Académique Documents
Professionnel Documents
Culture Documents
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
Data analysis
We used Poisson regression to examine dependence among Results
the four types of child abuse. Linear regression was used to Sample characteristics
estimate mean differences in body weight (kg) between those Mean age of participants was 55.7 y, 51% were women,
exposed and unexposed to abuse. We used logistic regression nearly one-quarter were ethnic minorities and over 40%
to estimate relative risks from abuse for BMI 30 and 40. were college educated (Table 1). Among women, the modal
Because odds ratios from logistic regression may overesti- number of births was two. Mean BMI of participants was
mate relative risks, we used predictive margins16 to estimate 27.4; 25% had a BMI 30, and 2.4% had a BMI 40. Nine
relative risks from predicted values produced by logistic percent of participants currently smoked, almost four-fifths
regression. Standard normal confidence intervals for log- spent 15 min=week in recreational activity, and nearly
relative risks were calculated using bootstrap methods with 60% consumed alcohol. During childhood, over 20% wit-
200 replications,17 then transformed back to the original nessed violence against their mother, 28% witnessed alcohol
relative risk scale. or drug abuse in their households, 20% had a mentally ill
In regression analyses we estimated differences in body household member, and 5% had a household member in
weight and relative risks from three models with increasing prison.
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 Distribution of abuse types
adjusted for household characteristics during childhood (vio- Two-thirds of participants reported one or more type of
lence against mother, alcohol=drug abuse, mental illness, childhood abuse (Table 2). The most common type was
member in prison) and adult characteristics (smoking verbal (47.3%), followed by physical (44.5%), fear of physical
status, physical activity, alcohol consumption, education, abuse (42.7%) and sexual (21.7%). Single types of abuse
employment status, and number of births (women)). For occurred much less than expected if the abuse types were
visual clarity we report the 95% confidence intervals only independent; 4702 participants were expected to report only
for model 2. one type of abuse, but only about half this number (2486)
For completeness we show results for model 3, in which did so. Similarly, 258 participants were expected to report all
we further adjusted for the other three types of abuse. This four types of abuse, but nearly five times this number (1264)
approach may underestimate the impact of each type of did so. The hypothesis of independence among abuse types
abuse if the four types of abuse co-occur. was strongly rejected (w2 ¼ 11 278, d.f. ¼ 11, P < 0.0001).
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
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
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 by patients who were sexually abused in childhood was an
(women)) are potential intervening variables because they attempt to manage the dysphoria related to negative child-
occurred after the childhood abuse, and may have been at hood experiences.24 These mechanisms could also account
least partially ‘caused’ by abuse in childhood. Thus by for maintenance of adult obesity, even if maltreatment in
adjusting for these variables we may have ‘over-controlled’ childhood did not initially cause obesity.
in our analysis, and the estimates presented in model 2 may Childhood abuse, especially sexual abuse, has been docu-
be biased downard. mented among obese persons seeking weight loss.25 Among
Lissau and Sørensen suggested that neglect ‘may cause a our sample reporting any abuse (67%), the fraction of obesity
psychological state that affects energy balance by altering cases attributable to abuse was 11% for BMI 30 and 23% for
behavior (overeating and physical inactivity) or hormone BMI 40. This suggests that for a substantial minority of
balances, influencing fat storage (corticosteriods, catechola- obese adults abused as children, their obesity may result
mines, or insulin).’7 Felitti suggested that compulsive eating from childhood experiences.
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.