to
76 for the threat of abuse dimension and 27
to 108 for the physical abuse dimension. The
higher the score was, the more violence that
was reported.
Internal consistency reliability estimates
in studies of abused women have ranged
from 0.89 to 0.91 for the threat of abuse dimension and from 0.91 to 0.94 for the
physical abuse dimension.5,11,12 In the present study, reliabilities (measured with Cronbach α coefficients) were 0.91 for the threat
of abuse dimension and 0.94 for the physical abuse dimension.
Stalking Victimization Survey. This 17-item
yes/no questionnaire was used to document
the frequency and type of stalking engaged in
by the perpetrator. The initial stalking survey
instrument consisted of 7 items (e.g., being
followed or spied on, being sent unsolicited
letters or written correspondence, or finding
the perpetrator standing outside one’s home,
school, or workplace) developed by Tjaden
and Thoennes13; 10 items were added from
the Sheridan14 HARASS instrument to form
the overall 17-item instrument used here. Examples of items added include threats by the
abuser to harm the children or to commit suicide if the woman left the relationship, leaving threatening notes on the woman’s car,
and threatening her family. The possible
score range was 0 to 17. In this study, reliability (Cronbach α coefficient) was 0.83.
Danger Assessment Scale. This instrument,
which consists of 15 items with a yes/no response format, assists women in determining
their potential risk of becoming a femicide
victim.15 All of the items refer to risk factors
that have been associated with murder in situations involving abuse. Examples of risk
factors include the abuser’s possession of a
gun, use of drugs, and violent behavior outside the home. The possible score range was
0 to 15. Scale reliability coefficients have
ranged from 0.60 to 0.86 in several
studies.16 In this study, the reliability (Cronbach α coefficient) was 0.67.
Worksite harassment. Eight yes/no questions were asked about worksite harassment.
Questions were derived from a congressional
report17 that reviewed studies of worksite harassment of women by intimate partners.
Questions focused on, for example, repeated
calls/visits to the woman’s worksite and difficulties experienced by the woman in regard
to going to work. The possible score range
was 0 to 8. Reliability (Cronbach α coefficient) was measured as 0.76.
To assist in maintaining contact with each
of the women, we formed a safe contact list
of at least 6 persons the woman granted permission for us to contact in the event she
could not be reached. This list consisted of
close relatives (i.e., mother, grandmother, sister, and adult children), neighbors, friends,
work colleagues, and other acquaintances. In
each case, name, relationship, address (home
and work), and contact telephone numbers
(i.e., home, work, and cellular) were listed.
When contacted, the person was told that
the woman was involved in a health study
and had given permission for the researcher
to contact individuals who may know of her
current address/telephone number(s). During all subsequent interviews with the
women, both their contact information and
that of each safe contact were reviewed and
updated.
The safe contact list proved the best
method for maintaining contact with the
women over the 18-month study period. We
ensured women’s safety in completing the
follow-up telephone interviews by establishing a convenient, private, and safe time for
these interviews. A safety protocol was used
for each follow-up telephone interview.
Women were reimbursed $20 for the first interview; $30 for the 3-month interview; $40
for the 6-month interview; $50 for the 12month interview; and $60 for the 18-month
interview. They were reimbursed an extra
$40 for completing all of the interviews.
Procedure
Data Analyses
Data collection began after institutional review board approval had been received and
consent had been obtained from the district
attorney’s office. Women meeting the study
criteria were escorted to a private room in the
offices of the family violence unit where the
investigators provided an explanation of the
study’s purpose, protocol, instruments, administration time, and follow-up schedules.
Women who agreed to take part in the study
signed an informed consent form, and the investigators administered the study instruments. Instruments were offered in English
and Spanish according to women’s language
preference. All measures focused on women’s
reports of violence and health status during
the preceding 3 months.
Means, standard deviations, and frequencies were used in descriptions of the demographic characteristics of the 150 women who
applied for protection orders. We conducted
independent t tests to determine whether the
women who were granted an order differed
significantly in terms of age or years of education from the women who were not granted
an order. Chi-square analyses were used to determine whether the groups of women differed significantly with respect to race/ethnicity,
income, employment status, or status of relationship with abuser.
Using Cohen’s power analyses and tables,18
we calculated the a priori power of our betweengroups repeated measures analysis of variance to produce a small-to-moderate multi-
614 | Research and Practice | Peer Reviewed | McFarlane et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
variate effect size by conducting a multivariate analysis of variance with 4 dependent
variables (i.e., the 4 score differences from intake scores). Given a significance level of .05,
150 participants, 4 dependent variables, 2
groups, and the goal of a small-to-moderate
effect size, we calculated the power of the
analysis as 91%. Assumptions of independent
observations, normality, and homogeneity of
(co)variance were examined. Results indicated
that the study’s robustness, procedure, number of participants, and sample size ratio satisfied these assumptions.
Violence scores for women who were
granted or not granted the protection order
were subjected to repeated measures analyses. We initially considered as covariates demographic characteristics that exhibited significant between-groups differences at intake
and were shown to be univariately associated
with the dependent variables. However, we
retained only significant covariates in the final
analyses. We conducted a 1-factor repeated
measures multivariate analysis of covariance
on SVAWS scores to accommodate the 2 dependent subscales (threats of violence and
physical violence scores). We performed 1factor repeated measures analyses of covariance (ANCOVAs) on danger, stalking, and
work harassment scores.
We calculated adjusted means, standard
deviations, and multivariate effect sizes
(0.02 = small, 0.10 = small to moderate,
0.15 = moderate, 0.35 = large).18 To achieve
a balance between type I and type II error,
we set the significance level at .025 for each
SVAWS subscale. In the case of within-group
(time, Group × Time interaction) contrasts, we
set significance levels at .006 for subscale
scores and .0125 for stalking, danger, and
worksite harassment scores.
RESULTS
The women were stratified into 2 groups:
those who were granted a 2-year protection
order (n = 81) and those who were not
granted such an order (n = 69). Reasons for
nonreceipt of the protection order were as follows: the woman dropped the order (n = 40),
inability to locate the abuser and serve papers
to appear in court (n = 18), and dismissal of
cases (n = 11).
Reasons for Not Being Granted a 2-Year
Protection Order
Forty women dropped the protection
order before their court date. Most did so
because they returned to the relationship
with the abuser or because the protection
order process was “too much of a hassle” or
“inconvenient.” To obtain a protection order,
applicants must be willing to arrive at the
district attorney’s office with proper photo
identification and complete paperwork, and
they are required to complete an interview
with a caseworker, be photographed, and
sign an affidavit. This process requires
about 2 to 3 hours. Applicants must wait
approximately 6 weeks for a court date and
then appear in court in front of a judge, at
which time the abuser may contest the protection order.
In addition, many women need to return to
the district attorney’s office at a later date
with additional required paperwork/witnesses
to the abuse. For some women, these trips to
the district attorney’s office mean work absences and loss of income. We did not ask the
participants in our study who had dropped
the protection order when they did so; however, at the 3-month interview, many women
reported dropping the order within the first 2
weeks after application.
Eighteen women were not granted a protection order because the abuser could not
be found and served papers to appear in
court. Eleven women were not granted the
order because their case was dismissed. Six
cases were dismissed by the district attorney’s
office owing to incomplete applications (e.g.,
required documents not being supplied).
Seven cases were dismissed by the judge, 2
because the protection order was contested
by the abuser and the remaining 5 because
the women did not appear in court. One of
these women committed suicide. As mentioned, the remaining 149 women completed
the 4 follow-up interviews, for a retention
rate of 99%.
Between-Groups Differences in
Demographic Characteristics and
Violence Scores
Frequencies, percentages, and the results
of tests assessing demographic differences
among women who were and were not
April 2004, Vol 94, No. 4 | American Journal of Public Health
granted a 2-year protection order are
shown in Table 1. Relationship status was
significantly (χ21 = 4.407, P = .036) associated with receipt of a protective order.
Slightly more than half of the women who
were granted a protection order were involved in relationships, as compared with
71% of women who were not granted the
order. No other significant differences were
found.
Adjusted means and standard deviations
for violence scores at intake and at 3, 6, 12,
and 18 months among women who were
granted (n = 81) and not granted (n = 69)
the 2-year protection order are shown in
Table 2. After adjustment for age, race/
ethnicity, and relationship status, results of
the multivariate analysis of variance focusing on SVAWS scores yielded a significant
(F8,1144 = 16.123, P < .001) multivariate
main effect for time. The magnitude of this
multivariate effect size was in the small-tomoderate range (0.10). Univariate tests revealed a significant main effect for both
SVAWS subscales: threats (F4,572 = 19.077,
P < .001) and physical abuse (F4,572 = 36.261,
P < .001). The group main effect and the
Group × Time interaction were not significant. The effect size between the groups
was small (0.02). Examination of withinsubject contrasts showed that intake scores
were significantly (P < .001) higher than
subsequent scores.
After adjustment for age and race/ethnicity,
repeated measures ANCOVAs showed significant (F4,141 = 16.17, P < .001, and F4,141 =
18.33, P < .001, respectively) effects over
time of stalking and danger scores. The time
effect size for stalking was in the medium-tolarge range (0.31), and the effect size for
danger was moderate (0.18). There was no
significant group main effect or significant
Group × Time interaction. Between-groups
effect sizes were zero or small (0.02). After
adjustment for relationship status, the repeated measures ANCOVA of work harassment scores also showed a significant (F4,80 =
13.88, P < .001) effect over time. There was
no significant group main effect or significant
Group × Time interaction. The betweengroups effect size was zero. Examination of
within-subject contrasts for the main effect of
time showed that intake scores were signifi-
McFarlane et al. | Peer Reviewed | Research and Practice | 615
 RESEARCH AND PRACTICE 
TABLE 1—Demographic Characteristics and Results From χ2 Tests of Independence
Assessing Differences Between Women Who Were Granted (n = 81) and Not Granted
(n = 69) a 2-Year Protection Order
Characteristic
Protection Order
No Protection Order
Total No.
Test Statistic (2 or t) (P)
Age, y, mean (SD)
Education, y, mean (SD)
Race/ethnicity, No. (%)
African American
White
Latino/Hispanic
Relationship status, No. (%)
Current spouse/boyfriend
Ex-spouse/friend
Family income, $, No. (%)
≥19 000
>19 000
English speaking, No. (%)
No
Yes
Employed, No. (%)
No
Yes
33.5 (9.2)
11.7 (3.0)
31.2 (9.1)
11.9 (2.6)
150
150
1.503a (.135)
0.503a (.615)
3.320b (.190)
31 (38.3)
22 (27.2)
28 (34.6)
18 (26.1)
18 (26.1)
33 (47.8)
49
40
61
4.407b (.036)
44 (54.3)
37 (45.7)
49 (71.0)
20 (29.0)
93
57
0.672b (.412)
22 (32.8)
45 (67.2)
22 (40.0)
33 (60.0)
44
78
0.831b (.362)
15 (18.5)
66 (81.5)
9 (13.0)
60 (87.0)
24
126
2.916b (.088)
8 (9.9)
73 (90.1)
2 (2.9)
67 (97.1)
10
140
a
t test.
χ test.
b 2
cantly (P < .001) higher than subsequent
scores.
Protection Order Violations
Finally, women were asked, at each interview, whether a violation of the 2-year protection order had occurred since the previous
interview. Among the 81 women granted a
protection order, 36 (44%) reported at least
1 violation over the 18 months of the study.
Violations were reported by 17 women (21%)
at 3 months, 16 women (20%) at 6 months,
20 women (25%) at 12 months, and 19
women (23%) at 18 months. Four women
(5%) reported a violation during each of the
4 time periods measured. Most violations involved nonadherence to the order to stay
200 ft from the woman’s home or workplace;
stalking, threats of violence, and a combination of these infractions were other examples
of violations. Women reporting a violation
also were asked whether they had called the
police. Among these 36 women, 21 (58%)
had called the police at least once to report a
violation.
DISCUSSION
The 149 women who took part in this
study reported significantly lower levels of
intimate partner violence, including worksite
harassment, up to 18 months after applying
for a protection order. Whether women were
granted or not granted the protection order
made no significant difference in terms of
the amount of violence they reported at the
time of application for the order or during
the subsequent 3, 6, 12, or 18 months.
Forty-four percent of the women granted a
2-year protection order reported at least 1
violation over the 18-month study period,
and half of these women reported the violation to the police.
This study followed women after they had
qualified for a protection order, irrespective
of whether or not they were granted the
order. Our results agree with those of others5,19 reporting significantly lower levels of
violence experienced by women seeking assistance from the justice system, irrespective
of the justice system outcome. One other
616 | Research and Practice | Peer Reviewed | McFarlane et al.
study, to our knowledge, involved the use of
victim interviews to measure levels of violence toward women granted and not
granted a protection order.20 Although this
study reported that violence frequency was
not significantly decreased by receipt of a
protection order, the study’s low response
rate and short follow-up period limited the
generalizability of the findings. Other researchers have focused only on women who
received an order of protection against the
abuser1–4,7 or have relied solely on police reports.6 Because fewer than half of abused
women ever report intimate partner violence
to law enforcement personnel,21 relying on
police reports may severely underrepresent
levels of violence experienced by women
both with and without a protection order.
Our findings of significant reductions in violence scores over time among all of our participants, regardless of receipt or nonreceipt
of the protection order, are consistent with
abuse intervention findings reported by social and health researchers. In one study,
abused women exiting a shelter and receiving home social support were compared, at 6
months, with abused women not receiving
such support; women in both groups reported decreases in physical abuse.22 In 2
health clinic studies involving comparisons of
abused women receiving intensive counseling
and outreach support and abused women offered a wallet-sized card listing community
abuse resources, women in both groups reported significantly lower levels of abuse at
6, 12, and 18 months postintervention.23,24
Although we found no other studies with
which to compare our results, the economic
implications of the significant decline in
worksite harassment experienced by abused
women after contact with the justice system
merit further research.
Do these findings indicate that the justice intervention of a protection order and the health
and social service interventions of counseling,
support, and referrals are no more of a deterrent to future violence than an abused
woman’s contact with assistance agencies?
When an abused woman decides to contact a
criminal justice, civil justice, health, or social
service agency, information about the abuse is
shared, and contact is made. Just as the privatization of domestic violence contributes to its
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 2—Adjusted Means and Standard Deviations for Violence Scores at Intake and 3, 6,
12, and 18 Months: Women Who Were Granted (n = 81) and Not Granted (n=69) a 2-Year
Protection Order
Measure and Group
SVAWS a
Threats of abusea
No order
Order
Physical abuseb
No order
Order
Stalking b
No order
Order
Danger b
No order
Order
Worksite harassment c
No order
Order
Intake,
Mean (SD)
3 Months,
Mean (SD)
6 Months,
Mean (SD)
12 Months,
Mean (SD)
18 Months,
Mean (SD)
44.7 (13.7)
47.5 (13.7)
21.6 (7.7)
23.1 (7.8)
20.7 (7.1)
22.9 (7.2)
21.4 (8.6)
24.9 (8.7)
21.9 (8.1)
22.7 (8.1)
49.2 (17.0)
48.5 (17.0)
27.7 (5.5)
29.2 (5.5)
27.54 (4.6)
28.46 (4.6)
27.2 (7.4)
31.1 (7.5)
28.3 (7.6)
29.2 (7.6)
7.7 (4.0)
7.0 (4.0)
2.2 (3.3)
3.0 (3.3)
1.4 (2.8)
1.8 (2.8)
1.9 (3.0)
2.4 (3.0)
2.1 (3.0)
1.6 (3.0)
7.1 (3.0)
7.1 (3.0)
1.5 (2.2)
2.2 (2.2)
1.2 (2.0)
1.7 (2.0)
1.1 (2.1)
2.0 (2.1)
1.4 (2.1)
1.6 (2.1)
3.7 (1.8)
4.3 (1.8)
2.1 (1.5)
2.1 (1.5)
1.5 (1.1)
1.2 (1.1)
1.4 (1.1)
1.6 (1.1)
1.3 (0.8)
1.3 (0.8)
Note. One participant committed suicide; analyses were performed on a sample of 149 participants. SVAWS = Severity of
Violence Against Women Scales.
a
Adjusted for age, race/ethnicity, and relationship.
b
Adjusted for age and race/ethnicity.
c
Adjusted for relationship status.
continuation, perhaps the contact and public
knowledge stemming from justice encounters
can prevent reoccurrence of violence. Perhaps
just as legal sanctions (e.g., requirements involving the use of helmets and seat belts) have
proven effective in reducing unintentional injuries, such sanctions can reduce the occurrence of intentional intimate partner violence.
An earlier qualitative study focusing on
why women seek civil orders of protection revealed a desire among women to regain some
measure of control in their lives by making
the abuse public.25 These women discussed
using the application for a protection order as
a “loudspeaker” to notify the abuser that the
law knew about his behavior. They viewed
the legal system as a force larger than themselves and as having power over the abuser
that they themselves had lost as a result of
the abuse. Moreover, they felt a need to have
the legal system both approve and reinforce
their decision to leave the abuser. The protection order becomes an announcement that
the abused woman refuses to “take it” any-
more and is acting on her own behalf. Our results appear to quantify these qualitative findings. Once a woman applied and qualified for
a protection order, a rapid and significant decline in violence scores occurred and was sustained for 18 months.
Our study involved limitations that are important to the generalizability of the findings.
Our sample was small and limited to women
from a single urban agency who were seeking assistance. Furthermore, we relied exclusively on self-reports, possibly leading to underreporting as a result of inadequate recall
or lack of voluntary disclosure. If we are to
learn more about the occurrence of intimate
partner violence in the absence of justice system contact, there is a need for future research with larger, representative samples of
abused women that include those who are
victimized but do not apply for a protection
order. In addition, replication is essential in
rural settings with diverse ethnic groups. Despite these limitations, our urban sample of
English- and Spanish-speaking women dem-
April 2004, Vol 94, No. 4 | American Journal of Public Health
onstrates the important effect of justice system contact in terms of reductions in future
episodes of violence.
CONCLUSIONS
Ensuring the safety of victims of intimate
partner violence is of utmost importance to
health care providers, justice agencies, shelter
workers, and other service providers. This
study clearly demonstrates that, irrespective
of whether or not a 2-year protection order
was granted, abused women who sought a
protection order reported significantly lower
levels of threats of abuse, physical abuse,
stalking, work harassment, and risk factors for
femicide at 3, 6, 12, and 18 months after
their initial contact with the justice system.
About the Authors
Judith McFarlane, Ann Malecha, and Julia Gist are with
the College of Nursing, Texas Woman’s University,
Houston. Kathy Watson is with the Baylor College of
Medicine, Houston, Tex. Elizabeth Batten is with the
Harris County District Attorney’s Office, Family Criminal Law Division, Houston. Iva Hall and Sheila Smith
are with the Nursing Department, Lamar University,
Beaumont, Tex.
Requests for reprints should be sent to Judith McFarlane, DrPH, Texas Woman’s University, College of Nursing, 1130 John Freeman Blvd, Houston, TX 77030 (e-mail:
[email protected]).
This article was accepted April 3, 2003.
Contributors
J. McFarlane conceived the study, supervised all aspects
of its implementation, and wrote the first draft of the article. A. Malecha assisted with supervision of all aspects
of the study, synthesized the analysis, and edited the
drafts. J. Gist assisted with data collection and managed
the study. K. Watson completed the data analyses. E.
Batten translated all instruments and collected and
coded data on Spanish speakers. I. Hall and S. Smith
collected data, coded data, and synthesized analyses.
All of the authors helped to conceptualize ideas, interpret findings, and review drafts of the article.
Acknowledgments
This project was supported by grant 200-WT-VX-0020
awarded by the National Institute of Justice, Office of
Justice Programs, US Department of Justice.
We wish to thank the Family Criminal Law Division
of the Harris County District Attorney’s Office for its
unflagging support and assistance in collecting the data
for this study. We also thank the 149 women who
shared their stories and maintained contact with the research team for 18 months.
Note. The points of view offered in this article are those
of the authors and do not necessarily represent the official
positions or policies of the US Department of Justice.
McFarlane et al. | Peer Reviewed | Research and Practice | 617
 RESEARCH AND PRACTICE 
Human Participant Protection
This study was approved by the Texas Woman’s University institutional review board. All participants provided
informed consent according to the guidelines specified
by the Texas Woman’s University institutional review
board.
References
1. Kaci JH. Aftermath of seeking domestic violence
protection orders: the victim’s perspective. J Contemp
Criminal Justice. 1994;10:201–219.
2. Keilitz SL, Hannaford PL, Efkeman HS. Civil Protection Orders: The Benefits and Limitations for Victims
of Domestic Violence. Williamsburg, Va: National Center
for State Courts; 1997.
3. Ptacek J. Battered Women in the Courtroom: The
Power of Judicial Responses. Boston, Mass: Northeastern
University Press; 1999.
4. Carlson JJ, Harris SD, Holden GW. Protective orders and domestic violence: risk factors for re-abuse.
J Fam Violence. 1999;14:205–226.
5. Gist J, McFarlane J, Malecha A, et al. Protection
orders and assault charges: do justice interventions reduce violence against women? Am J Fam Law. 2001;
15:59–71.
6. Holt VL, Kernie MA, Lumley T, Wolf ME, Frederick PR.. Civil protection orders and risk of subsequent
police-reported violence. JAMA. 2002;288:589–594.
7. Harrell A, Smith BE. Effects of restraining orders
on domestic violence victims. In: Buzawa ES, Buzawa
CG, eds. Do Arrests and Restraining Orders Work?
Thousand Oaks, Calif: Sage Publications; 1996:214–
242.
18. Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Hillsdale, NJ: Lawrence Erlbaum
Associates; 1988.
20. Grau J, Fagan J, Wexler S. Restraining orders for
battered women: issues of access and efficacy. In:
Schweber C, Feinman C, eds. Criminal Justice Politics
and Women: The Aftermath of Legally Mandated Change.
New York, NY: Haworth Press Inc; 1985:13–28.
21. Tjaden P, Thoennes N. Extent, Nature, and Consequences of Intimate Partner Violence. Washington, DC:
National Institute of Justice; 2000.
22. Sullivan C, Campbell R, Angelique H, Eby K,
Davidson W. An advocacy intervention program for
women with abusive partners: six-month follow-up.
Am J Community Psychol. 1994;22:101–122.
24. Parker B, McFarlane J, Soeken K, Silva C, Reel S.
Testing an intervention to prevent further abuse to
pregnant women. Res Nurs Health. 1999;22:59–64.
25. Fischer K, Rose M. When “enough is enough”:
battered women’s decision making around court orders
of protection. Crime Delinquency. 1995;41:414–429.
Texas Family Code, Title 4, ch 71 (1995).
10. Marshall LL. Development of the Severity of Violence Against Women Scales. J Fam Violence. 1992;7:
103–121.
11. Wiist W, McFarlane J. Severity of spousal and
intimate partner abuse to pregnant Hispanic women.
J Health Care Poor Underserved. 1998;9:248–261.
12. Wiist W, McFarlane J. Use of police by abused
pregnant Hispanic women. Violence Women. 1998;4:
677–693.
Planning for
Community-Oriented
Health Systems
By James E. Rohrer, PhD
23. McFarlane J, Soeken K, Wiist W. An evaluation of
interventions to decrease intimate partner violence to
pregnant women. Public Health Nurs. 2000;17:443–
451.
8. Klein AR. Re-abuse in a population of courtrestrained male batterers: why restraining orders don’t
work. In: Buzawa ES, Buzawa CG, eds. Do Arrests and
Restraining Orders Work? Thousand Oaks, Calif: Sage
Publications; 1996:192–213.
9.
2nd
Edition
19. McFarlane J, Willson P, Lemmey D, Malecha A.
Women filing assault charges on an intimate partner:
criminal justice outcome and future violence experienced. Violence Women. 2000;6:396–408.
T
he second edition of this APHA best
seller acquaints readers with the broad
field of community health planning.
System design and needs, performance
and quality assessment are addressed.
Chapters include planning doctrine, designing community systems, community
needs assessment, health system performance measurements, and more.
This book is an essential guide for practicing planners, hospital administrators,
community leaders, and planning board
members.
ISBN 0-87553-246-2
1999 ❚ 230 pages ❚ softcover
$23 APHA Members
$33 Nonmembers
Plus shipping and handling
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
13. Tjaden P, Thoennes N. Stalking in America: Findings From the National Violence Against Women Survey.
Washington, DC: US Dept of Justice; 1998.
14. Sheridan D. Measuring Harassment of Battered
Women [dissertation]. Portland, Ore: Oregon Health
Sciences University; 1998.
15. Campbell JC. Assessment of risk of homicide for
battered women. Adv Nurs Sci. 1986;8:36–51.
16. Campbell JC. Assessing Dangerousness: Violence by
Sexual Offenders, Batterers, and Child Abusers. Thousand
Oaks, Calif: Sage Publications; 1995.
17. Domestic Violence: Prevalence and Implications for
Employment Among Welfare Recipients. Washington,
DC: US General Accounting Office, Health, Education,
and Human Services Division; 1998. GAO/HEHS
publication 99-12.
618 | Research and Practice | Peer Reviewed | McFarlane et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Assessing the Long-Term Effects of the Safe Dates Program
and a Booster in Preventing and Reducing Adolescent
Dating Violence Victimization and Perpetration
| Vangie A. Foshee, PhD, Karl E. Bauman, PhD, Susan T. Ennett, PhD, G. Fletcher Linder, PhD, Thad Benefield, MS, Chirayath Suchindran, PhD
Adolescent dating violence is a public health
problem.1–12 The Safe Dates Project is a randomized controlled trial for testing the effects
of a school-based intervention on the prevention and reduction of dating violence among
adolescents. Findings reported earlier suggested that 1 month after intervention, Safe
Dates prevented and reduced dating violence
and positively changed cognitive mediating
variables that were based on program content.13 One year after the intervention, cognitive risk factor effects were maintained, but
behavioral effects disappeared.14 These findings are consistent with those from prevention
trials aimed at other adolescent problem behaviors that measured long-term effects: behavioral effects faded whereas effects on cognitive risk factors persisted.15–18
Three years after Safe Dates was implemented, a booster was implemented with a
random half of the original treatment group
adolescents. Boosters are intended to reinforce the content of original programs so as to
maintain or regain initial program effects.
They are typically a briefer version of the
original program but with the same theoretical base and are administered at least 1 year
after the original intervention. Boosters that
have been used with school-based programs
for preventing other adolescent problem behaviors have included newsletters followed
by telephone contact with the adolescent,19
magazines handed out to the adolescents in
school,20 and a reduced number of classroom
sessions.21,22 Our booster was a newsletter
mailed to the adolescents and a personal contact from a health educator by telephone.
The purposes of this article are to (1) examine the 4-year postintervention effects of
Safe Dates on dating violence perpetration
and victimization and (2) determine whether
the booster improved the effectiveness of Safe
Dates. Findings from evaluations of other
Objectives. This study determined 4-year postintervention effects of Safe Dates on
dating violence, booster effects, and moderators of the program effects.
Methods. We gathered baseline data in 10 schools that were randomly allocated to
a treatment condition. We collected follow-up data 1 month after the program and then
yearly thereafter for 4 years. Between the 2- and 3-year follow-ups, a randomly selected
half of treatment adolescents received a booster.
Results. Compared with controls, adolescents receiving Safe Dates reported significantly less physical, serious physical, and sexual dating violence perpetration and victimization 4 years after the program. The booster did not improve the effectiveness of
Safe Dates.
Conclusions. Safe Dates shows promise for preventing dating violence but the booster
should not be used. (Am J Public Health. 2004;94:619–624)
adolescent problem behavior interventions
support the potential for long-term program
effects23 and booster effects20,22,24 even after
original program effects have faded. This is
the first study to test the long-term effects of
an adolescent dating violence prevention program and to test whether a booster prevents
adolescent dating violence.
We examined the effects of Safe Dates and
the booster on psychological, physical, serious physical, and sexual dating violence victimization and perpetration. Because the effects of programs for preventing other
adolescent problem behaviors have been
found to vary by gender,25 race,19 and preprogram involvement in the problem behavior,15,20,21,26 we also determined if the effects
of Safe Dates and the booster were moderated by these variables.
METHODS
Design
Adolescents were eligible for this study if
they were enrolled in the 8th grade in the
fall of 1994 in 1 of the 10 public schools in
a rural North Carolina county. Baseline data
were collected in October 1994 (wave 1)
from 85.1% (n = 957) of eligible adolescents.
April 2004, Vol 94, No. 4 | American Journal of Public Health
The 10 schools were then matched by
school size. One member of each matched
pair was randomly assigned to receive either
Safe Dates or to serve as a control. Adolescents in the 5 treatment schools were exposed to Safe Dates from November 1994
through March 1995.
Safe Dates included a theater production
performed by students, a curriculum comprising 10 45-minute sessions taught by health
and physical education teachers, and a poster
contest based on curriculum content. Process
data suggested high program fidelity in treatment schools.13,14 For details on program development, content, and theoretical base, see
the 1996 report by Foshee et al.27
Follow-up data were collected from treatment and control adolescents at 1 month
(wave 2) and 1 year (wave 3) after Safe Dates
was completed. After wave 3, parents of adolescents who provided baseline data collection were recontacted to solicit permission for
continued adolescent participation, and 65%
(n = 620) of the parents consented to have
their child do so. Adolescents who had parental consent for continued participation completed questionnaires 2 years after Safe Dates
(wave 4), and then the original treatment
group adolescents were randomly allocated to
Foshee et al. | Peer Reviewed | Research and Practice | 619
 RESEARCH AND PRACTICE 
booster and nonbooster conditions. Hence,
the study design changed from 2 groups
(treatment and control) to 3 groups (treatment only, treatment plus booster, control).
Adolescents completed questionnaires again
3 years (wave 5) and 4 years (wave 6) after
Safe Dates was completed.
The booster was an 11-page newsletter
mailed to the adolescents’ homes and a personal contact by a health educator by telephone approximately 4 weeks after the mailing. The newsletter included information and
worksheets based on content from the Safe
Dates school curriculum. Examples of information presented include red flags that a relationship is abusive, effective communication
strategies, and tips for safe dating. Five worksheets were included. As 1 example, in a
large paper heart, adolescents wrote down
how they want to be treated by dating partners (e.g., respected, listened to, treated
equally), and in a circle with a line through it,
they wrote how they did not want to be
treated (e.g., lied to, threatened, ignored, humiliated). In another example, adolescents
considered the short- and long-term consequences of various abusive behaviors for the
victims and perpetrators.
Approximately 4 weeks after the mailing, a
health educator made a personal contact with
the adolescent by telephone. At that contact,
the health educator answered the adolescent’s
questions related to the newsletter, provided
additional information when needed, and determined if the adolescent read each informational component and completed the worksheets. The adolescent was mailed $10 after
the health educator determined that the
newsletter activities were completed. Approximately 82% of the adolescents assigned to receive the booster read the newsletter and
completed the worksheets.
The analyses for this article are limited to the
adolescents who completed baseline (wave 1)
and both wave 4 and wave 6 questionnaires
(n=460). Wave 4 data are required to assess
whether booster effects differ by prior involvement in dating violence, and wave 6 data are
required to assess booster and 4-year follow-up
effects of Safe Dates. Of the 460 adolescents,
201 were in the control group, 124 were in the
group that received only Safe Dates, and 135
were in the group that received Safe Dates and
TABLE 1—Baseline Characteristics of the Baseline Sample and the Study Sample:
North Carolina, 1994
Baseline Sample (n = 957)
Percentage or Mean
Female, %
White, %
Mean perpetration scores
Psychological
Physical
Serious physical
Sexual
Mean victimization scores
Psychological
Physical
Serious physical
Sexual
Standard Deviation
51.20
72.80
Study Sample (n = 460)
Percentage or Mean
Standard Deviation
58.50*
75.60
2.15
0.79
0.19
0.06
4.19
3.94
1.34
0.50
1.67
0.69
0.14
0.05
4.11
3.46
1.15
0.38
3.75
1.40
0.25
0.16
6.72
4.36
1.25
0.66
3.30
1.22
0.21
0.17
6.65
3.87
1.01
0.72
Note. Satterthwaite’s approximation for the degrees of freedom for the appropriate t test was used.
*P < .01
the booster. The only statistically significant difference between the study sample (n=460)
and the 957 8th graders who completed baseline questionnaires was gender; there were significantly more females in the study sample
(58.5%) than in the baseline sample (51.2%)
(P=.01) (Table 1).
(i.e., forced them to have sex, and forced
them to do something sexual that they did
not want to do). Parallel questions were used
to measure physical, serious physical, and sexual violence victimization. Adolescents were
asked to report acts perpetrated or received
that were not in self-defense.
Measures
Attrition Analyses
The 8 behavioral outcomes measured, 4
pairs of parallel perpetration and victimization outcomes, were anchored to the previous
year. The frequency of perpetrating each of
14 psychologically abusive acts (e.g., “damaged something that belonged to them,” “insulted them in front of others”) was summed
to form a composite score for psychological
abuse perpetration. A parallel procedure was
used to create a composite score for psychological abuse victimization. The frequency of
perpetrating each of 18 physically and sexually violent acts (e.g., “slapped them,” “kicked
them,” “hit them with my fist”) was summed
to form a composite score for physical violence perpetration. Serious physical violence
perpetration was defined by the sum of responses to a subset of 6 serious acts (i.e.,
choked, burned, hit with a fist, hit with something hard besides a fist, beat up, and assaulted with a knife or gun). Sexual violence
was defined by the sum of a subset of 2 acts
The outcome in our attrition analysis was
whether adolescents who completed a baseline questionnaire also completed wave 4 and
wave 6 questionnaires. Our attrition analysis
indicated that there were no significant interactions between treatment condition and baseline characteristics when predicting dropout
status and that the amount of attrition did not
differ for treatment and control groups. Gender and serious physical violence victimization
were associated at P < .05 with dropout status
in both treatment and control groups; males
were more likely than females to drop out of
the study (odds ratio [OR] = 1.69; 95% confidence interval [CI] = 1.13, 2.53), and the odds
of dropping out decreased with increased serious physical violence victimization (OR = 0.51
per unit; 95% CI = 0.30, 0.89).
620 | Research and Practice | Peer Reviewed | Foshee et al.
Analysis Strategy
Linear regression models were used to assess Safe Dates’ effects and booster effects,
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
and effect modifiers. Each of the 8 wave 6
outcome variables was regressed on treatment condition (0 = control and 1 = Safe
Dates but no booster), booster condition (0 =
Safe Dates and 1 = Safe Dates + booster), and
4 covariates: gender (0 = male, 1 = female),
race (0 = White and 1 = non-White), the wave
1 (baseline) value of the outcome variable,
and the wave 4 value of the outcome variable. The interactions of the treatment and
booster variables with the 4 covariates were
included. The interactions with gender and
race assessed whether program effects were
moderated by gender and race, respectively.
The interaction between the wave 1 value of
the outcome variable and treatment condition
assessed whether the effects of Safe Dates
were moderated by prior (i.e., in the previous
year) involvement in dating violence. The interaction between the wave 4 value of the
outcome variable and booster condition assessed whether the effects of the booster were
moderated by prior (i.e., in the year before
the booster) involvement in dating violence.
Models were reduced using a backward elimination procedure.
When statistically significant interactions
remained in the reduced models, we calculated the predicted mean of the outcome for
each intervention condition based on the parameters of the reduced models, and then calculated the difference in those predicted
means at each level of the moderator variable. For these analyses, prior involvement in
dating violence was reduced to 3 strata: no
prior involvement, the mean level of involvement (average prior involvement), and the
mean level of involvement plus 1 SD (high
prior involvement). Statistical tests were computed to determine whether there were statistically significant differences in predicted
means between the intervention conditions
for each level of the moderator.
RESULTS
We first present results concerning the
long-term effects of Safe Dates, followed by
results concerning the effects of the booster.
For each, we present the effects on perpetration followed by the effects on victimization.
Because neither race nor gender moderated
either Safe Dates or booster effects on any of
TABLE 2—Reduced Models When Predicting Perpetration of Dating Violence
Psychological
Intercept
Treatment (Safe Dates vs control)
Booster (Safe Dates + booster vs Safe Dates)
Gender
Race
Wave 1 outcome
Wave 4 outcome
Wave 1 outcome by treatment
Wave 1 outcome by booster
Wave 4 outcome by treatment
Wave 4 outcome by booster
Physical
Serious Physical
Sexual
SD
SD
SD
SD
2.33**
–1.07
0.40
–0.25
–0.18
0.04
0.13*
0.31*
–0.16
–0.14
0.34**
0.55
0.72
0.61
0.42
0.49
0.07
0.06
0.14
0.16
0.10
0.12
0 .08
–1.11*
0.70
–0.35
–0.22
0.14**
0.02
0.53
0.49
0.46
0.35
0.41
0.05
0.03
–0.01
–0.42**
0.21
–0.18
–0.07
–0.02
–0.01
0.17
0.16
0.14
0.11
0.13
0.05
0.03
–0.01
–0.10*
0.05
–0.08*
–0.02
–0.00
–0.05
0.05
0.05
0.05
0.04
0.04
0.05
0.03
Note. The wave 4 outcome-by-treatment and the wave 1 outcome-by-booster interactions are included in the models as
required because of the dummy coding of the treatment and booster variables, but they are conceptually meaningless.
Analyses controlled for the correlation between individuals in the same school by using SAS PROC MIXED with school
specified as a random effect.
*P < .05
**P < .01
the 8 outcomes, these interactions are not further considered in this article.
Safe Dates’ Effects on Perpetration
As shown in Table 2, adolescents who received only Safe Dates reported perpetrating
significantly less physical (β = −1.11, P = .02),
serious physical (β = −.42, P = .01), and sexual
(β = −.10, P = .04) dating violence perpetration
at the 4-year follow-up than those in the control group. Safe Dates’ effects on psychological abuse perpetration are moderated by
prior (wave 1) involvement in dating violence
(β = .31, P = .02). As noted in Table 3, in all 3
strata of prior psychological abuse perpetration, the Safe Dates group reported less psychological abuse perpetration than the control
group at follow-up. However, none of those
differences were statistically significant. The
likely reason for the significant interaction is
that the difference in the Safe Dates and control group predicted means is progressively
less as prior psychological abuse perpetration
status increases.
Safe Dates Effects on Victimization
As shown in Table 4, Safe Dates had a significant main effect on sexual victimization
(β = −.23, P = .01) in the expected direction
but no effect on psychological abuse victim-
April 2004, Vol 94, No. 4 | American Journal of Public Health
ization (β = −.35, P = .68), and the effects of
Safe Dates on physical and serious physical
victimization were moderated by prior
(wave 1) involvement with the behavior
(β = .34, P = .02; β = .59, P = .003, respectively). As noted in Table 3, in all 3 strata of
prior physical abuse victimization, the Safe
Dates group reported less physical abuse victimization at follow-up than the control
group. These differences were statistically significant when prior physical victimization
was average (P = .01) and high (P = .002) and
close to significant when there was no prior
physical victimization (P = .07). The pattern
was similar for serious victimization: in all 3
strata of prior serious physical victimization,
adolescents exposed only to Safe Dates reported significantly less victimization from serious dating violence than adolescents in the
control group did.
Booster Effects on Perpetration
As shown in Table 2, the booster did not
improve the effectiveness of Safe Dates in preventing physical (β = .70, P = .12), serious physical (β = .21, P = .14), or sexual (β = .05, P = .26)
dating violence perpetration, and prior (wave 4)
involvement in psychological abuse perpetration moderated the effect of the booster on
psychological abuse perpetration (β = .34,
Foshee et al. | Peer Reviewed | Research and Practice | 621
 RESEARCH AND PRACTICE 
and booster variables were dummy coded, we
were able to determine the differences in the
predicted means between the control and the
booster group from the estimates in Table 2.
We determined if those differences were statistically significant using linear contrasts with
SAS (SAS Institute Inc, Cary, NC). There were
no significant differences between the booster
and the control group in follow-up physical
(P = .38), serious physical (P = .16), or sexual
dating violence perpetration (P = .28). There
were also no significant differences between
those 2 groups in follow-up psychological
abuse perpetration in any of the strata of
prior (wave 4) psychological abuse perpetration. Thus, there were no situations in which
the booster group reported significantly more
perpetration at follow-up than controls.
TABLE 3—Differences in the Predicted Means on the Follow-Up Outcomes Between
Specified Intervention Groups and Significance Levels, Stratifying by Prior Involvement in
Dating Violence
Prior Involvementa
Follow-Up Outcome
Psychological abuse perpetration
Safe Dates mean minus control mean
Safe Dates + booster mean minus Safe Dates mean
Physical abuse victimization
Safe Dates mean minus control mean
Safe Dates + booster mean minus Safe Dates mean
Serious physical victimization
Safe Dates mean minus control mean
Safe Dates + booster mean minus Safe Dates mean
Sexual violence victimization
Safe Dates + booster mean minus Safe Dates mean
None
Average
High
–1.07
0.40
–0.86
0.88
–0.33
2.02*
–1.12
0.42
–1.53**
0.69
–2.74**
1.49
–0.45*
0.08
–0.50**
0.22
–0.66*
0.82*
0.05
0.14
0.52***
Note: Predicted means for each treatment condition were calculated based on the reduced models in Tables 2 and 4. The
differences in predicted means in the treatment conditions are presented in this table.
a
Prior involvement refers to involvement in the same types of dating violence as the follow-up outcome.
*P < 0.05
**P < 0.01
***P < 0.001
Booster Effects on Victimization
TABLE 4—Reduced Models When Predicting Victimization of Dating Violence
Psychological
Intercept
3.67***
Treatment (Safe Dates vs control)
–0.35
Booster (Safe Dates + booster vs Safe Dates) 0.68
Gender
–0.46
Race
–1.30
Wave 1 outcome
0.15**
Wave 4 outcome
0.30***
Wave 1 outcome by treatment
Wave 1 outcome by booster
Wave 4 outcome by treatment
Wave 4 outcome by booster
Physical
SD
0.80 0.47
0.86 –1.12
0.91 0.42
0.69 –0.48
0.80 –0.74
0.06 0.04
0.05 0.43***
0.34*
–0.10
–0.44***
0.21*
Serious Physical
Sexual
SD
SD
SD
0.64
0.62
0.59
0.42
0.49
0.08
0.06
0.14
0.16
0.08
0.11
0.01
–0.45*
0.08
–0.11
–0.29
–0.04
0.37***
0.59**
–0.24
–0.42***
0.47**
0.21
0.20
0.19
0.14
0.16
0.08
0.06
0.20
0.24
0.08
0.15
–0.16
–0.23**
0.05
0.02
–0.11
0.10*
0.28***
0.09
0.08
0.08
0.06
0.07
0.04
0.06
–0.28*** 0.08
0.50*** 0.11
Note: The wave 4 outcome-by-treatment and the wave 1 outcome-by-booster interactions are included in the models as
required because of the dummy coding of the treatment and booster variables, but they are conceptually meaningless.
Analyses controlled for the correlation between individuals in the same school by using SAS PROC MIXED with school
specified as a random effect.
*P < .05
**P < .01
***P < .001
P = .003 ). As can been seen in Table 3, those
adolescents high in prior psychological abuse
perpetration who were exposed to the booster
reported significantly more psychological
abuse perpetration at follow-up than those exposed only to Safe Dates (P = .03).
Next we compared the booster to the control group. Because of the way the treatment
622 | Research and Practice | Peer Reviewed | Foshee et al.
We first compared the booster to the Safe
Dates–only group. As shown in Table 4, there
were no effects of the booster on psychological abuse victimization (β = .68, P = .46), and
the effects of the booster on physical (β = .21,
P = .05), serious physical (β = .47, P = .002),
and sexual victimization (β = .50, P < .0001)
were all moderated by prior (wave 4) victimization. As noted in Table 3, in all 3 strata of
prior physical abuse victimization, adolescents
exposed to the booster reported more physical victimization at follow-up than those exposed only to Safe Dates; however, none of
these differences were statistically significant.
A similar pattern emerged when considering
serious physical and sexual victimization in
that in all 3 strata of prior victimization, adolescents exposed to the booster reported
more serious physical and sexual victimization at follow-up than adolescents who received only Safe Dates. Those differences
were statistically significant only when prior
involvement in dating violence was high.
Next we compared the booster to the control group. There were no significant differences between the booster and the control
group in follow-up psychological abuse victimization (P = .70). Within the strata of prior
(wave 4) physical, serious physical, and sexual violence victimization, the only significant
differences in the booster and control groups
were in serious victimization when there was
no prior serious victimization (P = .05) and
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
sexual victimization when there was no prior
sexual victimization (P = .03). In both cases,
those exposed to the booster reported significantly less victimization than controls. Thus,
there were no comparisons in which the
booster group reported significantly more
victimization at follow-up than controls, and
in 2 comparisons, the booster group reported
significantly less victimization at follow-up
than controls.
DISCUSSION
In this 4-year follow-up of Safe Dates, we
found significant treatment and control group
differences in the expected direction in physical, serious physical, and sexual dating violence perpetration and victimization. Although prior victimization moderated
program effects on physical and serious physical victimization, there were statistically significant program effects on those 2 victimization
variables at almost all strata of prior victimization. The program was equally effective for
males and females and for Whites and nonWhites. Compared with controls, adolescents
exposed to Safe Dates reported from 56% to
92% less dating violence victimization and
perpetration at follow-up.
It is unlikely that these favorable effects are
due to differential attrition, because we found
no evidence of greater attrition of high-risk
adolescents from our Safe Dates group than
from the control group, and the amount of attrition was the same in both groups.28 Because of the long period since program exposure, it is also unlikely that these changes
were the result of more socially desirable reporting of the outcomes by the treatment
than the control group. A likely explanation
for the favorable changes is that Safe Dates
caused the changes observed. Long-term effects may have been realized because Safe
Dates was offered at the beginning of the
adolescents’ dating careers (8th grade) and
included information and skills that could be
incorporated into individual dating practices
that continued through the high school years.
For example, adolescents were asked to actively consider how they wanted to be treated
by their dating partners, they analyzed the
negative consequences of being a perpetrator
and a victim of dating abuse, they learned ef-
fective ways of communicating with their
partners and for dealing with anger toward a
partner, and they learned how having unfair
gender-based expectations of partners could
lead to abuse. Specific to the prevention of
sexual dating violence, they analyzed verbal
and nonverbal cues that a partner is not
ready to have sex, were encouraged to be
clear with partners about sexual boundaries,
and discussed dating tips for protecting themselves from sexual dating violence and for respecting their partners.
The booster did not improve the effectiveness of Safe Dates. In fact, adolescents exposed to Safe Dates and the booster reported
significantly more psychological abuse perpetration and serious physical and sexual victimization at follow-up than those exposed only
to Safe Dates, but only when prior involvement in those forms of dating violence was
high. It is possible that the booster prompted
adolescents who were already being victimized to leave abusive relationships. Studies report that partner violence escalates when victims try to leave the abusive relationship.29–31
Boosters, because of their low intensity, may
be inappropriate for the secondary prevention
of dating violence. Leaving an abusive dating
partner can be complicated and dangerous,
and adolescents doing so may need support
from their family, friends, and community
agencies. A booster may motivate a victim to
leave the relationship but may need to be
paired with additional support to do that
safely and successfully.
Boosters have received substantial prominence. For example, both the National Cancer Institute and the Center for Substance
Abuse Prevention list boosters as essential
and effective elements of adolescent substance use prevention programs.32,33 However, only 3 studies on adolescent substance
use prevention rigorously evaluated the impact of a booster with an experimental design that allowed assessment of booster effects independent of original treatment
effects,20,22,24 and there have been no prior
studies testing the effectiveness of a booster
in preventing dating violence or other forms
of youth violence. Our findings suggest that
boosters could have negative effects. However, there were no situations in which the
booster group reported significantly more
April 2004, Vol 94, No. 4 | American Journal of Public Health
victimization or perpetration at follow-up
than the control group.
Attrition is the primary potential limitation
of this study. However, as mentioned earlier,
our analyses suggest that differential attrition
did not threaten the internal validity of the
study. It is also unlikely that attrition affected
external validity given the similarity of the
study sample to the baseline sample, which
because of the high response rate should approximate the characteristics of 8th graders in
the county. The study sample did have significantly more females than the baseline sample,
but given that program effects did not vary by
gender, this finding should not reduce the
generalizability or external validity of the
findings. These findings can be generalized
with a fair amount of confidence to other
rural counties with similar demographic characteristics. Relative to the United States as a
whole, when the study was conducted, the
county had an overrepresentation of minority
residents, lower-income households, and
more individuals with limited education.
Another potential limitation is reliance on
self-reports of dating violence. Previous analyses of these data, however, suggest that our
measures of dating abuse have high construct
validity: they correlate as expected with theoretically based constructs34; also as expected,
the prevalence of psychological abuse was
larger than the prevalence of physical abuse,
which was larger than the prevalence of serious physical and sexual abuse; and the prevalences of the various forms of dating abuse
were comparable to those found in other adolescent dating abuse studies.2,6 Also, consistent with almost all other studies of adolescent dating violence, gender was not
associated with physical dating violence victimization1,5,9 or perpetration3,6,35 but was associated with sexual dating violence victimization, with females reporting more sexual
dating violence victimization than boys.9
Safe Dates is being used in many geographically diverse areas, including inner-city urban
areas, rural areas, and countries besides the
United States. However, the only published
evaluations of the Safe Dates program have
been in this rural US sample.13,14 Future studies are needed to determine the effectiveness
of Safe Dates for adolescents living in other
locales. Also, from anecdotal reports we know
Foshee et al. | Peer Reviewed | Research and Practice | 623
 RESEARCH AND PRACTICE 
that the program is not always used in its entirety (the play, curriculum, and poster contest), yet the design of our evaluation did not
allow assessment of the effectiveness of individual components. Future evaluations need
to incorporate designs that allow assessment
of individual components and fewer curriculum sessions.
In conclusion, this is the first experimental
study to test the long-term effects of an adolescent dating violence prevention program
and to test the efficacy of a booster for preventing adolescent dating violence. Safe Dates
reduced dating violence as many as 4 years
after the program. The booster did not improve the effectiveness of Safe Dates. Neither
gender nor race moderated program effects,
but prior behavior moderated some effects.
These findings suggest that implementation of
the Safe Dates program to reduce dating violence is indicated but that the booster should
not be used.
About the Authors
Vangie A. Foshee, Karl E. Bauman, and Susan T. Ennett are
with the Department of Health Behavior and Health Education and Thad Benefield and Chirayath Suchindran are with
the Department of Biostatistics, School of Public Health,
University of North Carolina at Chapel Hill. G. Fletcher
Linder is with the Department of Sociology and Anthropology, James Madison University, Harrisonburg, Va.
Requests for reprints should be sent to Vangie Foshee,
PhD, Associate Professor, Department of Health Behavior
and Health Education, 317 Rosenau Hall, University of
North Carolina at Chapel Hill, Chapel Hill, NC
27599–7440 (e-mail: [email protected]).
This article was accepted May 9, 2003.
Contributors
V. Foshee conceived of the study, supervised all aspects
of its implementation, and prepared drafts of the manuscript. K. Bauman assisted with all aspects of designing
and conducting the study. S. Ennett contributed to the
analysis strategy, presentation of results, and interpretation of the findings. F. Linder managed all aspects of
the study. T. Benefield completed all the analyses for
the article with direct supervision by C. Suchindran and
V. Foshee. C. Suchindran designed the analysis strategy.
All authors contributed by conceptualizing ideas, interpreting findings, and reviewing drafts of the article.
Acknowledgment
jects. Active parental consent and adolescent assent
were obtained from all study adolescents.
Five- and six-year follow-up results from four seventhgrade smoking prevention strategies. J Behav Med.
1989;12:207–218.
References
19. Elder JP, Wildey M, de Moor C, et al. The longterm prevention of tobacco use among junior high
school students: classroom and telephone interventions.
Am J Public Health. 1993;83:1239–1244.
1. Avery-Leaf S, Cascardi M, O’Leary KD, Cano A.
Efficacy of a dating violence prevention program on attitudes justifying aggression. J Adolesc Health. 1997;21:
11–17.
2. Bergman L. Dating violence among high school
students. Social Work. 1992;37:21–27.
3. Henton J, Cate RM, Koval J, Lloyd S, Christopher S.
Romance and violence in dating relationships. J Fam Issues. 1983;4:467–482.
4. Foshee VA. Gender differences in adolescent dating abuse prevalence, types, and injuries. Health Educ
Res. 1996;11:275–286.
5. Malik S, Sorenson SB, Aneshensel CS. Community
and dating violence among adolescents: perpetration
and victimization. J Adolesc Health. 1997;21:291–302.
6. O’Keeffe NK, Brockopp K, Chew E. Teen dating
violence. Social Work. 1986;31:465–468.
7. O’Keefe M. Predictors of dating violence among
high school students. J Interpersonal Violence. 1997;12:
546–568.
8. Schwartz M, O’Leary SG, Kendziora KT. Dating
aggression among high school students. Violence Vict.
1997;12:295–305.
9. O’Keefe M, Treister L. Victims of dating violence
among high school students. Are the predictors different for males and females? Violence Against Women.
1998;4:195–223.
10. Halpern CT, Oslak SG, Young ML, Martin SL,
Kupper LL. Partner violence among adolescents in opposite-sex romantic relationships: findings from the National Longitudinal Study of Adolescent Health. Am J
Public Health. 2001;91:1679–1685.
11. Molidor C, Tolman RM. Gender and contextual
factors in adolescent dating violence. Violence Against
Women. 1998;4:180–194.
12. Silverman JG, Raj A, Mucci LA, Hathaway JE.
Dating violence against adolescent girls and associated
substance use, unhealthy weight control, sexual risk behavior, pregnancy, and suicidality. JAMA. 2001;286:
572–579.
13. Foshee VA, Bauman KE, Arriaga XB, Helms RW,
Koch GG, Linder GF. An evaluation of Safe Dates, an
adolescent dating violence prevention program. Am J
Public Health. 1998;88:45–50.
14. Foshee VA, Bauman KE, Greene WF, Koch GG,
Linder GF, MacDougall JE. The Safe Dates program:
1-year follow-up results. Am J Public Health. 2000;90:
1619–1622.
15. Ellickson PL, Bell RM, McGuigan K. Preventing
adolescent drug use: long-term results of a junior high
program. Am J Public Health. 1993;83:856–861.
This study was funded by the Centers for Disease Control and Prevention, Cooperative Agreement U81/
CCU409964.
16. Dusenbury L, Falco M, Lake A. A review of the
evaluation of 47 drug abuse prevention curricula available nationally. J Sch Health. 1997;67:127–132.
Human Participant Protection
17. Flay BR, Koepke D, Thomson SJ, Santi S, Best JA,
Brown KS. Six-year follow-up of the first Waterloo
school smoking prevention trial. Am J Public Health.
1989;79:1371–1376.
This study was reviewed and approved by the University of North Carolina, School of Public Health, institutional review board for the protection of human sub-
18. Murray DM, Pirie P, Leupker RV, Pallonen U.
624 | Research and Practice | Peer Reviewed | Foshee et al.
20. Dijkstra M, Mesters I, De Vries H, van Breukelen G,
Parcel GS. Effectiveness of a social influence approach
and boosters to smoking prevention. Health Educ Res.
1999;14:791–802.
21. Ellickson PL, Bell RM. Drug prevention in junior
high: a multi-site longitudinal test. Science. 1990;247:
1299–1305.
22. Botvin GJ, Renick NL, Baker E. The effects of
scheduling format and booster sessions on a broadspectrum psychosocial approach to smoking prevention. J Behav Med. 1983;6:359–379.
23. Spoth RL, Redmond C, Shin C. Randomized trial
of brief family interventions for general populations:
adolescent substance use outcomes 4 years following
baseline. J Consult Clin Psychol. 2001;69:627–642.
24. Botvin GJ, Baker E, Filazzola AD, Botvin EM. A
cognitive-behavioral approach to substance abuse prevention: one-year follow-up. Addict Behav. 1990;15:
47–63.
25. Klepp KI, Tell GS, Vellar OD. Ten-year follow-up
of the Oslo youth study smoking prevention program.
Prev Med. 1993;22:453–462.
26. Bauman KE, Foshee VA, Ennett ST, et al. The influence of a family program on adolescent tobacco and
alcohol use. Am J Public Health. 2001;91:604–610.
27. Foshee VA, Linder GF, Bauman KE, et al. The
Safe Dates Project: theoretical basis, evaluation design,
and selected baseline findings. Am J Prev Med. 1996;
12(suppl 5):39–47.
28. Resnicow K, Botvin G. School-based substance
use prevention programs: why do effects decay? Prev
Med. 1993;22:484–490.
29. Ellis D, Dekeseredy WS. Rethinking estrangement,
interventions, and intimate femicide. Violence Against
Women. 1997;3:590–609.
30. Wilson M, Daly M. Spousal homicide risk and estrangement. Violence Vict. 1993;8:3–16.
31. Bachman R, Saltzman LE. Violence against
Women: Estimates from the Redesigned Survey. Washington, DC: US Department of Justice, Office of Justice
Programs, Bureau of Justice Statistics; 1995:1–8.
32. Glynn TJ. Essential elements of school-based
smoking prevention programs. J Sch Health. 1989;59:
181–188.
33. Schinke S, Brounstein P, Gardners S. ScienceBased Prevention Programs and Principles, 2002. Rockville, Md: Center for Substance Abuse Prevention,
Substance Abuse and Mental Health Services Administrtion, 2002. DHHS publication (SMA) 03-3764.
34. Foshee VA, Linder F, MacDougall JE, Bangdiwala
S. Gender differences in the longitudinal predictors of
adolescent dating violence. Prev Med. 2001;32:
128–141.
35. Symons PY, Groër MW, Kepler-Youngblood P,
Slater V. Prevalence and predictors of adolescent dating violence. J Child Adolesc Psychiatr Nurs. 1994;7(3):
14–23.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Community Violence and Asthma Morbidity:
The Inner-City Asthma Study
| Rosalind J. Wright, MD, MPH, Herman Mitchell, PhD, Cynthia M. Visness, MA, MPH, Sheldon Cohen, PhD, James Stout, MD, MPH, Richard Evans,
MD, MPH, and Diane R. Gold, MD, MPH
In the United States, recent trends of increasing childhood asthma morbidity disproportionately affect urban children who are poor
and non-White. Known risk factors (e.g., air
pollutants, environmental and in utero tobacco smoke, viral infections, indoor allergens) do not fully explain these trends.1
Geographic variation in asthma outcomes
among large cities2 and among neighborhoods within cities3–5 has been observed.
Variation in asthma morbidity across urban
neighborhoods cannot be explained by socioeconomic factors alone. Many New York
City communities do not have elevated
asthma morbidity in spite of the fact that
they are comparably low on many socioeconomic indicators and have physical environmental exposures seemingly similar to other
high-risk neighborhoods. These findings indicate that other factors may mediate the effects of living in low–socioeconomic status
(SES) neighborhoods.
Health disparities research points to the influence of specific community characteristics,
conceptualized as neighborhood disadvantages, on residents’ health and well-being.6
Neighborhood disadvantage—characterized
by the presence of a number of communitylevel stressors, including poverty, underemployment, limited social capital, substandard
housing, and high crime and violence rates7—
is prevalent in many US urban communities.8
Studies of minority and low-income populations have shown a high prevalence of
children who experience9 and witness violence in the inner city.10–12 A prevalence
study in a Boston, Mass, pediatric primary
care clinic found that 10% of children younger than 6 years had witnessed a knifing or
a shooting and that 47% had heard gunshots in their neighborhoods.11 In Chicago,
Ill, investigators found that 42% of children
between the ages of 7 and 13 years had witnessed a shooting.13
Objectives. We examined the association between exposure to violence and asthma
among urban children.
Methods. We obtained reports from caretakers (n=851) of violence, negative life events,
unwanted memories (rumination), caretaker-perceived stress, and caretaker behaviors
(keeping children indoors, smoking, and medication adherence). Outcomes included caretaker-reported wheezing, sleep disruption, interference with play because of asthma, and
effects on the caretaker (nights caretaker lost sleep because of child’s asthma).
Results. Increased exposure to violence predicted higher number of symptom days
(P = .0008) and more nights that caretakers lost sleep (P = .02) in a graded fashion
after control for socioeconomic status, housing deterioration, and negative life events.
Control for stress and behaviors partially attenuated this gradient, although these variables had little effect on the association between the highest level of exposure to morbidity, which suggests there are other mechanisms.
Conclusions. Mechanisms linking violence and asthma morbidity need to be further explored. (Am J Public Health. 2004;94:625–632)
Exposure to violence may affect asthma
through many pathways.14 It may be related
to psychological stress experienced by those
who witness or are victims of violence,15
which may have an impact on asthma.16
Evolving research is exploring adverse psychological consequences among children
who grow up in violent neighborhoods.17,18
Health behaviors may be influenced by environmental factors, including high levels of
stress, violence, and unpredictable daily life
experiences. Exposure to violence (and
other determinants of neighborhood disadvantage) may influence impulse control and
risk-taking behavior, resulting in the adoption of coping behaviors (e.g., smoking) and
leading to increased exposure to a known
environmental trigger of asthma, (tobacco
smoke).19 Families who live in a violent environment may develop a fatalistic outlook
that undermines their ability to invest in the
future by complying with prescribed asthma
treatment.20
High crime rates are correlated with other
indicators of social disadvantage, including
poor-quality housing. Deteriorated housing
has been linked to high household cockroach
allergen levels,21 which in turn may increase
April 2004, Vol 94, No. 4 | American Journal of Public Health
asthma morbidity.22 Exposure to community
violence may influence behaviors that could
result in increased exposure to other known
environmental risk factors. Parents who live
in high-violence communities may restrict
their children’s outdoor activities, causing increased indoor-allergen exposure and higher
asthma morbidity. Individuals who live in
low-SES neighborhoods with high violence
rates also may experience other adverse life
events more frequently than their higher-SES
counterparts.23,24
We examined the association between exposure to community violence and caretaker-reported asthma symptoms and behaviors in the Inner-City Asthma Study (ICAS).
We hypothesized that families with children
who lived with higher levels of violence
would have increased asthma morbidity. We
examined factors that might be correlated
with violence (SES and other adverse life
events) and hypothesized mediating pathways, including measures of the psychological experience of stress (perceived stress and
intrusive memories), poor compliance with
medication regimens, and other caretaker
behaviors (keeping children indoors and
smoking).
Wright et al. | Peer Reviewed | Research and Practice | 625
 RESEARCH AND PRACTICE 
METHODS
The ICAS was conducted from August
1998 to July 2001. This study enrolled 937
children with asthma (aged 5 to 12 years)
and their caretakers to an intervention study
to reduce symptoms. Families were recruited from 7 cities: Boston; Chicago; New
York City (Manhattan and the Bronx); Dallas, Tex; Seattle, Wash; and Tucson, Ariz.
The study design has been detailed elsewhere.25 Eligibility required that the child
had at least 1 hospitalization or 2 emergency department visits for asthma during
the 6 months before screening. Census
tracts with 20% to 40% of households
below federal poverty guidelines were targeted. Many census tracts also were racially
segregated (i.e., Black–White): Boston
(39.4% vs 39%), Chicago (65.2% vs
18.5%), Manhattan (47.9% vs 22.1%), the
Bronx (38.7% vs 19.3%), Dallas (44.5% vs
30.1%), Seattle (8.6% vs 69.4%), and Tucson (3.5% vs 59.1%).
After informed consent was obtained,
trained bilingual interviewers administered
a baseline survey to the child’s primary
caretaker that included questions about demographics, asthma morbidity, home environmental characteristics, exposure to tobacco smoke, the child’s medication
regimen and any problems with adherence,
and psychosocial well-being measures described elsewhere.26
Assessment of Exposure to Violence
A community violence survey27 was administered to caretakers. Caretakers were
asked whether any of the following events
had occurred in their neighborhoods during
the past 6 months: (1) a fight in which a
weapon was used, (2) a violent argument
between neighbors, (3) a gang fight, (4) a
sexual assault or rape, and (5) a robbery
or mugging. Answers to these 5 items
were summed to produce the Adult Violence Score. Additionally, caretakers were
asked (1) whether the caretaker was afraid
that the child would be hurt by violence
in the neighborhood, and (2) whether
the caretaker did not let the child play
outside because of fear of violence in the
neighborhood.
Additional Measures of Stress
The Negative Life Events (NLE) instrument is a modified version of the List of Recent Experiences28,29 that has been shown to
have good test–retest reliability for the scale
(0.89 to 0.94) and for specific items (0.70).30
Participants indicated whether they had undergone any of the enumerated experiences
during the past 12 months and whether the
experience in question had a positive or a
negative impact. A few items (e.g., death of a
family member) were assumed to be consensually negative. Total NLE score was derived
by adding the number of negative experiences (either consensually rated or participant-rated).
The experience of unwanted thoughts and
memories (rumination) was ascertained for
each reported negative life event with this follow-up question: “In the last month, how
often did you experience unwanted thoughts,
memories, or images about this event?” Each
item was scored on a 5-point frequency scale
of “never” (0) through “very often” (4). A
maximum score was based on the highest frequency of unwanted thoughts and memories
reported for any experience (other than violence). Thus, if an individual reported 2 negative life events but experienced unwanted
thoughts and memories only in connection
with 1 of the events, the participant was classified on the basis of the higher frequency.
The 4-item Perceived Stress Scale (PSS4)31
measured the degree to which respondents
had felt that their lives were unpredictable,
uncontrollable, and overwhelming in the preceding month (reliability = 0.85). Each item
was scored on a 5-point frequency scale of
“never” (0) through “very often” (4), and an
overall/total score was obtained by summing
the items (maximum = 16). Higher scores indicated greater stress.
Sociodemographic Indicators
Socioeconomic indicators included household income, the presence of at least 1 employed adult in the household, and caretaker
level of education. Housing deterioration was
assessed by summing a number of problems
including water damage on walls or ceilings;
other evidence of leaks; damaged or rotting
windows; cracks or holes in floors; and
chipped, cracked, or peeling paint on walls or
626 | Research and Practice | Peer Reviewed | Wright et al.
windows. Race/ethnicity was categorized as
Hispanic, Black, or White/other.
Outcome Measures
Measures of morbidity included caretakerreported wheezing, sleep disruption, or interference with play activities caused by asthma
during the preceding 2 weeks and the impact
of the child’s asthma on the caretaker (number of nights caretaker lost sleep because of
child’s asthma). A measure of maximum
symptom days during the preceding 2 weeks
was defined as the number of days that the
child experienced wheezing, sleep disturbance, or disruption of play activities because
of asthma.
Analyses
A total of 851 children and their caregivers
had complete data for all covariates. We used
analysis of variance (ANOVA) to examine
mean outcome measures ([1] maximum
symptom days and [2] nights caretakers lost
sleep) by level of community violence. All
analyses were adjusted for study center (site
adjusted). A test for linear trend that used orthogonal polynomial coefficients was used to
determine the relationship between (1) the
Adult Violence Score and the mean asthma
morbidity score and (2) the Adult Violence
Score and the mean caretaker impact score.32
Control variables were added in a stepwise
fashion. We first added standard control variables, including SES, race/ethnicity, and a
composite measure of general condition of
the home, to ascertain whether associations
we found were spurious (i.e., is level of exposure to violence merely a marker of low SES,
race/ethnicity, or substandard housing stock,
each of which may increase exposure to physical environmental factors related to morbidity?). We then added total NLE score to test
whether the influence of violence on asthma
morbidity was in part caused by greater exposure to other adverse events. Next we added
hypothesized mediating variables. We introduced covariates into the linear model—individually or in sets—to determine whether they
modified the effect of violence on morbidity.
As covariates were added, we examined the
change in the ANOVA model sums of squares
related to the violence indicator. A substantial
decrease in the effect size of the association
between violence and the asthma morbidity
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
measure (i.e., percentage decrease in the violence sums of squares) would support 1 or
more of these mediating pathways. We identified 2 mediating pathways—caretaker behaviors and the psychological experience of
stress. Behaviors included (1) presence of
smokers in the household, (2) caretakers skipping medications, and (3) caretakers not allowing the child to play outside. The stressrelated predictors included (1) PSS4 score,
and (2) unwanted thoughts about stressful life
events (i.e., rumination over adverse events),
which is another measure of coping. A final
model was adjusted for site, SES, and all potential mediating variables. Mean outcome
measures adjusted for other covariates were
produced for each level of the violence score.
RESULTS
The frequency of caretaker-reported community violence varied across the sites: the
a
12
Violence exposure score
0
1
2
3
4
5
10
Mean Days
8
6
4
2
0
P for trend =
.0008
Site, race/
a
ethnicity, SES
.003
.003
.004
.009
+ Negative life
events
+ Behaviors
+ Stress
+ Behaviors and
stress
Models
5
b
Mean Nights
4
3
2
1
0
P for trend=
.02
Site, race/
a
ethnicity, SES
.06
.06
.07
.013
+ Negative life
events
+ Behaviors
+ Stress
+ Behaviors and
stress
Models
Note. Behaviors = caretaker behaviors, including smoking, keeping children indoors, and skipping medications.
Stress = Perceived Stress Scale and unwanted thoughts and memories. Each set of bars represents mean maximum symptom
days (or nights caretaker lost sleep) adjusted for control variables. All models are adjusted for site, race/ethnicity, and
socioeconomic status (SES) in addition to listed variates. Ps are for trends.
a
SES includes adjustment for household income, employment, caretaker education, and housing deterioration.
FIGURE 1—Mean (a) maximum symptom days and (b) nights caretaker lost sleep, by level
of adult exposure to violence: adjusted analyses.
April 2004, Vol 94, No. 4 | American Journal of Public Health
highest mean scores were reported in Chicago
and Manhattan (1.3 and 1.4, respectively);
Boston (0.9), the Bronx (1.2), and Dallas
(0.89) were close behind; and the lowest
mean scores were reported in Seattle (0.84)
and Tucson (0.65). Caretaker reports of violent events occurring in their neighborhoods
during the past 6 months were quite prevalent for certain categories of events: a fight in
which a weapon was used (28%), a violent
argument between neighbors (33%), a gang
fight (15%), a sexual assault or rape (9%), or
a robbery or mugging (21%). More than onethird of caretakers (38%) reported being
afraid their child would be hurt by violence
in the neighborhood and reported keeping
their children indoors owing to fear of violence (34%). Table 1 shows the mean exposure-to-violence scores and the outcome measures stratified by sociodemographic factors
and by control variables. Those caretakers
who had higher exposure-to-violence scores
were more likely to be minorities, were less
likely to report at least 1 employed adult in
the household, had more housing problems,
had greater perceived stress, ruminated more
about adverse life events, smoked more often,
kept their children indoors more often, and
skipped medications more often than caretakers who had lower scores.
Site-adjusted analyses showed a gradient
increase in mean maximum symptom days
with increasing exposure to violence (P =
.0006) (data not shown). Figure 1 shows the
associations between violence exposure and
mean (a) maximum symptom days or (b)
nights caretaker lost sleep, adjusted for control and hypothesized mediator variables. We
found no meaningful attenuation of the relationship between violence and caretaker-reported symptoms among children after we
controlled for sociodemographic factors (we
simultaneously adjusted for annual household
income, presence of at least 1 employed adult
in the home, caretaker education, housing deterioration score, and race/ethnicity) (P =
.0008). A similar graded relationship for exposure to violence and caretaker impact was
seen in these adjusted analyses (P = .02).
To assess whether level of exposure to violence was a marker for exposure to other adverse events, total NLE score was added to
the model. The graded relationship between
Wright et al. | Peer Reviewed | Research and Practice | 627
 RESEARCH AND PRACTICE 
TABLE 1—Mean Caretaker Violence Exposure Scores and Maximum Symptom Days, by Potential Control, Confounding,
and Mediating Variables: Inner-City Asthma Study (n = 851, baseline assessment), August 1998–July 2001
Characteristic
Caretaker race/ethnicity
Hispanic
African American
White/mixed/other
Household income, $
< 15 000
≥ 15 000
At least 1 employed adult in household
No
Yes
Caretaker education
High school graduate
Not a high school graduate
No. of housing problems
0
1
2
3
4
5
6
Smoking in household
No
Yes
NLEs score (quartiles)
1st
2nd
3rd
4th
Unwanted thoughts about adverse life events
Never
Almost never
Sometimes
Fairly often
Very often
PSS4 score (quartiles)
1st
2nd
3rd
4th
Ever skip medications
No
Yes
Afraid to let child play outside
No
Yes
n (%)
Mean Violence
Exposure Score
P
Mean Nights
Caretaker Lost Sleep
P
364 (42.8)
340 (40.0)
146 (17.2)
0.96
1.16
1.05
.1238
5.93
6.48
5.52
.1181
3.15
3.19
2.74
.5238
512 (60.2)
339 (39.8)
1.11
0.96
.0973
6.40
5.58
.0203
3.33
2.72
.0360
207 (24.3)
644 (75.7)
1.33
0.96
.0003
6.77
5.85
.0206
3.81
2.86
.0042
600 (70.5)
251 (29.5)
1.07
1.02
.5746
6.12
5.96
.6821
3.07
3.14
.8096
355 (41.7)
267 (31.4)
82 (9.6)
60 (7.0)
38 (4.5)
23 (2.7)
26 (3.1)
0.88
1.14
1.35
1.32
1.18
0.78
1.08
.0106
5.86
5.94
6.32
6.55
7.42
6.30
6.35
.6134
2.76
3.02
4.04
3.80
3.50
3.22
3.00
.1964
432 (50.8)
419 (49.2)
0.97
1.14
.0583
5.66
6.50
.0133
3.01
3.17
.5745
368 (43.2)
130 (15.3)
184 (21.6)
169 (19.9)
0.79
0.85
1.32
1.50
<.0001
5.70
5.58
6.74
6.54
.0435
2.77
2.58
3.69
3.51
.0218
296 (34.8)
53 (6.2)
191 (22.4)
135 (15.9)
176 (20.7)
0.74
0.81
1.07
1.04
1.65
<.0001
5.55
5.74
5.90
5.82
7.44
.0017
2.76
2.64
2.80
2.79
4.33
.0006
229 (26.9)
262 (30.8)
194 (22.8)
166 (19.5)
0.84
1.11
0.96
1.37
.0004
5.43
6.11
6.12
6.85
.0493
2.73
3.14
3.19
3.41
.4241
550 (64.6)
301 (35.4)
0.97
1.21
.0088
5.57
7.00
<.0001
2.83
3.56
.0146
559 (65.7)
292 (34.3)
0.82
1.51
<.0001
5.71
6.76
.0037
2.66
3.90
<.0001
P
Mean Maximum
Symptom Days
Note. NLE = Negative Life Events instrument; PSS4 = 4-item Perceived Stress Scale.
628 | Research and Practice | Peer Reviewed | Wright et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
violence and morbidity markers remained
significant.
Caretaker behaviors as mediating variables
were then considered. Adjustments for smoking, keeping the child indoors, and skipping
medications attenuated the gradient relationship between violence and morbidity markers. A significant trend remained for symptoms (P = .003), although the association was
borderline significant for caretaker impact
(P = .06). The decrease in impact of violence
was not uniform across the gradient. The
greatest absolute attenuation occurred in
groups with lower levels of exposure to violence (i.e., violence exposure score ≤ 3).
When we adjusted for measures of psychological stress, including the PSS4 score and
the frequency of unwanted thoughts and
memories, exposure to violence remained an
independent predictor of mean maximum
symptom days (P = .004). We found attenuation of the graded relationship between violence and morbidity markers after we added
these other measures of stress. Again, the
greatest attenuation occurred in the groups
with the lowest levels of exposure to violence.
Notably, when frequency of other negative
life events and the unwanted thoughts and
memories were added together in models
predicting asthma morbidity, negative life
events were no longer significant (P = .5), sug-
gesting that chronic reexperiencing of adverse
life events may have a greater impact than
discrete events.
In a final model adjusted for SES, other
negative life events, perceived stress, unwanted thoughts and memories, and caretaker behaviors, increased exposure to violence was still associated with greater mean
maximum symptom days (P = .009) and caretaker’s losing sleep (P = .13), with more attenuation of the gradient in the groups with the
lowest levels of exposure to violence.
To determine the relative contribution of
the standard control variables and the purported mediators in explaining the effect of
exposure to violence on morbidity, we examined the differences in the sums of squares associated with the violence exposure score
alone in site-adjusted models and the sums of
squares related to violence, and we adjusted
for each covariate as it was added. The percentage decrease in the violence exposure
sums of squares when each control variable
was added is shown in Table 2. Socioeconomic indicators and smoking in the home
explained relatively little of the effect of violence. Conversely, experiencing other negative life events and the occurrence of unwanted thoughts and memories of adverse
events individually explained the greatest proportion of change in the violence exposure
TABLE 2—Percentage Decrease in Violence Exposure Sums of Squares After Each Variable
Is Added Individually to the Model With Violence
sums of squares. An intermediate proportion
of change in the violence exposure sums of
squares was explained by caretaker-perceived
stress, skipping medications, and keeping the
child indoors more often.
In subsequent models that adjusted for
multiple variables, we again assessed by determining the percentage decrease in the
violence exposure sums of squares in the respective multivariate models the relative contribution of the combined covariates in explaining the effect of exposure to violence.
After we controlled for site, socioeconomic indicators, and race/ethnicity, 6% of the violence exposure effect was explained for symptoms and 11% for caretaker impact (i.e., on
the basis of a 6% and 11% decrease in the violence exposure sums of squares in the respective models). After all behaviors were
added with standard control variables (site,
SES, race/ethnicity), 34% of the effect of exposure to violence was explained for symptoms and 50% for caretaker impact. After the
other measures of stress were added with
standard controls, 36% of the violence exposure effect was explained for symptoms and
51% for caretaker impact. A fully adjusted
model including standard control variables,
other negative life events, stress measures
(perceived stress, unwanted thoughts and
memories), and the behavior variables accounted for 46% of violence exposure effect
for mean maximum symptom days and 70%
for nights caretaker lost sleep.
DISCUSSION
Decrease in Violence Exposure Sums of Squares, %
Added Variable
Socioeconomic status indicators
Income category
Household employment
Caretaker education
Housing problems
Caretaker behaviors
Smoking at home
Skipping medications
Caretaker will not let child play outside
Negative life events
Experience psychological stress
Unwanted thoughts
Perceived stress
Exposure to Violence Predicting
Maximum Symptom Days
3.0
7.0
0.37
1.2
Exposure to Violence Predicting
Nights Caretaker Lost Sleep
2.6
10.6
-0.31
1.9
7.7
14.0
18.1
25.2
1.7
12.5
37.3
30.6
31.4
18.1
40.7
16.7
April 2004, Vol 94, No. 4 | American Journal of Public Health
As in previous studies,10–13 a high prevalence of exposure to violence among the
inner-city families was found. Greater exposure to violence was independently associated
with asthma morbidity after simultaneous adjustment for income, employment status, caretaker education, housing problems, and other
adverse life events, which suggests that exposure to violence was not merely a marker for
these other factors. Psychological stress and
caretaker behaviors (keeping children indoors, smoking, and skipping medications)
partially explained the association between
higher exposure to violence and increased
asthma morbidity, although the greatest attenuation occurred among caretakers who re-
Wright et al. | Peer Reviewed | Research and Practice | 629
 RESEARCH AND PRACTICE 
ported lower levels of exposure to violence.
These findings suggest that other mechanisms
are operating between high-level exposure to
violence and childhood asthma morbidity.
The impact of exposure to violence on
asthma morbidity was, in part, attenuated
through psychological experiences of stress
(i.e., the degree to which participants felt that
their lives were uncontrollable, unpredictable,
or overwhelming and the occurrence of unwanted thoughts and memories in connection
with other adverse advents), supporting the
notion that exposure to violence is a pervasive stressor that adds to environmental demands imposed on an already vulnerable
population.33 Living in a violent environment
is associated with a chronic, pervasive atmosphere of fear and the perceived threat of violence.34,35 Families who live with violence are
more likely than those not exposed to violence to view their world and their lives as
being out of their control.36 Facing daily life
experiences in an unpredictable or an uncontrollable environment may predispose these
populations to suffer more deleterious effects
from stress.37
Psychological stress has been associated
with disturbed regulation of the hypothalamic-pituitary-adrenal (HPA) axis. An optimal
level of mediators is needed to maintain a
functional balance, and the absence of appropriate levels of glucocorticoids and catecholamines may allow immune mediators to
overreact, thereby increasing the risk of inflammatory disorders, such as asthma.38,39 In
this framework, exposure to violence may be
a psychosocial environmental factor that can
“get into the body” and result in long-term biological changes that contribute to asthma
morbidity.
Life events can have long-term effects on
stress through lasting psychological, behavioral, and physiological responses maintained
by recurrent unwanted thoughts about past
events.40 Caretakers who reported higher levels of exposure to violence were more likely
to ruminate. Ongoing rumination may have
an impact on problem-solving skills, may
erode perceived control, and may decrease
motivation to manage ongoing challenges, including management of a chronic illness such
as asthma.41 Caregivers who use ruminative
coping strategies may experience greater
stress and psychological comorbidity42 that
may more directly influence a child. Growing
evidence links caregiver stress to the stress responses of their offspring. Animal and human
studies suggest that caregiver stress may influence the stress response of the child and may
modify infant neuroendocrine function during
early development.43–45 It also is possible that
caretaker exposure to violence resulting in
posttraumatic stress symptoms (e.g., avoidance, rumination) may cue their children to
adopt less effective coping strategies, so that
the children themselves experience greater
stress.46 This area of study warrants further
research.
Poor adherence to medication regimens
partially explained the relationship between
exposure to violence and asthma. Coping with
a violent environment may have an impact
on compliance with therapy and with medical
follow-up. Living in a violent community has
been conceptualized as a barrier to keeping
appointments and to following prescribed exercise programs.47 Fearing to make a trip to a
pharmacy or a medical facility may lead to
lapses in prophylactic medication use, delayed
intervention, and higher morbidity. Ruminative coping may influence problem-solving behaviors, which may impede compliance.
Other unmeasured barriers to medication adherence may exist; for example, pharmacies
may be reluctant to remain open 24 hours a
day in high-crime communities. Exposure to
violence may have an impact on access to
medical care by diverting limited funds away
from primary care and asthma specialty clinics.48 Future research exploring other potential mediating pathways may contribute to
more effective intervention strategies targeting high-risk urban populations.
Keeping children indoors also mediated the
violence exposure and asthma relationship in
our study. Children who are kept indoors will
be more sedentary than those who go outside. This sedentariness may be linked to obesity, which has increased among US families
who live in poverty49 and has been linked to
asthma.50,51 Another reasonable hypothesis is
that children who are restricted from going
outside may have greater exposure to aeroallergens and increased likelihood of sensitization. Further research is needed to systematically examine this hypothesis.
630 | Research and Practice | Peer Reviewed | Wright et al.
Unexpectedly, smoking had little impact on
the association between exposure to violence
and asthma, perhaps because smoking is a
strategy to cope with stress52 that is related to
violence.19 This finding may reflect the fact
that smoking was considered a dichotomous
predictor, and we did not account for dose
(i.e., number of cigarettes per day) or misclassification of self-reported smoking.
Exposure to violence and asthma morbidity were related in a graded fashion, even
after we adjusted for socioeconomic indicators. The greatest absolute attenuation of the
gradient occurred at the lowest level of exposure to violence, after we controlled for potential mediators. The relationship between
the highest level of exposure to violence and
increased asthma morbidity was not influenced by caretaker behaviors, perceived
stress, or recurrent memories. Other factors
important in explaining the association at the
highest level of exposure to violence (i.e., that
covary with high rates of exposure to violence) may not have been measured or cannot be fully adjusted for when accounting for
individual-level SES factors. Crime and violence (or their absence) can be thought of as
indicators of collective well-being or social cohesion within a community,53,54 constructs increasingly linked to health.55
Neighborhood disadvantage, including
higher crime rates and community violence, is
enhanced in more racially segregated communities.56 Segregated minority group status may
predispose individuals to other pervasive
stressors (e.g., discrimination, institutionalized
violence, police injustice), a lack of infrastructure of the sort facilitating healthy living (e.g.,
fewer facilities for healthy physical recreation
or purchase of healthy foods), and other societal factors that link minorities with neighborhood disadvantage.57,58 Thus, individuals in
these communities may face multiple social
challenges simultaneously. Whereas individual
psychosocial stressors may have small effects,
cumulative stressors (at the individual and
ecological levels) can enormously increase the
likelihood of adverse health outcomes.59
Marginalized groups are disadvantaged
not only in their vulnerability to adverse
events but also in their access to coping resources.60,61 Parents who are worried about
their children’s safety may restrict their social
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
behavior; thus, the family’s ability to develop
support networks may be compromised (i.e.,
exposure to violence may lead to diminished
stress-buffering factors). These additional supports are especially important to the wellbeing of populations faced with the cumulative effects of many ecological stressors.
CONCLUSIONS
High crime rates and, thus, the real or perceived threat of violence are aspects of the
inner-city environment that have an impact
on the psychological and physiological functioning, as well as the health-promoting behaviors, of the inhabitants. Exposure to violence contributes to the environmental
demands that tax both individuals and the
communities in which they live. Systematic
exploration of an association between exposure to violence (an urban stressor) and
asthma may help us understand the rise in
morbidity and further our understanding of
the disproportionate asthma burden among
poor urban children.
About the Authors
Rosalind J. Wright is with the Beth Israel Deaconess Medical Center, Pulmonary and Critical Care Division, and
the Channing Laboratory, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School,
Boston, Mass. Herman Mitchell and Cynthia M. Visness
are with Rho Inc, Federal Systems Division, Chapel Hill,
NC. Sheldon Cohen is with the Department of Psychology,
Carnegie Mellon University, Pittsburgh, Pa. James Stout is
with the Department of Pediatrics, University of Washington School of Medicine, Seattle, Wash. Richard Evans is
with the Department of Pediatrics and Medicine, Northwestern University Medical School, Chicago, Ill. Diane R.
Gold is with the Channing Laboratory, Dept of Medicine,
Brigham and Women’s Hospital, Harvard Medical
School.
Requests for reprints should be sent to Rosalind J.
Wright, MD, MPH, Channing Laboratory, 181 Longwood Ave, Boston, MA 02115 (e-mail: [email protected]
channing.harvard.edu).
This article was accepted May 16, 2003.
Contributors
H. Mitchell, J. Stout, and R. Evans participated in the
conceptualization of the study and the implementation
of the Inner-City Asthma Study as principal investigators at their respective study sites. H. Mitchell, S. Cohen,
D. R. Gold, and R. J. Wright guided the inclusion of
stress measures in the parent study. H. Mitchell and
C. M. Visness conducted the analyses and were guided
by input from all of the authors. R. J. Wright synthesized
the analyses and led the writing. All of the authors interpreted the findings and reviewed drafts of the article.
Acknowledgments
Rosalind J. Wright was supported by National Institutes
of Health Mentored Clinical Scientist Development
award K08 HL07427. This study also was supported
by grants AI-39769, AI-39900, AI-39902, AI-39789,
AI-39901, AI-39761, AI-39785, and AI-39776 from
the National Institute of Allergy and Infectious Diseases,
National Institutes of Health, and the National Institute
of Environmental Health Sciences. The Fetzer Institute
supported the incorporation of the stress measures in
the parent study by S. Cohen, D. R. Gold, H. Mitchell,
and R. J. Wright.
The Inner-City Asthma Study is a collaboration of
the following institutions and investigators: Boston University School of Medicine, Boston, Mass—G. O’Connor
(principal investigator), S. Steinbach, A. Zapata, J. Cline,
and L. Schneider; Albert Einstein College of Medicine/
Jacobi Medical Center, Bronx, NY—E. Crain (principal investigator), L. Bauman, Y. Senturia, and D. Rosenstreich;
Children’s Memorial Hospital, Chicago, Ill—R. Evans III
(principal investigator), J. Pongracic, A. Sawyer, and K.
Koridek; University of Texas Southwestern Medical Center at Dallas, Tex—R. S. Gruchalla (principal investigator),
V. Gan, Y. Coyle, and N. F. Gorham; Mount Sinai School
of Medicine, New York, NY—M. Kattan (principal investigator), C. Lamm, M. Lippmann, E. Luder, M. Chassin,
and G. Xanthos; University of Washington School of
Medicine and Public Health, Seattle, Wash—J. Stout
(principal investigator), G. Shapiro, L. Liu, J. Koenig, M.
Lasley, S. Randels, and H. Powell; University of Arizona
College of Medicine, Tucson, Ariz—W. Morgan (principal
investigator), P. Enright, J. Goodwin, and T. Garcia; Data
Coordinating Center, Rho Inc, Chapel Hill, NC—H.
Mitchell (principal investigator), M. Walter, H. Lynn, S.
Hart, W. Tolbert, and E. Nuebler; Allergen Assay Laboratory, Harvard School of Public Health, Boston, Mass—
H. Burge, M. Muilenberg, and D. Gold; National Institute
of Allergy and Infectious Diseases, Bethesda, Md—M.
Plaut and E. Smartt; and National Institute of Environmental Health Sciences, Research Triangle Park, NC—G.
Malindzak.
Human Participant Protection
7. Attar BK, Guerra NG, Tolan PH. Neighborhood
disadvantage, stressful life events and adjustment in
urban elementary school children. J Clin Child Psychol.
1994;23:391–400.
8. Wilson W. The Truly Disadvantaged: The InnerCity, the Underclass, and Public Policy. Chicago, Ill: University of Chicago Press; 1987.
9. Finkelhor D, Dziuba-Leatherman J. Children as
victims of violence: a national survey. Pediatrics. 1994;
94(4 pt 1):413–420.
10. Schubiner H, Scott R, Tzelepis A. Exposure to violence among inner-city youth. J Adolesc Health. 1993;
14:214–219.
11. Groves BM, Zuckerman B, Marans S, Cohen DJ.
Silent victims. Children who witness violence. JAMA.
1993;269:262–264.
12. Fitzpatrick KM, Boldizar JP. The prevalence and
consequences of exposure to violence among AfricanAmerican youth. J Am Acad Child Adolesc Psychiatry.
1993;32:424–430.
13. Sheehan K, DiCara JA, LeBailly S, Christoffel KK.
Children’s exposure to violence in an urban setting.
Arch Pediatr Adolesc Med. 1997;151:502–504.
14. Wright RJ, Steinbach S. Violence: an unrecognized
environmental exposure that may contribute to greater
asthma morbidity in high risk inner-city populations.
Environ Health Perspect. 2001;109:1085–1089.
15. Breslau N, Davis G, Andreski P, Pertersen E. Traumatic events and posttraumatic stress disorder in an
urban population of young adults. Arch Gen Psychiatry.
1991;48:216–222.
16. Wright R, Rodriguez M, Cohen S. Review of psychosocial stress and asthma: an integrated biopsychosocial approach. Thorax. 1998;53:1066–1074.
17. Martinez P, Richters J. The NIMH community violence project, II: children’s distress symptoms associated with violence exposure. Psychiatry. 1993;56:
22–35.
18. Boney-McCoy S, Finkelhor D. Psychosocial sequelae of violent victimization in a national youth sample.
J Consult Clin Psychol. 1995;63:726–736.
This study was approved by the institutional review
boards of all participating institutions.
References
1. Wright RJ, Weiss ST. Epidemiology of allergic disease. In: Holgate S, Church M, Lichtenstein L, eds. Allergy. 2nd edition. London, England: Harcourt; 2001:
203–212.
2. Perrin J, Homer C, Berwick D, Woolf A, Freeman
J, Wennberg J. Variations in rates of hospitalization of
children in three urban communities. N Engl J Med.
1989;320:1183–1187.
3. Carr W, Zeitel L, Weiss K. Variations in asthma
hospitalizations and deaths in New York City. Am J
Public Health. 1992;82:59–65.
4. Lang D, Polansky M. Patterns of asthma mortality
in Philadelphia from 1969 to 1991. N Engl J Med.
1994;331:1542–1546.
5. Marder D, Targonski P, Orris P, Persky V, Addington W. Effect of racial and socioeconomic factors on
asthma mortality in Chicago. Chest. 1992;101(6 suppl):
426S–429S.
6. Wright RJ, Fisher EB. Putting asthma into context:
community influences on risk, behavior, and intervention. In: Kawachi I, Berkman LF, eds. Neighborhoods
April 2004, Vol 94, No. 4 | American Journal of Public Health
and Health. New York, NY: Oxford University Press;
2003:233–262.
19. Acierno R, Kilpatrick D, Rsesnick H, Saund C. Violent assault, posttraumatic stress disorder, and depression: risk factors for cigarette use among adult women.
Behav Modif. 1996;20:363–384.
20. Augustyn MS, Parker B, McAlister-Groves B,
Zuckerman B. Silent victims: children who witness violence. Contemp Pediatr. 1995;12:35–57.
21. Rauh VA, Chew GR, Gardfindel RS. Deteriorated
housing contributes to high cockroach allergen levels in
inner-city households. Environ Health Perspect. 2002;
11:323–327.
22. Rosenstreich DL, Eggleston P, Kattan M, et al.
The role of cockroach allergy and exposure to cockroach allergen in causing morbidity among inner-city
children with asthma. N Engl J Med. 1997;336:
1356–1363.
23. McLeod JD, Kessler RC. Socioeconomic status
differences in vulnerability to undesirable life events.
J Health Soc Behav. 1990;31:162–172.
24. Kessler RC. Stress, social status, and psychological
distress. J Health Soc Behav. 1979;20:259–272.
25. Crain EF, Walter M, O’Connor GT, et al. Home
and allergic characteristics of children with asthma in
Wright et al. | Peer Reviewed | Research and Practice | 631
 RESEARCH AND PRACTICE 
seven US urban communities and design of an environmental intervention: the Inner-City Asthma Study.
Environ Health Perspect. 2002;110:939–945.
44. Matthews KA, Woodall KL. Childhood origins of
overt type A behaviors and cardiovascular reactivity to
behavioral stressors. Ann Behav Med. 1988;10:71–79.
26. Wade S, Weil C, Holden G, et al. Psychosocial
characteristics of inner-city children with asthma: a description of the NCICAS psychosocial protocol. National Cooperative Inner-City Asthma Study. Pediatr
Pulmonol. 1997;24:263–276.
45. Liu D, Diorio J, Tannenbaum B, et al. Maternal
care, hippocampal glucocorticoid receptors, and hypothalamic-pituitary-adrenal response to stress. Science.
1997;277:1659–1662.
27. Thomson CC, Roberts K, Curran A, Ryan L,
Wright RJ. Caretaker-child concordance for child’s exposure to violence in a preadolescent inner-city population. Arch Pediatr Adolesc Med. 2002;156:818–823.
28. Henderson S. Neurosis and the Social Environment.
Sydney, Australia: Academic Press; 1981.
29. Cohen S, Tyrrell DAJ, Smith AP. Psychological
stress and susceptibility to the common cold. N Engl J
Med. 1991;325:606–612.
30. Steele GP, Henderson S, Duncan-Jones P. The reliability of reporting adverse experiences. Psychol Med.
1980;10:301–306.
31. Cohen S, Williamson G. Perceived stress in a
probability sample of the United States. In: Scacapan S,
Oskamp S, eds. The Social Psychology of Health. Newbury Park, Calif: Sage Publications; 1988:31–67.
32. Kleinbaum DG, Kupper LL, Muller KE, Nizam A.
Applied Regression Analysis and Other Multivariable
Methods. 3rd ed. Boston, Mass: PWS-KENT Publishing
Co; 1998.
46. Smith P, Perrin S, Yule W, Rabe-Hesketh S. War
exposure and maternal reactions in the psychological
adjustment of children from Bosnia-Hercegovina.
J Child Psychol Psychiatry. 2001;42:395–404.
47. Fong R. Violence as a barrier to compliance for
the hypertensive African-American. J Natl Med Assoc.
1995;87:203–207.
48. Robicsek R, Ribbeck B, Walker L. The cost of violence: the economy of health care delivery for non-accidental trauma in an urban southeastern community.
N C Med J. 1993;54:578–582.
49. Gortmacker S, Must A, Sobol A, Peterson K,
Colditz G, Dietz W. Television viewing as a cause of increasing obesity among children in the United States,
1986–1990. Arch Pediatr Adolesc Med. 1996;150:
356–362.
50. Camargo CJ, Field A, Colditz G, Speizer F. Body
mass index and asthma in children age 9–14. Am J
Respir Crit Care Med. 1999;159:A150.
33. Isaacs M. Violence: The Impact of Community Violence on African American Children and Families. Arlington, Va: National Center for Educational in Maternal
and Child Health; 1992.
51. Stenius-Aarniala B, Pousse T, Kvarnstrom J, Gronlund E, Ylikahri M, Mustajoki P. Immediate and long
term effects of weight reduction in obese people with
asthma: randomised controlled study. BMJ. 2000;320:
827–832.
34. Herman AA. Political violence, health, and health
services in South Africa. Am J Public Health. 1988;78:
767–768.
52. Beckham J, Roodman A, Shipley R, et al. Smoking
in Vietnam combat veterans with posttraumatic stress
disorder. J Trauma Stress. 1995;8:461–472.
35. Zapata BC, Rebolledo A, Atalah E, Newman B,
King MC. The influence of social and political violence
on the risk of pregnancy complications. Am J Public
Health. 1992;82:685–690.
53. Kennedy B, Kawachi I, Prothrow-Smith D,
Lochner K, Gupta V. Social capital, income inequality,
and firearm violent crime. Soc Sci Med. 1998;47:7–17.
36. Walters V, Charles N. “I just cope from day to
day”: unpredictability and anxiety in the lives of
women. Soc Sci Med. 1997;45:1729–1739.
37. Cohen S, Kessler R, Underwood-Gordon L. Measuring Stress: A Guide for Health and Social Scientists.
New York, NY: Oxford University Press; 1995.
38. Sternberg EM. Neural-immune interactions in
health and disease. J Clin Invest. 1997;100:
2641–2647.
39. Umetsu DT, McIntire JJ, Akbari O, Macaubas C,
DeKruyff RH. Asthma: an epidemic of dysregulated
immunity. Nat Immunol. 2002;3:715–720.
40. Baum A. Stress, intrusive imagery, and chronic
distress. Health Psychol. 1990;9:653–675.
41. Lyubomirsky S, Tucker KL, Caldwell ND, Berg K.
Why ruminators are poor problem solvers: clues from
the phenomenology of dysphoric rumination. J Pers Soc
Psychol. 1999;77:1041–1060.
42. Nolen-Hoeksema S. The role of rumination in depressive disorders and mixed anxiety/depressive symptoms. J Abnorm Psychol. 2000;109:504–511.
43. Vallee M, Mayo W, Dellu F, LeMoal M, Simon H,
Maccari S. Prenatal stress induces high anxiety and
postnatal handling induces low anxiety in adult offspring: correlation with stress-induce corticosterone secretion. J Neurosci. 1997;17:626–636.
54. Sampson R. Family management and child development: insight from social disorganization theory. In:
Cord JM, ed. Facts, Frameworks, and Forecasts: Advances
in Criminological Theory. Piscataway, NJ: Transaction
Publisher; 1992:63–93.
55. Lochner K, Kawachi I, Kennedy B. Social capital:
a guide to its measurement. Health Place. 1999;5:
259–270.
56. Sampson R, Raudenbush S, Earls F. Neighborhoods and violent crime: a multilevel study of collective efficacy. Science. 1997;277:918–924.
Local Public Health
Practice:
Trends & Models
By Glen P. Mays, PhD, MPH;
C. Arden Miller, MD; and
Paul K. Halverson, DrPH, MHSA
T
his book describes the varied spectrum of work done at the local public
health level, and how practitioners take
the lead in social justice today. The wide
array of public health department approaches, such as budgeting, staffing, services, involvement in personal health services, and their relationships with states
is disclosed.
This book is an incredible resource
for: local public health officers, administrators, and state and local health planners for use in their own local public
health practice.
ISBN 0-87553-243-8
2000 ❚ 281 pages ❚ softcover
$24.50 APHA Members
$35.00 Nonmembers
plus shipping and handling
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
LP01J7
57. Macintyre S, Ellaway A. Neighborhoods and
health: an overview. In: Kawachi I, Berkman LF, eds.
Neighborhoods and Health. New York, NY: Oxford University Press; 2003:20–42.
58. Kawachi I. Social capital and community effects
on population and individual health. Ann N Y Acad Sci.
1999;896:120–130.
59. Rutter M. Prevention of children’s psychosocial
disorders: myth and substance. Pediatrics. 1982;70:
883–894.
60. McLeod JD, Kessler RC. Socioeconomic status
differences in vulnerability to undesirable life events.
J Health Soc Behav. 1990;31:162–172.
61. Kessler RC. Stress, social status, and psychological
distress. J Health Soc Behav. 1979;20:259–272.
632 | Research and Practice | Peer Reviewed | Wright et al.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Social Disparities in Housing and Related Pediatric Injury:
A Multilevel Study
| Edmond D. Shenassa, ScD, Amy Stubbendick, MS, Mary Jean Brown, ScD, RN
It is commonly, but not universally, reported
that children residing in areas with concentrated poverty or a high concentration of minorities suffer unintentional injury at higher
rates than do other children.1–7 However,
concentrated poverty and concentration of
minorities cannot be considered causal; they
are correlates of more proximal determinants
of injury. Because young children often suffer
injuries at home,8 at least some of these more
proximal determinants are likely to be related
to housing conditions. To the extent that
housing conditions reflect conditions of the
larger residential area,9,10 they can be viewed
as community-level determinants of injury,
but more proximal community-level determinants than social conditions such as poverty.
Moreover, because housing conditions are relatively more proximal to injury than social
conditions, it is likely that the association between social conditions and injury is, at least
partially, mediated by housing conditions.
Other factors—particularly those at the individual level, such as parental supervision—are
also among the important correlates of pediatric injury. However, the focus of this study
was on housing conditions.
Age of housing is a likely determinant of
injury.8 Older houses are less likely to be in
compliance with building or sanitary codes
and may have substandard electrical and
heating systems, narrow stairwells, or other
safety hazards.11,12 Another likely housingrelated determinant of injury is whether a
house is owner occupied or rented. Inadequate or deferred maintenance can be a common problem in low-income rental properties,13,14 and high tenant turnover can increase
the number of people exposed to these hazards over time.
To date, no studies have examined the association between housing factors and injury
in which social conditions (e.g., concentrated
poverty) and individual-level determinants
(e.g., age, gender) have been simultaneously
Objectives. We conducted an ecologic analysis to determine whether housing characteristics mediate the associations between concentration of poverty and pediatric injury
and between concentration of racial minorities and pediatric injury and whether the association between housing conditions and pediatric injury is independent of other risks.
Methods. We created a hierarchical data set by linking individual-level data for pediatric injury with census data. Effect sizes were estimated with a Poisson model.
Results. After adjustment for owner occupancy and the percentage of housing built
before 1950, the association between concentration of poverty and pediatric injury was
attenuated. For concentration of racial minorities, only percentage of owner occupancy
had some mediating effect. In hierarchical models, housing characteristics remained independent and significant predictors of pediatric injury.
Conclusions. The association between community characteristics and pediatric injury
is partially mediated by housing conditions. Risk of pediatric injury associated with
housing conditions is independent of other risks. (Am J Public Health. 2004;94:633–639)
examined. We present the first multilevel,
population-based study of pediatric injury.
We also respond to calls for the study of nonfatal injuries,15–17 the most severe of which
require hospitalization. Injuries requiring hospitalization are often associated with high
treatment and rehabilitation costs18 and appear to have patterns and risks that are distinct from fatal injuries or injuries that do not
require hospitalization.16,19 As a consequence,
the study of nonfatal, hospitalized injuries
can provide information useful for their prevention. Falls and burns are among the most
prevalent causes of pediatric injury.20 In our
sample, they accounted for 59% of all hospitalized pediatric injuries; after adjustment of
hospital charges to true costs, the estimated
median cost was $4670 for burns and
$2760 for falls.21 As a consequence, these
types of injury were used to test the following
hypotheses: community-level owner occupancy and age of housing (measured at the
zip code level) mediate the association between concentration of poverty and concentration of minorities and risk for pediatric injury, and the association between pediatric
injury and community-level housing conditions is independent of individual- and other
community-level determinants of injuries requiring hospitalization.
April 2004, Vol 94, No. 4 | American Journal of Public Health
METHODS
Sources of Data
Data from 2 different sources were used in
this study. Information for all hospital discharges in the state of Illinois for 1990
through 2000 was abstracted from administrative hospital discharge data compiled as
part of an Illinois state mandate22 (as described elsewhere23), and the 1990 US census collected housing information by zip code,
including number of owner-occupied units,
number of residents living below the federal
poverty limit, number of housing units built
before 1950, and number of residents, by
race. We linked these 2 data sets by zip code
and created a hierarchical data structure with
both individual- and zip code–level data.
Diagnostic codes were based on the International Classification of Diseases, 9th Revision,24 in which E-codes reflect external
causes of injury, poisoning, or other adverse
events. An observation was defined as a fall if
1 of the principal or secondary diagnosis
codes contained an E-code between 880 and
888. An observation was defined as a burn if
it had an E-code between 890 and 899,
924.0 and 924.9, or 925.0. E-codes 925.1,
925.2, 925.8, and 925.9 were defined as
nondomestic burns and were excluded. Sixty-
Shenassa et al. | Peer Reviewed | Research and Practice | 633
 RESEARCH AND PRACTICE 
nine observations had 2 types of injury
E-codes, in which case the first recorded
E-code was used. Only children 6 years old
and younger were included in the analysis. A
total of 241 (2%) observations were excluded
because they could not be linked to the 1990
US census. Of these, 96 observations were in
zip codes created after 1990, and 29 observations were in zip codes that were either exclusively for post office boxes or for commemorative postal issues. The zip codes of
116 (1%) observations could not be identified
in the Postal Bulletin archives, and these
codes were also removed. Of the 1240 zip
codes listed for the state of Illinois in 1990,
12 had no households or no children 6 years
old or younger. Thus, 11 735 observations in
1228 zip codes remained for analysis.
Statistical Analyses
Data for this study involved counts of discharged injuries nested within a zip code. Because counts of injuries at the zip code level
are bounded at zero and approximate a binomial distribution with a large number of trials and a small probability of success, the
data assumed the shape of a Poisson distribution. Therefore, Poisson regression was used
to model the rate of injury as a function of
individual and community variables. An assumption of the Poisson distribution is that
each individual observation is an independent event. In this study, some observations
may have represented the same person or
perhaps children from the same family, thus
violating this assumption. We examined the
extent of this violation, and our analysis indicated that multiple discharges for the same
child were rare. Random samples of 1000
falls and 1000 burns were matched by birth
date (month, day, and year), gender, and insurance type. Only 2 falls and 10 burns
matched with 1 other observation on all 3
characteristics. When we used only birth
date and gender, 3 falls and 13 burns had
multiple observations. Thus, only about 1%
of the falls and burns might have been repeat
discharges of the same person. Moreover, the
low probability of falls or burns in the population further limited the likelihood that the
assumption of independence was violated.
For these reasons, we assumed that the
events were independent. Also, whereas vio-
lation of independence may have led to overestimation of injury rates, lack of independence will not bias rate ratios so long as the
extent of overestimation is comparable across
subgroups. Finally, offsets for rate calculations were based on the 1990 census population of children by age and gender.
The hierarchical structure of the data led
to correlation between counts that were
nested within zip codes. Thus, estimates of
rates and rate ratios of injuries and their associated 95% confidence intervals were calculated using Generalized Estimating Equations,
a method that allows for specification of the
Poisson distribution and accommodates correlated data.25 An exchangeable correlation
structure was assumed, and standard errors
for the parameters were based on the empirical estimator of the covariance matrix of the
estimated coefficients. Rate ratios and their
95% confidence intervals were calculated in
SAS with the GENMOD26 procedure.
Chi-squared goodness-of-fit tests27 indicated some lack of fit, so the standardized
Pearson residuals were examined, and 3 outlying zip codes that were exerting undue influence on the coefficients were removed.
Next, residual plots versus the predicted values and the covariates were examined. No
systematic biases were detected that would
indicate nonlinearity.
The distributions of concentrated poverty
and concentration of minorities were also divided into decile indicator variables, and saturated models were fit. Because plots of the
point estimates and confidence intervals
evinced a slight curvature, both of these variables were divided into tertiles. This procedure was informed by data from the 1990
US Census. In 1990, approximately 12% of
the population was African American28; this
proportion was unchanged in 2000.29 Between 1990 and 2000, the percentage of
children living in poverty declined from approximately 20% to 16%.30 Thus, the 3 categories for each of the census indicators corresponded roughly to below average, average,
and above average for concentrated poverty
and percentage African American in the
United States.
Multivariate analyses followed the study
aims. First, hierarchical models were fit to examine whether owner occupancy and age of
634 | Research and Practice | Peer Reviewed | Shenassa et al.
housing mediated the association between
concentration of poverty or concentration of
minorities and pediatric injury. For example,
the mediating effect of owner occupancy was
examined by entering this variable in a model
that, in addition to the individual-level variables, included either zip code–level concentrated poverty or concentrated minority. A
significant change in the coefficient for concentrated poverty or concentrated minority
indicated that its association with injury
was mediated by owner occupancy (see
D’agostino31 for the significance test). Multivariate hierarchical models also were developed both with and without interactions.
RESULTS
In the state of Illinois, from January 1,
1990, through September 30, 2000, there
were 11 735 hospital discharges of children
with nonfatal injuries coded as occurring at
home. The annual incidence rates for the 2
most prevalent types of injury, falls (43.5%)
and burns (15.2%), were 3.93/10 000 population aged 6 years or younger (95% confidence interval [CI] = 3.83, 4.04) and 1.37/
10 000 population aged 6 years or younger
(95% CI = 1.31, 1.44), respectively. Fifty-eight
percent of burns occurred among children
aged 1 through 2 years, whereas falls were
fairly evenly distributed across age groups.
The median length of hospital stay for burns
and falls was 5 and 2 days, respectively.
Bivariate analyses of individual-level variables indicated that compared with children
aged 5 through 6 years, infants (< 1 year)
were more likely to suffer an injury resulting
from a fall, and toddlers (aged 1 through 2
years) were most likely to suffer a burn.
Males were at a higher risk for both types of
injury. Bivariate analyses of the zip code–level
variables for tenancy and age of housing
demonstrated the change in risk for falling or
being burned following a 10% increase in the
continuous explanatory variable. For every
10% increase in the proportion of owneroccupied units, risk for falling decreased by
16% and risk for being burned decreased by
27%. A 10% increase in the proportion of
housing built before 1950 was associated
with a 17% increase in risk for falling and a
34% increase in risk for being burned. Chil-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 1—Hierarchical Fall and Burn Models
(95% CIa)
Burn Rate Ratio
(95% CIa)
2.13
1.11
0.97
1.00
(1.94, 2.34)
(1.01, 1.23)
(0.88, 1.07)
...
8.16
8.28
2.02
1.00
(6.77, 9.83)
(6.92, 9.92)
(1.64, 2.49)
...
1.45
1.00
(1.35, 1.55)
...
1.32
1.00
(1.18, 1.48)
...
0.94
1.10
(0.90, 0.99)
(1.06, 1.15)
0.92
1.11
(0.84, 1.00)
(1.04, 1.18)
1.05
1.02
1.00
(0.86, 1.27)
(0.83, 1.24)
...
2.10
1.79
1.00
(1.56, 2.83)
(1.36, 2.36)
...
1.92
1.43
1.00
(1.55, 2.36)
(1.17, 1.74)
...
2.64
1.24
1.00
(1.84, 3.79)
(0.88, 1.74)
...
Fall Rate Ratio
Individual
Age, y
<1
1–2
3–4
5–6 (reference group)
Gender
Male
Female (reference group)
Zip Code
Percentage owner-occupied housingb
Percentage housing built before 1950b
Concentrated poverty
High
Middle
Low (reference group)
African American population
High
Middle
Low (reference group)
a
Wald confidence interval.
Rate ratios per 10% increase in the value of the independent variable.
b
dren residing in zip codes with the highest
concentrations of poverty, compared with residents of zip codes with the lowest concentrations of poverty, were significantly more likely
to sustain a fall or burn. Similarly, children residing in zip codes with the highest concentrations of minorities were significantly more
likely to sustain a fall or burn than residents
of zip codes with the lowest concentrations of
minorities.
Next we examined the mediating effect of
housing conditions on risk for injury associated with concentrated poverty and a concentration of minorities (Figures 1a–2b). The association between concentrated poverty and
risk for falling or being burned was mediated
by both owner occupancy and age of housing; the mediation was considerably larger
and significant for the top tertile. When both
housing conditions were included in the
model for falls, the risk ratio for areas with a
medium concentration of poverty became insignificant, and the risk ratio for areas with a
high concentration of poverty remained only
marginally significant. For burns, the risk ra-
tios for middle and high levels of poverty
were reduced but remained significant when
both housing conditions were included in the
model.
For concentrated minority populations,
only owner occupancy appeared to be a mediator. Again, the mediation was larger and
significant for the top tertile. Inclusion of age
of housing actually increased the risk ratio for
the middle tertile and only slightly reduced
the risk ratio for the top tertile. The full
model for falls (i.e., including both housing
conditions) rendered significant the associations for areas with both medium and high
minority concentration. The full model for
burns evinced a significant association for
areas with a high minority concentration, but
the risk ratio for areas with a medium concentration became insignificant.
In the final hierarchical models (Table 1),
both owner occupancy and age of housing remained significant predictors of both types of
injury. We also found that owner occupancy
modified risk for injury associated with concentration of poverty and minority concentra-
April 2004, Vol 94, No. 4 | American Journal of Public Health
tion. In particular, owner occupancy was
more protective in areas with higher concentrations of poverty and was less protective in
areas with higher minority concentration and
old housing.
DISCUSSION
Our findings indicate that the communitylevel concentration of owner-occupied housing and age of housing are significantly associated with rates of nonfatal hospitalized
pediatric injury. This is in line with earlier
findings that a variety of negative health outcomes, including fatalities caused by house
fires, all-cause mortality, and elevated lead
blood levels, are significantly more prevalent
among renters than among homeowners.1,2,32,33 This study further illuminates the
association between housing conditions and
health in 2 important ways. First, our results
indicate that housing conditions mediate the
association between community characteristics, such as concentrated poverty, and pediatric injury. Second, the results of our hierarchical analyses demonstrate that the
association between housing conditions and
pediatric injury is independent of both individual- and other community-level determinants of injury.
However, the association between these
community characteristics and injury does
remain significant after control for housing
conditions. That the association between
concentrated poverty and injury is only partially mediated by housing conditions may
be in part the result of the fact that during
most of the period covered by this study,
some of the zip codes with concentrated
poverty also had a concentration of federally
owned or subsidized housing, which must
meet minimum maintenance standards. Partial mediation of the association between
concentration of minorities and injury is consistent with the findings of previous studies
of racial disparities in health. These studies
found that a significant proportion of healthrelated racial disparities remain unexplained
even after control for social factors, probably
because of the myriad of pathways by which
race relations can influence health and because of the difficulties in capturing all of
these influences in 1 study.34–39
Shenassa et al. | Peer Reviewed | Research and Practice | 635
 RESEARCH AND PRACTICE 
3.00
a
High concentration of poverty
Medium concentration of poverty
2.50
2.00
Rate Ratio
2.00
1.50
1.40
1.46 a
1.43 a
1.20
1.20 a
1.01a
1.00 a
1.00
0.50
0.00
Crude poverty
Poverty + old housing
Poverty + owner
occupancy
Poverty +
old housing and
owner occupancy
Models
9.00
b
High concentration of poverty
Medium concentration of poverty
8.00
7.00
6.44
Rate Ratio
6.00
5.00
4.20 a
3.91a
4.00
3.00
3.39 a
2.84
2.40
2.02
2.02
2.00
1.00
0.00
Crude poverty
Poverty + old housing
Poverty + owner
occupancy
Poverty +
old housing +
owner occupancy
Models
a
The estimate is significantly different from that of the crude model.
FIGURE 1—Old housing and owner occupancy as mediators of risk of (a) falls and (b) burns
associated with poverty.
The implications of these findings should
be considered in light of the study’s limitations and strengths. A common shortcoming
of ecologic studies such as this is the use of
broadly defined and heterogeneous geographic areas (zip codes in our case) as the
unit of analysis.40,41 This misspecification can
dilute the effect of interest and result in underestimation of the true effects. This misspecification also can compromise the accu-
racy by which community-level variables
allow measurement of the underlying construct of interest. Given the heterogeneity of
zip codes, underestimation of the relative
risks and a degree of residual confounding
are both possibilities in this study.
The main threat to validity of any ecological study is ecological fallacy, and we addressed this issue both methodologically and
logically. In addition to controlling for key
636 | Research and Practice | Peer Reviewed | Shenassa et al.
confounders, we limited our inferences at the
ecological level (i.e., inferences regard group
rates and not individuals’ risk).42,43 As outlined below, the validity of our conclusions
are bolstered by their logical plausibility, an
important criterion of validity for any study.44
Arguably the most prominent source of ecological fallacy that remains unaccounted for in
this study is parental child supervision and its
correlates such as race/ethnicity and income.
Clearly the link between poor supervision of
children and injury (e.g., placing high chairs
close to windows without screens) can masquerade as an effect of housing conditions. (It
can be logically argued that given the same
degree of parental supervision, children residing in well-maintained homes would be less
likely to suffer injuries.) It is also reasonable to
question whether housing conditions are
themselves proxies for poverty and compromised ability to supervise children or are proxies for racial/ethnic norms about parenting.
In regard to poverty, although several
developmental models evince a negative correlation between income and the quality of
parenting,45,46 material deprivation is not uniformly linked with poor child health outcomes
or parenting practices. For example, lowincome Hispanic women can have better
pregnancy outcomes than wealthier residents
of the same area, a paradox that may be attributable to social cohesion and the presence
of extended family members among this population.47–49 We suggest that social cohesion
and the presence of extended family members
also can promote better child supervision.
In regard to racial/ethnic differences in
parenting, the literature generally indicates
that correlates of poor parenting skills and supervision are not differentially distributed by
race/ethnicity. Although certain parenting
practices are more prevalent among various
racial/ethnic groups, these differences are not
necessarily correlated with poor developmental outcomes.50 Moreover, the influence of
race/ethnicity is often confounded by socioeconomic status and recency of immigration.51,52 However, it appears that the effect of
socioeconomic status is stronger and absorbs
that of race/ethnicity.53,54
Another persuasive piece of evidence for
the independent contribution of housing conditions to risk for pediatric injury is the signif-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
3.50
a
High minority concentration
Medium minority concentration
3.00
2.61
2.50
Rate Ratio
2.28
1.94 a
2.00
1.77 a
1.55
1.50
1.42
1.39
1.24
1.00
0.50
0.00
Crude minority
concentration
Minority concentration + Minority concentration +
old housing
owner occupancy
Models
9.00
Minority
concentration +
old housing +
owner occupancy
b
High minority concentration
8.00
Medium minority concentration
7.00
Rate Ratio
6.00
5.58
5.00
4.36
4.00
2.94 a
3.00
2.00
1.25
1.41
1.01
1.00
0.00
Crude minority
concentration
Minority concentration + Minority concentration +
old housing
owner occupancy
Models
a
3.16a
1.18
Minority
concentration +
old housing +
owner occupancy
The estimate is significantly different from that of the crude model.
FIGURE 2—Old housing and owner occupancy as mediators of risk of (a) falls and (b) burns
associated with minority concentration.
icant and sustained reduction in pediatric falls
and burns that is attributed to the installation
of window guards and sprinklers.55,56 These
passive measures address building defects
and not parenting skills. In these cases, even
when a high chair is placed next to a window,
the window guard will prevent a fall regardless of the parents’ income or skills. Thus, we
suggest that the reported mediation by hous-
ing conditions is not simply a reflection of
ecological fallacy. Future studies could be
more persuasive by adjusting for more individual-level correlates of injury.
Finally, because our results are from
racially segregated geographic areas, the
generalizability of our findings may be limited to such areas. However, racial segregation is unfortunately a defining characteristic
April 2004, Vol 94, No. 4 | American Journal of Public Health
of many US cities.35,39 Subsequently, our
findings are likely to be generalizable to a
broad cross-section of residential areas in the
United States.
A strength of this study is our use of both
individual- and community-level data in a
hierarchical design. To our knowledge, this is
the first study of injury that has used such a
design. Another strength is our analysis of incident cases of acute health outcomes. This
feature discounts 2 competing explanations
that often limit inference regarding the effect
of place on health. First is the argument that
the health outcomes attributed to a place
may be simply long-term consequences of
individual-level characteristics.57 Although
this argument can be persuasive in the case
of prevalent chronic illnesses, the argument
that acute events such as injury among children age 6 years or younger are long-term
consequences of individual-level characteristics is far less compelling. The second competing explanation, “social drift,”57 suggests
that those in poor health, through loss of economic status, “drift” to poorer residential
areas. In this study, by examining incident
cases among young children, we have rendered social drift an unlikely explanation for
the disparate concentration of injury in poor
and wealthy areas.
The association between housing conditions and pediatric injury has both immediate
and long-term implications. The immediate
implications regard the efficacy of intervention and prevention programs.58 These programs can increase their efficiency and efficacy by considering both housing and
socioeconomic characteristics of the community. Although not the focus of this study, we
did find evidence of some interesting interactions. For example, we found that owner occupancy is less protective in areas with a high
proportion of old housing. We also found
owner occupancy to be more protective in
areas with a higher concentration of poverty.
This indicates that where a high proportion of
homes were built before 1950, interventions
should target old housing, whereas in highpoverty areas, the first priority should be
non–owner-occupied homes.
The long-term implications are predicated
on replication of these findings in longitudinal
studies that consider more individual-level
Shenassa et al. | Peer Reviewed | Research and Practice | 637
 RESEARCH AND PRACTICE 
variables, including parenting practices, conducted with smaller units of analysis. If such
studies, as we suspect, further support our
conclusions, then the long-term implications
pertain to remediation of social disparities in
health through remediation of differences in
housing. Housing is an indicator (at times
even a leading indicator) of conditions prevalent in the larger residential area. For example, age of housing reflects neighborhood economic conditions and can foretell future
economic patterns. In 1990, housing units in
high-poverty US census tracts, compared with
low-poverty census tracts, were on average
11 years older and had correspondingly
lower market value.9 The units’ low market
values are a disincentive for maintaining
them and a harbinger of worse physical conditions and further reductions in value. Concentration of units with low market value can
initiate a cycle of private and institutional divestments, such as “redlining” by financial institutions, which can lead to the deterioration
of neighborhoods’ social, financial, and physical resources.9,10 In turn, poor housing conditions have important and wide-ranging health
implications.9,33,59–61
Vigorous application of already existing
programs62 to improve housing conditions
may have benefits beyond the immediate
residences and their occupants. This study
indicates that social disparities in health
may be addressed, at least partially, through
remediation of social disparities in housing—
a remediation that also can benefit future
generations.63,64
About the Authors
Edmond D. Shenassa is with the Department of Community Health and the Centers for Behavioral and Preventive
Medicine, Brown Medical School/Miriam Hospital, Providence, RI. Amy Stubbendick is with the Department of Biostatistics, Harvard School of Public Health, Boston, Mass.
Mary Jean Brown is with the Department of Maternal and
Child Health, Harvard School of Public Health.
Requests for reprints should be sent to Edmond D.
Shenassa, ScD, Brown Medical School, Department of
Community Health and Centers for Behavioral and Preventive Medicine, One Hoppin Street, Suite 500, Providence, RI 02903 (e-mail: [email protected]).
This article was accepted March 21, 2003.
Contributors
E. D. Shenassa contributed to the conceptual development of the study, the data analysis plan, and the writing of the article. A. Stubbendick contributed to the de-
velopment of the data analysis plan, the analysis of
data, and the Methods and Results sections of the article. M. J. Brown contributed to the conceptual development of the study, the data analysis plan, and the writing of the article.
Acknowledgments
Supported in part by grants from National Center for
Healthy Housing (M.J. Brown and E.D. Shenassa) and the
Harvard Injury Control Research Center (E.D. Shenassa).
We thank Drs Kimberly Lochner and Stephen
Gilman for their insightful comments on an earlier version of the manuscript, and Meredith Eastman for her
invaluable assistance with data management.
Human Participant Protection
This study was exempt from human subject review because the secondary data used for the analysis did not
contain personal data.
References
1. Baker S, O’Neill B, Ginsburg M, Guohua L. The
Injury Fact Book. 2nd ed. New York, NY: Oxford University Press; 1991.
2. Mallone S, Istre G, Rosenberg M, et al. Surveillance and prevention of residential-fire injuries. N Engl
J Med. 1996;335(1):27–31.
3. Istre G, McCoy M, Osborn L, Barnard J, Bolton A.
Deaths and injuries from house fires. N Engl J Med.
2001;344:1911–1916.
4. Office of Population Censuses and Surveys. Occupational Mortality: Childhood Supplement. London:
HMSO; 1988.
epidemiology of child and adolescent injuries. Future
Child. 2000;10(1):23–52.
16. Runyan C, Gerken E. Epidemiology and prevention of adolescent injury: A review and research
agenda. JAMA. 1989;262:2273–2279.
17. Peel NM, Kassulke DJ, McClure RJ. Population
based study of hospitalised fall related injuries in older
people. Inj Prev. 2002;8:280–283.
18. Centers for Disease Control and Prevention. Cost
of injury – United States: a report to Congress, 1989.
MMWR Morb Mortal Wkly Rep. 1989;38(43):
743–746.
19. Pickett W, Harting L, Brison R. A populationbased study of hospitalized injuries in Kingston, Ontario, identified via the Canadian Hospitals injury reporting and prevention program. Chronic Dis Can.
1997;18:61–69.
20. Rivara F. Pediatric injury control in 1999. Pediatrics. 1999;103(4):883–888.
21. Haddis AC, Teutsch SM, Schaffer PA, and Dunet
DO, eds. Prevention Effectiveness: A Guide to Decision
Analysis and Economic Evaluation. London: Oxford University Press; 1996.
22. Illinois Health Financing Reform Act. Public Act
83–1243.
23. Shenassa E, Catlin S, Buka S. Lethality of firearms
relative to other suicide methods: a population based
study. J Epidemiol Community Health. 2003;57:
120–124.
24. International Classification of Diseases, 9th Revision. Geneva, Switzerland: World Health Organization;
1980.
25. Liang K, Zeger, S. Longitudinal data analysis using
generalized linear models. Biometrika. 1986;73:13–22.
5. Hippisley-Cox J, Groom L, Kendrick D, Coupland
C, Webber E. Cross sectional survey of socioeconomic
variations in severity and mechanism of childhood injuries in Trent 1992–7. BMJ. 2002;324:132–1138.
26. SAS/STAT User’s Guide, Version 8. Cary, NC: SAS
Institute Inc; 1999.
6. Mierley M, Baker S. Fatal house fires in an urban
population. JAMA. 1983;249(11):1466–1468.
27. Agresti A. An Introduction to Categorical Data
Analysis. New York, NY: Wiley; 1996.
7. Lyons R, Delahunty A, Heaven M, McCabe M,
Howard A, Nash P. Incidence of childhood fractures.
BMJ. 2000;320:149.
28. Shea M. Dynamics of Economic Well-Being: Poverty, 1990 to 1992. US Bureau of Census Current Population Reports P70–42. Washington, DC: US Government Printing Office; 1995.
8. Macfarland A. Child deaths from accidents: place
of accident. Popul Trends. 1979;15:10–15.
9. Jargowsky P. Poverty and Place: Ghettos, Barrios
and the American City. New York, NY: Russell Sage
Foundation; 1996.
10. Wilson W. When Work Disappears: The World of
the New Urban Poor. New York, NY: Vintage Books;
1997.
11. Runyan C, Bangdiwala S, Linzer M, Sacks J, Butts
J. Risk factors for fatal residential fires. N Engl J Med.
1992;327(12):859–863.
12. Ashley DL, Menkedick JR, Wooton JV. Healthy
Home Issue: Injury. External Review Draft, Version 2.
Washington, DC: US Department of Housing; 2002.
13. Cummins S, Jackson R. The built environment
and children’s health. Pediatr Clin North Am. 2001;
48(2):1241–1256.
14. Ian R. Deaths from injuries in house fires: Fanning
the flames of child health advocacy? BMJ. 1995;311:
1381–1382
15. Grossman D. The history of injury control and the
638 | Research and Practice | Peer Reviewed | Shenassa et al.
29. Census 2000: General Population Characteristics.
Washington, DC: US Bureau of the Census; 2002.
Available at: http://www.census.gov. Accessed November 21, 2002.
30. Dalaker J. Poverty in the United States: 2000. US
Census Bureau Current Population Reports, Series
P60–214. Washington, DC: Government Printing Office; 2001.
31. D’Agostino RB Sr, Grundy S, Sullivan LM, Wilson
P. Validation of Framingham coronary heart disease
prediction score: results of a multiple ethnic groups investigation. JAMA. 2001;286:180–187.
32. Sargent J, Brown MJ, Freeman J, et al. Childhood
lead poisoning in Massachusetts communities: its association with sociodemographic and housing characteristics. Am J Public Health. 1995;85:528–534.
33. Filakti H, Fox J. Differences in mortality by housing tenure and by car access from the OPCS Longitudinal Study. Popul Trends. 1995;81:27–30.
34. Otten MW, Teutsch SM, Williamson DF, Marks
JM. The effects of known risk factors on the excess
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
mortality of black adults in the United States. JAMA.
1990;263(6):845–850.
35. Massey D, Denton N. American Apartheid: Segregation and the Making of the Underclass. Cambridge,
Mass: Harvard University Press; 1996.
36. Jackson S, Anderson R, Johnson N, Sorlie P. The
relation of residential segregation to all cause mortality:
a study in black and white. Am J Pub Health. 2000(4);
90:615–616.
37. Massey DS, Gross AB, Eggers ML. Segregation,
the concentration of poverty and the life chances of individuals. Soc Sci Res. 1991;20:397–420.
38. Lanphear BP, Weitzman M, Eberly S. Racial differences in urban children’s environmental exposure to
lead. Am J Public Health. 1996;86:1460–1463.
39. Polednak AP. Segregation, Poverty, and Mortality in
Urban African Americans. New York, NY: Oxford University Press; 1997.
40. Krieger N. Use of census-based aggregate variables to proxy for socioeconomic group: evidence from
national samples. Am J Epidemiol. 1999;150(8):892–
896.
41. Pickett K, Pearl M. Multilevel analyses of neighborhood socioeconomic context and health outcomes:
a critical review. J Epidemiol Community Health. 2001;
55:111–122.
42. Susser M. The logic in ecological: II. The logic of
design. Am J Public Health. 1994;84(5):830–835.
43. Morgenstern H. Uses of ecologic analysis in epidemiologic research. Am J Public Health. 1982;72:1336–
1344.
44. Cohen B. In defense of ecologic studies for testing
a linear no threshold theory. Am J Epidemiol. 1994;
139:765–768.
45. Brody GH, Flor DL. Maternal resources, parenting
practices, and child competence in rural, single-parent
African American families. Child Dev. 1998;69:803–
816.
46. McLoyd VC. The impact of economic hardship on
African American families and children: psychological
distress, parenting, and socioemotional development.
Child Dev. 1990;61:311–346.
47. Balcazar H, Aoyama C, Xi C. Interpretative views
on Hispanics’ prenatal problems of low birth weight
and prenatal care. Public Health Rep. 1991;106:420–
426.
52. García Coll C, Magnusson K. Cultural influences
on child development: are we ready for a paradigm
shift? In: Matsen AS, ed. Cultural Processes in Child Development. The Minnesota Symposia on Child Psychology.
Vol. 29. Mahwah, NJ: Erlbaum; 1999:1–24.
Third
Edition
53. García Coll C, Meyer E, Britten S. Ethnic and minority parenting. In: Bornstein MH, ed. Handbook of
Parenting. Vol. 2. Biology and Ecology of Parenting.
Mahwah, NJ: Erlbaum; 1995:189–210.
54. Duncan GJ, Brooks-Gunn J. Consequences of Growing Up Poor. New York, NY: Russell Sage Foundation;
1997.
55. Bradley RH, Corwyn RF, McAdoo HP, García Coll
C. The home environments of children in the United
States, Part I: variations by age, ethnicity, and poverty
status. Child Dev. 2001;72(6):1844–1867.
56. Spiegel CN, Lindaman FC. Children can’t fly: a
program to prevent childhood morbidity and mortality
from window falls. Am J Public Health. 1977;67:1143–
1147.
57. Diez Roux AV. Invited commentary: places, people, and health. Am J Epidemiol. 2002;155(6):516–519.
58. Shenassa E. Delivering the goods: the relevance of
classification accuracy to the design of community intervention and prevention programs. J Community Psychol. 2002;30:197–210.
59. Evans G, Chan H-Y, Wells N, Saltzman H. Housing quality and mental health. J Consult Clin Psychol.
2000;68(3):526–530.
60. Calsyn R, Winter J. Social support, psychiatric
symptoms and housing: a causal analysis. J Community
Psychol. 2002;30(3):247–259.
61. Clark C, Bornschein P, Succop S, Que Hee S,
Hammond P, Peace B. Condition and type of housing
as an indicator of potential environmental lead exposure and pediatric blood lead level. Environ Res. 1985;
38:46–52.
NEW!
Standards for Health
Services in Correctional
Institutions
Developed by APHA Task Force on
Correctional Health Care Standards
T
he long-awaited third edition of
Standards for Health Services in
Correctional Institutions is now available.
The third edition defines the scope of
services that are necessary to provide adequate care, basing these standards upon
principles of public health and constitutional standards developed through litigation. The book has been cited as the
standard for jail and prison health services in state and federal court decisions.
The new edition includes significant
changes including expansion of both the
mental health section, and children and
adolescents section.
The new Standards is easy to use and
the most comprehensive and inclusive set
of standards for health services in correctional institutions. It is an essential reference for anyone in the field of corrections.
62. Brown MJ, Gardner J, Sargent J, Swartz K, Hu H,
Timperi R. The effectiveness of housing policies in reducing children’s lead exposure. Am J Public Health.
2001;91(4):621–624.
ISBN 0-87553-029-X
Softcover ❚ 182 pages
$26.50 APHA Members ❚ $37.95 Nonmembers
63. Rose G. Sick individuals and sick populations. Int J
Epidemiol. 1985;14:32–38.
American Public Health Association
64. Shenassa E. Society, physical health and modern
epidemiology. Epidemiology. 2001;12:467–470.
Plus shipping and handling
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
PRS03J5
48. Gorman B. Racial and ethnic variation in low
birthweight in the United States: individual and contextual determinants. Health Place. 1999;5(3):195–207.
49. Palloni A, Morenoff JD. Interpreting the paradoxical in the Hispanic paradox: demographic and epidemiologic approaches. Ann N Y Acad Sci. 2001;954:
140–174.
50. Ogbu JU. A cultural ecology of competence
among inner-city blacks. In: Spencer ME, Brookins GK,
Allen WR, eds. Beginnings: The Social and Affective Development of Black Children. Hillsdale, NJ: Erlbaum;
1985:45–66.
51. García Coll C, Lamberty G, Jenkins R, et al. An integrative model for the study of developmental competencies in minority children. Child Dev. 1996;67:
1891–1914.
April 2004, Vol 94, No. 4 | American Journal of Public Health
Shenassa et al. | Peer Reviewed | Research and Practice | 639
 RESEARCH AND PRACTICE 
Parental Social Determinants of Risk for Intentional Injury:
A Cross-Sectional Study of Swedish Adolescents
| Karin Engström, PhD, MPH, Finn Diderichsen, MD, PhD, and Lucie Laflamme, PhD
The contribution of intentional injury to the
overall burden of trauma, mortality, and morbidity increases substantially during adolescence. In Sweden, for example, the incidence
of self-inflicted injury among girls aged
15–19 years is close to that of traffic-related
injury.1 Additionally, strong inverse relationships have been found between household
(parental) socioeconomic status (SES) and injury risk among adolescents2–10—in particular,
interpersonal violence–related injury2,3,10and
self-inflicted injury.4,6
Apart from household SES, few household
social and economic characteristics have been
documented in relation to intentional injuries
during adolescence. However, it seems that living in a single-parent home11–13 and being in a
family that receives welfare benefits14 both
have an impact on risk level. By contrast, the
effect of parental country of birth is unclear.3
Furthermore, population density has been
found to be associated with injury caused by
interpersonal violence among young people.15
The fact that studies have not commonly
considered several family-related social characteristics simultaneously limits our understanding of the true effect of any particular
family social circumstance on injury risk during adolescence.2 A recent Swedish national
study revealed that among adolescents, the
combination of living in a single-parent home,
receiving welfare benefits, and not having a
parent born in Sweden reduces the association of household SES with risk of intentional
injury but not with risk of traffic-related injury.1 Yet, the manner in which these characteristics operate separately and in association
with one another remains unclear.
Our study investigated this question more
closely. We considered family-related social
attributes in conjunction with population density, and we measured the individual and the
combined effects of these factors on risk of
self-inflicted and interpersonal violence–
related injury among Swedish adolescents.
Objectives. We investigated the effect of family social and economic circumstances
on intentional injury among adolescents.
Methods. We conducted a cross-sectional register study of youths aged 10 to 19 years
who lived in Sweden between 1990 and 1994. We used socioeconomic status, number
of parents in the household (1- or 2-parent home), receipt of welfare benefits, parental country of birth, and population density as exposures and compiled relative risks and populationattributable risks (PARs) for self-inflicted and interpersonal violence–related injury.
Results. For both genders and for both injury types, receipt of welfare benefits showed
the largest crude and net relative risks and the highest PARs. The socioeconomic
status–related PAR for self-inflicted injury and the PAR related to number of parents in
the household for interpersonal violence–related injury also were high.
Conclusions. Intentional-injury prevention and victim treatment need to be tailored to
household social circumstances. (Am J Public Health. 2004;94:640–645)
METHODS
Creation of the Data Set
A cross-sectional design was employed
and the study was based on a data set we
created by linking records from 14 Swedish
national registers (the population register, 2
censuses, 5 annual income registers, 5 annual hospital discharge registers, and the
causes-of-death register). The study population consisted of all adolescents aged 10 to
19 years who lived in Sweden at some point
between 1990 and 1994. Subjects were
identified through the Swedish National
Population Register, and gender and age
were established through the national census of 1990.
Adolescents were linked to parents using
the national censuses of 1985 and 1990 to
document parental social and economic attributes and population density. Subjects were
matched with the adult or adults they lived
with (including biological and nonbiological
parents); those subjects who could not be
linked to a parent and whose parents did not
reside in Sweden at the time of the 1990 census were excluded (about 5.2%). All household information was taken from the 1990
census, with the exception of information on
receipt of welfare benefits, which was extracted from the annual income registers.
640 | Research and Practice | Peer Reviewed | Engström et al.
Adolescent SES was determined on the
basis of the highest parental SES in the household in accordance with the dominance principle.16 The Swedish socioeconomic classification provides a measure of class on the basis
of occupation.17,18 It divides individuals in the
labor force into self-employed and employed.
The former group is further divided into
farmers and other self-employed persons; the
latter group is divided into manual workers
and nonmanual workers, who in turn are subgrouped according to the average educational
level required for any particular occupation.
In our study, all adolescents were allocated to
1 of 6 categories of household SES: high/
intermediate-level nonmanual workers, assistant nonmanual workers, skilled manual
workers, unskilled manual workers, selfemployed persons (farmers and other selfemployed), and other (students, housewives,
persons living on early-retirement pensions,
and the long-term unemployed).
With regard to number of parents in the
household (1- or 2-parent home), we assigned
the single-parent home characteristic to adolescents who were living with a parent who was
not cohabiting with another adult. During the
study period, approximately 15% to 20% of
all children in Sweden lived with a single parent, and approximately 30% to 50% of these
children were born to single-parent families.19
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
A household was regarded as having received welfare benefits if anyone in the household received benefits at least once during
the study period.
We also included parental country of birth
to assess whether subjects had at least 1
Swedish-born parent. An adolescent was considered to have a parent born in Sweden if 1
or both parents were born in the country.
Population density was calculated as the
number of inhabitants (in 1990) within a
30-kilometer radius of the most heavily populated district within a particular municipality.
For our study, 2 categories were created:
high–population density areas (i.e., the 3 main
Swedish urban areas of Stockholm/Södertälje,
Gothenburg, and Malmö/Lund/Trelleborg),
and low–population density areas (i.e., the
rest of Sweden).
This material was then linked to the annual
National Hospital Discharge registers for the
years 1990 to 1994 and to the national
causes-of-death registry for the years 1991 to
1994. Non-fatal (but requiring 1 or more
nights of hospitalization) and fatal (2.6%) intentional injuries were examined together. We
avoided double counting of subjects in both
types of registers by excluding from the hospital discharge registers any subject who had
the same diagnosis in both register data sets
within 2 months. Coverage of the hospital discharge registers was estimated to be nearly
complete; however, about 4.5% of subjects
either lacked information on the external
cause of injury or had no personal identification number.20
In accordance with the International Classification of Diseases, 9th Revision (ICD-9),21,22
injuries were divided into 2 categories: interpersonal violence–related injury (E960–
E969) and self-inflicted injury (E950–E959).
Person-years were compiled as follows: subjects who lived in Sweden a whole year
contributed 1 person-year; those who
moved from Sweden, or who were born or
died, contributed 1 half year for that year.
Person-years (denominator in the relativerisk calculations) and injuries (numerator)
were summed for the 5-year study period
(1990–1994). Table 1 shows the distribution of person-years across categories of social characteristics. It also shows injury incidences per 100 000 person-years across
TABLE 1—Injury Incidence per 100 000 Person-Years, by Selected Household
Characteristics: Sweden, 1990–1994a
Injury Incidence per 100 000
Person-Years
Person-Years
Characteristic
Household socioeconomic status
High/intermediate-level nonmanual
workers
Assistant nonmanual workers
Skilled manual workers
Unskilled manual workers
Self-employed persons
Other
No. parents in household
2-parent home
1-parent home
Receipt of welfare benefits
No
Once or more
Parental country of birth
1 or both parents born in Sweden
No parent born in Sweden
Population density
Low
High
Total
Girls
Self-Inflicted
Boys
Interpersonal
Violence–Related
Girls
Boys
Girls
Boys
843 705
885 876
111
26
9
57
275 404
282 805
292 772
154 774
13 853
285 019
298 899
305 868
164 411
146 576
129
145
210
133
288
30
30
44
34
74
13
17
27
14
41
74
92
99
69
150
1 553 839
33 415
1 739 833
346 816
121
277
28
62
13
31
68
128
1 745 102
242 887
1 830 954
255 694
110
414
26
91
11
53
64
176
1 809 384
178 606
1 894 430
192 218
137
251
32
48
15
27
73
128
1 544 749
443 241
162 097
465 678
140
170
33
38
15
19
67
116
1 987 990
2 086 648
147
34
16
78
a
Fatal injuries were measured for 1991 to 1994 only; the proportion of fatal injuries (2.6%) and the injury incidence are
therefore somewhat underestimated.
social characteristics by diagnostic group and
by gender. As expected, boys experienced
more interpersonal-violence injuries and
girls experienced more self-inflicted injuries.1,23 For boys and girls together, selfinflicted injuries outnumbered injuries
caused by interpersonal violence. Additionally, a large proportion of the interpersonal
violence–related injuries were likely to have
been perpetrated by strangers or acquaintances (i.e., nonfamily members.)10
Data Analysis
We performed all analyses separately for
boys and for girls, and we controlled for age
category (10–14 years and 15–19 years) in
all instances. Relative risks (RRs) with 95%
confidence intervals (CIs) were computed for
each social characteristic independently.
High/intermediate-level nonmanual workers,
April 2004, Vol 94, No. 4 | American Journal of Public Health
2-parent home, not having received welfare
benefits, having at least 1 parent born in Sweden, and living in a low–population density
area were used as reference categories.
We then performed multivariate regression
analyses, with all social characteristics in a
single model, to establish the importance of
each measure when we controlled for the
others. Population density was included only
when RRs had been found to be significant in
the former set of analyses. Logistic regression
was used to compute the RRs.
Finally, population-attributable risks were
calculated to assess the reduction (percentage)
in injury risk that would be achieved assuming all groups on 1 variable had the same risk
level as the group with the lowest risk level.24,25
For our study, population-attributable risks
were calculated with the RRs from the multivariate regressions.
Engström et al. | Peer Reviewed | Research and Practice | 641
 RESEARCH AND PRACTICE 
TABLE 2—Relative Risk (RR) for Intentional Injury, by Selected Household Characteristics,
Adjusted for Age: Sweden, 1990–1994
Self-Inflicted Injury, RR (95% CI)
Characteristic
Girls
Household socioeconomic status
High/intermediate-level
1.00
nonmanual workers
Assistant nonmanual workers
1.16 (1.02, 1.31)
Skilled manual workers
1.32 (1.18, 1.48)
Unskilled manual workers
1.90 (1.72, 2.10)
Self-employed persons
1.19 (1.02, 1.38)
Other
2.69 (2.39, 3.02)
No. of injuries
2917
No. parents in household
2-parent home
1.00
1-parent home
2.31 (2.13, 2.50)
No. of injuries
2805
Receipt of welfare benefits
No
1.00
Once or more
3.98 (3.68, 4.30)
No. of injuries
2917
Parental country of birth
1 or both parents born
1.00
in Sweden
No parent born in Sweden
1.92 (1.74, 2.12)
No. of injuries
2917
Population density
Low
1.00
High
1.22 (1.12, 1.32)
No. of injuries
2917
Interpersonal Violence–Related
Injury, RR (95% CI)
Boys
1.00
Girls
1.00
Boys
1.00
1.13 (0.88, 1.45) 1.39 (0.94, 2.06)
1.17 (0.92, 1.49) 1.79 (1.24, 2.56)
1.69 (1.37, 2.09) 2.88 (2.11, 3.94)
1.30 (0.97, 1.73) 1.51 (0.94, 2.42)
2.91 (2.32, 3.66) 4.48 (3.18, 6.29)
702
320
1.30 (1.11, 1.53)
1.63 (1.41, 1.89)
1.75 (1.52, 2.02)
1.21 (0.99, 1.48)
2.73 (2.33, 3.20)
1622
1.00
1.00
2.29 (1.95, 2.69) 2.42 (1.91, 3.06)
664
310
1.00
1.94 (1.74, 2.16)
1538
1.00
1.00
3.80 (3.25, 4.45) 5.03 (4.02, 6.29)
702
320
1.00
2.95 (2.65, 3.29)
1622
1.00
1.00
1.00
1.56 (1.25, 1.94) 1.83 (1.35, 2.49)
702
320
1.85 (1.62, 2.12)
1622
1.00
1.00
1.16 (0.97, 1.37) 1.21 (0.94, 1.55)
702
320
1.0
1.76 (1.58, 1.95)
1622
Note. CI = confidence interval.
RESULTS
Main Effects
Table 2 shows important differences in RRs
with regard to all characteristics. Results for
self-inflicted injury and for injury caused by interpersonal violence were fairly comparable.
For both diagnosis groups, and for both boys
and girls, the differences were greatest for receipt of welfare benefits, with the highest RR
for interpersonal violence among girls whose
families received welfare benefits (RR = 5.03;
95% CI = 4.02, 6.29).
Particularly high RRs were found for both
male and female adolescents whose families
were classified as “other” for household SES
(students, housewives, persons living on early-
retirement pensions, and the long-term
unemployed) compared with adolescents
whose parents were high/intermediate-level
nonmanual workers, but RRs also were high
among adolescents from unskilled-manualworker families and, to a lesser degree, from
skilled-manual-worker families.
Compared with living in a low–population
density area, living in a high–population density area entailed an excess risk (although a
lower risk than for other attributes) of interpersonal violence–related injury for teenaged
boys and self-inflicted injury for teenaged girls.
Combined Effects
The RRs derived from the multivariate
analyses are shown in Table 3. The expected
642 | Research and Practice | Peer Reviewed | Engström et al.
reductions in RR were comparable in size for
the 2 diagnostic groups, with the important
exception of household SES. In spite of these
reductions, RR for both types of injury remained higher for adolescents whose families
received welfare benefits than for adolescents
whose families did not. Interestingly, the RR
for girls was higher than that for boys in the
case of interpersonal violence–related injury
(3.71 vs 2.24), although the CIs did overlap.
Furthermore, the net effect of living in a
single-parent home, as opposed to living with
2 adults, remained significantly higher for
both boys and girls, with an excess risk of
about 60% for self-inflicted injury and about
40% for interpersonal violence.
Decreases in RRs were most considerable
for household SES and parental country of
birth. With regard to household SES, only
girls from unskilled-manual-worker families
remained at noticeably higher risk for selfinflicted injury than did the comparison
group, whereas girls and boys both from families classified as “other” and from unskilledmanual-worker families and skilled-manualworker families remained at higher risk of
injury caused by interpersonal violence. Having no parent born in Sweden was no longer
a risk factor for intentional injury for boys, although it remained a risk factor for girls in
the case of self-inflicted injury (albeit substantially lower than before controlling for the
other social and economic characteristics).
Living in a high–population density area, as
opposed to a low–population density area, still
entailed excess risk of interpersonal violence–
related injury for boys and excess risk of selfinflicted injury for girls.
Population-Attributable Risks
Population-attributable risks (expressed as
percentages) are shown in Table 4. The risks
varied from 0 to 29.8 for self-inflicted injury
and from 0 to 32.1 for interpersonal violence–
related injury. Population-attributable risks
were quite similar across diagnoses for all
characteristics except household SES.
Receipt of welfare benefits was the characteristic with the highest population-attributable
risk for the 2 types of injury. It was closely
followed by household SES for interpersonal
violence and by number of parents in the
household for self-inflicted injury. Interestingly,
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 3—Relative Risk (RR) for Intentional Injury, by Selected Household Characteristics,
Adjusted for All Other Household Characteristics and Age: Sweden, 1990–1994
Self-Inflicted Injury, RR (95% CI)
Characteristic
Household socioeconomic status
High/intermediate-level nonmanual
workers
Assistant nonmanual workers
Skilled manual workers
Unskilled manual workers
Self-employed persons
Other
No. of parents in household
2-parent home
1-parent home
Receipt of welfare benefits
No
Once or more
Parental country of birth
1 or both parents born in Sweden
No parent born in Sweden
Population density
Low
High
No. of injuries
Interpersonal Violence–Related
Injury, RR (95% CI)
Girls
Boys
Girls
Boys
1.00
1.00
1.00
1.00
0.99 (0.87, 1.12)
1.08 (0.96, 1.22)
1.19 (1.06, 1.33)
1.14 (0.98, 1.32)
1.16 (1.00, 1.35)
0.99 (0.77, 1.27)
0.98 (0.76, 1.25)
1.11 (0.88, 1.40)
1.24 (0.92, 1.66)
1.26 (0.93, 1.70)
1.21 (0.81, 1.80)
1.46 (1.01, 2.10)
1.81 (1.29, 2.54)
1.41 (0.88, 2.26)
2.29 (1.51, 3.47)
1.19 (1.01, 1.40)
1.53 (1.31, 1.78)
1.36 (1.16, 1.58)
1.22 (0.99, 1.50)
1.38 (1.12, 1.71)
1.00
1.56 (1.43, 1.71)
1.00
1.00
1.00
1.64 (1.37, 1.96) 1.36 (1.04, 1.76) 1.38 (1.22, 1.56)
1.00
3.20 (2.92, 3.50)
1.00
1.00
1.00
2.99 (2.47, 3.62) 3.71 (2.84, 4.84) 2.24 (1.96, 2.55)
1.00
1.17 (1.04, 1.31)
1.00
1.00
1.00
0.92 (0.72, 1.18) 0.91 (0.66, 1.28) 1.11 (0.95, 1.29)
1.00
1.14 (1.04, 1.24)
2805
...
...
664
...
...
310
1.00
1.72 (1.54, 1.92)
1538
Note. CI = confidence interval.
TABLE 4—Population-Attributable Risk for Intentional Injury, by Selected Household
Characteristics: Sweden, 1990–1994a
Self-Inflicted,
Injury, %
Characteristic
Receipt of welfare benefits
Household socioeconomic status
No. of parents in household
Parental country of birth
No. of injuries
Interpersonal
Violence–Related Injury, %
Girls
Boys
Girls
Boys
29.8
8.5
13.1
3.5
2805
26.6
8.1
14.4
...
664
32.1
31.1
9.5
...
310
22.6
21.1
9.7
2.4
1538
a
The population-attributable risk for each household characteristic was based on relative risks obtained after we controlled
for all other household characteristics and age.
for both diagnostic groups, the populationattributable risk related to parental country of
birth was extremely low.
DISCUSSION
Main Findings
We found considerable differences in risk
for intentional injury among Swedish adoles-
cents for each household characteristic we investigated, which is in line with the findings
of earlier studies.2–4,6,10–15 For both boys and
girls, receipt of welfare benefits showed the
largest RR differences and was followed by
household SES, number of parents in the
household, and parental country of birth. In
general, girls had somewhat higher RRs than
did boys, but CIs overlapped.
April 2004, Vol 94, No. 4 | American Journal of Public Health
As expected, simultaneous consideration of
all characteristics led to RR reductions for all
characteristics and for both diagnostic groups.
The most remarkable reductions occured for
parental country of birth (RRs became negligible for both types of intentional injuries) and
for household SES (mainly for self-inflicted injury). After we controlled for other characteristics, only adolescent girls from unskilledmanual-worker families showed a higher risk
of self-inflicted injuries as compared with the
reference group. Nevertheless, adolescent
boys and girls from both unskilled-manualworker and skilled-manual-worker families,
and also those from families classified as
“other,” still showed an excess risk of injury
caused by interpersonal violence.
As might be expected, populationattributable risks were highest for receipt of
welfare benefits—for both boys and girls and
for both types of intentional injuries. The
risks of self-inflicted injury and interpersonal
violence–related injury could be reduced by
23% to 30% (depending on gender and diagnosis) if adolescents from families who received welfare benefits lived with circumstances similar to those of families who did
not. Alternatively, the risk of self-inflicted
injury could be reduced by 13% to 14%
if the living circumstances of adolescents
from single-parent homes mirrored those of
2-parent homes. Likewise, injuries related to
interpersonal violence could be reduced by at
least 21% if adolescents from all household
SES categories lived with circumstances similar to those of children whose parents were
high/intermediate-level nonmanual workers.
The finding of a large net effect of receipt
of welfare benefits after we controlled for all
other attributes may be surprising, because
the Swedish welfare system is designed in
such a way that welfare allocations are sufficiently high to prevent individuals and families from living in poverty. As a consequence,
receipt of welfare benefits is generally not
strongly related to individual financial poverty
in Sweden.26 Furthermore, compared with
children in many other European countries,
few Swedish children are considered to be
poor in absolute or relative terms.27,28 Nevertheless, the fact that wage earners can count
on a reallocation of wealth to compensate for
economic shortfall does not eliminate the dis-
Engström et al. | Peer Reviewed | Research and Practice | 643
 RESEARCH AND PRACTICE 
comfort and the uncertainty they experience
when faced with any such shortfall. Additionally, there are good reasons to believe that receipt of welfare benefits, as well as indicating
financial strain on a family, also may signal
the presence of a variety of related dysfunctional conditions in the household, such as alcohol abuse, depression, and aggression.
It is important to note that during the study
period, Sweden was facing an economic recession. This recession meant that more people were dependent on welfare benefits and,
among these people, more beneficiaries were
newly exposed to such a situation because of
unemployment.29 Accordingly, it was difficult
to assess whether the effects we observed
were circumstantial or were intrinsic to being
in a family in need of state subsidy.
Furthermore, in light of our results, it may
be hypothesized that receipt of welfare benefits mediates household SES and intentional
injury and does so to a greater extent for selfinflicted injuries than for interpersonal
violence–related injuries. Additionally, the fact
that RRs and the population-attributable risk
for household SES remained high in the case
of injuries caused by interpersonal violence indicates that household SES had a true impact
on injury risk that cannot fully be explained
by the confounding effects of welfare benefits
or other family characteristics. A large proportion of the injuries caused by interpersonal violence during adolescence were not sustained
in the household.10
Welfare benefits also may have mediated
some of the effect of number of parents in the
household on the risk of self-inflicted injury.
However, single parents in Sweden need less
financial support than in many other countries
because of Sweden’s labor market policies and
subsidized public child care.30 The presence
of this “safety net” may explain, in part, why
the net RRs of living with 1 parent remained
significant for both boys and girls and for both
diagnoses. This observation, in turn, suggests
that single parenthood increased the risk for
intentional injury among adolescents for reasons that cannot be reduced to the economic
burden borne by parents. Still, it should be
emphasized that population-attributable risks
were quite low for that family characteristic.
One aspect highlighted by our results was that
not having a Swedish-born parent had a low
population-attributable risk when we controlled for other family social and economic
characteristics. It is reasonable to suppose
that factors such as receipt of welfare benefits, household SES, and population density
were mediators of the originally observed excess risk of intentional injury. There is, in
fact, evidence that Swedish immigrants are
educationally overqualified for their work to
a greater extent than are Swedish-born
workers31; a high proportion of immigrants
settle in the country’s 3 largest city areas.
However, the low population-attributable risk
associated with not having a Swedish-born
parent may be a reflection of a weak association between parental country of birth and
the risk for intentional injury.
Our study is silent regarding the mechanisms that underlie the social patterning of intentional injury during adolescence. In particular, other intrafamilial risk factors that were
not considered, or even the experience of earlier episodes of maltreatment within the family,32–34 may have had an aggravating effect
on the risk of self-inflicted and interpersonal
violence–related injuries in social groups already at high risk.10,32 Likewise, contextual
factors related to adolescents’ living circumstances outside the home (e.g. the school, peer
groups, youth culture) also may have modified—
either protected against or aggravated—the effect of family social characteristics.35–38
Study Strengths and Limitations
Our data have very good population coverage, and gaps in data caused by lack of information about social characteristics (5.2%) or
injuries (4.5%) were few.
The first limitation of our study lies in the
manner in which the household status of some
adolescents was determined, which may have
resulted in misclassification of SES and other
social characteristics. Only 1 household was
identifiable for adolescents who spent equal
time living in the separate homes of each parent, because children in Sweden are registered
at a single address. In the worst case, some
adolescents were allocated to a different SES
category than they should have been. Nevertheless, the number of cases is so small that it
cannot significantly alter our results.19
Another concern lies in the underreporting
inherent in register-based studies of inten-
644 | Research and Practice | Peer Reviewed | Engström et al.
tional injury.10 More importantly, because it is
not possible to establish whether the degree
of underreporting is comparable across categories of the social characteristics we considered, some uncertainty remains about the
RRs we compiled.1,23 For instance, if a lower
propensity exists among adolescents from
families with lower SES to seek hospital care
when intentionally injured, RRs will be underestimated. And, in contrast, if the propensity is greater, RRs will be overestimated. The
same reasoning applies to the other family
characteristics. Unfortunately, we had no opportunity to assess the direction of underreporting bias in the various family characteristics, nor do we know whether the propensity
on the part of hospital staff to keep injured
adolescents in the hospital varies according to
the adolescent’s social group or whether there
are diagnostic inaccuracies at the hospital (either the victim “does not tell” or the hospital
“does not see”) that are unevenly distributed
across social groups.39,40 However, it can be
stated that there is no evidence of such discrimination by hospital staff in Sweden.41
Also, it should be stressed that intentional
injuries, particularly self-inflicted ones, make
up diagnosis-related groups in which individual victims can appear several times in hospital discharge registers. In our study, the number of injury occasions was used regardless of
the number of so-called “repeaters.” It was beyond the scope of our study to investigate
whether repeaters were more prevalent in
some social groups than in others. Nor did we
investigate whether the likelihood of dying following an injury varied with social status.
These questions are worth investigating in future studies.
CONCLUSIONS
Our study highlights the importance of a
variety of social and economic characteristics
of an adolescent’s family when studying the
association between parental SES and risk for
self-harm or for violence perpetrated by others. Undeniably, the mechanisms that underlie the relationship between household SES
and risk for intentional injury are complex.
The relationship is likely to be mediated by
the receipt of welfare benefits in the case of
self-inflicted injuries.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
For long-lasting effects to be achieved in the
prevention of intentional injury—and for the
treatment of victims to be successful—there
may be a need to supplement populationbased interventions with other interventions
tailored to social circumstances particular to
some households.
About the Authors
The authors are with the Karolinska Institutet, Department
of Public Health Sciences, Division of Social Medicine,
Norrbacka, Stockholm, Sweden.
Requests for reprints should be sent Karin Engström,
PhD, MPH, Karolinska Institutet, Department of Public
Health Sciences, Division of Social Medicine, Norrbacka,
S-171 76 Stockholm, Sweden (e-mail: [email protected]
phs.ki.se).
This article was accepted January 7, 2003.
years, based on coroners’ and medical records. Br J
Psychiatry. 1999;175:271–276.
7. Laflamme L. Social Inequality in Injury Risks.
Knowledge Accumulated and Plans for the Future. Stockholm, Sweden: Swedish National Institute of Public
Health; 1998:33.
8. Laflamme L, Diderichsen F. Social differences in
the traffic-injury risks in childhood and youth—a literature review and a research agenda. Inj Prev. 2000;6:
293–298.
9. Laflamme L, Engström K. Socioeconomic differences in Swedish children and adolescents injured in
road traffic incidents: cross-sectional study. BMJ. 2002;
324:396–397.
10. Trocmé N, Lindsey D. What can child homicide
rates tell us about the effectiveness of child welfare services? Child Abuse Negl. 1996;20:171–184.
11. Finkelhor D, Hotaling G, Lewis LA, Smith C.
Sexual abuse in a national survey of adult men and
women: prevalence, characteristics, and risk factors.
Child Abuse Negl. 1990;14:19–28.
Contributors
12. Golding AM. Understanding and preventing violence: a review. Am J Public Health. 1995;109:91–97.
K. Engström conceived the study, built the data set, performed the analyses, and participated in the writing of
the article. F. Diderichsen assisted in refining core ideas
and helped with the analyses. L. Laflamme helped refine core ideas and interpret the results and participated
in the writing of the article.
13. Sauvola A, Räsänen PK, Joukamaa MI, Jokelainen J,
Järvelin M-R, Isohanni MK. Mortality of young adults in
relation to single-parent family background. A prospective study of the northern Finland 1966 birth cohort.
Eur J Public Health. 2001;11:284–286.
Acknowledgments
The project was sponsored by the Swedish Council for
Working Life and Social Research and by the Swedish
National Institute of Public Health.
We express our gratitude for the contributions made
by personnel at the National Board of Health and Welfare in Sweden, where the registers used for our study
are maintained.
14. Daniel JH, Hampton RL, Newberger EH. Child
abuse and accidents in black families: a control comparative study. Am J Orthopsychiatry. 1983;53:645–653.
15. Hjern A, Bremberg S. Social aetiology of violent
deaths in Swedish children and youth. J Epidemiol Community Health. 2002;56:688–692.
16. Eriksson R. Social class of men, women and families. Sociology. 1984:18:500–514.
17. Andersson LG, Eriksson R, Wärneryd B. To describe the social structure [in Swedish]. Statistisk Tidskrift. 1981:19:113–136.
of socio-economic inequalities in health: an overview of
available measures illustrated with two examples from
Europe. Soc Sci Med. 1997;44:757–771.
26. Whitehead M, Burström B, Diderichsen F. Social
policies and the pathways to inequalities in health: a
comparative analysis of lone mothers in Britain and
Sweden. Soc Sci Med. 2000;50:255–270.
27. Bradbury B, Jäntti N. Child poverty across the industrialised world: evidence from the Luxembourg Income Study. In: Vleminckx K, Smeeding TM, eds. Child
Well-Being, Child Poverty and Child Policy in Modern
Nations. What Do We Know? Bristol, England: Policy
Press; 2001.
28. Diderichsen F. Income maintenance policies. Determining their potential impact on socioeconomic inequalities in health. In: Mackenbach J, Bakker M, eds.
Reducing Inequalities in Health. A European Perspective.
New York, NY: Routledge; 2002.
29. Socialstyrelsen. Long-Term Recipients of Welfare
Benefits During the 1990s (in Swedish). Stockholm,
Sweden: Socialstyrelsen; 1999.
30. Burström B, Diderichsen F, Östlin P, Östergren P-O.
National experiences: Sweden. In: Mackenbach J,
Bakker M, eds. Reducing Inequalities in Health. A European Perspective. New York, NY: Routledge; 2002.
31. Duvander A-Z. Do country-specific skills lead to
improved labor market positions? An analysis of employment and labor market returns to education
among immigrants in Sweden. Work and Occupations.
2001;28:210–233.
32. Fergusson DM, Lynskey MT. Physical punishment/
maltreatment during childhood and adjustment in
young adulthood. Child Abuse Negl. 1997;21:617–630.
33. Silverman AB, Reinherz HZ, Giaconia RM. The
long-term sequelae of child and adolescent abuse: a
longitudinal community study. Child Abuse Negl. 1996;
20:709–723.
Ethical approval for this study was obtained from the
Karolinska Institute’s research ethics committee.
18. Reports on Statistical Co-ordination. Swedish
Socio-Economic Classification [in Swedish, with English
summary]. Örebro, Sweden: Statistics Sweden; 1983.
34. Tschumper A, Narring F, Meier C, Michaud PA.
Sexual victimization in adolescent girls (age 15–20
years) enrolled in post-mandatory schools of professional training programmes in Switzerland. Acta Paediatr. 1998;87:212–217.
References
19. Barnombudsmannen, Statistics Sweden. Up to
18—Facts on Children and Adolescents [in Swedish].
Halmstad, Sweden: Bulls tryckeri AB; 1995.
35. Feldmann S, Elliot G. At the Threshold—The Developing Adolescent. Cambridge, Mass: Harvard University
Press; 1993.
20. Centre of Epidemiology. The Swedish Hospital Discharge Register 1987–1996. Quality and Contents.
Stockholm, Sweden: Swedish National Board of Health
and Welfare; 1997.
36. Meeus W, Goede M, Kox W, Hurrelmann K, eds.
Adolescence, Careers and Cultures. Berlin, Germany:
Walter de Gruyter; 1992.
Human Participant Protection
1. Engström K, Diderichsen F, Laflamme L. Socioeconomic differences in injury risk in childhood and adolescence: a nation-wide study of intentional and unintentional injuries in Sweden. Inj Prev. 2002;8:137–142.
2. Cubbin C, Smith GS. Socioeconomic inequalities
in injury: critical issues in design and analyses. Annu
Rev Public Health. 2002;23:349–375.
3. Engström K. Social Inequality in Violence-Related
Injuries. Knowledge Accumulated, Research Needs, and
Alternatives for Prevention. Stockholm, Sweden: Swedish
National Institute of Public Health; 1999.
4. Fergusson DM, Woodward LJ, Horwood LJ. Risk
factors and life processes with the onset of suicidal behaviour during adolescence and early adulthood. Psychol Med. 2000;30:23–39.
5. Hasselberg M, Laflamme L, Ringbäck Weitoft G.
Socio-economic differences in road-traffic injuries during childhood and youth—a closer look at different
kinds of road users. J Epidemiol Community Health.
2001;55:858–862.
6. Hawton K, Houston K, Shepperd R. Suicide in
young people. Study of 174 cases, aged under 25
21. Socialstyrelsen. Classification of Diseases 1987.
Systematic Catalogue. Swedish version of International
Classification of Diseases, Ninth Revision (ICD-9). Stockholm, Sweden: Liber/Allmänna förlaget; 1986.
22. World Health Organization. Manual of the International Statistical Classification of Diseases, Injuries, and
Causes of Death (9th Revision). Geneva, Switzerland:
World Health Organization; 1977.
23. Engström K, Laflamme L. Socio-economic differences in intentional injuries. A national study of
Swedish male and female adolescents. Acta Psychiatr
Scand. 2002;106:26–29.
24. Kunst AE, Mackenbach JP. Measuring SocioEconomic Inequalities in Health. Copenhagen, Denmark:
World Health Organization Regional Office for Europe;
1995.
25. Mackenbach J, Kunst A. Measuring the magnitude
April 2004, Vol 94, No. 4 | American Journal of Public Health
37. Sibereisen RK, Todt E, eds. Adolescence in Context.
The Impact of Family, School, Peers, and Work in Adjustment. New York, NY: Springer Verlag; 1994.
38. West P. Health inequalities in the early years: is
there equalisation in youth? Soc Sci Med. 1997:44:
833–858.
39. Benson DE, Swann A, O’Toole R, Turbett JP. Physicians’ recognition of and response to child abuse:
Northern Ireland and the USA. Child Abuse Negl. 1991;
15:57–67.
40. Tygart CE. Student interpersonal public school violence: some explanations derived from adult interpersonal violence theory. Int J Group Tensions. 1991;21:
85–99.
41. Whitehead M, Evandrou M, Haglund B, Diderichsen F. As the health divide widens in Sweden and Britain, what’s happening to access to care? BMJ. 1997;
325:1006–1009.
Engström et al. | Peer Reviewed | Research and Practice | 645
 RESEARCH AND PRACTICE 
A Matched Case–Control Study Evaluating the Effectiveness
of Speed Humps in Reducing Child Pedestrian Injuries
| June M. Tester, MD, MPH, George W. Rutherford, MD, Zachary Wald, MCP, and Mary W. Rutherford, MD
Pedestrian injuries caused by automobile collisions are a leading cause of death among
children aged 5 to 14 years.1 The demographic characteristics of children injured by
automobiles have remained the same over
the past 20 years, with boys, children between the ages of 5 and 9 years, and children
living in neighborhoods of low socioeconomic
status (SES) at highest risk.2–4
Children en route to school or at play in
front of their homes are exposed to roads and
street traffic. Modifying traffic patterns is a
passive and sustainable public health intervention that may make children’s living environments safer.5 Traffic patterns can be modified with a number of engineering strategies
that fall under the rubric of “traffic calming.”
Distinct from speed limit signs or stop signs,
traffic calming measures such as speed
humps, street closures, median barriers, and
traffic circles are successful in providing longterm safety for pedestrians and motorists because they are physical structures with designs that are self-enforcing rather than
requiring police enforcement.6–8
For years, European countries such as Denmark, the Netherlands, and Great Britain, as
well as Australia and New Zealand, have implemented and tested the effects of traffic
calming.6 A report published in British Columbia summarized 43 international studies
that demonstrated reductions in collision frequency rates ranging from 8% to 100% after
implementation of traffic calming measures.6
A Danish study showed that, in comparison
with control streets, 72% fewer injuries occurred on experimental streets incorporating
a variety of traffic calming measures in addition to new speed zoning requirements.9
As a result of the successful efforts in other
countries, there is developing interest in traffic calming in the United States, and the Federal Highway Administration, in cooperation
with the Institute of Transportation Engineers,
has initiated a national traffic calming techni-
Objectives. We evaluated the protective effectiveness of speed humps in reducing child
pedestrian injuries in residential neighborhoods.
Methods. We conducted a matched case–control study over a 5-year period among
children seen in a pediatric emergency department after being struck by an automobile.
Results. A multivariate conditional logistic regression analysis showed that speed
humps were associated with lower odds of children being injured within their neighborhood (adjusted odds ratio [OR] = 0.47) and being struck in front of their home (adjusted
OR = 0.40). Ethnicity (but not socioeconomic status) was independently associated with
child pedestrian injuries and was adjusted for in the regression model.
Conclusions. Our findings suggest that speed humps make children’s living environments safer. (Am J Public Health. 2004;94:646–650)
cal assistance project.6 However, the majority
of safety studies focusing on traffic calming
measures have assessed accident statistics before and after installation, and there is no
available hospital-based information on the
specific effects of these interventions on childhood pedestrian injury.
Oakland has historically been one of the
most dangerous cities in California in which
to be a pedestrian, exhibiting, for example,
the highest rate of pedestrian fatalities among
the state’s cities in 1995.10 In that year, after
a series of child pedestrian deaths, the Oakland Pedestrian Safety Project was formed.
This multidisciplinary alliance addressed child
and senior pedestrian injuries occurring in the
city of Oakland and advocated for installation
of speed humps. Over the 5-year period
1995 to 2000, Oakland installed about 1600
speed humps on residential streets. In this
study, we examined the effect of residing on a
street with speed humps on the odds of child
pedestrian injuries in Oakland.
METHODS
We conducted a matched case–control
study among Oakland residents younger than
15 years over the 5-year period March 1,
1995, to March 1, 2000. Case patients were
children who were seen in the emergency department at Children’s Hospital Oakland after
646 | Research and Practice | Peer Reviewed | Tester et al.
having been struck and injured by an automobile on a residential street. Since this hospital receives all pediatric ambulance trauma
transports (including deaths on the scene)
from the city of Oakland, it was considered
an appropriate choice to target child pedestrians injured in Oakland. Case patients were
each compared with 2 respective controls
matched in regard to age and gender. The
purpose of the study was to determine
whether these children who had been struck
by automobiles were any less likely to live
near a speed hump than their peers who
lived in the same city boundaries but visited
the emergency room that day for a reason
other than being hit by a car.
We identified case patients retrospectively
from a trauma database using International
Classification of Diseases (9th Revision)11
E-code E814.7 (motor vehicle traffic accident
involving collision with a pedestrian). Cases
were limited to those involving children younger than 15 years who were residents of the
city of Oakland and who were injured or died
as a result of the collision. We reviewed
charts and emergency medical service data
sheets to eliminate parking lot injuries, injuries involving bicyclists who had been misclassified as pedestrians, and injuries suffered
by children in driveway rollover collisions. In
addition, we reviewed traffic report data from
the Oakland Police Department, primarily to
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
confirm locations of collisions. When necessary, we reviewed original traffic reports for
further clarification.
We also restricted our analysis to children
injured or killed within 0.25 mi (0.4 km) of
home and used a street atlas12 to determine
whether the injury occurred on the street
block of the child’s residence (defined by
Mueller et al.2 as the “index street”), within a
0.25-mi radius (about 5 blocks, considered
the “surrounding neighborhood”2), or at a
more distant location within Oakland. The
type of street on which a child lived was classified with the street atlas as well.12 Only children residing on minor roads (residential
streets) were eligible for the study, because
speed humps are installed only on such roads.
Living on a street with a speed hump, or
within 1 block of a speed hump, was our
principal predictor variable. We used data
from the Department of Traffic Engineering
in Oakland to determine the exact locations
and dates of installation of speed humps (Department of Traffic Engineering, unpublished
data, 1995–2000). Speed humps that were
located on the other sides of primary or secondary roads (arteries) or were installed after
the date of the injury were not considered.
As mentioned, we matched each case patient, according to age, gender, and date of
emergency department visit, with 2 controls
seen in the emergency department that same
day for a reason other than being struck by a
car. We identified all eligible controls of the
same sex and with the same year of birth as
the case patient from the daily log and randomly selected 2 such individuals. In situations
in which there were fewer than 2 control patients born in the same year as the case patient, we made a random decision to search
the 1 year above or below the age of the case
patient, and then 2 years above or below and
so on, until a suitable control was identified.
Ninety-three percent of all controls were within
2 years of age of their respective case patients.
Controls were restricted to Oakland residents living on residential streets. We collected information on ethnicity and insurance
status (classified as private, public, or self-pay)
from medical records. In addition, we categorized the SES of patient and control households, using 1990 census data on median
household income within the case patient or
control’s census tract, as low ($0–$15 736),
medium ($15 737–$30 115), or high (more
than $30 115).13 Finally, we examined the
records of case patients and controls to ascertain the presence of certain preexisting diagnoses, such as cerebral palsy, mental retardation, paraplegia, and developmental delay,
that would have affected their walking ability
and, thus, their potential to be exposed as
pedestrians to automobile traffic.
Statistical analyses were performed with
Stata software (Stata Corp, College Station,
Tex). We used McNemar matched pairs analyses in examining the 200 case–control pairs
(100 case patients each matched to 2 controls). When a factor is truly protective
against disease, there are more case–control
pairs in which the case lacks (and the control
has) this protective factor than the converse.
Separate univariate analyses focused on ethnicity, census tract household income, and insurance status to determine whether they
were independent predictors of child pedestrian injuries. Once significant (P < .05) variables were determined, we constructed a multivariate conditional logistic regression model
that included only these variables.
RESULTS
We identified 236 individuals who had been
seen in the emergency department during the
study period and had been assigned an E-code
of E814.7. We eliminated 52 potential case patients because they (1) were not Oakland residents at the time of admission, (2) were injured
outside Oakland, (3) were more than 14 years
of age, (4) were bicyclists who had been misclassified as pedestrians, or (5) had been injured
by an automobile backing up within a driveway
or parking lot. We eliminated an additional 84
potential patients because they either lived on
an artery street or had been injured outside of
their neighborhood, yielding a final study sample of 100 case patients.
Case patients and controls were similar in
terms of age, gender, insurance status, median household income, and proportion with
an underlying premorbid neurodevelopmental disease (Table 1). Case patients were
more likely to be Asian or of Hispanic ethnicity. The odds of Asian children having
been involved as a pedestrian in an accident
April 2004, Vol 94, No. 4 | American Journal of Public Health
were 5.8 times as high as those for White
children (P = .018), and the odds of Latino
children having been involved were 4.3
times as high (P = .038). Admitting diagnoses
of controls are available on request from the
authors.
Unadjusted odds ratios (ORs) derived from
a matched pairs analysis showed a protective
effect of speed humps. In comparison with
children living more than a block from a
speed hump, those living within a block of a
speed hump were significantly less likely to
be injured as pedestrians within their neighborhood (14% vs 23%; OR = 0.50; 95%
confidence interval [CI] = 0.27, 0.89)
(Table 2). Among the 100 case patients, 49
were actually hit on the block in front of
their home (index street). As a subset, these
children were even less likely to have a
nearby speed hump than their controls (12%
vs 24%; OR = 0.38; 95% CI = 0.15, 0.90)
(Table 2).
We performed multivariate logistic regression analyses using both predictor variables
and included race and ethnicity in the model.
After control for race and ethnicity, speed
humps were associated with significantly
lower odds of children being injured in their
neighborhood (adjusted OR = 0.47; 95% CI =
0.24, 0.95) and being struck on the block immediately in front of their home (adjusted
OR = 0.40; 95% CI = 0.15, 1.06) (Table 2).
DISCUSSION
In our observational study, we found that
children who lived within a block of a speed
hump had significantly lower odds of being
struck and injured by an automobile in their
neighborhood. Living within a block of a
speed hump was associated with a roughly
2-fold reduction in the odds of injury within
one’s neighborhood (adjusted OR = 0.47).
This protective effect was even more pronounced among the subset of children who
were injured on the block immediately in
front of their house (index street). Children
living within a block of a speed hump exhibited a 2.5-fold reduction in the odds of being
injured on their street (adjusted OR = 0.4).
These results highlight the effectiveness of
speed humps in reducing child pedestrian
injuries.
Tester et al. | Peer Reviewed | Research and Practice | 647
 RESEARCH AND PRACTICE 
have essentially 2 prevention strategies at our
disposal: we can protect children from fastmoving traffic by modification of either their
behavior or their traffic environment. There
have been multiple attempts to modify children’s behavior, including school training programs,17 “traffic clubs” designed to educate
parents and children about safe behavior on
streets,18 simulation games,19 and communitylevel interventions.20 For the most part, however, these educational efforts have been unable to exert meaningful changes in the longterm behavior of children, largely owing to
the developmental limitations of preschoolaged children.20 As a result, a great deal of attention has shifted to environment modification and the promise it holds for affecting
child pedestrian injury rates.
TABLE 1—Demographic Characteristics of Case Patients and Controls
Male, No. (%)
Age, y, mean (SD)
Ethnicity, %
White
Black
Native American/other
Hispanic
Asian
Insurance status
Private insurance
Public insurance
Self-pay
Household income, $ (census tract)
High (> 30 115)
Medium (15 737–30 115)
Low (0–15 736)
Premorbid diagnosisb
Mild mental retardation
Developmental delay
Case Patients
(n = 100)
Controls
(n = 200)
Odds Ratio
Pa
68 (68)
6.8 (3.5)
136 (68)
6.6 (3.7)
...
...
...
.63
3 (3)
49 (49)
11 (11)
22 (22)
15 (15)
16 (8)
117 (58.5)
21 (10.5)
31 (15.5)
15 (7.5)
Reference
2.4
3.2
4.3
5.8
.187
.115
.038
.018
17 (17)
78 (78)
5 (5)
43 (21.5)
147 (73.5)
10 (5)
Reference
1.3
1.3
.366
.717
12 (12)
75 (75)
13 (13)
39 (19.5)
136 (68)
25 (12.5)
Reference
1.8
1.7
.105
.265
Focus on Neighborhood Injury
1 (1)
0 (0)
1 (0.5)
3 (1.5)
...
...
Note. A univariate analysis of age, ethnicity, insurance status, household income, and presence of a premorbid diagnosis
showed that only ethnicity was independently associated with child pedestrian injury.
a
All P values were obtained from conditional logistic regression analyses, except for age, which was obtained with a 2-tailed
test of means.
b
Case patients and controls were screened for the presence of any of the following premorbid diagnoses: cerebral palsy,
mental retardation, quadriplegia, paraplegia, and developmental delay.
Exposure to Traffic
Increased exposure to traffic (especially
traffic at high volume and speed) is a known
risk factor for child pedestrian injury. Stevenson and colleagues showed that an increase
in volume of 100 vehicles per hour is associated with an incremental increase of about
2.0 in the odds of pedestrian injury.14 Average speeds traveled on streets are also associated with risk of injury, and at least 2 studies
have demonstrated that a higher proportion
of vehicles exceeding the posted speed limit is
associated with higher odds of child pedestrian injuries.14,15 In addition to the type of
street, the number of streets that children
cross on their way to school seems to affect
their risk.16
Need for Passive Environment
Modification
Given the relationship between exposure to
traffic and risk of child pedestrian injuries, we
TABLE 2—Odds of Pedestrian Injury Within a Child’s Neighborhood and Odds of Injury on a
Child’s Index Street of Residence When Child’s Home Is Within 1 Block of a Speed Hump:
Multivariate Model
Neighborhood injury
Index street injury
Case Patients
(n = 100), No. (%)
Control Subjects
(n = 200), No. (%)
OR (95% CI)a
Adjusted OR (95% CI)b
14 (14)
6 (12)
46 (23)
24 (24)
0.50 (0.27, 0.89)
0.38 (0.15, 0.90)
0.47 (0.24, 0.95)
0.40 (0.15, 1.06)
Note. OR = odds ratio; CI = confidence interval.
a
Calculated from McNemar matched pairs analysis.
b
Calculated from multivariate model including ethnicity.
648 | Research and Practice | Peer Reviewed | Tester et al.
The deliberate focus of our study was on
pedestrian injuries occurring in a child’s own
neighborhood (defined here as within a
0.25-mi radius of the child’s home) as opposed to all injuries, including those occurring at more distant sites. We focused on
such injuries because although children leave
their neighborhoods with adults (and often in
automobiles), most of their unsupervised
time is likely to be near home. In addition,
the traffic calming methods we examined
can be applied only to residential streets. One
8-year study that examined fatal head injuries revealed that injuries to pedestrians
were the most common cause of fatal head
injuries and that 53% of those injured were
playing in the street at the time of the injury.
Of the 135 accidents that fell into this category, only 1 involved a child who had been
under adult supervision at the time of the accident (the remaining children had been supervised by siblings or other children).
The same study showed that 80% of fatal
pedestrian injuries had taken place within 1
mi (1.6 km) of the child’s home.21 Among the
184 children we initially identified for this
study, 125 (68%) were eligible for the study
because their injury occurred within 0.25 mi
of home (the other children were eliminated
because they lived on arterial streets). Therefore, our data suggests that roughly two thirds
of injuries occur within the 0.25 mi surrounding a child’s home. Passive interventions that
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
reduce child pedestrian injuries are likely to
be of greater benefit in areas where children
are prone to spend time without adults.
In our study, SES was not a significant independent predictor of child pedestrian injury. Mueller and colleagues found that living
in a census tract with a median household income level below $20 000 was associated
with 7.0-fold higher odds of injury than living
in a census tract with a median income level
above $30 000.2 Other research points toward an association between increasing rates
of pedestrian injury and lower SES, as approximated by census tract of residence,4 spatial modeling of census tract and other data
with a geographic information system,22 and
more indirect indicators of lower SES such as
living near a convenience store, gas station, or
fast food store.15
It is possible that, in our population,
“overmatching” was the reason SES was not
found to be an independent risk factor. Case
patients were not matched with controls on
SES, but if lower SES is associated with
both increased odds of injury2 and increased
odds of an emergency department visit,23
choosing controls from the emergency department may have resulted in overmatching
in terms of SES.
Limitations
Our study involves potential methodological limitations. For example, limiting measurement to speed humps on a child’s street
ignores the potential protective effect of speed
humps around the corner from a child’s
house. Thus, by measuring speed humps lateral to an index street (rather than in a 1-block
radius), we may have underestimated the relevant rate of exposure to this intervention,
which would have affected our estimation of
the intervention’s protective impact.
There are also limitations involved with
our study sample. While relying on emergency department visits ensured that we incorporated higher severity injuries (including
deaths), injuries that were not reported to
the emergency medical services (and for
which children may have been taken by
their family to their regular doctor) would
have been missed. This would mean that our
sample underrepresented lower acuity injuries. It is also possible that our sample un-
derrepresented younger children, in that
children younger than 5 years are more
likely to be hit in their driveway (often by a
backing automobile)24,25; we excluded children in this age group from our study because such injuries are not related to the
flow of street traffic.
Finally, it is possible that significant confounding factors were not addressed in this
study. Some research suggests that the
presence of sidewalks is not a significant
contributor to odds of injury,2,15 and other
research suggests that the presence of sidewalks is a strong risk factor, with an odds
ratio of 11.0.14 We would have liked to control for the presence of sidewalks, but there
were no reliable retrospective data on sidewalk or curb presence available to do so.
Also, since much of the earlier literature
points to lower SES as a risk factor for child
pedestrian injury, the reason for our inability to reproduce this relationship may have
been that the factors we used to approximate SES—census tract household income
and medical insurance status—are inappropriate proxies for SES.
CONCLUSIONS
We found that speed humps were associated with a 53% to 60% reduction in the
odds of injury or death among children struck
by an automobile in their neighborhood.
These findings invite additional research on
the protective effects of traffic calming interventions and offer a framework for studying
pedestrian injuries in relation to physical interventions implemented within a localized
geographic region. Further confirmation of
the protective effects of speed humps would
be useful and could be augmented by additional information on stop signs or other factors that would affect slowing distances on either side of a speed hump. Our study provides
direct observational evidence that speed
humps are associated with a reduction in the
odds of childhood pedestrian injuries and
supports the installation of speed humps by
traffic engineering departments.
About the Authors
At the time of the study, June Tester was a medical student
at the University of California, San Francisco, and an
April 2004, Vol 94, No. 4 | American Journal of Public Health
MPH candidate at the University of California, Berkeley.
George W. Rutherford is with the Department of Epidemiology and Biostatistics at the University of California, San
Francisco, School of Medicine. Zachary Wald is with California Walks, Oakland, Calif. Mary W. Rutherford is with
the Children’s Hospital and Research Center at Oakland.
Requests for reprints should be sent to June M. Tester,
MD, MPH, who is now at Children’s Hospital Oakland,
747 52nd St, Oakland, CA 94609 (e-mail: [email protected]
post.harvard.edu).
This article was accepted March 2, 2003.
Contributors
J. M. Tester conceived the study, performed all analyses,
and led the writing of the article. G. W. Rutherford assisted in data analyses, interpretation of findings, and
revisions of the article. Z. Wald contributed to conceptualization of ideas as well as reviews of the article.
M. W. Rutherford contributed to the study design and
interpretation of the findings.
Acknowledgments
We are grateful for the assistance of the medical records personnel of the Children’s Hospital and Research
Center at Oakland, with special thanks to Eve Magee
and Midge Maritzen. Also, we would like to acknowledge the wonderful assistance of Lieutenant David Kozicki, of the Oakland Police Department Traffic Division,
for providing traffic report information for this study,
and the invaluable help of Henry Choi, who provided
data on speed hump installations in Oakland.
Human Participant Protection
This study was reviewed and approved by the institutional review board of Children’s Hospital and Research
Center at Oakland. Informed consent was not required
by the review board because patients did not need to
be contacted for this retrospective data analysis.
References
1. Grossman D. The history of injury control and the
epidemiology of child and adolescent injuries. Future
Child. 2000;10:23–52.
2. Mueller B, Rivara FP, Lii S, Weiss NS. Environmental factors and the risk for childhood pedestrianmotor vehicle collision occurrence. Am J Epidemiol.
1990;132:550–560.
3. Pless I, Verreault R, Arsenault L, Frappier JY, Stulginskas J. The epidemiology of road accidents in childhood. Am J Public Health. 1987;77:358–360.
4. Rivara F. Demographic analysis of childhood
pedestrian injuries. Pediatrics. 1985;76:375–381.
5. Rivara F. Pediatric injury control in 1999: where
do we go from here? Pediatrics. 1999;103:883–888.
6. Ewing R. Traffic Calming: State of the Practice.
Washington, DC: Institute of Transportation Engineers;
1999.
7. Roundabouts Are Becoming More Familiar on US
Roads, Not Just for Safety Reasons: Status Report. Arlington, Va: Insurance Institute for Highway Safety; 2000;
35(5):1–6.
8. Appleyard D. Livable Streets. Berkeley, Calif: University of California Press; 1981.
9.
Engel U, Thomsen LK. Safety effects of speed re-
Tester et al. | Peer Reviewed | Research and Practice | 649
 RESEARCH AND PRACTICE 
ducing measures in Danish residential areas. Accid
Anal Prev. 1992;24:17–28.
The Emergence of AIDS
10. California Cities Pedestrian Injuries/Fatalities Comparisons: Annual Report. Sacramento, Calif: Statewide
Integrated Traffic Records System; 1999.
The Impact on Immunology,
Microbiology and Public Health
Edited by Kenneth H. Mayer, MD,
and H. F. Pizer
11. International Classification of Diseases, 9th Revision. Geneva, Switzerland: World Health Organization;
1980.
12. The Thomas Guide: Alameda County. Irvine, Calif:
Thomas Bros. Maps; 2000.
T
13. US census data, 1990. Available at: http://
factfinder.census.gov. Accessed November 1, 2001.
14. Stevenson M, Jamrozik KD, Spittle J. A case control study of traffic risk factors and child pedestrian injury. Int J Epidemiol. 1995;25:957–964.
15. Kraus J, Hooten EG, Brown KA, Peek-Asa C,
Heye C, McArthur DL. Child pedestrian and bicyclist
injuries: results of community surveillance and a casecontrol study. Inj Prev. 1996;2:212–218.
ISBN 0-87553-176-8
2000 ❚ 350 pages ❚ softcover
$25.00 APHA Members
$36.00 Nonmembers
plus shipping and handling
16. Rao R, Hawkins M, Guyer B. Children’s exposure
to traffic and risk of pedestrian injury in an urban setting. Bull N Y Acad Med. 1997;74:65–80.
17. Rivara F, Booth CL, Bergman AB, Rogers LW,
Weiss J. Prevention of pedestrian injuries to children:
effectiveness of a school training program. Pediatrics.
1991;88:770–775.
his unique book highlights the lessons learned from and
about AIDS over the past 20 years, and highlights the
knowledge that may advance worthwhile strategies for
combating HIV and AIDS in the future. The 11 chapters include: The Virus Versus the Immune System, How
Infectious is Infectious, The Race against Time: The
Challenge for Clinical Trials, Sex and Drugs and the Virus,
and more.
This book is an ideal reference for Infectious disease specialists ❚ Epidemiologists ❚ Public health practitioners
❚ Clinicians ❚ All those concerned with AIDS.
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
18. West R, Sammons P, West A. Effects of a traffic
club on road safety knowledge and self-reported behavior of young children and their parents. Accid Anal
Prev. 1993;25:609–618.
19. Renaud L, Suissa S. Evaluation of the efficacy of
simulation games in traffic safety education of kindergarten children. Am J Public Health. 1989;79:307–
309.
Children’s Environmental
Health
20. Klassen T, MacKay JM, Moher D, Walker A, Jones AL.
Community-based injury prevention interventions. Future Child. 2000;10:83–110.
21. Sharples P, Storey A, Aynsley-Green A, Eyre JA.
Causes of fatal childhood accidents involving head injury in Northern Region, 1979–86. BMJ. 1990;301:
1193–1197.
by Dona Schneider and Natalie Freeman
T
22. LaScala E, Gerber D, Grunewald PJ. Demographic
and environmental correlates of pedestrian injury collisions: a spatial analysis. Accid Anal Prev. 2000;32:
651–658.
23. Shah-Canning DAJ, Bauchner H. Care-seeking patterns of inner-city families using an emergency room.
Med Care. 1996;34:1171–1179.
24. Roberts I, Norton R, Jackson R. Driveway-related
child pedestrian injuries: a case control study. Pediatrics. 1995;95:405–408.
25. Winn D, Agran PF, Castillo DN. Pedestrian injuries to children younger than 5 years of age. Pediatrics. 1991;88:776–782.
EA01J7
ISBN 0-87553-241-1
2000 ❚ 149 pages ❚ softcover
$13.50 APHA Members
$19.95 Nonmembers
plus shipping and handling
he health of our children is a critical issue facing our society today. The toll of childhood death and disability
extends well beyond the individual child to affect all of us.
This book empowers readers by providing clear information
about environmental threats and what we can do to prevent
them.
The six chapters include Infectious Diseases in the
Environment; Injuries and Child Health; The Legacy of
Lead; Environmental Chemicals and Pests; Childhood
Asthma; and Reducing Environmental Health Risks. An
Appendix of activities to do with children is included.
Pediatricians, child health care practitioners and parents
will find this book an invaluable resource.
American Public Health Association
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
FAX: (301) 843-0159
650 | Research and Practice | Peer Reviewed | Tester et al.
CE01J7
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Housing First, Consumer Choice, and Harm Reduction
for Homeless Individuals With a Dual Diagnosis
| Sam Tsemberis, PhD, Leyla Gulcur, PhD, and Maria Nakae, BA
Current rates of homelessness in New York City
are the highest ever documented.1 A small percentage of this population remains chronically
homeless, either living on the streets or other
public places or intermittently using emergency
rooms, shelters, jails, and other short-term services, but never successfully ending their homelessness.2 Members of this chronically homeless
group typically have a history of mental illness,3
compounded by substance use disorders.4,5,6 Although much is known about the chronically
homeless, these individuals continue to elude
existing program efforts.
The predominant service delivery model designed to address the needs of this chronically
homeless population, called the Continuum of
Care, consists of several program components.
It begins with outreach, includes treatment and
transitional housing, and ends with permanent
supportive housing. The purpose of outreach
and transitional residential programs is to enhance clients’ “housing readiness” by encouraging the sobriety and compliance with psychiatric treatment considered essential for successful
transition to permanent housing. This approach
assumes that individuals with severe psychiatric
disabilities cannot maintain independent housing before their clinical status is stabilized. Furthermore, the model presumes that the skills a
client needs for independent living can be
learned in transitional congregate living. Research in psychiatric rehabilitation indicates,
however, that the most effective place to teach
a person the skills required for a particular environment is within that actual setting.7
Consumers’ perception of the Continuum of
Care offers another divergent perspective. Consumers experience the Continuum as a series
of hurdles—specifically, ones that many of them
are unable or unwilling to overcome. Consumers who are homeless regard housing as an
immediate need, yet access to housing is not
made available unless they first complete treatment. By leveraging housing on participation
and treatment, continuum program require-
Objectives. We examined the longitudinal effects of a Housing First program for homeless, mentally ill individuals’ on those individuals’ consumer choice, housing stability,
substance use, treatment utilization, and psychiatric symptoms.
Methods. Two hundred twenty-five participants were randomly assigned to receive
housing contingent on treatment and sobriety (control) or to receive immediate housing without treatment prerequisites (experimental). Interviews were conducted every 6
months for 24 months.
Results. The experimental group obtained housing earlier, remained stably housed,
and reported higher perceived choice. Utilization of substance abuse treatment was
significantly higher for the control group, but no differences were found in substance use
or psychiatric symptoms.
Conclusions. Participants in the Housing First program were able to obtain and maintain independent housing without compromising psychiatric or substance abuse symptoms. (Am J Public Health. 2004;94:651–656)
ments are incompatible with consumers’ priorities and restrict the access of consumers who
are unable or unwilling to comply with program terms.
In addition, most consumers prefer to live in
a place of their own rather than in congregate
specialized housing with treatment services onsite.8,9 Most programs have rules that restrict
clients’ choices and that when violated are used
as grounds for discharging the consumer from
the program. For example, despite having attained permanent housing, clients who relapse
and begin to drink mild or moderate amounts
of alcohol, may be evicted if the program has
strict rules about sobriety maintenance. The
chronically homeless population is characterized by its frequent inability to gain access to
existing housing programs. Individuals in this
group often have multiple disabling conditions,
especially psychiatric conditions and substance
abuse.10 Most programs are poorly equipped to
treat people with dual diagnoses, let alone prepared to address their housing needs.11 Treatment requires time and commitment and is
often not available if a program is under pressure to move clients along a continuum.12
The loss of control over one’s life resulting
from housing instability, frequent psychiatric
hospitalizations, and intermittent substance
abuse treatment leaves some consumers mis-
April 2004, Vol 94, No. 4 | American Journal of Public Health
trustful of the mental health system and unwilling to comply with demands set by providers.13
Others prefer the relative independence of life
on the streets to a fragmented treatment system
that inadequately treats multiple diagnoses or
addresses housing needs.14,15 Paradoxically, consumers’ reluctance to use traditional mental
health and substance abuse services as a condition of housing only confirms providers’ perceptions that these individuals are “resistant” to
treatment, not willing to be helped, and certainly not ready for housing.16
The Housing First model was developed by
Pathways to Housing to meet the housing and
treatment needs of this chronically homeless
population. The program is based on the belief
that housing is a basic right and on a theoretical foundation that includes psychiatric rehabilitation and values consumer choice.17 Pathways
is designed to address the needs of consumers
from the consumer’s perspective.18 Pathways
encourages consumers to define their own
needs and goals and, if the consumer so wishes,
immediately provides an apartment of the consumers’ own without any prerequisites for psychiatric treatment or sobriety. In addition to an
apartment, consumers are offered treatment,
support, and other services by the program’s
Assertive Community Treatment (ACT) team.
ACT is a well defined community based inter-
Tsemberis et al. | Peer Reviewed | Research and Practice | 651
 RESEARCH AND PRACTICE 
disciplinary team of professionals that includes
social workers, nurses, psychiatrists, and vocational and substance abuse counselors who are
available to assist consumers 7 days a week 24
hours a day. The Pathways program has made
two modifications to the standard ACT model:
a nurse practitioner was included to address
the considerable number of health problems,
and a housing specialist was added to coordinate the housing services. Although housing
and treatment are closely linked, they are considered separate domains, and consumers in
the program may accept housing and refuse
clinical services altogether without consequences for their housing status. There are 2
program requirements: tenants must pay 30%
of their income (usually Supplemental Security
Income [SSI]) toward the rent by participating
in a money management program, and tenants
must meet with a staff member a minimum of
twice a month. These requirements are applied
flexibly to suit consumers’ needs.21
Consistent with the principles of consumer
choice, Pathways uses a harm-reduction approach in its clinical services to address alcohol
abuse, drug abuse, and psychiatric symptoms
or crises. At its core, harm reduction is a pragmatic approach that aims to reduce the adverse
consequences of drug abuse and psychiatric
symptoms.22 It recognizes that consumers can
be at different stages of recovery and that effective interventions should be individually tailored to each consumer’s stage.23 Consumers
are allowed to make choices—to use alcohol or
not, to take medication or not—and regardless
of their choices they are not treated adversely,
their housing status is not threatened, and help
continues to be available to them.
Continuum of Care supportive housing programs subscribe to the abstinence–sobriety
model based on the belief that without strict
adherence to treatment and sobriety, housing
stability is not possible. But studies examining
the model’s effectiveness report only modest results in achieving housing stability for individuals who are chronically homeless and mentally
ill.24 Alternatively, the approach used by the
Pathways program assumes that if individuals
with psychiatric symptoms can survive on the
streets then they can manage their own apartments. The program posits that providing a person with housing first creates a foundation on
which the process of recovery can begin. Hav-
ing a place of one’s own may—in and of itself—
serve as a motivator for consumers to refrain
from drug and alcohol abuse.
The purpose of this study was to compare
the effectiveness of the Housing First model
with that of programs that used the Continuum
of Care model for individuals who are chronically homeless and mentally ill.
We tested the following hypotheses: (1) the
experimental (Housing First) group would report greater consumer choice over time than
the control (Continuum of Care) group; (2) the
experimental group would (a) exhibit lower
rates of homelessness than the control group
and (b) achieve and sustain greater residential
stability than the control group; (3) the experimental group would exhibit rates of substance
use similar to or lower than those of the control
group; (4) the experimental group would participate in fewer substance-abuse treatments over
time than the control group (i.e., because substance abuse treatment is not a precondition for
the Housing First model, it is expected that
there will be a lower rate of service utilization
for the experimental group); and (5) the experimental group would exhibit rates of psychiatric
symptoms similar to or lower than those of the
control group.
METHODS
Participants
The 225 participants were randomized into
2 groups. One hundred twenty-six participants
(56%) were assigned to the control group—and
entered programs that followed the Continuum
of Care model—and 99 (44%) were assigned to
the experimental group and to a program that
used the Housing First model. The control
group was intentionally oversampled, anticipating that a higher number of control group participants may remain homeless and prove more
difficult to locate for follow up interviews. The
sample comprised 2 subgroups: an original
street sample of 157 participants who met eligibility criteria, and a second group of 68 individuals recruited from 2 state psychiatric hospitals. To meet eligibility criteria, the first group
had to have spent 15 of the past 30 days on
the street or in other public places (shelters
were not included), exhibited a history of
homelessness over the past 6 months, and had
an Axis I diagnosis25 of severe mental illness.
652 | Research and Practice | Peer Reviewed | Tsemberis et al.
Diagnoses were based on previous records
from service providers or, in cases in which records were unavailable, on an interview with an
independent psychiatrist. Although a diagnosis
or history of alcohol or substance abuse disorders was not an eligibility criterion, according
to clinical records 90% of all the participants
also had a diagnosis or history of alcohol or
substance abuse disorders. The street sample
was recruited through service agency staff referral of eligible clients who were interested in
study participation. The second group met the
same entry criteria for homelessness and mental illness immediately before hospitalization as
did the street sample.
Because of administrative problems, 12 participants in the experimental condition were
not assigned a Pathways apartment, and 7 control participants were erroneously assigned a
Pathways apartment. Excluding these 19 participants reduced the number of control participants to 119 (58%) and the number of experimental participants to 87 (42%).
As can be seen in Table 1, the final sample
consisted of 162 (79%) men and 44 (21%)
women whose average age was 41.3 years.
More than half of the participants (n=110,
53%) were diagnosed with a psychotic disorder. Seventeen percent (n=35) had become
homeless before the age of 18 years. The
longest period ever homeless, on average, was
4.5 years. Fifty-one percent (n=114) of the participants were literally homeless (staying in the
streets or public spaces) at the time of the baseline interview. Another 36% entered the study
from psychiatric institutions but had been
homeless before hospitalization. After randomization, there were no significant differences between groups for baseline demographic characteristics such as gender, age, education, race,
diagnosis, or amount of time homeless.
Procedures
After completing their baseline interviews,
participants were interviewed every 6 months.
Interviewers were blind to participants’ assignment for baseline interviews but not for followup interviews. Data for the complete 24-month
period were collected between December
1997 and January 2001. During each interim
period, 5-minute telephone calls were conducted primarily to maintain contact with participants and establish their whereabouts. Par-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 1—Participant Characteristics at
Baseline (n = 206)
No. (%)
Study group
Experimental
Control
Gender
Female
Male
Age, y
18–30
31–40
41–50
51–60
≥ 61
Education
8th grade or less
Some high school
Finished high school
Completed general equivalency
diploma
Vocational/trade/business school
Some college
College degree
Graduate degree
Race/ethnicity
White (not Hispanic)
Black (not Hispanic)
Hispanic
Mixed/other/unknown
Diagnosis
Psychotic
Mood—depressive
Mood—bipolar
Other
Unknown
Residence at baseline
Streets/subways/parks/abandoned
building/drop-in centers
Shelter/safe haven
Psychiatric hospital
Other
87 (42)
119 (58)
44 (21)
162 (79)
39 (19)
59 (29)
62 (30)
36 (17)
10 (5)
21 (10)
66 (32)
34 (17)
16 (8)
5 (2)
49 (24)
10 (5)
4 (2)
55 (27)
84 (41)
30 (15)
37 (18)
110 (53)
29 (14)
29 (14)
10 (5)
28 (14)
114 (51)
13 (6)
80 (36)
18 (8)
ticipants were paid for all interviews. Six-month
interviews were conducted in a variety of locations, including the research office, the participant’s apartment/residential location, or a public place such as a cafe or restaurant. When it
was not possible for interviews to be conducted
face-to-face (e.g., the participant had moved out
of state), interviews were conducted by telephone. For participants in psychiatric hospitals
and correctional facilities, research interviewers
made onsite visits. The questions asked during
each interview period remained the same. The
follow-up rates by time period were as follows:
88% at 6 months, 87% at 12 months, 84% at
18 months, and 78% at 24 months. These follow-up rates do not include individuals who
were missing at certain time points but who
were located subsequently and for whom residential data was collected at a later point. Thus,
the follow-up rates reported here are based on
conservative calculations.
Measures
A modified version of Consumer Choice, a
16-item, 5-point scale developed by Srebnik,
Livingston, Gordon, and King,26 was used to
determine (1) how important it was for the participant to have choice at baseline and (2) how
much choice the participant actually had, at
subsequent time points, in their location, neighbors and housemates, visitors, and so forth.
We measured residential status with a 6month residential follow-back calendar developed by New Hampshire Dartmouth Research
Center.27 The interviewer assessed the participant’s location for each day during the past 6
months on a day-by-day basis. From this information, we calculated the proportion of time
spent homeless as well as the proportion of
time spent in stable housing.
Following the interview, the interviewer
coded the participant’s residential location according to several distinct residential categories.
For the purpose of analyses, homelessness was
considered as living on the streets, in public
places, or in shelter-type accommodations. Residential stability was defined as residing in one’s
own apartment; or having a room or studio
apartment in a supportive housing program, a
group home, a boarding home, or a long-term
transitional housing program; or living longterm with parents, friends, or other family
members. The number of days spent in any of
the locations categorized as “homeless” or “stably housed” was summed and divided by the
total number of days of residency reported at
the interview.
We measured alcohol and drug use with the
Drug and Alcohol Follow-Back Calendar.28,29
Participants reported the number of drinks
consumed each day, as well as the number of
days that selected drugs were used during a
April 2004, Vol 94, No. 4 | American Journal of Public Health
6-month period. We used an alcohol use variable (measuring the total number of drinks)
and a drug use variable (measuring the total
number of days of drug use) for each 6-month
period in the analyses.
We measured substance abuse treatment
service utilization with a modified shorter version of the Treatment Services Inventory.30 In
the interview, participants were asked whether
they received any substance abuse treatment
during the past 2 weeks. Drug and alcohol
treatment services use was indicated by an average of 7 items including questions asking
whether the participant had received treatment
in a detoxification program or other program;
consulted with a counselor to talk about substance problems; or attended Alcoholics Anonymous, Narcotics Anonymous, or any other
self-help group.
Psychiatric symptoms were measured with
the Colorado Symptom Index,31 a 15-item
questionnaire including items assessing psychotic symptoms as well as symptoms related
to mood and suicidality.
Data Analysis
Repeated-measures analysis of variance
(ANOVA) was used to examine group differences, during the 2-year follow-up period, for
hypothesis 1 (consumer choice), hypothesis 2
(housing stability assessed as 2 separate outcomes: proportion of time stably housed and
proportion of time homeless), and hypothesis 3
(substance abuse assessed as 2 separate outcomes: alcohol abuse and drug abuse). In cases
in which repeated-measures ANOVAs yielded
significant results, t tests were conducted to
compare group differences at each time point.
Group differences were then plotted and
graphed for the 2 groups across time.
To appropriately examine differences in substance abuse treatment services use, hypothesis
4 was tested with a subsample of participants
who were not on the streets but who were in
some type of service-related program: namely,
experimental participants who were currently
housed by the Housing First program and control participants who were housed by one of
the Continuum of Care programs. Control participants were included in this analysis if they
reported that they lived most recently in one of
the following places at the time of the interview: shelters, supportive housing programs,
Tsemberis et al. | Peer Reviewed | Research and Practice | 653
 RESEARCH AND PRACTICE 
drop-in centers, safe havens, detoxification facilities, crisis housing, intermediate care, boarding
houses, transitional housing, group homes, alcohol/drug-free facilities, and treatment/recovery
programs. Because participants’ residential status changed from one time point to the next,
the subsample also changed; we therefore had
to conduct separate t tests for each time point.
Because there were 5 time points, we used a
Bonferroni adjusted α of .025 to account for
Type I error.
Power Analysis
To retain 80% power to detect an effect that
explains 4% of the variance in the context of
an equation (with 5 covariates) that explains
25% of the variance, we needed to retain 68%
of the original sample; moreover, power for repeated-measures analyses would be higher.32
Our retention rates were substantially above
this figure, so we did not anticipate any problems in the power to detect group differences.
group’s perceptions were more stable than
were those of the control group. As can be seen
from Figure 1, subsequent univariate analyses
showed significant differences at 6, 12, 18, and
24 months, with the experimental group reporting significantly more choice than the control group.
Residential Stability
Repeated-measures ANOVA results showed
a significant Time × Group status effect. Participants in the experimental condition had significantly faster decreases in homeless status and
increases in stably-housed status relative to participants in the control condition (F4,137 =10.1,
P<.001; F4,137 =27.7, P<.001). As can be seen
from Figures 2 and 3, subsequent univariate
analyses showed significant differences at 6, 12,
18, and 24 months, with the experimental
group reporting less time spent homeless and
more time spent stably housed compared with
the control group.
Substance Use
RESULTS
Consumer Choice
Results from repeated-measures ANOVA
showed that there was a significant time ×
group status effect, indicating that participants
in the experimental condition perceived their
choices to be more numerous than did participants in the control condition (F4,112=8.91,
P <.001 ). Additionally, the experimental
Repeated-measures analyses showed no significant differences in either alcohol or drug
use between the 2 groups by time condition
(F4,136 =1.1, P=.35 for alcohol use; F4,136 =.98,
P=.42 for drug use).
Substance Abuse Treatment Utilization
Five t tests were conducted with an adjusted
α level of .025. As can be seen from Figure 4,
these univariate analyses showed significant dif-
5
4.5
Consumer Choice
4
3.5
3
2.5
2
1.5
Experimental
Control
1
0.5
0
Baseline
6
12
18
24
Months
Note. At baseline, participants were asked how much choice they would like to have. Subsequent time-points assess how
much choice participants actually have.
FIGURE 1—Consumer choice in housing: baseline–24 months.
654 | Research and Practice | Peer Reviewed | Tsemberis et al.
ferences at 6, 18, and 24 months (P<.025)
and at 12 months (P<.05), with the Continuum group reporting significantly higher use of
substance abuse treatment programs than the
Housing First group. In addition, a decrease in
service use occurred among the Housing First
group and an increase occurred among the
Continuum group over time.
Psychiatric Symptoms
Repeated-measures analyses showed no
significant differences psychiatric symptoms
between the 2 groups by time condition
(F4,137 = .348, P = .85).
DISCUSSION
Our results attest to the effectiveness of using
the Housing First approach in engaging, housing, and keeping housed individuals who are
chronically homeless and dually diagnosed. The
Housing First program sustained an approximately 80% housing retention rate, a rate that
presents a profound challenge to clinical assumptions held by many Continuum of Care
supportive housing providers who regard the
chronically homeless as “not housing ready.”
More important, the residential stability
achieved by the experimental group challenges
long-held (but previously untested) clinical assumptions regarding the correlation between
mental illness and the ability to maintain an
apartment of one’s own. Given that all study
participants had been diagnosed with a serious
mental illness, the residential stability demonstrated by residents in the Housing First program—which has one of the highest independent housing rates for any formerly homeless
population—indicates that a person’s psychiatric
diagnosis is not related to his or her ability to
obtain or to maintain independent housing.
Thus, there is no empirical support for the practice of requiring individuals to participate in psychiatric treatment or attain sobriety before
being housed.
Participants’ ratings of perceived choice—
one of the fidelity dimensions of the Housing
First program—show that tenants at Pathways
experience significantly higher levels of control
and autonomy in the program. This experience
may contribute to their success in maintaining
housing and to most consumers’ choice to participate in treatment offered by the ACT team
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
1
Proportion of Time Spent Homeless
0.9
0.8
Experimental
Control
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Baseline
6
12
18
24
Months
FIGURE 2—Proportion of time spent homeless: baseline–24 months.
1
Proportion of Time Stably Housed
0.9
0.8
0.7
0.6
Experimental
Control
0.5
0.4
0.3
0.2
0.1
0
Baseline
6
12
18
24
Months
FIGURE 3—Proportion of time stably housed: baseline–24 months.
after they were housed. In addition, contrary
to the fears of many providers and policymakers, housing consumers without requiring sobriety as a precondition did not increase the
use of alcohol or drugs among the experimental group compared with the control group.
Providing housing first may motivate consumers to address their addictions to keep
their housing, so that providing housing before
treatment, may better initiate and sustain the
recovery process.
Our findings indicate that ACT programs
that combine a consumer-driven philosophy
with integrated dual diagnosis treatment based
on a harm-reduction approach positively affect
residential stability and do not increase substance use or psychiatric symptoms. In addition,
because the ACT teams were providing ser-
April 2004, Vol 94, No. 4 | American Journal of Public Health
vices directly, substance abuse treatment services use was significantly lower for Housing
First residents than for Continuum of Care residents. Because treatment for substance abuse is
required, along with sobriety, by the Continuum of Care model, it is not surprising that individuals in the control group show greater use
of treatment services. However, despite the
control group’s higher use of services, their levels of alcohol or drug use were not different
from those of the experimental group. This disconnect between drug treatment services use
and levels of drug use suggests that the control
group may be using treatment facilities as
short-term housing.
One limitation of the study is that selfreports of the use of alcohol and drugs and
treatment services can be susceptible to reporting bias. Several studies have shown that
among people who are homeless and dually
diagnosed, there is a high rate of discrepancy
between self-reports and client observation for
substance use and for utilization of substance
abuse treatment services.33,34 Memory error,
nondisclosure, social desirability concerns, and
intentional misrepresentation can lead to reporting errors. Powerful systemic reasons for
underreporting also exist. For example, participants enrolled in Continuum of Care residential programs, for which sobriety is mandatory,
may be inclined to underreport the amount of
drugs and alcohol consumed out of fear that
such information may reach a caseworker or
staff member and lead to the loss of their
housing. Errors in self-reporting could be reduced if other measures (e.g., case manager’s
reports, laboratory drug tests) could be incorporated into a multiple-measure data report.
In conclusion, the outcomes achieved provide grounds for the rejection of the erroneous
assumptions underlying the ubiquitous Continuum of Care model, the elimination of treatment requirements as a precondition for housing, and the support of initiatives adopting a
Housing First approach to end homelessness
and increase integration into the community
for individuals with psychiatric disabilities living
on our streets.
About the Authors
The authors are with Pathways to Housing, Inc, New
York, NY.
Tsemberis et al. | Peer Reviewed | Research and Practice | 655
Average No. Substance Abuse Treatment Services Used
 RESEARCH AND PRACTICE 
a program for homeless persons labeled psychiatrically
disabled. Hum Organ. 1998;57:8–20.
2.5
17. Tsemberis SJ, Moran L, Shinn M, Asmussen SM,
Shern DL. Consumer preference programs for individuals who are homeless and have psychiatric disabilities:
a drop-in center and a supported housing program. Am
J Community Psychol. 2003;32:305–317.
Experimental
Control
2
18. Shern DL, Tsemberis S, Anthony W, et al. Serving
street-dwelling individuals with psychiatric disabilities:
outcomes of a psychiatric rehabilitation clinical trial.
Am J Public Health. 2000;90:1873–1878.
1.5
1
19. Teague GB, Bond GR, Drake RE: Program fidelity
in assertive community treatment: development and
the use of a measure. Am J Orthopsychiatry. 1998;68:
216–232.
0.5
20. Stein LI, Santos AB. Assertive Community Treatment of Persons with Severe Mental Illness. New York,
NY: WW Norton; 1998.
0
Baseline
6
12
18
24
21. Tsemberis S, Asmussen S. From streets to homes:
Pathways to Housing consumer preference supported
housing model. Alcohol Treatment Q. 1999;17:113–131.
Months
FIGURE 4—Average number of substance abuse treatment services used: baseline–24 months.
Requests for reprints should be sent to Sam Tsemberis,
Pathways to Housing, 55 West 125th St, 10th Floor, New
York, NY 10027 (e-mail: [email protected]
org).
This article was accepted April 30, 2003.
Contributors
S. Tsemberis oversaw all aspects of the study and
preparation of the article. L. Gulchur completed data
collection and the statistical analysis. M. Nakae assisted
with data analysis and literature review.
Acknowledgments
This study was funded in part by Substance Abuse and
Mental Health Services Administration (SAMHSA)
grant 4UD9SM51970–03–2, SAMHSA/Center for
Substance Abuse Treatment grant 1KD1TI12548–01,
and the New York State Office of Mental Health grant
C005345.
We thank Marybeth Shinn, Ana Stefancic, Ronni Michelle Greenwood, and Nicole Schafer as well as the
study participants for their assistance.
Human Participant Protection
The protocol was approved by the institutional review
boards of Pathways to Housing, Inc. and New York
University. Informed consent was obtained from all
participants.
References
1. Bloomberg plans more housing aid for the homeless. New York Times. June 18, 2002:A1
2. Hopper K, Jost J, Hay T, et al. Homelessness,
mental illness and the institutional circuit. Psychiatr
Serv. 1997;48:659–665.
3. Sullivan G, Burnam A, Koegel P. Pathways to
homelessness among the mentally ill. Social Psychiatry.
2000;35:444–450.
4. Drake RE, McHugo GJ, Clark RE, et al. Assertive
community treatment for patients with co-occurring severe mental illness and substance use disorder: a clinical trial. Am J Orthopsychiatry. 1998;68:201–215.
5. Johnson TP, Freels SA, Parsons JA, et al. Substance abuse and homelessness: social selection or social adaptation? Addiction. 1997;92:437–445.
6. McCarty D, Argeriou M, Huebner RB, et al. Alcoholism, drug abuse, and homelessness. Am Psychol.
1991;46:1139–1148.
7. Anthony WA, Blanch A. Supported employment
for persons who are psychiatrically disabled: an historical and conceptual perspective. Psychosoc Rehabil J.
1989;12:55–81.
8. Carling PJ. Housing and supports for persons with
mental illness: emerging approaches to research and
practice. Hosp. Community Psychiatry. 1993;44:
439–449.
9. Ridgeway P, Zipple AM. Challenges and strategies
for implementing supported housing. Psychosoc Rehabil
J. 1990;13:115–120
10. Kuhn R, Culhane DP. Applying cluster analysis to
test a typology of homelessness by pattern of shelter
utilization: results form the analysis of administrative
data. Am J Community Psychol. 1998;26:207–232.
11. Hurlburt HS, Hough RL, Wood PA. Effects of substance abuse on housing stability and homeless mentally ill persons in supported housing. Psychiatr Serv.
1996;47:731–736.
12. Minkoff K. Developing standards of care for individuals with co-occurring psychological and substance
use disorders. Psychiatr Serv. 2001;52:597–599.
13. Howie the Harp. Independent living with support
services: the goals and future for mental health consumers. Psychosoc Rehabil J. 1990;13:85–89.
14. Asmussen SM, Romano J, Beatty P, et al. Old answers for today’s problems: helping integrate individuals who are homeless with mental illnesses into existing
community-based programs. Psychosoc Rehabil J. 1994;
17:17–34.
15. Osher FC, Drake RE. Reversing a history of
unmet needs: approaches to care for persons with cooccurring addictive and mental disorders. Am J Orthopsychiatry. 1996;66:4–11.
16. Lovell AM, Cohn S. The elaboration of “choice” in
656 | Research and Practice | Peer Reviewed | Tsemberis et al.
22. Inciardi JA, Harrison LD. Introduction: the concept
of harm reduction. In: Inciardi JA, Harrison LD, eds.
Harm Reduction: National and International Perspectives.
Thousand Oaks, CA: Sage Publications, 2000:2–19.
23. Prochaska JO, DiClemente CC, Norcross JC. In
search of how people change: applications to addictive
behaviors. Am Psychol. 1992;47:1102–1114.
24. Lipton FR, Siegel C, Hannigan A, Samuels J,
Baker S. Tenure in supportive housing for homeless
persons with severe mental illness. Psychiatr Serv.
2000;51:479–486.
25. Diagnostic and Statistical Manual of Mental Disorders. 4th ed. Washington DC: American Psychiatric Association; 1994.
26. Srebnik D, Livingston J, Gordon L, et al. Housing
choice and community success for individuals with serious and persistent mental illness. Community Ment
Health J. 1995;31:139–152.
27. New Hampshire Dartmouth Psychiatric Research
Center. Residential Follow-Back Calendar. Lebanon, NH:
Dartmouth Medical School; 1995.
28. New Hampshire Dartmouth Psychiatric Research
Center. Drug and Alcohol Follow-Back Calendar.
Lebanon, NH: Dartmouth Medical School; 1995.
29. Sobell LC, Sobell MB, Leo GI, et al. Reliability of
a timeline method: assessing normal drinker’s reports
of recent drinking and a comparative evaluation across
several populations. Br J Addict. 1988;83:393–402.
30. McLellan AT, Alterman AI, Woody GE, et al.
Treatment Services Review. Philadelphia, PA: University
of Pennsylvania; 1992.
31. Ciarolo JA, Edwards DW, Kiresuk TJ, Newman
FL, Brown TR. Colorado Symptom Index. Washington,
DC: National Institute of Mental Health; 1981.
32. Cohen J, Cohen P. Applied Regression/Correlation
Analysis for Behavioral Sciences. New York, NY: John
Wiley & Sons; 1983
33. Calsyn RJ, Morse GA, Klinkenberg WD, et al. Reliability and validity of self-report data of homeless
mentally ill individuals. Eval Program Plann. 1997;20:
47–54.
34. Goldfinger SM, Schutt RK, Seidman LJ, et al. Selfreport and observer measure of substance abuse among
homeless mentally ill persons in the cross-section and
over time. J Nerv Ment Dis. 1996;184:667–672.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Housing Characteristics and Children’s Respiratory
Health in the Russian Federation
| John D. Spengler, PhD, Jouni J. K. Jaakkola, MD, DSc, PhD, Helen Parise, ScD, Boris A. Katsnelson, MD, DSc, Larissa I. Privalova, MD, DSc,
and Anna A. Kosheleva, MS
Numerous studies have associated indoor
housing factors with increased prevalence of
respiratory symptoms in children as well as
adults.1–4 Yet there are few studies from the
Russian Federation or the former Soviet
Union, where a large percentage of the population live in concrete apartment buildings,
in which water and heat are supplied by district heating systems and gas is used for
cooking. Furthermore, most Russian families
benefit from a state health care system that
provides pre- and postnatal care.5 It therefore behooves us to examine housing factors
such as smoking, moisture, indoor combustion sources (e.g., gas cooking, tobacco use),
and ventilation on the health of school-aged
children living in contemporary Russian
housing.
METHODS
Study Population
The study population comprised 5951
8- to 12-year-old children in 9 Russian cities.
Eight cities of the Sverdlovsk Oblast region
and the city of Cherepovets in the Vologda
Oblast participated in the study. Cities were
selected to participate in a cross-sectional
study of air pollution and children’s health. In
4 cities, 2 areas were selected—1 to represent
a more polluted area and 1 a less polluted
area; in 5 cities, only 1 area was included.
Within each area, 1 or 2 elementary
schools were selected for participation. The
principals of the selected schools were informed about the study and agreed to participate. Teachers were given verbal and written
instructions, questionnaires, envelopes, and
forms to record questionnaire distribution
and collection. Parents were invited to a parents’ night where teachers explained the
study and the conditions of consent. Teachers
were instructed not to urge parents to fill out
the questionnaire, as compliance was strictly
Objectives. We studied housing characteristics, parental factors, and respiratory
health conditions in Russian children.
Methods. We studied a population of 5951 children from 9 Russian cities, whose
parents answered a questionnaire on their children’s respiratory health, home environment, and housing characteristics. The health outcomes were asthma conditions, current wheeze, dry cough, bronchitis, and respiratory allergy.
Results. Respiratory allergy and dry cough increased in association with the home
being adjacent to traffic. Consistent positive associations were observed between some
health conditions and maternal smoking during pregnancy, many health conditions and
lifetime exposure to environmental tobacco smoke (ETS), and nearly all health conditions and water damage and molds in the home.
Conclusions. Vicinity to traffic, dampness, mold, and ETS are important determinants
of children’s respiratory health in Russia. (Am J Public Health. 2004;94:657–662)
voluntary. Parents who wished to participate
completed the questionnaire either in the
classroom or at home, and returned it (via the
child) to the teacher in a sealed envelope.
There was a 98% response rate.
The questionnaires, which were identified
by identification number only, were reviewed by the field coordinators for quality
assurance and for encoding written replies.
The questionnaires were then sent to the
Harvard School of Public Health for optical
scanning using internal consistency checks to
identify questionnaires requiring additional
verification.
The questionnaire had been modified
from previous European and North American questionnaires, which originated from
respiratory health questionnaires of the British Medical Research Council and the American Thoracic Society.6 The questionnaire was
composed of the following: the child’s personal characteristics; the child’s respiratory
health, presence of atopic diseases, and number of infections during the past year; parents’ education and job category (as an indicator of socioeconomic status); parents’
smoking habits as well as respiratory and allergic diseases; and details of the home environment and building characteristics. Details
of health, housing characteristics, and socio-
April 2004, Vol 94, No. 4 | American Journal of Public Health
economic factors were adjusted for the current Russian conditions.
Health Outcomes
Twenty health outcomes were derived from
the questionnaire (Table 1). We focused on
the children’s current symptoms and conditions of allergy and eye irritation. Some of the
health outcomes were composite variables derived from multiple questions. Believing that
asthma may be underdiagnosed or not clearly
remembered by the parent, we defined a
composite variable called “asthma-like symptoms” that included wheezing and shortness
of breath. Other outcomes examined the consistency of associations. When parents reported hearing their child wheeze for 3 or
more consecutive days or using medication
for wheezing in the past year, the child was
classified as having severe wheezing within
the last 12 months. A child was classified as
having current wheeze if within the past 12
months their wheezing caused shortness of
breath, woke them at night, or occurred with
exercise in addition to any of the conditions
described above for severe wheezing.
Exposure Assessment
Exposure assessment was based on questionnaire information on housing character-
Spengler et al. | Peer Reviewed | Research and Practice | 657
 RESEARCH AND PRACTICE 
posure. We calculated crude ORs and 95%
confidence intervals (CIs) based on the Mantel-Haenszel test statistics. We estimated the
adjusted ORs in logistic regression analysis.
The ORs were adjusted for the covariates
described above. The results from the adjusted logistic regression analyses are reported in this paper.
TABLE 1—Prevalence of Respiratory Symptoms and Other Conditions
Term
Ever asthma
Current asthma
Severe asthma
Ever wheeze
Current wheeze
Current severe wheeze
Current asthma-like symptoms
Severe asthma-like symptoms
Ever cough
Persistent cough
Ever phlegm
Persistent phlegm
Dry cough
Persistent dry cough
Upper respiratory infection
Severe upper respiratory infection
Current bronchitis
Any allergy
Respiratory allergy
Eye irritation
Description
Frequency, %
Ever told by doctor that child has asthma
Doctor-diagnosed asthma with parental reporting of shortness of breath
or wheeze, or use of asthma medication within past 12 months
Hospitalization for asthma/regular medication for asthma within
past 12 months
Wheezing heard from distance without a cold
Wheezing heard from distance with or without a cold, shortness of
breath with wheezing, awakening at night by wheezing, wheezing
with exercise, or use of medication or hospitalization within past
12 months for wheezing
Wheezing heard from a distance for ≥ 3 consecutive days,
hospitalization or medication use for asthma within past 12 months
Asthma symptoms or asthma medication use, awakening by asthma,
wheezing upon exercise, or hospital care for wheezing within
past 12 months
Current asthma-like symptoms with routine medication use within
past 12 months
Usual cough day or night
Cough ≥ 3 consecutive months within past 12 months
Wet cough or phlegm produced without a cold
Wet cough or phlegm ≥ 3 consecutive months within past 12 months
Ever cough without phlegm
Persistent cough without phlegm for ≥ 3 consecutive months within
past 12 months
Acute upper respiratory infection within past 12 months
Two or more acute upper respiratory infections within past 12 months
Doctor-diagnosed bronchitis within past 12 months
Doctor-diagnosed allergy, reported hay fever, or pollenosis
Hay fever or doctor-diagnosed allergies to airborne substances (e.g., dust,
animals, molds, pollens, air pollution, environmental tobacco smoke)
Eye irritation sometimes
1.9
1.5
istics. Questions inquired about the age of
the building, the type of construction, and
its proximity to traffic. Apartment-related
factors included heating and cooking methods, presence of ventilation, and geographic
orientation and size. Respondents reported
on smoking within the apartment, water
damage, presence of mold, and the number
of occupants. Ancillary information on
cleaning frequency, parental occupational
exposures to chemicals, and parental income, as well as variables related to nonrespiratory health outcomes, was collected.
Density indicators were derived from information on apartment size and number of
occupants.
0.8
3.1
13.4
RESULTS
10.1
10.3
2.2
25.7
5.5
7.0
1.5
17.8
3.6
76.9
24.2
8.3
33.2
8.0
20.8
Covariates Used for Adjustment
Univariate analyses explored several potential confounders. Gender, age, preterm birth,
parental atopy, parents’ education in a specialty field beyond high school, and smoking
variables were used as core adjusting covariates in logistic regression unless the variable
of interest was smoking itself. Additional
models that included income, presence of
furry pets, and sharing a bedroom as adjusting variables were explored but not presented
because our basic findings were not altered.
Statistical Methods
The odds ratio (OR) was used as a measure of effect between the outcome and ex-
658 | Research and Practice | Peer Reviewed | Spengler et al.
The response rate across schools varied
from 96% to 99% and averaged 98%
overall. Half of the buildings/homes were
constructed within the past 20 years, and
70% of the buildings were concrete highrises. Eighty-five percent of the respondents
lived in single-family apartments, of which
50% were smaller than 40 m2. Seventyfour percent of the children shared a bedroom. District heating plants provided heat
and hot water requirements for 95% of the
apartments/homes. Only 5% had a combustion heat source within their home. Gas
was the cooking fuel for 80% of the homes,
and 73% had no mechanical means of
venting exhaust. Only 5% of the housing
units had gas water heaters. Sixty percent
of the respondents reported that their
apartments did not face roadways.
Almost 60% of the families reported having
a furry pet at home. Toxic substance exposures
of parents at the workplace were reported for
21.7% of the children, and 1.9% had parents
with the potential for bringing toxic material
home as they did not change their clothes at
their workplace. Cleaning of homes was infrequent; nearly 80% said they cleaned less than
once per month, with only 3.2% cleaning
weekly. Water damage was reported for
22.4% of the living units, and 10.4% reported
water damage within the past 12 months. The
appearance or detection of molds within the
past 12 months occurred in 10% of the homes.
A small percentage of mothers (4.2%) admitted smoking during pregnancy. Environmental
tobacco smoke (ETS) exposure at home at various stages of the child’s life—less than 2 years
of age, 2 to 6 years of age, and currently—
occurred for 45%, 51%, and 46% of the children, respectively.
Additional variables were derived from
smoking responses, occupancy, and size of
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 2—Adjusted ORs for Respiratory Symptoms and Housing Conditions
Current doctor-diagnosed asthma
Current wheeze
Current asthma-like symptoms
Persistent cough
Persistent phlegm
Current dry cough
Persistent dry cough
Current upper respiratory infection
Current bronchitis
Any allergy
Respiratory allergy
Eye irritation
Building Age
(> 40 y vs ≤ 10 y)
OR (95% CI)
Building Material
(Concrete vs Wood)
OR (95% CI)
Density of Children
(4th vs 1st Quartile)
OR (95% CI)
Area of Residence
(> 60 m2 vs < 25 m2)
OR (95% CI)
Traffic Near Apartment
(Medium vs None)
OR (95% CI)
Cooking Fuel
(Gas vs No Gas)
OR (95% CI)
1.59 (0.75, 3.33)
1.37 (1.04, 1.80)
1.41 (1.04, 1.92)
1.78 (1.12, 2.60)
1.83 (0.83, 4.06)
1.16 (0.91, 1.48)
1.65 (0.98, 2.78)
1.02 (0.82, 1.27)
1.24 (0.90, 1.73)
0.89 (0.74, 1.08)
0.97 (0.69, 1.36)
0.93 (0.74, 1.16)
1.28 (0.46, 3.62)
1.00 (0.72, 1.38)
1.08 (0.74, 1.56)
1.24 (0.73, 2.10)
0.69 (0.30, 1.57)
1.13 (0.83, 1.54)
1.40 (0.70, 2.79)
0.94 (0.72, 1.23)
0.98 (0.66, 1.45)
1.34 (1.05, 1.72)
2.61 (1.45, 4.72)
1.20 (0.91, 1.60)
1.40 (0.50, 3.86)
1.34 (0.92, 1.96)
1.31 (0.82, 2.09)
2.55 (1.44, 4.51)
3.49 (1.29, 9.47)
1.84 (1.31, 2.57
2.08 (1.03, 4.17)
1.28 (0.95, 1.72)
1.31 (0.82, 2.08)
0.87 (0.66, 1.14)
1.23 (0.78, 1.93)
0.91 (0.67, 1.25)
0.60 (0.24, 1.48)
0.99 (0.71, 1.37)
1.16 (0.86, 1.69)
0.48 (0.30, 0.79)
0.27 (0.11, 0.72)
0.54 (0.40, 0.72)
0.66 (0.36, 1.19)
0.90 (0.68, 1.17)
0.93 (0.60, 1.44)
1.04 (0.81, 1.32)
0.99 (0.65, 1.52)
0.76 (0.58, 0.99)
0.89 (0.41, 1.91)
1.22 (0.94, 1.57)
1.28 (0.96, 1.69)
2.27 (1.64, 3.14)
2.21 (1.20, 4.06)
1.30 (1.04, 1.63)
2.46 (1.66, 3.62)
1.02 (0.83, 1.25)
1.19 (0.87, 1.61)
1.09 (0.91, 1.32)
1.41 (1.04, 1.90)
1.31 (1.06, 1.61)
2.28 (1.04, 5.01)
1.06 (0.86, 1.31)
1.19 (0.94, 1.52)
1.19 (0.85, 1.65)
0.85 (0.48, 1.53)
1.04 (0.86, 1.25)
1.31 (0.86, 1.99)
0.91 (0.77, 1.08)
1.13 (0.87, 1.47)
1.06 (0.92, 1.23)
1.08 (0.83, 1.40)
0.98 (0.83, 1.16)
Note. OR = odds ratio; CI = confidence interval.
the living unit. Exposure to ETS sometime
during the child’s life occurred for 63% of
the children. We hypothesized that internal
sources of air pollution, including airborne
pathogens, might result in higher concentrations that vary inversely to volume of the residence or directly with crowding factors,
based on occupant density. The area of the
residence was separated by quartiles as a
proxy for volume. The number of children
and total number of occupants were divided
by reported floor area and divided into quartiles to create 2 indicators of crowding.
The ORs for housing conditions are shown
in Table 2. The results suggest that living in
apartments more than 40 years old might increase the risk of wheeze. Similar results were
seen for ever phlegm, ever cough, and persistent cough. The ORs for buildings aged 10 to
20 and 20 to 40 years, compared with buildings less than 10 years old, showed no evidence for a trend by age of residence. Only allergy (any or respiratory) showed an
association with concrete versus wooden
houses or apartments. Reporting of cough and
phlegm conditions was significantly higher for
more crowded housing and significantly less
for larger residences. Findings are consistent
when density is calculated by total number of
occupants or just total children per area of the
home. There were no observable trends over
the 4 quartiles, and higher prevalences were
observed for the quartile of most densely
crowded residences. A protective effect of a
larger apartment/home was seen only for
cough and phlegm symptoms and was more
pronounced for the larger-area apartments
(> 60 m2) versus the smaller units (< 25 m2).
Those reporting a self-defined medium exposure to traffic outside their residence had
higher prevalence of both respiratory allergy
and eye irritation (nonsignificant). However,
the cough and phlegm symptoms showed a
significant positive association with traffic, with
an apparent trend from light to medium traffic.
Health outcomes were examined for internal
heating, gas cooking, gas water heaters, and
whether or not exhaust ventilation made any
difference in response rates. Only 12 families
had unvented gas water heaters, so these results
were not reported. Although gas cooking and a
combustion heating device were positively associated with increased symptoms, none reached
significance. Having some form of exhaust ventilation reduced the risk for respiratory allergy
and dry cough, but only the latter was significant (OR=0.77 [95% CI=0.64, 0.93]). For
completeness, we examined other symptoms for
the influence of combustion and exhaust ventilation and found that doctor-diagnosed asthma
and current asthma had a significant positive
association with gas cooking. The adjusted ORs
were 2.28 (95% CI=1.04, 5.01) for current
asthma and 2.12 (95% CI=1.09, 4.11) for
April 2004, Vol 94, No. 4 | American Journal of Public Health
doctor-diagnosed asthma. Although the severity
of asthma and the various wheeze-related outcomes all had positive adjusted ORs, none were
significant at the 95% CI.
In examining all smoking variables, we
found for the most part that all adjusted ORs
across all outcomes showed positive associations with smoking exposure variables. Current
dry cough showed significant associations, as
did ever cough, persistent cough, and persistent
dry cough. Experiencing a respiratory tract infection within the past year was associated with
ETS exposure sometime in the child’s life but
not necessarily with current smoking in the
home. Doctor-diagnosed bronchitis was
strongly associated with lifetime ETS exposure
(OR=1.26 [95% CI=1.10, 1.44]), but not for
bronchitis within the past year. Table 3 presents
the adjusted ORs and 95% CIs for 12 of the
health variables and 2 of the smoking variables
(smoking during pregnancy and the composite
variable of any ETS exposure during the child’s
life). Other smoking variables similar to smoking during pregnancy showed few statistically
positive associations, unlike the composite variable of ever being exposed to ETS.
The housing conditions with the strongest
and most consistent associations with health
outcomes were reported moisture (water
damage) and the presence of molds on surfaces. Table 4 presents the adjusted ORs for
reported water and mold conditions within
Spengler et al. | Peer Reviewed | Research and Practice | 659
 RESEARCH AND PRACTICE 
with asthma symptoms (OR = 1.06 [95% CI =
0.78, 1.44]).
Examining the relationship between parental exposures to toxic material at work and
their children’s symptoms yielded interesting
results. Even though only about 2% of the responding parents had workplace exposures,
there were significant associations with dry
cough (OR = 2.35 [95% CI = 1.54, 3.59]),
persistent dry cough (OR = 2.18 [95% CI =
1.05, 4.55], and severe wheezing (OR = 1.76
[95% CI = 1.03, 3.02]). Reported frequency
of house cleaning revealed no consistent or
significant relationships.
TABLE 3—Adjusted ORs for Respiratory Symptoms and Tobacco Smoke Exposure During
Child’s Lifetime
Tobacco Smoke Exposure
Current doctor-diagnosed asthma
Current wheeze
Current asthma-like symptoms
Persistent cough
Persistent phlegm
Current dry cough
Persistent dry cough
Current upper respiratory infection
Current bronchitis
Any allergy
Respiratory allergy
Eye irritation
In Utero OR (95% CI)
Lifetime Exposure OR (95% CI)
2.07 (0.85, 5.03)
1.41 (0.97, 2.06)
1.44 (0.95, 2.16)
0.95 (0.51, 1.75)
0.54 (0.13, 2.29)
1.24 (0.86, 1.78)
1.17 (0.58, 2.38)
0.72 (0.52, 1.00)
1.57 (1.00, 2.45)
0.90 (0.66, 1.23)
1.10 (0.66, 1.83)
1.09 (0.77, 1.54)
1.22 (0.75, 2.03)
1.12 (0.94, 1.34)
1.15 (0.95, 1.40)
1.34 (1.02, 1.75)
1.31 (0.78, 2.19)
1.35 (1.15, 1.58)
1.53 (1.08, 2.16)
1.21 (1.06, 1.38)
1.05 (0.85, 1.29)
1.18 (1.04, 1.33)
1.04 (0.84, 1.28)
1.22 (1.06, 1.41)
DISCUSSION
Note. OR = odds ratio; CI = confidence interval.
TABLE 4—Adjusted ORs for Respiratory Symptoms and Water Damage and Presence of
Molds Within the Past 12 Months
Current doctor-diagnosed asthma
Current wheeze
Current asthma-like symptoms
Persistent cough
Persistent phlegm
Current dry cough
Persistent dry cough
Current upper respiratory infection
Current bronchitis
Any allergy
Respiratory allergy
Eye irritation
Water Damage OR (95% CI)
Presence of Molds OR (95% CI)
1.37 (0.69, 2.70)
1.53 (1.19, 1.95)
1.77 (1.36, 2.30)
1.51 (1.06, 2.16)
2.15 (1.18, 3.93)
1.35 (1.08, 1.69)
1.33 (0.85, 2.09)
1.23 (0.98, 1.55)
1.52 (1.14, 2.03)
1.26 (1.05, 1.52)
1.30 (0.95, 1.77)
1.21 (0.98, 1.50)
2.82 (1.63, 4.88)
1.52 (1.19, 1.94)
1.98 (1.53, 2.55)
1.88 (1.35, 2.63)
2.46 (1.38, 4.38)
1.40 (1.12, 1.76)
1.53 (0.99, 2.35)
1.74 (1.35, 2.25)
1.70 (1.28, 2.27)
1.51 (1.25, 1.82)
1.50 (1.11, 2.02)
1.42 (1.15, 1.76)
Note. OR = odds ratio; CI = confidence interval.
the last 12 months. Prevalence of symptoms
increased from 35% to almost 100% when
mold was present in the home. The association was slightly stronger for mold conditions
than for just water damage. All health outcomes were more strongly associated with reported mold and water damage within the
past 12 months compared with ever having
water damage or molds in the living unit. The
association was weaker for molds being reported in the child’s bedroom.
Having any furry pet was strongly protective for respiratory allergy (OR = 0.61 [95%
CI = 0.50, 0.74]) but less so for severe wheezing (OR = 0.84 [95% CI = 0.74, 1.01]) and
current bronchitis (OR = 0.86 [95% CI =
0.67, 1.00]). Having a furry pet was strongly
negatively associated with current asthma
(OR = 0.40 [95% CI = 0.25, 0.64]), whereas
having a cat was specifically associated with
higher rates of doctor-diagnosed asthma
(OR = 3.29 [95% CI = 1.01, 10.72]) but not
660 | Research and Practice | Peer Reviewed | Spengler et al.
Consistent with similar health surveys conducted in the United States and Europe, conditions of mold and dampness in living areas are
strongly associated with increased respiratory
symptoms. In an examination of all published
literature, a Nordic scientific review panel concluded that the presence of dampness in a
home increased the reporting of cough,
wheeze, and respiratory symptoms by 40%
over a reference population.3 The risk appears
to be similar for Russian housing. However, the
prevalence of moisture and mold in the housing stock is approximately half of the prevalence reported for surveys done in the United
States and Canada.7,8 Jacob et al.9 showed that
high counts of Cladosporium and Aspergillus
spores in house dust were associated with increased risk of allergic sensitization. Their results suggest that higher spore counts, particularly in the winter, are likely to increase the
prevalence of allergic symptoms in children.
Cook and Strachan10 conducted pooled
analysis of ORs for parental smoking on
asthma, wheeze, chronic cough, chronic
phlegm, and shortness of breath symptoms in
children exposed to ETS. Our Russian results
for ever cough, persistent cough, dry cough,
and persistent dry cough are similar to the
pooled ORs for cough (OR = 1.35 [95%
CI = 1.13, 1.62]). Also, our findings for
phlegm were similar to the pooled ORs of
1.31 (95% CI = 1.08, 1.59). Asthma and
wheeze, although both significantly associated
with parental smoking in the pooled analysis,
were not significantly associated with any
measure of ETS exposure over the child’s life.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Gilliland et al.11 reported that in utero exposure to maternal smoking without subsequent
postnatal ETS exposure significantly increased
the association with doctor-diagnosed asthma,
asthma symptoms, and asthma severity later in
a child’s life, as well as most of the wheezing
outcomes. Our ORs for asthma and wheeze
outcomes were all positively associated with
smoking during pregnancy (approximately 2.0)
but did not reach P < 0.05 significance. Given
the lack of specificity to the smoking questions
asked in this survey, it is not possible to ascertain the separate influence of maternal versus
paternal smoking or even age-related responses seen in other studies. Gilliland et al.11
similarly showed that current and previous
ETS exposure was not associated with asthma
prevalence but was consistently associated
with various wheezing variables.
Apelberg et al.12 performed a meta-analysis
of the studies on the effect of early exposure to
household pets on the development of asthma
and asthma-related symptoms. Inappropriate
time sequence of the exposure and outcome information, typical for cross-sectional studies,
was an important source of heterogeneity and
an indication of potential selection bias. In studies ensuring a meaningful temporal relation between exposure and outcome, the pooled risk
estimates for both asthma (fixed-effects OR=
1.11 [95% CI=0.98, 1.25]; P=0.04) and
wheeze (OR=1.19 [95% CI=1.05, 1.35]; P=
0.03) indicated a small effect. However, the effect was limited to studies with a median study
population age greater than 6 years. In younger
children, the effect appeared protective for
wheezing (OR=0.80 [95% CI=0.59, 1.08];
P=0.38). The authors concluded that the observed lower risk among exposed compared to
unexposed young children is consistent with a
protective effect in this age group, but could
also be explained by selection bias.
In a prospective study of asthma incidence
in adolescents, McConnell et al.13 reported a
relative risk of 1.6 (95% CI = 1.0, 2.5) for having a furry pet at home. The present study was
cross-sectional and did not inquire precisely
when the pet had been present. Therefore, the
negative associations between the presence of
pets and the risk of asthma and allergies could
be a result of either avoidance or removal of
pets in families with children allergic to respiratory allergens and with asthma problems.
Exposure to nitrogen dioxide (NO2) from gas
cooking is a common experience for the majority of children in this survey. From studies
that measured NO2 indoors, it can be inferred
that concentrations will be higher in the absence of exhaust vents. Yet examining possible
interactive effects for cooking fuel and ventilation offered no evidence for increased association with the inferred exposure gradient (e.g.,
not using gas but having ventilation versus
having gas but not having ventilation). Garrett
et al.14 reported that gas stoves increased the
risk of respiratory symptoms in children
(OR = 2.3 [95% CI = 1.0, 5.2]), whereas the association with direct measures of NO2 was
marginal. Shima and Adachi15 reported that
the prevalence of bronchitis wheeze and
asthma significantly increased with indoor NO2
exposures among girls but not among boys.
They also showed that wheeze and asthma incidence were associated with outdoor NO2 but
not indoor NO2. Examining the effects of NO2
from gas heaters in school rooms, Pilotto et
al.16 found significant increases in sore throats,
colds, and absences from school when hourly
peak exposures exceeded 80 ppb compared
with background levels of 20 ppb. In a large
study of respiratory infections among 1000 infants in Albuquerque, New Mexico, Samet et
al.17 found no associations for either gas stove
or NO2 levels measured in the kitchen or the
child’s bedroom.
This study of Russian schoolchildren and
housing factors poses some interesting observations for further investigation. Only 40% of
the respondents reported either light or medium traffic outside their homes/apartments.
Persistent cough, phlegm, and dry cough, as
well as the prevalence of severe upper respiratory infection in children, were positively
associated with medium traffic loadings compared with not living along any roadways.
Clustered apartment complexes removed
from roadways are common in many Russian
cities; it is a situation unlike that of any study
reporting associations with asthma and respiratory symptoms for children living close to
heavily traveled roads. In US, Western European, and Japanese studies, the reference
group always has approximate exposure to
some road traffic.15,18–24 In this Russian study,
none of the asthma or wheeze variables
showed an association with subjective report-
April 2004, Vol 94, No. 4 | American Journal of Public Health
ing of traffic exposure. Nevertheless, the possible association of vehicle exhaust on chronic
cough and phlegm cannot be dismissed.
Just 1.9% of the children had 1 or both
parents reporting occupational contact with
potentially toxic substances and not leaving
their work clothes at the job site. Another
19.75% had parents who might be exposed
but leave their clothing at work. Consistently
positive associations were found for wheezing
(severe and current) and most of the coughs
(persistent dry cough, dry cough, usual cough)
as well as asthma-like symptoms and general
atopy. These observations suggest that compounds may be carried home on clothing or
absorbed in fibers of clothing, leading to children being exposed at home. Many metallurgical and chemical production facilities are located in Cherepovets and throughout
Sverdlovsk Oblast. It is likely that some parents are heavily exposed to potentially irritating or toxic materials at work.
Although doctor-diagnosed asthma rates for
Russian children were substantially lower than
rates children are currently experiencing in the
West, the rates are consistent with reports from
former Soviet Bloc countries. Jedrychowski et
al.25 reported doctor-diagnosed asthma among
1129 9-year-old children living in Krakow,
Poland, as 1.9% for girls and 2.4% for boys in
1995. The International Study of Asthma and
Allergies in Childhood showed that the variation in the prevalence of asthma and selfreported asthma symptoms between different
countries is striking.26 In these comparisons
the prevalences of all the studied indices of
asthma were lower in Eastern than in Western
Europe. Corresponding figures for asthma
symptoms from video questionnaires were 2%
in Russia and between 12% and 20% in the
United Kingdom, the United States, Canada,
New Zealand, and Australia. Differences in access to health care, diagnostic practice, and environmental and dietary factors are plausible
explanations for the large variation in the
prevalences of asthma between Russia and
Western Europe/North America.
Rates for other conditions and symptoms in
Russian children are comparable to rates reported from studies conducted in the United
States. In the Harvard 24 Cities Study of air
pollution and children’s health in 24 North
American towns,27 33% of the children re-
Spengler et al. | Peer Reviewed | Research and Practice | 661
 RESEARCH AND PRACTICE 
ported some atopy. The reporting of current
asthma symptoms within the past year ranged
from 3% to 11% of the children across 24
communities, with persistent wheeze ranging
from 4% to 12%. For Russian children, asthmalike symptoms were 10%, with a higher rate of
current wheeze (13.4%). Chronic cough in the
24 Cities Study ranged from 4% to 9%, which
is consistent with the 5.5% noted in our study.
Parents reported chronic bronchitis in the past
year at rates of 3% to 10% across the 24 Cities
Study, and a rate of 8.3% for our Russian children. The respiratory symptoms in children associated with ETS exposure, water damage,
and presence of molds in the Russian housing
study were consistent with reports on housing
conditions in many other countries.
Russian housing is characterized by large,
concrete high-rise structures. Apartments are
similar in layout and size, and are served by
district hot water for heating and gas for
cooking; mechanical ventilation and air conditioning is a rarity. Our study reporting prevalences of conditions and associated health
symptoms provides important insights that
are applicable to millions of children living in
similar housing.
About the Authors
John D. Spengler is with the Environmental Science and
Engineering Program, Harvard School of Public Health,
Boston, Mass. Jouni J. K. Jaakkola is with the Institute of
Occupational Health, The University of Birmingham, Edgbaston, United Kingdom. Helen Parise is with the Department of Mathematics and Statistics, Boston University,
Boston. Boris A. Katsnelson, Larissa I. Privalova, and
Anna A. Kosheleva are with the Ural Region Environmental Epidemiology Center, Ekaterinburg, Russia.
Requests for reprints should be sent to John D. Spengler,
PhD, Environmental Science and Engineering Program,
Harvard School of Public Health, PO Box 15677, Landmark Ctr, Rm 406 W, 401 Park Dr, Boston, MA 02215
(e-mail: [email protected]).
This article was accepted April 16, 2003.
Contributors
J. D. Spengler designed the study and led the analysis
and the writing of the article. J. J. K. Jaakkola helped design and oversee the study and analysis, and participated
in the writing of the article. H. Parise and A. A. Kosheleva
analyzed the data. B. A. Katsnelson and L. I. Privalova
helped design and conduct the study, managed staff,
checked records, and participated in data analysis.
Acknowledgments
This study was supported by a World Bank loan to the
Russian Federation and administered under the environmental epidemiology component of the Centre for
Preparation and Implementation of International Pro-
jects on Technical Assistance, managed by Vladislav
Furman, PhD, with assistance from Victor Kislitsin,
PhD, and Natalia Lebedeva, MD, DSc. Analyses were
partially supported by the National Institute of Environmental Health Sciences (NIEHS) Center for Environmental Health at the Harvard School of Public Health
(grant ES000002); the contents of our analyses are
solely the responsibility of the authors and do not necessarily represent the official views of the NIEHS.
We thank the parents who gave their time to complete questionnaires and the teachers who distributed
and efficiently collected the surveys. We are indebted
to our colleagues in the Ural Region Environmental
Epidemiology Center, who, under the direction of
Sergey Kuzmin, MD, DSc, conducted a successful comprehensive study of air pollution and children’s health
in 9 Russian cities. The advice and contributions of Olga
Malykh, MD, CSc, Boris Nikonov, MD, DSc, and
Vladimir Gurvitch, MD, CSc, were greatly appreciated.
Haluk Ozkaynak, PhD, and Thomas Dumyahn, MS,
both of the Harvard School of Public Health at the time
of the study, contributed substantially to the design and
management of the environmental epidemiology assistance program with Russia.
Human Participant Protection
Parents were informed that participation was voluntary.
No personal identifiers were used in our data files, and
all questionnaires have been destroyed. Data were not
collected from children.
References
1. Office of the Surgeon General. The Health Consequences of Involuntary Smoking: A Report of the Surgeon
General. Rockville, Md: US Public Health Service, Department of Health and Human Services, Office of the
Surgeon General; 1986.
2. Samet JM, Spengler JD, eds. Indoor Air Pollution:
A Health Perspective. Baltimore, Md: Johns Hopkins
University Press; 1991.
3. Bornehag C-G, Blomquist G, Gyntelberg F, et al.
Dampness in buildings and health. Nordic interdisciplinary review of the scientific evidence on associations between exposure to “dampness” in buildings and health
effects (NORDDAMP). Indoor Air. 2001; 11:72–86.
4. Smith KR, Samet JM, Romieu I, Bruce N. Indoor
air pollution in developing countries and acute lower
respiratory infections in children. Thorax. 2000;55:
518–532.
5. Jaakkola JJ, Cherniack M, Spengler JD, et al. Use of
health information systems in the Russian Federation
in the assessment of environmental health effects. Environ Health Perspect. 2000;108:589–594.
6. Ferris BG. Epidemiology Standardization Project
(American Thoracic Society). Am Rev Respir Dis. 1978;
118(6 pt 2):1–120.
7. Spengler J, Neas L, Nakai S, et al. Respiratory
symptoms and housing characteristics. Indoor Air.
1994;4:72–82.
8. Dales RE, Burnett R, Zwanenburg H. Adverse
health effects among adults exposed to home dampness
and molds. Am Rev Respir Dis. 1991;143:505–509.
9. Jacob B, Ritz B, Gehring U, et al. Indoor exposure
to molds and allergic sensitization. Environ Health Perspect. 2002;110:647–653.
10. Cook DG, Strachan DP. Health effects of passive
smoking. 3. Parental smoking and prevalence of respi-
662 | Research and Practice | Peer Reviewed | Spengler et al.
ratory symptoms and asthma in school age children.
Thorax. 1997;52:1081–1094.
11. Gilliland FD, Li YF, Peters JM. Effects of maternal
smoking during pregnancy and environmental tobacco
smoke on asthma and wheezing in children. Am J
Respir Crit Care Med. 2001;163:429–436.
12. Apelberg BJ, Aoki Y, Jaakkola JJ. Systematic review:
Exposure to pets and risk of asthma and asthma-like
symptoms. J Allergy Clin Immunol. 2001;107:455–460.
13. McConnell R, Berhane K, Gilliland F, et al. Indoor
risk factors for asthma in a prospective study of adolescents. Epidemiology. 2002;13:288–295.
14. Garrett MH, Hooper MA, Hooper BM, Abramson
MJ. Respiratory symptoms in children and indoor exposure to nitrogen dioxide and gas stoves. Am J Respir
Crit Care Med. 1998;158:891–895.
15. Shima M, Adachi M. Effect of outdoor and indoor
nitrogen dioxide on respiratory symptoms in schoolchildren. Int J Epidemiol. 2000;29:862–870.
16. Pilotto LS, Douglas RM, Attewell R, Wilson SR.
Respiratory effects associated with indoor nitrogen
dioxide exposure in children. Int J Epidemiol. 1997;26:
788–796.
17. Samet JM, Lambert WE, Skipper BJ, et al. Nitrogen dioxide and respiratory illnesses in infants. Am Rev
Respir Dis. 1993;148:1258–1265.
18. Brunekreef B, Janssen NA, de Hartog J, Harssema H,
Knape M, van Vliet P. Air pollution from truck traffic
and lung function in children living near motorways.
Epidemiology. 1997;8:298–303.
19. English P, Neutra R, Scalf R, Sullivan M, Waller L,
Zhu L. Examining associations between childhood
asthma and traffic flow using a geographic information
system. Environ Health Perspect. 1999;107:761–767.
20. Oosterlee A, Drijver M, Lebret E, Brunekreef B.
Chronic respiratory symptoms in children and adults
living along streets with high traffic density. Occup Environ Med. 1996;53:241–247.
21. van Vliet P, Knape M, de Hartog J, Janssen N,
Harssema H, Brunekreef B. Motor vehicle exhaust and
chronic respiratory symptoms in children living near
freeways. Environ Res. 1997;74:122–132.
22. Venn AJ, Lewis SA, Cooper M, Hubbard R, Britton J. Living near a main road and the risk of wheezing illness in children. Am J Respir Crit Care Med.
2001;164:2177–2180.
23. Weiland SK, Mundt KA, Ruckmass A, Keil U. Selfreported wheezing and allergic rhinitis in children and
traffic density on street of residence. Ann Epidemiol.
1994;4:243–247.
24. Wjst M, Reitmeir P, Dold S, et al. Road traffic and
adverse effects on respiratory health in children. BMJ.
1993;307:596–600.
25. Jedrychowski W, Flak E, Mroz E, eds. Effects of
Poor Air Quality on the Health of Krakow Children.
Krakow, Poland: Jagiellonian University, Collegium
Medicum; 1998.
26. The International Study of Asthma and Allergies
in Childhood (ISAAC) Steering Committee. Worldwide
variation in prevalence of symptoms of asthma, allergic
rhinoconjunctivitis, and atopic eczema: ISAAC. Lancet.
1998;351:1225–1232.
27. Dockery DW, Cunningham J, Damokosh AI, et al.
Health effects of acid aerosols on North American children: respiratory symptoms. Environ Health Perspect.
1996;104:500–505.
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
Metropolitan-Area Estimates of Binge Drinking
in the United States
| David E. Nelson, MD, MPH, Timothy S. Naimi, MD, MPH, Robert D. Brewer, MD, MSPH, Julie Bolen, PhD, and Henry E. Wells, MS
Alcohol use results in approximately 100 000
deaths each year in the United States, which
makes it the third leading cause of preventable death in the country.1 Binge drinking is an
especially hazardous pattern of alcohol consumption that causes a substantial proportion
of alcohol-related deaths.2–5 Adverse effects of
binge drinking include intentional injuries
(e.g., interpersonal violence and suicide), unintentional injuries (e.g., motor vehicle crashes
and drownings), fetal alcohol syndrome/effect,
and unintended pregnancy.6–8
Binge drinking most commonly occurs
among males, younger persons, and persons
residing in urban or suburban areas,9,10 and
the occurrence is generally higher in the Midwest, the Northeast, and the West.9–11 Recent
studies have shown that binge drinking and its
related health consequences have increased in
the past few years,12–14 as binge drinking increased in 19 states while it declined in only 3
states from 1991 through 1999.13
Although national and state-based surveys
obtain data on alcohol measures on a regular
basis, there are only limited data about binge
drinking at the local level. Independent surveys of adults in Los Angeles, Calif,15,16 and
Harlem, NY,17 obtained selected data on alcohol use, and a few states have used the Behavioral Risk Factor Surveillance System
(BRFSS) to generate health district or city estimates.18–21 The lack of local data is unfortunate. Local data can empower communities to
address public health issues, to track progress
toward Healthy People 2010 alcohol-related
goals,22,23 and to improve planning and evaluation efforts to prevent alcohol abuse. However, because conducting surveys is timeconsuming and expensive, most local health
departments lack the resources to collect or
analyze survey data.
We used reweighted BRFSS data24 to examine current alcohol use and binge drinking
in 120 US metropolitan areas. The purposes
of our study were to (1) estimate the preva-
Objectives. We estimated adult binge drinking prevalence in US metropolitan areas.
Methods. We analyzed 1997 and 1999 Behavioral Risk Factor Surveillance System
data for 120 metropolitan areas in 48 states and the District of Columbia.
Results. The prevalence of binge drinking varied substantially across metropolitan
areas, from 4.1% in Chattanooga, Tenn, to 23.9% in San Antonio, Tex, (median = 14.5%).
Seventeen of the 20 metropolitan areas with the highest estimates were located in the
upper Midwest, Texas, and Nevada. In 13 of these areas, at least one third of persons
aged 18 to 34 years were binge drinkers. There were significant intrastate differences
for binge drinking among metropolitan areas in New York, Tennessee, and Utah.
Conclusions. Metropolitan-area estimates can be used to guide local efforts to reduce
binge drinking. (Am J Public Health. 2004;94:663–671)
lence of binge drinking in metropolitan areas,
(2) determine if there were differences between metropolitan and statewide estimates
of binge drinking, (3) assess intrastate differences in binge drinking for states with data
from 2 or more metropolitan areas, (4) estimate the proportion of current drinkers who
were binge drinkers by metropolitan area,
and (5) identify the demographic subgroups
within metropolitan areas that have the highest overall binge-drinking estimates and are at
greatest risk for experiencing adverse effects
from this type of alcohol use.
METHODS
Data were obtained from the BRFSS, a
state-based system of adult health surveys coordinated by the Centers for Disease Control
and Prevention.11,25 Health risk factor and
preventive health services data were obtained
through telephone surveys of randomly selected adults aged 18 years and older. Data
were collected annually for some measures
and biennially for others; questions on alcohol use were included in odd-numbered-year
surveys during the 1990s.25 Current drinking
was defined as consuming 1 or more alcoholic beverages in the past month; binge
drinking was defined as consuming 5 or more
alcoholic beverages on at least 1 occasion in
the past month, which was determined by the
question, “Considering all types of alcoholic
April 2004, Vol 94, No. 4 | American Journal of Public Health
beverages, how many times during the past
month did you have 5 or more drinks on an
occasion?”
We analyzed data from 1997 and 1999,
the most recent years with available metropolitan-level data on binge drinking. The
overall sample size was 133 048 in 1997 and
159 921 in 1999; median state sample sizes
were 2340 in 1997 (range = 1505 to 4923)
and 2939 in 1999 (range = 1248 to 7543).
Median state response rates, on the basis of
persons actually reached by telephone, were
76.8% in 1997 and 68.4% in 1999, with individual state response rates ranging from
44.5% to 95.1%.
Self-reported county of residence was used
to classify respondents as residents of metropolitan areas in accordance with standard
census definitions.26 Response rates by metropolitan area were unavailable, as all rates
were calculated by state. Metropolitan-level
estimates are composed of groups of counties,
oftentimes encompassing counties in more
than one state. Metropolitan-level data were
not merged with state data because the state
data were weighted to state populations and
metropolitan-level data were weighted to
county populations. See Table 1 for metropolitan city designations.
All analyses were conducted with SAS27
and SUDAAN28 software. To increase sample sizes and to improve the precision of estimates, we pooled 1997 and 1999 data and
Nelson et al. | Peer Reviewed | Research and Practice | 663
 RESEARCH AND PRACTICE 
TABLE 1—Binge-Drinking Prevalence and 95% Confidence Intervals for US Metropolitan
Areas, by Region: 1997 and 1999
Metropolitan Area and Region
Northeast
Bergen–Passaic, NJ
Boston–Worcester–Lawrence–Lowell–Brockton, Mass–NH
Burlington, Vt
Hartford, Conn
Middlesex–Somerset–Hunterdon, NJ
Monmouth–Ocean, NJ
Nassau–Suffolk, NY
New Haven–Bridgeport–Stamford–Waterbury–Danbury, Conn
New York, NY
Newark, NJ
Philadelphia, Pa–NJ
Pittsburgh, Pa
Portland, Me
Providence–Warwick–Pawtucket, RI
Rochester, NY
Springfield, Mass
Median (range)
Midwest
Bismarck, ND
Cedar Rapids, Iowa
Chicago, Ill
Cincinnati, Ohio–Ky–Ind
Cleveland–Lorain–Elyria, Ohio
Columbus, Ohio
Davenport–Moline–Rock Island, Iowa–Ill
Des Moines, Iowa
Detroit, Mich
Duluth–Superior, Minn–Wis
Evansville–Henderson, Ind–Ky
Fargo–Moorhead, ND–Minn
Fort Wayne, Ind
Gary, Ind
Grand Forks, ND–Minn
Grand Rapids–Muskegon–Holland, Mich
Indianapolis, Ind
Kansas City, Mo–Kan
Lincoln, Neb
Milwaukee–Waukesha, Wis
Minneapolis–St Paul, Minn–Wis
Omaha, Neb–Iowa
Rapid City, SD
Sioux Falls, SD
South Bend, Ind
St Louis, MO–IL
Wichita, Kan
Median (range)
n
Binge Drinking, %
820
6682
1891
1765
682
796
867
2542
2588
1261
2657
1463
708
5186
425
819
10.0 (7.5, 12.5)
17.6 (16.2, 19.0)
19.6 (17.1, 22.1)
14.3 (12.2, 16.4)
12.5 (9.2, 15.8)
14.1 (10.9, 17.3)
11.6 (9.0, 14.2)
14.2 (12.3, 16.1)
9.3 (8.0, 10.6)
11.2 (9.0, 13.4)
14.6 (12.9, 16.3)
16.3 (14.0, 18.6)
15.3 (12.0, 18.6)
14.6 (13.3, 15.9)
15.5 (11.5, 19.5)
20.1 (15.3, 24.9)
14.5 (9.3, 20.1)
543
404
1793
1784
986
881
421
1088
1960
495
656
814
589
582
453
550
1123
2475
756
1112
5971
2346
771
1314
510
1371
1098
18.6 (15.2, 22.0)
20.6 (15.7, 25.5)
16.8 (14.8, 18.8)
12.5 (10.2, 14.8)
8.8 (6.4, 11.2)
11.0 (7.8, 14.2)
21.1 (14.8, 27.4)
18.2 (15.5, 20.9)
20.0 (17.9, 22.1)
20.3 (16.0, 24.6)
13.6 (8.9, 18.3)
17.7 (14.6, 20.8)
14.1 (10.4, 17.8)
16.3 (11.8, 20.8)
23.4 (17.9, 28.9)
18.1 (14.6, 21.6)
15.1 (12.5, 17.7)
15.6 (13.5, 17.7)
20.1 (15.7, 24.5)
22.7 (19.4, 26.0)
16.2 (15.0, 17.4)
17.2 (15.2, 19.2)
16.9 (13.7, 20.1)
21.4 (18.8, 24.0)
14.5 (10.6, 18.4)
17.2 (14.8, 19.6)
11.0 (8.7, 13.3)
17.2 (8.8, 23.4)
Continued
664 | Research and Practice | Peer Reviewed | Nelson et al.
limited analyses to the 120 metropolitan
areas with 400 or more respondents for
both alcohol measures; sample sizes ranged
from 404 to 6682. We used intercensal estimates to reweight data by the age, gender,
and racial/ethnic distributions for each metropolitan area. Missing or unknown data
were excluded from all calculations. Data
were not age standardized, because we
wanted to provide actual binge-drinking estimates for each area.
Metropolitan-level data were available
from metropolitan areas in 48 states and the
District of Columbia (data not available for
Alaska and New Hampshire, where there
were no metropolitan areas with sufficient
sample sizes). To provide additional context,
we pooled state estimates of binge drinking
from 1997 and 1999. State estimates were
based on data from all state respondents, including those living within metropolitan
areas.
We grouped metropolitan areas by census
region (Northeast, Midwest, South, and
West) and by state, and we calculated regional and national median and range values. To determine the relationship between
estimates of current drinking and binge
drinking, we calculated the proportion of
current drinkers who were binge drinkers
for each metropolitan area. For the 20 metropolitan areas with the highest levels of
binge drinking, we conducted analyses of
binge-drinking estimates stratified by age
(18–34 years and ≥ 35 years), gender, race/
ethnicity (White, Black, Hispanic, other), education level (≤ high school, > high school),
and income (< $25 000, $25 000–$49 999,
and ≥ $50,000). To improve precision, these
analyses were conducted only for subpopulations with at least 50 respondents.
We used 2-sample t tests29 to assess
whether differences between statewide and
metropolitan estimates and differences between intrastate metropolitan-area estimates
were statistically significant. For the 20
areas with the highest levels of binge drinking, we also used 2-sample t tests to examine differences in estimates by demographic
groups within each metropolitan area. Because of the large number of comparisons,
differences were considered statistically significant only when 99% confidence inter-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
TABLE 1—Continued
South
Atlanta, Ga
2176
10.8 (9.2, 12.4)
Augusta–Aiken, Ga–SC
429
9.2 (4.7, 13.7)
Austin–San Marcos, Tex
510
22.6 (17.8, 27.4)
Baltimore, Md
2969
11.9 (10.4, 13.4)
Baton Rouge, La
466
16.3 (12.8, 19.8)
Biloxi–Gulfport–Pascagoula, Miss
469
13.5 (9.6, 17.4)
Birmingham, Ala
937
12.9 (10.4, 15.4)
Charleston, WVa
758
8.7 (6.4, 11.0)
Charleston–North Charleston, SC
719
13.7 (10.4, 17.0)
1153
11.8 (9.3, 14.3)
449
4.4 (2.4, 6.4)
732
12.8 (10.0, 15.6)
Charlotte–Gastonia–Rock Hill, NC–SC
Chattanooga, Tenn
Columbia, SC
Dallas, Tex
Daytona Beach, Fla
Dover, Del
1123
15.6 (12.7, 18.5)
407
14.7 (10.4, 19.0)
1446
13.0 (11.0, 15.0)
Fayetteville–Springdale–Rogers, Ark
454
9.9 (6.9, 12.9)
Ft Lauderdale, Fla
850
11.8 (9.3, 14.3)
Ft Worth–Arlington, Tex
535
19.4 (15.1, 23.7)
977
9.6 (7.1, 12.1)
Greensboro–Winston-Salem–High Point, NC
Greenville–Spartanburg–Anderson, SC
Hagerstown, Md
Houston, Tex
1223
8.8 (6.9, 10.7)
422
8.8 (5.3, 12.3)
1221
17.5 (14.7, 20.3)
Huntington–Ashland, WVa–Ky–Ohio
986
6.5 (4.2, 8.8)
Jackson, Miss
626
11.1 (8.1, 14.1)
Jacksonville, Fla
769
12.6 (9.7, 15.5)
Johnson City–Kingsport–Bristol, Tenn–Va
573
6.9 (4.1, 9.7)
Knoxville, Tenn
698
7.2 (5.1, 9.3)
Lexington, Ky
743
10.0 (7.3, 12.7)
Little Rock–North Little Rock, Ark
996
11.5 (9.1, 13.9)
Louisville, Ky–Ind
1490
12.4 (10.3, 14.5)
Memphis, Tenn–Ark–Miss
1271
9.4 (7.4, 11.4)
Miami, Fla
974
10.0 (7.6, 12.4)
Mobile, Ala
468
13.2 (9.7, 16.7)
Nashville, Tenn
1240
10.4 (8.3, 12.5)
New Orleans, La
1002
16.4 (13.6, 19.2)
1275
14.1 (11.5, 16.7)
Norfolk–Virginia Beach–Newport News, Va–NC
Oklahoma City, Okla
Orlando, Fla
1557
9.6 (7.7, 11.5)
715
14.6 (11.6, 17.6)
Raleigh–Durham–Chapel Hill, NC
835
11.4 (8.9, 13.9)
Richmond–Petersburg, Va
932
15.9 (12.8, 19.0)
San Antonio, Tex
Tampa–St Petersburg–Clearwater, Fla
496
23.9 (18.8, 29.0)
1112
12.8 (10.5, 15.1)
Tulsa, Okla
1130
7.4 (5.4, 9.4)
Washington, DC–Md–Va–WVa
7276
11.5 (10.1, 12.9)
West Palm Beach–Boca Raton, Fla
Wilmington–Newark, Del–Md
Median (range)
746
12.9 (10.1, 15.7)
2545
16.3 (13.9, 18.7)
11.9 (4.4, 23.9)
Continued
April 2004, Vol 94, No. 4 | American Journal of Public Health
vals (CI) for differences excluded the null
value.
We used logistic regression analyses to examine the independent association of binge
drinking with age, gender, education, or race/
ethnicity (on the basis of odds ratios [OR] and
95% CI) for the 20 areas with the highest
levels of binge drinking. Income was not included in these models because of collinearity
with education.
We mapped binge-drinking estimates with
ArcGIS 8.2 software (Environmental Research Systems, Inc, Redlands, Calif) for all
120 metropolitan areas, as well as by state, to
better understand regional patterns.30 Cutpoints for metropolitan and state estimates
were based on quartile ranges.
RESULTS
Considerable variability was found in overall binge-drinking estimates by metropolitan
area. The median metropolitan estimate was
14.5%, which ranged from 4.4% in Chattanooga Tenn, to 23.9% in San Antonio, Tex
(Table 1). Estimates were generally highest in
the Midwest, intermediate in the West and the
Northeast, and lowest in the South (Table 1
and Figure 1). However, substantial variation
in binge drinking was found in all 4 regions
that ranged from a 2-fold difference in the
Northeast to a 5-fold difference in the South.
State binge-drinking estimates ranged from
7.4% in Tennessee to 25.2% in Wisconsin
(median = 14.8%) (Figure 2) and were significantly different from metropolitan estimates
in Montana, Tennessee, and Utah (data not
shown). Three of 40 states—New York, Tennessee, and Utah—had statistically significant
intrastate differences in binge drinking among
metropolitan areas.
Nationally, the proportion of current drinkers who were binge drinkers was 26.7%,
which ranged from 16.9% in Chattanooga,
Tenn, to 42.2% in Grand Forks, ND–Minn; regional median estimates ranged from a low of
23.6% in the Northeast to a high of 30.4% in
the Midwest (data not shown). The geographic
pattern for this measure was generally similar
to the overall pattern for binge-drinking prevalence. However, Provo–Orem, Utah; Salt Lake
City, Utah; and Johnson City–Kingsport–Bristol, Tenn–Va were in the lowest quartile for
Nelson et al. | Peer Reviewed | Research and Practice | 665
 RESEARCH AND PRACTICE 
TABLE 1—Continued
West
Albuquerque, NM
Billings, Mont
Boise City, Idaho
Casper, Wyo
Cheyenne, Wyo
Colorado Springs, Colo
Denver, Colo
Eugene–Springfield, Ore
Honolulu, Hawaii
Las Cruces, NM
Las Vegas, Nev–Ariz
Los Angeles–Long Beach, Calif
Oakland, Calif
Orange County, Calif
Phoenix–Mesa, Ariz
Pocatello, Idaho
Portland–Vancouver, Ore–Wash
Provo–Orem, Utah
Reno, Nev
Riverside–San Bernardino, Calif
Sacramento, Calif
Salem, Ore
Salt Lake City–Ogden, Utah
San Diego, Calif
San Francisco, Calif
San Jose, Calif
Santa Fe, NM
Seattle–Bellevue–Everett, Wash
Spokane, Wash
Tacoma, Wash
Tucson, Ariz
Median (range)
National median (range)
2064
490
2150
667
759
433
1664
448
2390
493
1893
2061
612
642
1235
671
2768
527
1531
724
420
451
2398
739
431
409
419
2887
529
715
827
15.2 (13.1, 17.3)
10.9 (7.7, 14.4)
17.0 (14.8, 19.2)
15.1 (12.0, 18.2)
14.5 (11.7, 17.3)
13.7 (10.0, 17.4)
16.5 (14.5, 18.5)
14.7 (9.8, 19.6)
14.0 (12.4, 15.6)
16.9 (13.0, 20.8)
18.7 (15.9, 21.5)
15.3 (13.4, 17.2)
14.8 (11.4, 18.2)
16.5 (13.2, 19.8)
9.3 (7.2, 11.4)
14.4 (11.5, 17.3)
14.9 (13.3, 16.5)
4.6 (0.1, 9.1)
9.2 (4.7, 13.7)
9.2 (4.7, 13.7)
17.0 (12.5, 21.5)
9.6 (6.3, 12.9)
10.4 (9.0, 11.8)
14.9 (12.0, 17.8)
17.3 (13.0, 21.6)
16.0 (11.8, 20.2)
15.9 (11.2, 20.6)
14.9 (13.2, 16.6)
19.6 (15.5, 23.7)
13.5 (10.6, 16.4)
8.6 (6.3, 10.9)
14.9 (4.6, 20.0)
14.5 (4.4, 23.9)
Note. CI = confidence interval.
binge drinking and were in the highest quartile for the proportion of current drinkers who
also were binge drinkers.
The 20 metropolitan areas with the highest
binge-drinking estimates are shown in Table 2
and Figure 2. Twelve areas were in 7 states in
the upper Midwest (Iowa, Michigan, Minnesota, Nebraska, North Dakota, South
Dakota, and Wisconsin), 3 were in Texas, and
2 were in Nevada. Not surprisingly, there was
a strong correlation between high statewide
estimates and metropolitan areas with the
highest levels of binge drinking.
Binge drinking was significantly more common among persons aged 18 to 34 years than
among those aged 35 years and older in 18 of
the 20 metropolitan areas with the highest
binge-drinking estimates. Among persons aged
18 to 34 years, estimates ranged from 28.8%
in Fargo–Moorhead, Minn–ND, to 44.2% in
Springfield, Mass (Table 2); in 13 areas, at least
one third of persons in this age group were
binge drinkers. Among adults aged 35 years
and older, binge-drinking estimates ranged
from 9.5% in Springfield, Mass, to 18.3% in
San Antonio, Tex. Men were statistically more
666 | Research and Practice | Peer Reviewed | Nelson et al.
likely to report binge drinking than were
women in 18 of the 20 metropolitan areas
with the highest overall estimates (Table 2).
More than one third of men were binge drinkers in San Antonio, Tex; Milwaukee–Waukesha,
Wis; and Grand Forks, ND–Minn.
Comparisons of binge drinking by race/
ethnicity were limited, because sample sizes
were sufficient in only 4 metropolitan areas
for Blacks and in only 7 areas for Hispanics.
We found no significant differences in the
prevalence of binge drinking by race/ethnicity
in these areas (data not shown). Similarly,
analyses of binge-drinking prevalence by education level for all 20 areas with the highest
estimates revealed no significant differences
(data not shown). With the exception of
Burlington, Vt, where binge drinking was
higher among persons who had income levels
below $25 000 compared with those who
had incomes above $50 000, there were no
other significant differences in binge drinking
by income level (data not shown).
Logistic regression models confirmed the
strong association between age and gender
with binge drinking for nearly all metropolitan areas (Table 3), with a median OR for
persons aged 18 to 34 years of 3.76
(range = 1.80–8.21) relative to persons aged
35 years and older and with a median OR
for men of 3.72 (range = 2.50–8.11) relative
to women. Significantly higher odds ratios for
binge drinking among younger adults were
found in Springfield, Mass, and Spokane,
Wash, compared with Austin–San Marcos,
Tex. There were no significant differences in
OR between metropolitan areas for binge
drinking among men.
In San Antonio, Tex; Sioux Falls, SD;
Davenport–Moline–Rock Island, Iowa–Ill;
and Spokane, Wash, there was some evidence
that lower levels of education were associated
with binge drinking, although 95% CIs were
close to 1.00 in each area. No association was
found between levels of education and binge
drinking in the remaining areas. Logistic models confirmed the absence of an independent
association between race/ethnicity and binge
drinking in areas with 50 or more Black or
Hispanic respondents (data not shown), because a significant difference was found only
among blacks in Las Vegas, Nev (OR = 1.78;
95% CI = 1.03, 3.08).
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
FIGURE 1—Geographic distribution of binge-drinking prevalence estimates for US metropolitan areas: 1997 and 1999.
DISCUSSION
To our knowledge, this is the first study
that comprehensively reports binge-drinking
estimates for US metropolitan areas. We
found that binge drinking was especially common in metropolitan areas located in the
upper Midwest, Texas, and Nevada. The proportion of current drinkers who reported
binge drinking exceeded 30% in many US
metropolitan areas. Analyses of areas with the
highest binge-drinking estimates showed that
at least one third of persons aged 18 to 34
years in 13 metropolitan areas were binge
drinkers, as were more than one third of men
in San Antonio, Tex; Grand Forks, ND–Minn;
and Milwaukee–Waukesha, Wis. In the 20
areas with the highest binge-drinking estimates, we found that age and gender were
strongly and independently associated with
binge drinking.
Our results indicate that metropolitan
binge-drinking estimates vary across regions,
within regions, and within individual states.
Although there were few differences when
we compared metropolitan estimates with
their corresponding state estimates, our ability
to detect such differences was limited, because state estimates were based on information provided by all respondents within states,
which included persons within metropolitan
areas. Future studies are needed that compare metropolitan and nonmetropolitan
binge-drinking estimates within states.
Our results were consistent with nationally
representative adult data from the National
Household Survey on Drug Abuse (NHSDA),
which demonstrated higher binge-drinking estimates among young adults, men, and residents in the upper Midwest, as well as the
general lack of differences by educational
level.9,10 In contrast to our findings, NHSDA
April 2004, Vol 94, No. 4 | American Journal of Public Health
data demonstrated that adult binge-drinking
estimates were higher for Whites and Hispanics than for Blacks. Although we did not find
similar racial/ethnic differences, we used only
a limited number of metropolitan areas for
such comparisons. Our metropolitan estimates
for binge drinking were generally similar to
previous BRFSS-developed estimates for metropolitan areas in Wisconsin, Idaho, and
Massachusetts.18–21
There are several possible explanations for
the substantial variation in binge drinking
across metropolitan areas. Because binge
drinking among adults varies inversely with
age,6,9,10 metropolitan areas with younger populations are likely to have higher estimates.
For example, several metropolitan areas with
high binge-drinking estimates, such as Grand
Forks, ND–Minn; Austin–San Marcos, Tex;
Lincoln, Neb; and Burlington, Vt, have major
state universities. However, the OR for binge
Nelson et al. | Peer Reviewed | Research and Practice | 667
 RESEARCH AND PRACTICE 
TABLE 2—Binge Drinking Prevalence (%), by Age and Gender, and 95% Confidence Intervals
for the 20 Metropolitan Areas With the Highest Levels of Binge Drinking: 1997 and 1999
Age Group, y
Metropolitan Area
1. San Antonio, Tex
2. Grand Forks, ND–Minn
3. Milwaukee–Waukesha, Wis
4. Austin–San Marcos, Tex
5. Sioux Falls, SD
6. Davenport–Moline–Rock Island, Iowa–Ill
7. Cedar Rapids, Iowa
8. Duluth–Superior, Minn–Wis
9. Lincoln, Neb
10. Springfield, Mass
11. Detroit, Mich
12. Reno, Nev
13. Spokane, Wash
14. Burlington, Vt
15. Ft Worth–Arlington, Tex
16. Las Vegas, Nev–Ariz
17. Bismarck, ND
18. Des Moines, Iowa
19. Grand Rapids–Muskegon–Holland, Mich
20. Fargo–Moorhead, ND–Minn
N
496
453
1112
510
1314
421
404
495
756
819
1960
1531
529
1891
535
1893
543
1088
550
814
Region
South
Midwest
Midwest
South
Midwest
Midwest
Midwest
Midwest
Midwest
Northeast
Midwest
West
West
Northeast
South
West
Midwest
Midwest
Midwest
Midwest
Overall
Sex
≥ 35
18–34
a
23.9 (18.8, 29.0)
23.4 (17.9, 28.9)
22.7 (19.4, 26.0)
22.6 (17.8, 27.4)
21.4 (18.8, 24.0)
21.1 (14.8, 27.4)
20.6 (15.7, 25.5)
20.3 (16.0, 24.6)
20.1 (15.7, 24.5)
20.1 (15.3, 24.9)
20.0 (17.9, 22.1)
35.3 (26.6, 44.0)
38.1 (25.3, 39.9)a
32.6 (25.3, 39.9)a
29.9 (21.0, 38.8)
35.9 (30.3, 41.5)a
28.4 (18.5, 38.3)
36.2 (26.1, 46.3)a
36.9 (26.7, 47.1)a
35.1 (27.9, 42.3)a
44.2 (33.4, 55.0)a
33.3 (28.8, 37.8)a
20.0 (17.2, 22.8)
19.6 (15.5, 23.7)
19.6 (17.1,22.1)
19.4 (15.1, 23.7)
18.7 (15.9, 21.5)
18.6 (15.2, 22.0)
18.2 (15.5, 20.9)
18.1 (14.6, 21.6)
17.7 (14.6, 20.8)
a
33.4 (27.2, 39.6)
38.9 (30.0, 47.8)a
35.2 (29.9, 40.5)a
33.1 (24.2, 42.0)a
30.6 (24.0, 37.2)a
29.6 (22.2, 37.0)a
33.0 (27.1, 38.9)a
30.4 (23.0, 37.8)a
28.8 (22.7, 34.9)a
Men
Women
18.1 (11.6, 24.6)
13.2 (8.2, 18.6)
18.3 (14.8, 21.8)
18.2 (12.8, 23.6)
14.2 (11.7, 16.7)
17.9 (9.7, 26.1)
13.8 (8.9, 18.7)
13.1 (9.5, 16.7)
11.9 (6.3, 17.5)
9.5 (6.0, 13.0)
b
34.2 (26.6, 41.8)
34.5 (25.3, 43.7)b
36.5 (30.9, 42.1)b
32.5 (24.8, 40.2)b
30.9 (26.5, 35.3)b
29.4 (19.1, 39.7)
30.5 (22.3, 38.7)b
28.9 (21.6, 36.2)b
29.4 (21.9, 36.9)b
27.0 (19.1, 34.9)
14.1 (7.3, 20.9)
11.8 (6.9, 17.7)
10.2 (7.6, 12.8)
13.4 (7.4, 19.4)
12.9 (10.0, 15.8)
13.5 (6.8, 20.2)
10.6 (6.5, 14.7)
13.6 (8.6, 18.6)
11.0 (7.4, 14.6)
14.2 (8.3, 20.1)
14.1 (11.9, 16.3)
14.1 (11.2, 17.0)
10.0 (6.6, 13.4)
10.6 (8.6, 12.6)
12.0 (8.0, 16.0)
13.7 (11.2, 16.2)
13.9 (10.2, 17.6)
11.5 (8.9, 15.1)
11.9 (8.2, 15.6)
11.4 (8.3, 14.5)
31.0 (27.3, 34.7)b
29.1 (18.2, 40.0)b
32.8 (25.8, 29.8)b
29.2 (25.2, 33.2)b
31.1 (23.8, 38.4)b
27.6 (22.9, 32.3)b
25.9 (20.1, 31.7)b
28.9 (24.1, 33.7)b
28.3 (22.2, 34.4)b
29.2 (23.4, 35.0)b
10.2 (8.3, 12.1)
10.9 (8.0, 13.8)
7.5 (4.1, 10.9)
10.8 (7.9, 13.7)
8.0 (4.7, 11.3)
9.8 (7.3, 12.3)
12.0 (7.8, 16.2)
8.8 (6.2, 11.4)
8.7 (5.4, 12.0)
7.7 (5.0, 10.4)
a
Difference of estimate for persons aged 18–34 years was statistically significant (P < 0.01) compared with persons aged 35 years and older.
Estimate for men was statistically significantly different (P < 0.01) from estimate for women.
b
drinking in Austin–San Marcos, Tex, was the
lowest among all 20 areas in the 18- to 34year age group, which suggests that factors besides age distribution may account for higher
estimates in this metropolitan area.
Other factors also influence local alcohol
estimates. Sociocultural norms, such as religious beliefs, are likely to be influential.6 For
example, alcohol use is proscribed for Mormons, and many members of Southern Baptist
churches abstain from alcohol. This may help
to explain low binge-drinking estimates for
metropolitan areas in Utah and in certain parts
of the South. Country of origin, level of acculturation, alcohol availability, price, alcohol outlet density, and type and extent of alcoholrelated legislation and level of enforcement
(e.g., beverage service practices, drinking and
driving laws) also may contribute to local variation of binge-drinking estimates.
Previous research suggests that there may
be a distinct drinking culture, especially
among males, in parts of the United States,
where abstention rates are high and where alcohol is less widely available (so-called “dry”
areas, e.g., parts of the South and the Rocky
Mountain States) such that the proportion of
current drinkers who binge drink in these
areas may still be quite high.6,31,32 We found
some evidence of this, especially in Tennessee
and Utah metropolitan areas, but even so, the
proportion of alcohol users who binge drink
was typically highest in areas with the highest
prevalence of current alcohol use.
Overall, the BRFSS provides an efficient way
to perform alcohol-related surveillance at the
metropolitan level, and the local variation
found in our study demonstrates the importance of conducting local analyses on an ongoing basis. The BRFSS uses a standardized
methodology, relies on an existing infrastructure, and requires no new data collection,
which results in cost savings for local health departments. Furthermore, BRFSS data on alcohol consumption, including binge drinking, has
been collected annually (rather than in odd
668 | Research and Practice | Peer Reviewed | Nelson et al.
years only) since 2001. The increasing focus
on binge drinking as an important public health
problem, coupled with the growing demand for
state and local surveillance data, underscores
the need to further develop the BRFSS as a
vital component of the public health infrastructure in the United States.
Our study has several limitations. The results probably underestimate the extent of
binge drinking6,33,34 with social desirability35
and possibly with noncoverage (younger persons are less likely to have household telephones36,37 and are more likely to drink alcohol9,10), which may have had some effect on
our estimates.35 Survey interview mode effects can affect estimates, although a study
that compared BRFSS binge-drinking estimates with household survey estimates in
Michigan found little difference.38 Estimates
for women may be conservative, because others have used “4 alcoholic beverages or more
on 1 occasion in the past month” to define
binge drinking for women.2
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
FIGURE 2—State binge-drinking estimates and the 20 US metropolitan areas with the highest binge-drinking levels: 1997 and 1999.
Differences by metropolitan area may be
the result of variations in demographics
other than age, race/ethnicity, and education, such as employment and social class.
Furthermore, our estimates were for entire
metropolitan areas, but binge drinking is
likely to vary within individual areas as well
(e.g., central cities vs suburbs). We were unable to assess response rates by metropolitan
area, as all rates were calculated by state.
Typical of other telephone surveys in the
late 1990s, response rates declined over the
study period, and the effect of this decline
on our estimates is unknown.39,40 Because
we pooled data, we could not examine
trends between 1997 and 1999; nevertheless, in spite of pooling, the number of respondents was small for certain areas and
subpopulations, which reduced the precision
of some estimates.
About the Authors
CONCLUSIONS
The adoption of public health measures is
needed to address the problem of binge
drinking. Effective measures include increasing alcohol excise taxes, enforcing the minimum drinking age, and developing comprehensive community-based programs that
include education, enforcement, and community mobilization.41,42 In addition, clinicians
should screen all adult and adolescent patients for alcohol abuse in accordance with
recommendations by the US Preventive Services Task Force.43 Screening and brief intervention strategies for alcohol abuse, including
binge drinking, are effective and can reduce
costs in primary care settings.44,45 Through
this combination of interventions, binge
drinking and its attendant health and social
consequences can be prevented.
April 2004, Vol 94, No. 4 | American Journal of Public Health
David E. Nelson is with the Office on Smoking and Health,
National Center for Chronic Disease Prevention and
Health Promotion, Centers for Disease Control and Prevention, Atlanta, Ga. Timothy S. Naimi, Robert D. Brewer,
and Julie Bolen are with the Division of Adult and Community Health, National Center for Chronic Disease Prevention and Health Promotion. Henry E. Wells is with Research Triangle Institute, Atlanta, Ga.
Requests for reprints should be sent to David E. Nelson,
MD, MPH, Centers for Disease Control and Prevention,
4770 Buford Hwy, NE, Mail Stop K-50, Atlanta, GA
30341 (e-mail: [email protected]).
This article was accepted May 13, 2003.
Contributors
D. E. Nelson, T. S. Naimi, R. D. Brewer, and J. Bolen conceived and designed the study, interpretated the data,
and wrote the article. H. E. Wells conducted the data
analyses.
Acknowledgments
We thank the state Behavioral Risk Factor Surveillance
System coordinators for their assistance with data collection. We also thank Paul Mowery, MS, and Glenn
Nelson et al. | Peer Reviewed | Research and Practice | 669
 RESEARCH AND PRACTICE 
Findings. Available at: http://www.samsha.gov/oas/
nhsda.htm. Accessed July 13, 2002.
TABLE 3—Odds Ratios and 95% Confidence Intervals for Binge Drinking, by Age, Sex, and
Level of Education, for the 20 Metropolitan Areas With the Highest Levels of Binge
Drinkinga: 1997 and 1999
Metropolitan Area
Age Group, y
18–34b
Sex
Menc
Level of Education
≤ High Schoold
1. San Antonio, Tex
2. Grand Forks, ND–Minn
3. Milwaukee–Waukesha, Wis
4. Austin–San Marcos, Tex
5. Sioux Falls, SD
6. Davenport–Moline–Rock Island, Iowa–Ill
7. Cedar Rapids, Iowa
8. Duluth–Superior, Minn–Wis
9. Lincoln, Neb
10. Springfield, Mass
11. Detroit, Mich
12. Reno, Nev
13. Spokane, Wash
14. Burlington, Vt
15. Ft Worth–Arlington, Tex
16. Las Vegas, Nev–Ariz
17. Bismarck, ND
18. Des Moines, Iowa
19. Grand Rapids–Muskegon–Holland, Mich
20. Fargo–Moorhead, ND–Minn
2.29 (1.23, 4.26)
4.02 (2.21, 7.34)
2.44 (1.56, 3.81)
1.80 (0.97, 3.34)
3.80 (2.69, 5.37)
2.19 (1.11, 4.34)
3.89 (2.08, 7.30)
3.73 (2.12, 6.53)
4.38 (2.28, 8.40)
8.21 (4.34, 15.51)
3.21 (2.41, 4.27)
3.33 (2.24, 4.93)
7.80 (4.32, 14.11)
4.98 (3.66, 6.78)
3.18 (1.72, 5.89)
2.82 (1.91, 4.18)
2.63 (1.61, 4.27)
4.43 (3.00, 6.54)
3.99 (2.36, 6.75)
3.78 (2.41, 5.91)
3.31 (1.71, 6.43)
3.89 (2.08, 7.26)
5.15 (3.51, 7.54)
3.15 (1.70, 5.84)
3.07 (2.17, 4.34)
2.81 (1.31, 6.03)
4.04 (2.17, 7.53)
2.65 (1.47, 4.78)
3.74 (2.21, 6.32)
2.50 (1.34, 4.67)
4.03 (3.06, 5.31)
3.42 (2.32, 5.03)
8.11 (4.22, 15.58)
3.61 (2.54, 5.12)
5.14 (2.83, 9.34)
3.69 (2.53, 5.39)
2.69 (1.61, 4.50)
4.81 (3.16, 7.32)
4.94 (2.88, 8.48)
5.22 (3.22, 8.48)
1.82 (1.00, 3.30)
1.01 (0.52, 1.97)
0.87 (0.57, 1.32)
0.90 (0.48, 1.71)
1.59 (1.12, 2.24)
2.18 (1.03, 4.61)
0.74 (0.41, 1.36)
0.88 (0.49, 1.56)
1.12 (0.59, 2.10)
0.80 (0.43, 1.50)
0.98 (0.73, 1.31)
1.11 (0.76, 1.63)
1.78 (1.02, 3.12)
0.92 (0.65, 1.31)
1.30 (0.73, 2.29)
1.06 (0.73, 1.55)
0.73 (0.44, 1.21)
0.90 (0.60, 1.35)
1.14 (0.66, 1.96)
1.16 (0.71, 1.89)
11. Powell-Griner E, Anderson JE, Murphy W. Stateand sex-specific prevalence of selected characteristics—
Behavioral Risk Factor Surveillance System, 1994 and
1995. MMWR Surveill Summ. 1997;46:1–31.
12. Centers for Disease Control and Prevention. Alcohol involvement in fatal motor vehicle crashes. MMWR
Morb Mortal Wkly Rep. 2001;50:1064–1065.
13. Nelson DE, Bland S, Powell-Griner E, et al. State
trends for health risk factors and receipt of clinical preventive services among adults during the 1990s.
JAMA. 2002;287:2659–2667.
14. Naimi TS, Brewer RD, Mokdad A, Denny C,
Serdula MK, Marks JS. Binge drinking among US
adults. JAMA. 2003;289:70–75.
15. Los Angeles County Dept of Health Services. The
Health of Angelenos: A Comprehensive Report of the
Health of the Residences of Los Angeles County. Los Angeles, Calif: Los Angeles County Dept of Health Services; 2000. Available at: http://lapublichealth.org/ha/
reports/angelenos/hofa.pdf. Accessed July 13, 2002.
16. Simon PA, Wold CM, Cousineau MR, et al. Meeting the data needs of a local health department: the
Los Angeles county health survey. Am J Public Health.
2001;91:1950–1952.
17. Fullilove RE, Fullilove MT, Northridge ME, et al.
Risk factors for excess mortality in Harlem. Findings
from the Harlem Household Survey. Am J Prev Med.
1999;16(suppl 3):22–28
18. Idaho Dept of Health and Welfare. Idaho Behavioral Risk Factor Surveillance System, 1998 Survey Data.
Boise, Idaho: Idaho Dept of Health and Welfare. Available at: http://www2.state.id.us/dhw/vital_stats/brfss.
Accessed July 13, 2002.
a
On the basis of logistic regression models that included age, sex, education level, and race/ethnicity as independent
variables.
b
Reference group = people 35 years old and older.
c
Reference group = women.
d
Reference group = more than high school.
Laird, PhD, for their input on statistical analyses and
Jim Holt for preparing the maps (Figures 1 and 2).
Human Participant Protection
No protocol or institutional review board approval was
needed for this study, because data were collected
anonymously (no individual identifiers) from a public
health surveillance system in which adults voluntarily
consented to telephone interviews.
References
1. McGinnis JM, Foege WH. Actual causes of death
in the United States. JAMA. 1993;270:2207–2212.
2. Wechsler H, Austin SB. Binge drinking: the five/
four measure. J Stud Alcohol. 1998;59:122–124.
3. Wechsler H, Nelson TF. Binge drinking and the
American college student: what’s five drinks? Psychol
Addict Behav. 2001;15:287–291.
19. Utah Dept of Health. Utah Behavioral Risk Factor
Surveillance System 1989–2000. Salt Lake City, Utah:
Utah Dept of Health; 1997 and 1999 local health
data. Available at: http://health.utah.gov/ibisq/brfss.
Accessed July 13, 2002.
a comparison of acute and chronic causes. Med J Aus.
2001;174:281–284.
6. US Dept of Health and Human Services. Tenth
Special Report to the US Congress on Alcohol and Health.
Washington, DC: US Dept of Health and Human Services, Public Health Service, National Institutes of
Health, National Institute on Alcohol Abuse and Alcoholism; 2000.
7. National Institute on Alcohol Abuse and Alcoholism. Ninth Special Report to Congress on Alcohol and
Health. Bethesda, Md: National Institute on Alcohol
Abuse and Alcoholism; 1997. NIH publication
97–4017.
8. Shultz JM, Rice DP, Parker DL, Goodman RA,
Stroh G, Chalmers N. Quantifying the disease impact of
alcohol with ARDI software. Public Health Rep. 1991;
106:443–450.
4. Centers for Disease Control and Prevention. Alcoholrelated mortality and years of potential life lost—United
States, 1987. MMWR Morb Mortal Wkly Rep. 1990;
39:173–178.
9. Substance Abuse and Mental Health Administration. Summary of Findings from the 1998 National
Household Survey on Drug Abuse. Rockville, Md: Substance Abuse and Mental Health Administration, Office
of Applied Studies; 1999. National Household Survey
on Drug Abuse Series H-10.
5. Chikritzhs TN, Jonas HA, Stockwell TR, Heale PF,
Dietze PM. Mortality and life-years lost due to alcohol:
10. Substance Abuse and Mental Health Administration. 1999 National Household on Drug Abuse Summary
670 | Research and Practice | Peer Reviewed | Nelson et al.
20. Mass Dept of Public Health. Behavioral Risk Factor
Survey 1994–1996 City Reports. Boston, Mass: Mass
Dept of Public Health; 1998. Available at: http://www.
state.ma.us/dph/bhsre/cdsp/index.htm. Accessed July
13, 2002.
21. Wis Dept of Health and Family Services. County
and Regional Prevalence Estimates of Behavioral Risk
Factors and Health Screening Practices, Wisconsin, Combined Years of Data: 1989–1994 and 1993–1998.
Madison, Wis: Wis Dept of Health and Family Services;
2001. Available at: http://www.dhfs.state.wi.us/stats/
pdf/corgnlbrfs8998.pdf. Accessed July 13, 2002.
22. Leviton LC, Snell E, McGinnis M. Urban issues in
health promotion strategies. Am J Public Health. 2000;
90:863–866.
23. US Dept of Health and Human Services. Healthy
People 2010, 2nd ed. Washington, DC: US Dept of
Health and Human Services; 2000.
24. Centers for Disease Control and Prevention. Cigarette smoking in 99 metropolitan areas—United States,
2000. MMWR Morb Mortal Wkly Rep. 2001;50:
1107–1113.
25. Nelson DE, Holtzman D, Waller M, et al. Objectives and design of the Behavioral Risk Factor Surveillance System. Proceedings of the 1998 American Statisti-
American Journal of Public Health | April 2004, Vol 94, No. 4
 RESEARCH AND PRACTICE 
cal Association Section on Survey Research Methods, Dallas, Tex, 9–13 August 1998. Alexandria, Va: American
Statistical Association; 1998:214–218.
26. US Office of Management and Budget. Metropolitan Areas Defined. Washington, DC: Office of Management and Budget; June 30, 1993 (1990 CPHS11).
27. SAS Institute. SAS/STAT User’s Guide, Version 8.
Cary, NC: SAS Institute Inc; 1999.
London R. Brief physician advice for problem drinkers.
JAMA. 1997;277:1039–1045.
45. Fleming MF, Mundt M, Pfrench MT, Manwell LB,
Stauffacher EA, Barry KL. Brief physician advice for
problem alcohol drinkers: long-term efficacy and
benefit-cost analysis. A randomized controlled trial in
community-based primary care settings. Alcohol Clin
Exp Res. 2002;26:36–43.
28. Research Triangle Institute. SUDAAN User’s Manual, Release 8.0. Research Triangle Park, NC: Research
Triangle Institute; 2000.
29. Schenker N, Gentleman JF. On judging the significance of differences by examining the overlap between
confidence intervals. Am Stat. 2001;55:182–186.
30. Slocum, TA. Thematic Cartography and Visualization. Upper Saddle River, NJ: Prentice-Hall; 1999.
31. Hilton ME. Regional diversity in United States
drinking practices. Br J Addict. 1988;83:519–532.
32. Room, R. Region and urbanization as factors in
drinking practices and problems. In: Kissin B, Begleiter
H, eds. The Pathogenesis of Alcoholism: Psychosocial Factors. New York, NY: Plenum Press; 1983:555–604.
33. Midanik LT. Validity of self-reported alcohol use: a
literature review and assessment. Br J Addict. 1988;83:
1019–1029.
34. Feunekes GIJ, van Veer P, van Staveren WA, et al.
Alcohol intake assessment: the sober facts. Am J Epidemiol. 1999;150:105–112.
The Spirit of the
Coalition
35. Hingson R, Strunin L. Commentary: validity, reliability, and generalizability in studies of AIDS knowledge, attitudes, and behavioral risks based on subject
self-report. Am J Prev Med. 1993;9:62–64.
By Bill Berkowitz, PhD, and
Tom Wolff, PhD
36. US Bureau of the Census. Statistical Brief: Phoneless in America. Washington, DC: US Dept of Commerce, Economics and Statistics Administration; 1994.
T
he Spirit of Coalition is about creating
and maintaining local community
coalitions. It teaches practitioners about
community building by providing the
“nitty gritty” details of what makes coalitions work. The first-hand accounts, told
by public health practitioners, illustrate
how coalitions can be built and sustained,
leading to measurable, lasting results.
Chapters include how coalitions get
started, promoting and supporting the
coalition, structure, funding, pitfalls, and
much more.
Who will benefit by reading this
book? Public Health Workers ❚
Community Organizers ❚ Government
Leaders ❚ Public Health Educators.
37. Thornberry OT, Massey JT. Trends in United
States telephone coverage across time and subgroups.
In: Groves RM, Biemer PP, Lyberg LE, Massey JT,
Nichols WL, eds. Telephone Survey Methodology. New
York, NY: Wiley; 1988:25–49.
38. Anda RF, Dodson DL, Williamson DF, et al.
Health promotion data for state health departments:
telephone versus in-person survey estimates of smoking
and alcohol use. Am J Health Promotion. 1989;4:
32–36.
39. Singer E, Van Hoewyck J, Maher MP. Experiments
with incentives in telephone surveys. Public Opinion Q.
2000;64:171–188.
40. Tuckel P, O’Neill H. The vanishing respondent in
telephone surveys. J Advert Res. 2002;42(5):26–48.
ISBN 0-87553-244-6
2000 ❚ 264 pages ❚ softcover
$24.00 APHA Members
$30.00 Nonmembers
41. Holder HD, Gruenewald PJ, Ponicki WR, et al.
Effect of community-based interventions on high-risk
drinking and alcohol-related injuries. JAMA. 2000;
284:2341–2347.
42. Cook PJ, Moore MJ. The economics of alcohol
abuse and alcohol-control policies. Price levels, including excise taxes, are effective at controlling alcohol consumption. Raising excise taxes would be in the public
interest. Health Aff. 2002;21:120–133.
plus shipping and handling
American Public Health Association
43. US Preventive Services Task Force. Guide to Clinical Preventive Services. 2nd ed. Baltimore, Md: Williams
& Wilkins, 1996:567–582.
Publication Sales
Web: www.apha.org
E-mail: [email protected]
Tel: (301) 893-1894
SC01J7
FAX: (301) 843-0159
44. Fleming MF, Barry KL, Manwell LB, Johnson K,
April 2004, Vol 94, No. 4 | American Journal of Public Health
Nelson et al. | Peer Reviewed | Research and Practice | 671
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement