International Journal of Obesity (2002) 26, 1075–1082

ß 2002 Nature Publishing Group All rights reserved 0307–0565/02 $25.00

www.nature.com/ijo

PAPER
Body weight and obesity in adults and self-reported
abuse in childhood
DF Williamson1*, TJ Thompson1, RF Anda1, WH Dietz1 and V Felitti2
1
Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Atlanta,
Georgia, USA; and 2Department of Preventive Medicine, Southern California Permanente Medical Group (Kaiser Permanente), San
Diego, California, USA

BACKGROUND: Little is known about childhood factors and adult obesity. A previous study found a strong association between
childhood neglect and obesity in young adults.
OBJECTIVE: To estimate associations between self-reported abuse in childhood (sexual, verbal, fear of physical abuse and
physical) adult body weight, and risk of obesity.
DESIGN: Retrospective cohort study with surveys during 1995 – 1997.
PATIENTS: A total of 13 177 members of California health maintenance organization aged 19 – 92 y.
MEASUREMENTS: Body weight measured during clinical examination, followed by mailed survey to recall experiences during
first 18 y of life. Estimates adjusted for adult demographic factors and health practices, and characteristics of the childhood
household.
RESULTS: Some 66% of participants reported one or more type of abuse. Physical abuse and verbal abuse were most strongly
associated with body weight and obesity. Compared with no physical abuse (55%), being ‘often hit and injured’ (2.5%) had a
4.0 kg (95% confidence interval: 2.4 – 5.6 kg) higher weight and a 1.4 (1.2 – 1.6) relative risk (RR) of body mass index
(BMI)
30. Compared with no verbal abuse (53%), being ‘often verbally abused’ (9.5%) had an RR of 1.9 (1.3 – 2.7) for
BMI
40. The abuse associations were not mutually independent, however, because the abuse types strongly co-occurred.
Obesity risk increased with number and severity of each type of abuse. The population attributable fraction for ‘any mention’ of
abuse (67%) was 8% (3.4 – 12.3%) for BMI
30 and 17.3% ( 7 1.0 – 32.4%) for BMI
40.
CONCLUSIONS: Abuse in childhood is associated with adult obesity. If causal, preventing child abuse may modestly decrease
adult obesity. Treatment of obese adults abused as children may benefit from identification of mechanisms that lead to
maintenance of adult obesity.
International Journal of Obesity (2002) 26, 1075 – 1082. doi:10.1038=sj.ijo.0802038

Keywords: adult obesity; attributable fraction; body mass index; child abuse; relative risk

Introduction that obesity results from an inability to balance food intake

The etiology of adult obesity is poorly understood. Although
about 50% of variation in body weight is inherited, obesity
etiology is complex involving interactions among multiple
genes, environmental factors and behaviors.1 One view is
*Correspondence: DF Williamson, Senior Biomedical Research Scientist,
Division of Diabetes Translation (K-10), CDC, 4770 Buford Hwy NE,
Atlanta, GA 30341-2717, USA.
E-mail: drw1@cdc.gov
Received 25 September 2001; revised 1 March 2002;
accepted 31 March 2002
with physical activity in a culture that aggressively promotes
food consumption and sedentary living.2,3 Although family
environment is considered a key factor in the development
of child obesity,4 few studies have examined the impact of
childhood factors on adult obesity. Most studies have used
clinical samples lacking normal-weight controls.5,6
A notable exception is the work of Lissau and Sørensen,
who carried out a 10 y follow-up of a random sample of 9 –
10-y-old children in Copenhagen.7 At baseline the children’s
teachers and school nurses reported their impressions of the
students’ family structure, parental support and general
hygiene; at follow-up the heights and weights of the parti-
cipants were re-measured. After controlling for body weight
 Adult obesity and abuse in childhood
DF Williamson et al
1076
in childhood, children characterized as ‘dirty and neglected’
had 10 times the risk of becoming obese adults as those with
‘average’ childhood hygiene. The authors noted that in an
earlier study on heritability of adult obesity,8 genetic factors
approximately doubled the risk of adult obesity.7
Lissau’s and Sørensen’s study, however, may not apply to
other aspects of child maltreatment, such as child abuse, nor
to obesity risk in older adults. The purpose of this study was
to investigate relationships between abuse in childhood and
adult body weight and risk of obesity. We studied adults
(mean age 56 y) enrolled in the Kaiser Permanente health
maintenance organization (HMO) in San Diego, California.
Methods
Study population
Data are from the Adverse Childhood Experiences (ACE)
Study, whose objective is to assess the impact of childhood
experiences on adult health behaviors and outcomes.
Annually, more than 45 000 adult members of the Kaiser
Permanente HMO undergo a standardized medical examina-
tion at the organization’s health appraisal clinic. Approxi-
mately 80% of members are evaluated at that clinic,
primarily for preventive health assessments rather than
symptom=illness-based care. The ACE study sampled all
adult members aged
19 y examined at the clinic during
two time periods: August 1995 – March 1996 (wave 1) and
June 1997 – October 1997 (wave 2). One week after their
clinic visit, participants were mailed a questionnaire about
their childhood experiences and current health behaviors.
Prior publications from the ACE Study9 – 11 included only
respondents to the wave 1 survey, which had a response rate
of 70% (9508=13 494); the wave 2 survey had a response rate
of 65% (8667=13 330). Thus, the response rate for the two
waves combined was 68% (18 175=26 824). Because 754 per-
sons responded to both waves, the unduplicated number of
respondents was 17 421. All questions appeared on both
survey waves.
The ACE Study was approved by the institutional review
boards of the Southern California Permanente Medical
Group and of Emory University, and by the Office of
Protection from Research Risks, National Institutes of
Health.
Study variables
Questions about childhood experiences on the mailed ques-
tionnaire pertained to the respondent’s first 18 y of life. Data
were collected on four types of child abuse: sexual, verbal,
fear of physical and physical. For each type of abuse we
defined three or four categories, with one category classified
as severe.
Abuse variables. Four questions from Wyatt12 were adapted
to assess contact sexual abuse. The questions were introduced
with the statement, ‘Some people, while they are growing up in
International Journal of Obesity
their first 18 y of life, had a sexual experience with an adult or
someone at least 5 y older than themselves. These experiences
may have involved a relative, family friend or stranger. During
the first 18 y of life, did an adult, relative, family friend, or
stranger ever . . . (1) Touch or fondle your body in a sexual way?,
(2) Have you touched their body in a sexual way?, (3) Attempt
to have any type of sexual intercourse with you (oral, anal or
vaginal)?, (4) Actually have any type of sexual intercourse with
you (oral, anal or vaginal)?’
The categories of sexual abuse were none, touched only,
attempted intercourse and intercourse (the last defined as
severe).
Questions on verbal, fear of physical and physical abuse
were adapted from the Conflict Tactics Scale,13 in which
response categories are never, once or twice, sometimes,
often, or very often. ‘Once or twice’ and ‘sometimes’, were
combined as ‘sometimes’, ‘often’ and ‘very often’ were com-
bined as ‘often’.
Verbal abuse was assessed by asking, ‘How often did a
parent, step-parent, or adult living in your home swear at
you, insult you, or put you down?’ Categories were none,
sometimes and often (severe).
Fear of physical abuse was assessed by asking, ‘(1) How
often did a parent, step-parent, or adult living in your home
act in a way that made you afraid that you might be
physically hurt? and (2) How often did a parent, step-
parent, or adult living in your home threaten to hit you or
throw something at you, but didn’t do it?’ Categories were
none, sometimes and often (severe).
Physical abuse was assessed by asking, ‘(1) How often did a
parent, step-parent, or adult living in your home push, grab,
slap or throw something at you? and (2) How often did a
parent, step-parent, or adult living in your home hit you so
hard that you had marks or were injured?’ Categories were
none; hit, not injured; sometimes hit and injured, and often
hit and injured (the last two categories defined severe).
We also created four dichotomous abuse variables: any
mention of abuse; mention of one or more types of severe
abuse; mention of all four types of abuse, regardless of
severity; and mention of all four types of abuse, with all
types severe.
Weight and height. During the clinic examination, height
to the nearest inch and weight to the nearest pound were
measured while enrollee wore an examination gown without
shoes. Body mass index (BMI) was computed by dividing
weight in kg by height in m2. We defined obesity as
BMI
30, and severe obesity as BMI
40.14
Covariates. Age was coded as: 19 – 34, 35 – 49, 50 – 64, or

65 y. Ethnicity was white, black, Hispanic, Asian or other.
We used four dichotomous variables (violence against
mother, alcohol=drug abuse, mental illness, household
member in prison) to assess level of dysfunction in the
childhood household.15 Smoking status was coded as
never, former or current, and alcohol consumption as

none, 0.1 – 0.5, 0.6 – 1.0 and
1.0 drinks per day. Recrea-
tional physical activity was coded as none, 1 – 15, 16 – 30,
31 – 60, or
60 min per week. Education was classified as less
than high school, high school graduate, some college, or
college graduate, and employment status as unemployed,
employed full-time (
35 h=week), employed part-time (1 –
34 h=week), retired, or homemaker=student. Among women,
number of births was coded as 0, 1, 2, 3, 4 or
5.
Exclusions
We excluded 796 respondents who were pregnant at the time
of the mail surveys and 128 who had missing values for
weight and height. Additional exclusions included: 1190 for
the abuse variables; 424 for characteristics of the childhood
household environment; 184 for age, race or sex; and 1522
for smoking status, physical activity, alcohol, education,
employment or live births. These exclusions left 13 177 or
76% of the eligible sample.
Observations with missing values for the abuse variables
were re-coded in two ways: ‘not abused’ or ‘severely abused’.
The analysis was repeated but results were very similar
regardless of whether missing values were excluded or re-
coded.
Data analysis
We used Poisson regression to examine dependence among
the four types of child abuse. Linear regression was used to
estimate mean differences in body weight (kg) between those
exposed and unexposed to abuse. We used logistic regression
to estimate relative risks from abuse for BMI
30 and
40.
Because odds ratios from logistic regression may overesti-
mate relative risks, we used predictive margins16 to estimate
relative risks from predicted values produced by logistic
regression. Standard normal confidence intervals for log-
relative risks were calculated using bootstrap methods with
200 replications,17 then transformed back to the original
relative risk scale.
In regression analyses we estimated differences in body
weight and relative risks from three models with increasing
levels of control for covariates. Model 1 included categories
for a single type of abuse and was adjusted for survey wave
(linear regression models were also adjusted for height and
height2), sex, age and ethnicity. Model 2 was additionally
adjusted for household characteristics during childhood (vio-
lence against mother, alcohol=drug abuse, mental illness,
member in prison) and adult characteristics (smoking
status, physical activity, alcohol consumption, education,
employment status, and number of births (women)). For
visual clarity we report the 95% confidence intervals only
for model 2.
For completeness we show results for model 3, in which
we further adjusted for the other three types of abuse. This
approach may underestimate the impact of each type of
abuse if the four types of abuse co-occur.
Adult obesity and abuse in childhood
DF Williamson et al
1077
Population attributable fractions (PAF)18 and exposed
attributable fractions (EAF) were estimated for each of the
four dichotomous abuse variables. Predictive margins from
logistic regression models were used to estimate adjusted PAF
and EAF.19 We used bootstrap methods to calculate 95%
confidence intervals for attributable fractions. The PAF is
an estimate of the proportion of disease cases (eg BMI
30)
in the population that would be prevented by eliminating
the exposure (eg child abuse). The EAF is an estimate of the
proportion of cases, among those exposed, that would be
prevented if they were not exposed. The assumption under-
lying attributable fractions is that exposure causes the out-
come.
Interaction between the abuse variables and survey wave,
sex, age group (  55 vs > 55 y), and ethnic group (white,
black, Hispanic, Asian, other) was assessed by comparing
models with and without the interaction terms. For linear
regression we used the multiple partial F-test, and for logistic
regression we used the chi-square test of difference in log-
likelihoods. We found no evidence of interaction by survey
wave, sex, age, or ethnic group.
We used SAS statistical software20 for the Poisson
(GENMOD), linear (REG), and logistic (LOGISTIC) regression
analyses; GAUSS21 was used for predictive margin and boot-
strap estimation.
Results
Sample characteristics
Mean age of participants was 55.7 y, 51% were women,
nearly one-quarter were ethnic minorities and over 40%
were college educated (Table 1). Among women, the modal
number of births was two. Mean BMI of participants was
27.4; 25% had a BMI
30, and 2.4% had a BMI
40. Nine
percent of participants currently smoked, almost four-fifths
spent  15 min=week in recreational activity, and nearly
60% consumed alcohol. During childhood, over 20% wit-
nessed violence against their mother, 28% witnessed alcohol
or drug abuse in their households, 20% had a mentally ill
household member, and 5% had a household member in
prison.
Distribution of abuse types
Two-thirds of participants reported one or more type of
childhood abuse (Table 2). The most common type was
verbal (47.3%), followed by physical (44.5%), fear of physical
abuse (42.7%) and sexual (21.7%). Single types of abuse
occurred much less than expected if the abuse types were
independent; 4702 participants were expected to report only
one type of abuse, but only about half this number (2486)
did so. Similarly, 258 participants were expected to report all
four types of abuse, but nearly five times this number (1264)
did so. The hypothesis of independence among abuse types
was strongly rejected (w2 ¼ 11 278, d.f. ¼ 11, P < 0.0001).
International Journal of Obesity
 Adult obesity and abuse in childhood
DF Williamson et al
1078
Table 1 Characteristics of the Adverse Childhood Experiences (ACE) Table 2 Distribution of types of abuse by number of types reported in
Study samplea the Adverse Childhood Experiences (ACE) Study

Characteristic Value Number of types of abuse

a
reported
Social and demographic
Women 50.5 1 2 3 4 Row Total
Age, y 55.7 (14.7)a
No abuse reported — — — — 4419 (33.5b)
Ethnicity:
Sexual:
White 76.8
Touched only 366 233 210 530 1339 (10.2)
Black 4.5
Touched þ attempted 138 119 107 321 685 (5.2)
Hispanic 10.6
intercourse
Asian 6.3
Intercourse 164 121 138 413 836 (6.3)
Other 1.8
Education:
2860 (21.7)
Less than high school 6.0
Verbal:
High school graduate 16.4
Sometimes 869 1070 2261 780 4980 (37.8)
Some college 36.4
Often 50 100 624 484 1258 (9.5)
College graduate 41.2
Employment:
6238 (47.3)
Unemployed 19.9
Fear of Physical Abuse:
Employed full-time 47.6
Sometimes 339 1087 2406 839 4671 (35.4)
Employed part-time 12.7
Often 9 36 487 425 957 (7.3)
Retired 16.6
Homemaker=student 3.3
5628 (42.7)
Number of births (women):
Physical:
0 14.0
Hit, not injured 452 839 1447 428 3166 (24.0)
1 15.0
Hit, injured sometimes 99 307 1299 670 2375 (18.0)
2 33.2
Hit, injured often 0 16 153 166 335 (2.5)
3 21.4
4 10.0
2710 (20.6)

5 6.4
Column total: 2486 1964 3044 1264 13177 (100)
Anthropometry b
(18.9 ) (14.9) (23.1) (9.6)
Weight, kg 78.0 (17.5)
a
Height, m 1.68 (0.10) Column totals differ because they are specific for the number of types of
BMI, kg=m2 27.4 (5.1) abuse reported.
b
BMI category: Numbers in parentheses are percentage of grand total (13 177).
< 20 3.0
20 – < 25 31.5
25 – < 30 40.8
30 – < 35 17.0 For each type of abuse, the frequency of severe abuse rose
35 – < 40 5.3 with increasing number of types of abuse reported. For

40 2.4 example, among those exposed to only physical abuse, 99
Health practices (18%) reported being injured, while among those reporting
Smoking: all four types of abuse, 836 (66%) reported being injured.
Never 49.2
Former 42.1 This pattern was repeated for sexual, verbal and fear of
Current 8.7 physical abuse.
Recreational physical activity, min=week:
None 20.7
1 – 15 58.1
16 – 30 11.9 Abuse and adult body weight
31 – 60 7.4 All types of abuse were associated with increased weight in
> 60 1.9 adulthood (Table 3). After adjustment for all covariates
Alcohol consumption, drinks=day:
(model 2), often hit and injured was associated with the
None 41.5
0.1 – 0.5 54.2 largest increase in body weight (4.0 kg). The other types of
0.6 – 1.0 3.8 abuse were associated with weight increases about half as
> 1.0 0.5 large. Because the types of abuse are highly correlated, each
Household dysfunction in childhood abuse association decreased in magnitude after adjustment
Violence against mother: for the three other types of abuse (model 3).
Sometimes 18.9
Often 3.6
Alcohol=drug abuse in household 27.6
Mental illness in household 19.7 Abuse and adult obesity
Household member in prison 4.7
After adjustment for all covariates (Table 4, model 2), the risk
a
n ¼ 13 177. Age, weight, height, and BMI (body mass index) are continuous of BMI
30 increased most strongly for being often hit and
variables presented as mean (standard deviation). All other values are percentages. injured ( þ 39%). The other types of abuse were associated

International Journal of Obesity

 Adult obesity and abuse in childhood
DF Williamson et al
1079
Table 3 Mean differences in adult body weight associated with type of The largest population attributable fraction (PAF) of
childhood abuse in the adverse childhood experiences (ACE)a BMI
30, 8%, was for abuse defined as ‘any mention’. The
Type of abuse n Model 1 Model 2 Model 3 largest EAF of BMI
30, 19.4%, was for abuse defined as ‘four
types-severe’. The confidence interval around the EAF, how-
Sexual ever, was wide.
None (referent) 10317 0 0 0
Touched only 1339 0.9 0.6 (7 0.2 – 1.4)
b
0.4 The four definitions of abuse were associated with
Touched þ attempted 685 1.8 1.4 (0.2 – 2.5) 1.0 increases of about 20 – 90% in risk of BMI
40. The largest
intercourse PAF of BMI
40 was 17.3% for ‘any mention’ of abuse, and
Had intercourse 836 2.8 2.2 (1.2 – 3.2) 1.7 the largest EAF of BMI
40 was 48.3% for ‘four types-severe’.
Verbal The confidence intervals around the attributable fractions
None (referent) 6939 0 0 0 were wide, however.
Sometimes 4980 0.8 0.9 (0.4 – 1.5) 0.4
Often 1258 3.2 2.8 (1.9 – 3.8) 2.2

Fear of physical abuse Discussion

None (referent) 7549 0 0 0
Sometimes 4671 1.0 1.3 (0.8 – 1.8) 0.4
We found that adults who reported sexual and verbal abuse,
Often 957 2.8 2.2 (1.2 – 3.3) 7 0.8 fear of physical abuse, and physical abuse in childhood were,
on average, 0.6 – 4.0 kg heavier than adults who did not
Physical
None (referent) 7301 0 0 0 report abuse in childhood. These weight increases were
Hit, not injured 3166 0.6 0.8 (0.2 – 1.4) 0.2 translated into higher risks of adult obesity. Increases in
Hit=injured sometimes 2375 1.9 1.9 (1.2 – 2.6) 0.8 risk of BMI
30 ranged from 6 to 39%, and increases in
Hit=injured often 335 4.7 4.0 (2.4 – 5.6) 2.9
risk of BMI
40 ranged from 6 to 88%. Frequent verbal abuse
a
n ¼ 13 177. Estimates are from linear regression with current weight (kg) as and frequent physical abuse with injury were most strongly
the dependent variable and represent the difference in body weight between associated with increased risk of obesity in adulthood. In
the referent group and those with the specific type of abuse. Model 1 is addition, the risk of obesity increased with the number of
adjusted for survey wave, height, age, race and sex. Model 2 is additionally
adjusted for household dysfunction in childhood, smoking, alcohol, physical
types of severe abuse.
activity, education, employment status, and parity. Model 3 is additionally We also estimated the fraction of adult obesity cases that
adjusted for the other three types of abuse. See Methods for definitions of the were attributable to childhood abuse. In our study popula-
variables. tion, 8% of cases of BMI
30 and 17% of cases of BMI
40
b
95% confidence limits in parentheses.
were attributable to abuse in childhood. Surprisingly, we
were unable to identify any other published estimates of
the fraction of adult obesity attributable to child abuse, or
with risk increases of less than 30%. When each type of attributable to any other exposure. In one study of interest
abuse was adjusted for the other three types, the relative risks an estimated 17% of the cases of bulimia in US women was
decreased because of correlation among the abuse types attributable to childhood sexual abuse.22
(model 3). We found a weaker association between adult obesity and
Risk of BMI
40 was more strongly related to abuse than child abuse than the association found by Lissau and Sør-
risk of BMI
30. Being often verbally abused had the largest ensen between adult obesity and child neglect.7 The impact
increase in risk of 88%. Being often hit and injured increased of childhood exposures on adult body weight may weaken
the risk by 71%, intercourse and attempted intercourse over time; Lissau and Sørensen studied young adults aged
increased risk by 42 and 37%, while often being in fear of 20 y, while mean age in our study was 56 y. Their study was
physical abuse increased risk by 34%. Adjustment for the prospective, while our retrospective study required partici-
other three types of abuse (model 3) reduced the relative risk pants to recall childhood experiences that often occurred
for often fearing physical abuse by about 45%, but had less decades in the past. Thus, our findings may be attenuated by
impact on the other three types of abuse. failure to recall abuse in childhood by memory lapses, or
The risks of BMI
30 and
40 both increased with from repression of those memories. Furthermore, mechan-
increasing number of severe types of abuse (Figure 1). Expo- isms by which childhood neglect results in adult obesity may
sure to all four types of severe abuse (1%) had RR for be more powerful than mechanisms of abuse. In the child
BMI
30 and
40 of 1.46 (1.16 – 1.85) and 2.54 (1.21 – maltreatment literature, neglect is defined as acts of omis-
5.35). There was no evidence that risk of obesity increased sion including failure to provide for basic biological needs,
with increasing number of non-severe types of abuse (data abandonment, or lack of supervision; abuse is defined as acts
not shown). of commission by intentionally inflicted behaviors that can
harm a child.23 It is likely, however, that neglect and abuse
are not independent and also tend to co-occur.
Attributable fractions for adult obesity Our findings may also be attenuated because we adjusted
The four dichotomous abuse variables were associated with for participants’ adult characteristics. These characteristics
increases of about 15 – 25% in the risk of BMI
30 (Table 5). (smoking status, physical activity, alcohol consumption,

International Journal of Obesity

 Adult obesity and abuse in childhood
DF Williamson et al
1080
Table 4 Relative risk of BMI
30 and BMI
40 associated with type of childhood abuse in the Adverse Childhood Experiences (ACE)
Studya

Type of abuse n BMI
30 (%) Model 1 Model 2 Model 3

BMI
30
Sexual
None (referent) 10 317 23.8 1 1 1
Touched only 1339 26.1 1.08 1.07 (0.97 – 1.17) 1.05
Touched þ attempted intercourse 685 26.9 1.10 1.07 (0.95 – 1.22) 1.04
Had intercourse 836 33.4 1.36 1.29 (1.16 – 1.45) 1.25

Verbal
None (referent) 6939 23.1 1 1 1
Sometimes 4980 25.6 1.07 1.09 (1.03 – 1.16) 1.05
Often 1258 30.7 1.28 1.29 (1.16 – 1.42) 1.18

Fear of physical abuse

None (referent) 7549 23.0 1 1 1
Sometimes 4671 26.3 1.10 1.12 (1.04 – 1.20) 1.05
Often 957 31.2 1.29 1.26 (1.11 – 1.42) 1.03

Physical
None (referent) 7301 23.0 1 1 1
Hit, not injured 3166 25.5 1.08 1.09 (1.02 – 1.17) 1.03
Hit=injured sometimes 2375 27.8 1.14 1.13 (1.05 – 1.23) 1.04
Hit=injured often 335 35.8 1.45 1.39 (1.19 – 1.62) 1.20

BMI
40
Sexual
None (referent) 10 317 2.1 1 1 1
Touched only 1339 2.9 1.15 1.12 (0.81 – 1.55) 1.07
Touched þ attempted intercourse 685 4.7 1.49 1.37 (0.89 – 2.12) 1.30
Had intercourse 836 4.0 1.62 1.42 (0.96 – 2.10) 1.33

Verbal
None (referent) 6939 1.9 1 1 1
Sometimes 4980 2.5 1.11 1.15 (0.88 – 1.48) 1.10
Often 1258 5.1 1.84 1.88 (1.34 – 2.65) 1.97

Fear of physical abuse

None (referent) 7549 2.1 1 1 1
Sometimes 4671 2.7 1.20 1.21 (0.97 – 1.52) 1.10
Often 957 4.0 1.40 1.34 (0.88 – 2.04) 0.73

Physical
None (referent) 7301 2.1 1 1 1
Hit, not injured 3166 2.5 1.13 1.14 (0.86 – 1.52) 1.01
Hit=injured sometimes 2375 3.0 1.18 1.12 (0.83 – 1.52) 0.91
Hit=injured often 335 5.4 1.85 1.71 (1.02 – 2.89) 1.37
a
n ¼ 13 177. Estimates are from logistic regression-derived predictive margins with 95% confidence limits in parentheses. Model 1 is adjusted only for
survey wave, age, race and sex. Model 2 is additionally adjusted for household dysfunction in childhood, smoking, alcohol, physical activity,
education, employment status and parity. Model 3 is additionally adjusted for the other three types of abuse. See Methods for definitions of the
variables. BMI indicates body mass index.

education, employment status, and number of births
(women)) are potential intervening variables because they
occurred after the childhood abuse, and may have been at
least partially ‘caused’ by abuse in childhood. Thus by
adjusting for these variables we may have ‘over-controlled’
in our analysis, and the estimates presented in model 2 may
be biased downard.
Lissau and Sørensen suggested that neglect ‘may cause a
psychological state that affects energy balance by altering
behavior (overeating and physical inactivity) or hormone
balances, influencing fat storage (corticosteriods, catechola-
mines, or insulin).’7 Felitti suggested that compulsive eating
by patients who were sexually abused in childhood was an
attempt to manage the dysphoria related to negative child-
hood experiences.24 These mechanisms could also account
for maintenance of adult obesity, even if maltreatment in
childhood did not initially cause obesity.
Childhood abuse, especially sexual abuse, has been docu-
mented among obese persons seeking weight loss.25 Among
our sample reporting any abuse (67%), the fraction of obesity
cases attributable to abuse was 11% for BMI
30 and 23% for
BMI
40. This suggests that for a substantial minority of
obese adults abused as children, their obesity may result
from childhood experiences.

International Journal of Obesity


Figure 1 Relative risks of having a BMI
30 (white bars) or a BMI
40
(grey bars) by number of severe types of severe abuse; the referent group
is those reporting no types of severe abuse. Relative risks are adjusted for
all covariates including the number of types of abuse that occurred at
non-severe levels. Tests for linear trend were statistically significant for
BMI
30 (P < 0.001) and for BMI
40 (P ¼ 0.037).
Causality, however, cannot be established from a single
observational study. Causality can only be established using
external judgements about the plausibility of a study’s find-
ings, the putative role of bias in the findings, and the
replication of the findings over repeated studies by indepen-
dent investigators. Thus, our use of attributable fractions has
limited value if exposure to childhood abuse does not cause
adult obesity. In this study we did not know when obesity
first occurred. If obese participants were also obese as chil-
dren26 they may have been abused because they were
obese.27 If so, control for childhood obesity would have
reduced the associations in our study. Childhood weights
and heights, however, were not collected. Control for child-
hood obesity, however, may be inappropriate if the first
occurrence of obesity is not well established, and may
inappropriately attenuate the relationship between child
abuse and adult obesity. It is noteworthy that Lissau and
Sørensen7 controlled for child obesity in their analysis but
Adult obesity and abuse in childhood
DF Williamson et al
1081
found no impact on the association between childhood
neglect and adult obesity.
It is possible that obese adults are more likely than
normal-weight adults to falsely report abuse or they may
be more sensitive reporters of actual abuse. We were unable
to identify any studies to support either of these possibilities,
however. Only 51% of all available adult members of Kaiser
Permanente were included in our study sample. Thus it is
possible that extremely obese persons and persons most
seriously abused in childhood were under-represented in
our study. If non-response occurred more frequently in
those who were both obese and abused then we have under-
estimated the association of childhood abuse with adult
obesity.
In theory, prospective studies would avoid these potential
biases. In practice, however, prospective studies are difficult
to conduct properly because of misclassification of abuse.
Studies that use legal records to classify children as abused
may misclassify cases that never reach the criminal justice
system as ‘not abused’, thus biasing results toward the null.28
Parental or caretaker reports would be unlikely alternatives,
given the social and legal implications of self-identification
as a child abuser. In contrast, studies of neglect can use more
objective reports of teachers and health workers for exposure
classification.
Our study has several strengths. First, study participants
were unaware that their body weight, measured at a routine
clinical exam, would be related to their later reports of
childhood experiences. Because participants did not know
study intent, we believe there was no systematic over-report-
ing of abuse among obese participants. Second, the study’s
large sample size allowed control for 14 separate covariates,
each specified as a categorical variable. This approach mini-
mizes residual confounding. Third, in terms of study vari-
ables the participants were similar to those in the broader
general population. For example, in our study the mean BMI
was 27.4, and the prevalence of BMI
30 and
40 was 24.7
and 2.4%, respectively. In US adults aged
20 y these

Table 5 Childhood abuse relative risks (RR), population attributable fraction (PAF%) and exposed attributable fraction (EAF%) for adult BMI
30 and

40, using dichotomous definitions of childhood abuse: Adverse Childhood Experiences (ACE) Studya
b
Definition Prevalence (%) of abuse Prevalence (%) among BMI
30 RR (95% CI) PAF% (95% CI) EAF% (95% CI)

BMI
30
Any mention 66.5 70.0 1.13 (1.05 – 1.21) 8.0 (3.4 – 12.3) 11.4 (5.1 – 17.3)

1 type — severec 27.6 32.5 1.18 (1.10 – 1.26) 4.8 (2.5 – 7.1) 15.1 (9.0 – 20.8)
4 types — any severity 2.9 3.7 1.22 (1.04 – 1.43) 0.6 ( 7 0.3 – 1.5) 17.9 (3.7 – 30.0)
4 types — severe 1.0 1.4 1.24 (0.94 – 1.64) 0.2 ( 7 0.6 – 1.0) 19.4 ( 7 6.9 – 39.1)

BMI
40
Any mention 66.5 75.4 1.30 (0.99 – 1.70) 17.3 ( 7 1.0 – 32.4) 23.0 ( 7 0.6 – 41.1)

1 type — severec 27.6 37.4 1.20 (0.93 – 1.55) 6.2 ( 7 3.0 – 14.6) 16.7 ( 7 7.3 – 35.4)
4 types — any severity 2.9 6.5 1.80 (1.12 – 2.88) 2.9 ( 7 0.7 – 6.3) 44.4 (11.0 – 65.3)
4 types — severe 1.0 3.1 1.93 (0.96 – 3.90) 1.5 ( 7 1.4 – 4.3) 48.3 ( 7 4.5 – 74.4)
a
n ¼ 13 177. Referent groups: for ‘any mention’, ‘no abuse’. For ‘
1 type-severe’, ‘no abuse þ none severe’. For ‘4 types — any severity’, ‘no abuse þ 1 – 3 types’. For
‘4 types-severe’, ‘no abuse þ 1 – 3 types-severe þ 1 – 4 types-non-severe’.
b
Fully adjusted for all study covariates (model 2). See Methods for definitions of variables.
c
‘Severe’ abuse included sexual abuse with intercourse, often verbally abused, often in fear of physical abuse, and sometimes=often hit and injured.

International Journal of Obesity

 Adult obesity and abuse in childhood
DF Williamson et al
1082
estimates are 26.3, 22.5 and 2.8%.29,14 In our study 22% had
a history of sexual abuse in childhood, compared with 27%
in US adults.30 A history of physical abuse with injury was
reported by 21% of our study participants, the same as for
women, but below the 31% for men in the population of
Ontario, Canada.31
We found strong evidence that different types of child
abuse co-occur, with nearly three-quarters of the abused
reporting two or more types of abuse. The strong co-occur-
rence of different types of abuse in this and other studies32
questions the appropriateness of studying only single types
of abuse, and of adjusting the effects of individual types for
each other.
In summary, we found that abuse in childhood is related
to increased body weight and risk of obesity in middle age.
The proportion of cases of obesity that could be prevented
with elimination of childhood abuse was modest, however.
Still, given the multi-factorial etiology of obesity it is unli-
kely that any exposure, be it genetic, behavioral or environ-
mental, will account for a large proportion of cases in the
population. Our study suggests that clinical management of
obese adults who were abused as children may benefit from
understanding the mechanisms by which child abuse leads
to maintenance of adult obesity. This understanding may be
achieved by increased availability of psychological assistance
for such patients.
References
1 Comuzzie AG, Allison DB. The search for obesity genes. Science
1998; 280: 1374 – 1377.
2 Hill JO, Peters JC. Environmental contributions to the obesity
epidemic. Science 1998; 280: 1371 – 1374.
3 Stamler J. Epidemic obesity in the United States. Arch Intern Med
1993; 153: 1040 – 1044.
4 Dietz WH. Prevention of childhood obesity. Pediatr Clin N Am
1986; 33: 823 – 833.
5 Felitti VJ, Williams SA. Long-term follow-up and analysis of more
than 100 patients who each lost more than 100 pounds. Perma-
nente J 1998; 2: 17 – 21.
6 Maddi SR, Khoshaba DM, Persico M, Bleecker F, VanArsdall G.
Psychosocial correlates of psychopathology in a national sample
of the morbidly obese. Obes Surg 1997; 7: 397 – 404.
7 Lissau I, Sørensen TIA. Parental neglect during childhood and
increased risk of obesity in young adulthood. Lancet 1994; 343:
324 – 327.
8 Stunkard AJ, Sørensen TIA, Hanis C, Teasdale TW, Chakraborty R,
Schull WJ, Schulsinger F. An adoption study of human obesity.
New Engl J Med 1986; 314: 193 – 198.
9 Felitti VJ, Anda RF, Nordenberg D, Williamson DF, Spitz AM,
Edwards V, Koss MP, Marks JS. 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.
10 Dietz PM, Spitz AM, Anda RF, Williamson DF, McMahon PM,
Santelli JS, Nordenberg D, Felitti VJ, Kendrick JS. Unintended
pregnancy among adult women exposed to abuse or household
dysfunction during their childhood. JAMA 1999; 282: 1359 –
1364.
International Journal of Obesity
11 Anda RF, Croft JB, Felitti VJ, Nordenberg D, Giles WH, William-
son DF, Giovino GA. Adverse childhood experiences and smoking
during adolescence and adulthood. JAMA 1999; 282: 1652 – 1658.
12 Wyatt GE. The sexual abuse of Afro-American and White-Amer-
ican women in childhood. Child Abuse Negl 1985; 9: 507 – 519.
13 Straus M, Gelles RJ. Physical violence in American families: risk
factors and adaptations to violence in 8,145 families. Transaction
Press: New Brunswick, NJ; 1990.
14 Flegal KM, Carroll MD, Kuczmarski RJ, Johnson CL. Overweight
and obesity in the United States: prevalence and trends, 1960 –
1994. Int J Obes Relat Metab Disord 1998; 22: 39 – 47.
15 Schoenborn CA. Exposure to alcoholism in the family: United States,
1988. Advance Data From Vital and Health Statistics, no. 205. US
Department of Health and Human Services publication (PHS) 95-
1880. National Center for Health Statistics: Hyattsville, MD;
1995.
16 Graubard BI, Korn EL. Predictive margins with survey data.
Biometrics 1999; 55: 652 – 659.
17 Efron B, Tibshirani RJ. An introduction to the bootstrap. Chapman
and Hall: New York; 1993.
18 Rockhill B, Newman B, Weinberg C. Use and misuse of popula-
tion attributable fractions. Am J Pub Health 1998; 88: 15 – 19.
19 Greenland S, Drescher K. Maximum likelihood estimation of the
attributable fraction from logistic models. Biometrics 1993; 40:
865 – 872.
20 SAS Institute. SAS version 8.00 for Windows. SAS Institute: Cary,
NC; 2000.
21 Aptech Systems. GAUSS Mathematical and statistical system.
Aptech Systems Inc.: Maple Valley, WA; 1996.
22 Wonderlich SA, Wilsnack RW, Wilsnack SC, Harris TR. Childhood
sexual abuse and bulimic behavior in a nationally representative
sample. Am J Public Health 1996; 86: 1082 – 1086.
23 Lutzker JR, Boyle CL. Preventing physical and sexual abuse. In:
Innovative strategies for preventing psychological problems.
Glenwick D, Jason L (Eds). Springer: New York; 2002.
24 Felitti VJ. Childhood sexual abuse, depression, and family dys-
function in adult obese patients: a case control study. South Med J
1993; 86: 732 – 736.
25 King TK, Clark MM, Pera V. History of sexual abuse and obesity
treatment outcome. Addict Behav 1996; 21: 283 – 290.
26 Serdula MK, Ivery D, Coates RJ, Freedman DS, Williamson DF,
Byers T. Do obese children become obese adults? A review of the
literature. Prev Med 1993; 22: 167 – 177.
27 Wadden TA, Stunkard AJ. Social and psychological consequences
of obesity. Ann Intern Med 1985; 103(6 pt 2): 1062 – 1067.
28 Widom CS, Weiler BL, Cottler LB. Childhood victimization and
drug abuse: a comparison of prospective and retrospective find-
ings. J Consult Clin Psychol 1999; 67: 867 – 880.
29 Kuczmarski RJ, Flegal KM, Campbell SM, Johnson CL. Increasing
prevalence of overweight among US adults: the National Health
and Nutrition Examination Surveys, 1960 to 1991. JAMA 1994;
272: 205 – 211.
30 Finkelhor D, Hotaling G, Lewis IA, Smith C. Sexual abuse in a
national survey of adult men and women. Child Abuse Negl 1990;
14: 19 – 28.
31 MacMillan HL, Fleming JE, Trocme N, Boyle MH, Wong M,
Racine YA, Beardslee WR, Offord DR. Prevalence of child physical
and sexual abuse in the community: results from the Ontario
Health Supplement. JAMA 1997; 278: 131 – 135.
32 Ney PG, Fung T, Wickett AR. The worst combinations of child
abuse and neglect. Child Abuse Neglect 1994; 18: 705 – 714.