Cumulative Risk and Population Attributable Fraction in Prevention
Caroline H. Davis, David P. MacKinnon, Amy Schultz, and Irwin Sandler
Department of Psychology, Arizona State University
Compares the use of relative risk versus population attributable fraction in determin-
ing the target population for multirisk prevention programs in psychology. Results
show that relative risk generally increases as a function of cumulative risk. Guided by
this measure, prevention programs should target populations with the largest cumula-
tive risk. However, relative risk does not account for the prevalence of a particular
level of cumulative risk in the population. Therefore, because the largest cumulative
risk is experienced by only a small portion of the population, prevention programs
guided by this measure will not always have the greatest public health benefit to re-
duce the incidence of problem outcomes in the population. On the other hand, the
population attributable fraction, which does take into account the prevalence of a
particular level of cumulative risk, does not increase appreciably after a cumulative
risk of one, two, or three because the majority of people in the population will experi-
ence these levels of cumulative risk. Guided by this measure, prevention programs
that target the higher proportion of people who have a more moderate level of risk
would have the maximum impact on the population. National data sets from Great
Britain (the British Births Cohort Study [BCS]) and the United States (National Lon-
gitudinal Study of Youth [NLSY]) are used to explore this pattern of effects.
For prevention programs to deliver their services ef-
fectively, they must determine the largest possible tar-
get group that will benefit from their services. Typi-
cally, this is the group with the most important risk
factors for the outcome variable the program is de-
signed to prevent. The importance of a risk factor is of-
ten determined by measures of association such as re-
gression coefficients, correlations, relative risk, and
odds ratios, which quantify the relation between a risk
factor and an outcome variable for an individual. Pre-
vention programs are based on these risk factors
(Hawkins, Catalano, & Miller, 1992). For example, a
program designed to prevent childhood behavior prob-
lems might select risk factors based on problems given
bereavement, divorce, poverty, and parental depres-
sion. If divorce posed the highest relative risk for child-
hood behavior problems, then the program would tar-
get children who experienced divorce before children
who experienced any of the other less important risk
factors. This approach is economical in that the cost of
delivering the prevention is limited to those individuals
in the population who are at the greatest risk of devel-
oping childhood behavior problems.
The drawback of this approach is that measures of
association such as those listed previously fail to take
into account the prevalence of a risk factor in the popu-
lation when determining its importance. This has im-
plications for community-level decisions about select-
ing what risk factors to target to have the maximum
impact on the public health of their citizens. A pre-
vention program that determines its target population
based on the importance of a risk factor as defined
solely by the strength of association between a risk fac-
tor and an outcome while ignoring the prevalence of
that risk factor may find that the maximum impact on
Journal of Clinical Child and Adolescent Psychology
2003, Vol. 32, No. 2, 228–235
Copyright © 2003 by
Lawrence Erlbaum Associates, Inc.
228
This research was supported by the National Institute of Mental
Health Grant P30 MH39246–16 to Irwin Sandler to fund a Preven-
tive Intervention Research Center at Arizona State University.
For the use of the BCS 5-year follow-up data we acknowledge J.
Golding as depositor; N. Butler, S. Dowling, and A. Osborn as prin-
cipal investigators; the Medical Research Council as sponsor; and
the UK Data Archive as distributor. The above bear no responsibility
for their further analysis or interpretation. For the use of the BCS
10-year follow-up data we acknowledge J. M. Bynner as depositor;
N. Butler and J. M. Bynner as principal investigators; the Joseph
Rowntree Memorial Trust, the Department of Education and Sci-
ence, the Department of Health and Social Security, the Manpower
Services Commission, and the National Institute of Child Health and
Development as sponsors; and the UK Data Archive as distributor.
The above bear no responsibility for their further analysis or inter-
pretation. For the use of the BCS 16-year follow-up data we ac-
knowledge J. M. Bynner as depositor; N. Butler and J. M. Bynner as
principal investigators; the Home Office, the Cancer Research Cam-
paign, Beechams, Kelloggs, Westland, HTV, Channel 4, Allied Ly-
ons, the WT Grant Foundation, the Sir J. Knott Settlement, the Hay-
ward Foundation, the Daily Star, the New Moorgate Trust, the
Lankelly Foundation, the Laura Ashley Trust, and other public and
private bodies of private donations as sponsors; and the UK Data Ar-
chive as distributor. The above bear no responsibility for their further
analysis or interpretation.
Requests for reprints should be sent to Caroline H. Davis, Depart-
ment of Psychology, Arizona State University, Tempe, AZ 85283.
E-mail: caroline.davis@asu.edu