RISK FACTORS FOR CHILDHOOD OVERWEIGHT: A PROSPECTIVE STUDY
FROM BIRTH TO 9.5 YEARS
W. STEWART AGRAS, MD, LAWRENCE D. HAMMER, MD, FIONA MCNICHOLAS, MD, AND HELENA C. KRAEMER, PHD
hypothesized risk factors drawn from research reports.
Five independent risk factors for childhood overweight were found. The strongest was parent overweight, which was
mediated by child temperament. The remaining risk factors were low parent concerns about their child’s thinness, persistent
child tantrums over food, and less sleep time in childhood. Possible mechanisms by which each of these factors influence weight
gain are outlined. Two different pathways to childhood overweight/obesity were found, depending on degree of parental
There is evidence of considerable interaction between parent and child characteristics in the development of
overweight. Several of the identified risk factors are amenable to intervention possibly leading to the development of early
prevention programs. (J Pediatr 2004;145:20-5)
obesity in both children and adults. Although not all cases of adult obesity begin in childhood, a considerable proportion do so;
hence thedevelopmentofeffectivepreventionprogramsin childhood isimportant.This,in
turn, depends on the identification of modifiable risk factors for childhood overweight.
Risk factors for childhood overweight include parental obesity, socioeconomic status
(SES), birth weight, physical activity, and diet. Potential risk factors include child and
parent behavior and child temperament. Parental obesity probably contributes both genetic
and family environmental influences for childhood overweight.2-5Childhood SES is
inversely related to overweight and obesity and is a risk factor for both.4Race is a risk factor
for overweight independent of SES, with blacks and Hispanics having a higher risk than
whites.5Higherbirth weightand rapidgrowthin the first few months of lifeare risk factors
for overweight both in children and adults.6,7Measures of activity have been inconsistently
related to adiposity in childhood,9,10although some studies have found a modest
relation.11-14Despite the evident importance of dietary intake in obesity, caloric and
macronutrient intakes have been inconsistently associated with overweight both in
childhood and adolescence.10This may be due to difficulties in assessing diet, and the fact
that a small variation in caloric intake, which may not be measurable, would over time lead
to overweight. Learning within the family environment may lead to food likes and dislikes
and may affect caloric intake and activity levels.
In the current exploratory study, we prospectively assessed many of the established
and hypothesized risk factors for the development of childhood overweight.
To ascertain risk factors for the development of overweight in children at 9.5 years of age.
This was a prospective study of 150 children from birth to 9.5 years of age, with assessment of multiple
hildhood overweight continues to increase in prevalence, with theproportion of children above the 85thpercentile of body
mass index (BMI) reaching 31% in the most recent National Health and Nutrition Examination.1This will lead to
increases in adult overweight and obesity together with the consequent health problems associated with overweight and
See editorial, p 3.
From the Departments of Psychiatry
and Pediatrics, Stanford University
School of Medicine, Stanford, Califor-
Dr McNicholas is now at University
College, Dublin, Ireland.
Supported by grant HD-25492 from
the National Institute of Child Health
Submitted for publication May 19, 2003;
last revision received Feb 20, 2004;
accepted Mar 5, 2004.
Reprint requests: W. S. Agras, MD,
Department of Psychiatry, Stanford
University School of Medicine, 401
Quarry Rd, Stanford, CA 94305-5733.
0022-3476/$ - see front matter
SES Socioeconomic Status
Newborn infants (n = 216) and their parents were
recruited from the well newborn nurseries at a university
hospital, community hospital, and a health maintenance
organization in the San Francisco Bay Area. Recruitment
and eligibility procedures have been detailed elsewhere.15This
study was approved by the Stanford University Committee for
the Protection of Human Subjects. Both parents gave written
consent to participate in this study and also gave permission
for the participation of their infant. Of the original 216 infants
whose parents consented to participate, 150 (74 boys and 76
girls) were followed to 9.5 years.
These data included parent age, education, and ethnic
Parent Weight Status
Parent height and weight (for mothers, these were
higher of the parent’s baseline BMIs was used to characterize
Infant Weight Status
The weight and length of the infant were assessed at the
first laboratory visit at 2 weeks of age.
INFANT FEEDING BEHAVIORS. Infant sucking behavior was
recorded during a feed in the laboratory at 2 and 4 weeks of
age, following the methods detailed previously.8Based on
preliminary analyses, sucking pressure and burst duration were
used to characterize sucking behavior. The number of feeds
per day was acquired during a 24-hour period in which
mothers weighed their infants before and after each feeding on
a highly sensitive scale.
Early Weight Gain
Infant weight and length were assessed at birth and 6
months of age to determine early rate of growth.
Maternal Expectations of Infant Feeding
The Maternal Feeding Attitudes Scale, which assesses
the degree of ‘‘maternal pushiness’’ in infant feeding, was
obtained atthe firstlaboratoryvisit, togetherwith ameasureof
maternal perception of actual and desired infant body habitus
by using the Infant Body Habitus Scale, which presents
drawings of male and female infants from thin to plump. Both
the Maternal Feeding Attitudes Scale and Infant Body
Habitus Scale have a high test-retest reliability.16
Parent/Infant Feeding Practices
These data were collected from the first month of life
through the use of a monitoring form completed by mothers
for 3 days each month until the child was weaned from both
breast and bottle. These records were either mailed or
telephoned into the laboratory. If the record was mailed,
aresearch assistant telephonedthemothertoascertainthatthe
record was correct and to prompt the mother for further
information if necessary. Variables derived from these records
included time of introduction to solid food, duration of breast
feeding, duration of bottle feeding, and time of weaning from
any form of suckling defined as the later of breast and bottle
Parent Eating Behaviors
These measures were obtained from the Three-Factor
Eating Questionnaire, which measures dietary restraint,
hunger, and disinhibition.17The test-retest reliability of the
subscales is high. The Eating Disorder Inventory was used to
assess perfectionism, a trait associated with disordered
Maternal Return to Work
The date of maternal return to work, which has been
shown to affect infant feeding,15was obtained.
Childhood Eating Behaviors
Behaviors such as picky eating, rapid eating, tantrums
over food, and nonnutritive food uses were assessed at ages 2,
3, 4, and 5 years through the use of parental questionnaires.
Each of these measures was stable between ages 2 and 5 years,
and the annual assessments were averaged. A measure of
overinterest in food was not stable, hence each year was
entered separately. The eating behavior of the child was also
directly observed during laboratory sessions at 3 and 5 years of
age, as detailed previously.19Variables derived from these
observations were total eating time, number of bites per
weaning from breast and bottle was obtained from the
monthly feeding reports.
Twenty-Four-Hour Child Caloric Intake
Caloric intake was assessed during weekdays at 3 and 5
years of age. A cooler containing a variety of foods and drinks
was delivered to the participant’s home. The cooler contained
approximately 8000 kcal, with 59.9% carbohydrate, 14.2%
protein, and 25.9% fat. Parents were instructed to feed their
child only from the cooler, including any meals or snacks that
would be eaten out of the home. If the child ate an item not
included in the cooler, or if another child ate something from
the cooler, the parent was instructed to note this on a food log,
to return any uneaten portion of food not contained in the
cooler along with the food carton or wrapper, and to note the
items eaten by another child. All uneatenfood and drinkswere
replaced in special containers labeled by food type, and the
coolers were picked up from the home the next day.
Risk Factors for Childhood Overweight: A Prospective Study
from Birth to 9.5 Years
Activity level was assessed in two ways: first, by means of
an accelerometer that the children wore during the day for at
least 24 hours at 3 and 5 years of age,20and second, indirectly
by annual parental report of the child’s usual duration of TV
watching at ages 2 through 5 years.
Temperament was assessed at 5 years of age with the
Child Behavior Questionnaire,21which has a high test-retest
The length of the child’s sleep was assessed by annual
parent report at ages 2 through 5 years. This measure was
stable from 3 to 5 years, which data were averaged. The 2-year
data were not stable and were entered separately.
Maternal Weight Gain
Maternal weight and height were obtained at the 5.5-
year visit, allowing a calculation of weight gain over a 5-year
period. Insufficient data were collected from fathers to allow
for a similar calculation.
Parent Concerns about Child’s Weight
Concerns about their children’s weight (overweight and
underweight) were assessed by annual questionnaire at ages 2
through 5 years.
Parent Feeding Behaviors
Parent behaviors in feeding their children included
nonnutritive food uses and the use of food limits, by using an
annual questionnaire at ages 2 through 5 years. The scales for
parental concerns and nonnutritive food uses were stable over
that period, and the results were averaged; however, the
measure of food limits varied across time and hence each year
was entered separately.
Laboratory Measures of Parental Behaviors
Maternal prompting of their child’s eating was directly
observed during a standardized laboratory meal at 3 and 5
years of age, as described previously.19
The parental authority questionnaire was used to
measure parenting style, including authoritarian, authorita-
tive, and permissive styles of behavior.22
Forthisstudy, overweightwasdefinedasBMI above the
85thpercentile for age and sex. The first stage of analysis used
logistic regression to examine the association between each
possible risk factor and the binary outcome above or below the
85thpercentile of BMI at 9.5 years. Only variables found
statistically significant (P < .05) were pursued further. The
analytic strategy used the linear model described by Kraemer
et al.23,24Possible risk factors were ordered in terms of time.
First, within each time frame (birth, infancy, childhood), all
pairs of variables were considered to identify proxy risk factors,
overlapping risk factors, and independent risk factors. Second,
by using the time line from birth through childhood, the same
approach was used to identify any further proxy risk factors,
moderators (a risk factor that interacts with outcome and that
may explain on whom or under what conditions another risk
factor works), and mediators (a risk factor that is correlated
with a preceding risk factor and that explains some of the
variance in outcome and may explain why and how another
risk factor works).
To make the results more accessible to clinical
consideration and application, recursive partitioning methods
were used with the identified risk factors to develop specific
Table. Results of recursive partitioning showing numbers and percentages of children between 85th and 95th
percentiles and above 95th percentile by the six risk groups, together with the number at risk for each group,
and for the two levels of parental BMI
85th to 95th
Parental BMI <27.5 n = 98 No tantrums, high concerns about
thinness, n = 53
No tantrums, low concerns about
thinness, n = 29
Persistent tantrums, n = 16
High concerns about thinness,
n = 17
Low concerns about thinness,
n = 12
Low concerns about thinness,
irritable impulsive personality,*
n = 19
2 (3.8)0 (0)
6 (20.7)2 (6.9)
2 (11.8)Parental BMI >27.5 n = 52
4 (33.6)0 (0)
15 (78.9) 7 (36.8)
*Four participants missing the personality measure.
22Agras et al The Journal of Pediatrics?July 2004
rules for identification of high and low risk children. Recursive
partitioning is nonparametric and distribution free that
identifies in this case subgroups of children at high and low
risk for overweight. A series of optimal cut-points are selected,
continuing with each subsequent branch until the P value is
<.01. For further details, see Kraemer.25
The demographics for the study sample and for those
who dropped out of the study for whom an adequate data set
was not available were evaluated. There was only one
significant difference between these two groups, with the
mothers of dropouts being somewhat less highly educated
(v2= 11.3, P = .004).
Primary Risk Factor Analysis
Of the 150 children in this study, 38 (25.3%) were above
the 85thpercentile of BMI at 9.5 years of age, including 14
(9.0%) above the 95thpercentile. The primary analysis
identified 5 risk factors that are shown in the Figure, together
with the significant correlates for each factor. The strongest
relation was with parental BMI (W = 13.7, P < .001), which
was mediated by temperament of the child assessed as
a combination of approach and impulsivity (W = 6.6, P <
.01). This factor is correlated with an active personality and
anger/frustration, indicating highly emotional behavior. The
remaining risk factors were low parental concern about their
child’s thinness (W = 5.2, P < .02); children with persistent
tantrums over food (W = 9.3, P < .002); and children’s hours
of sleep at ages 3 to 4 years (W = 6.6, P < .01).
To understand the mechanism through which low
parental concern about their child’s weight may affect
overweight, we used stepwise regression with the following
potential independent measures: early feeding measures
(sucking pressure, burst duration, number of bursts, number
of sucks, 24-hour intake, and number of feeds per day in the
first month); early growth (birth weight, weight gain in the
first 6 months, BMI at 12 and 24 months); and maternal
attitudes (Infant Body Habitus Scale and Maternal Feeding
Attitudes Scale). The most powerful predictor of low concerns
about thinness was weight gain in the first 6 months of life
(B = ?0.3, t(118) = ?3.6, P = .001), followed by birth weight
(B = 0.24, t(118) = ?2.8, P = .006) and maternal preference
for a thin baby (B = 0.19, t(118) = 2.8, P = .006). It appears
that this risk factor reflectsa parent with a preference fora thin
child faced with a developmental pattern of higher birth
weight, rapid weight gain in the first 6 months of life, and as
shown by the correlates of this risk factor (Figure), an eating
pattern comprising rapid eating and a high interest in food.
Tantrums over food were positively related to over-
weight, with persistent tantrums about 3 times more likely to
occur in those who would become overweight. Although 82%
of children had such tantrums at some time during childhood,
only 19% had persistent tantrums reported annually from 2
through 5 years of age. Hours of sleep reported annually from
3 through 5 years were negatively related with overweight;
children who were to become overweight were reported to
sleep about 30 minutes less on average than those who would
remain normal weight. This difference was almost entirely due
to shorter daytime sleep, with only 5 minutes’ difference in
Figure. Risk factors for overweight children at 9.5 years of age together with the variables that correlate significantly with each risk factor.
Risk Factors for Childhood Overweight: A Prospective Study
from Birth to 9.5 Years
Recursive partitioning was used to identify subgroups
with a differential risk of having a BMI at or above the 85th
percentile at 9.5 years of age. The 5 identified risk factors were
entered into this analysis. The results of this partitioning are
shown in the Table. The first cut-point was into maximal
parental BMI above and below a value of 27.5 kg/m2,
suggestingtwodifferentpathways fostering childhood obesity.
A child with an overweight/obese parent had a 48.1% chance
of becoming overweight or obese, whereas a child with
normal-weight parents had a 13.3% chance of becoming
overweight or obese. For parents who were not overweight, if
their child had persistent temper tantrums over food during
years 2 through 5, 31.2% would become overweight compared
with 9.8% for those without tantrums. For children with
overweight parents with low concern about their child’s
thinness, reflecting the developmental pattern described
above, 62.9% would become overweight/obese compared with
17.6% for those without such concerns. If, in addition, the
child had the highly emotional temperament, 78.9% would
become overweight/obese compared with 33.6% for those
without this temperamental profile. The percentages of
children above the 85th and 95th percentiles for each of these
6 risk groups are also shown in the Table. Within each main
division (groups 1 through 3 and 4 through 6) indicated by
parental weight below or above a BMI of 27.5, the percentages
Risk factors for obesity must result in a positive energy
balance through 1 or more of 3 pathways: a higher caloric
intake than required to sustain normal growth; lower caloric
output through physical activity; or an alteration in metabo-
lism that affects caloric balance. Confirming previous
observations,2,3,10the most potent risk factor in this study
was parental overweight. Parental overweight has a direct
effect on childhood overweight, as has been described
previously, as well as an effect mediated by the child’s
temperament. One possibility is that temperament and
overweight are linked genetically. An alternative mechanism
is that overweight parents may overcontrol the behavior of
such children,26,27for example, excessively prompting food
use. This in turn may increase caloric intake, as shown in
concern about their child’s thinness. The mechanism behind
this relation appears to be parents who prefer a thin child but
are faced with an infant with a high birth weight, rapid weight
gain in the first 6 months of life, rapid eating, and a high
interest in food. This combination of parental preference and
a child with an avid feeding pattern may lead to overcontrol of
the child’s feeding behaviors, disrupting the child’s learning of
self-control thus furthering the effects of the avid feeding
Persistent tantrums over food during childhood is
a further risk factor. Tantrums are correlated with some of
the attributes of the highly emotional personality (Figure).
Parents faced with an emotional child who has tantrums over
food may feed the child to reduce the frequency of tantrums
rather than use nonfood methods to reduce such tantrums.
Hence, this risk factor is likely to lead to weight gain through
increased caloric consumption.
The final risk factor is a difference in sleep pattern, with
overweight children sleeping about 30 minutes less than
nonoverweight children. This difference has previously been
reported in a cross-sectional study in early childhood.30In the
present study, the difference in sleep length was almost
entirely accounted for by less daytime sleep. Correlates of this
sleep pattern suggest that these children have low activity
levels in childhood, affecting weight through reduced caloric
expenditure. Children with low activity may sleep less during
the day because they are less tired.
The recursive partitioning analysis suggests that the
pathways to childhood overweight differ, depending on
parental overweight. For the thinner parent, the highest risk
group is for the parent confronted with a child who has
persistent tantrums over food, perhaps leading to reactive
overfeeding of the child. For the overweight parent, the
developmental pattern associated with low concern about
their child’s thinness combined with a child with a highly
emotional personality confers the highest risk for overweight
Some risk factors for childhood overweight described in
breast-feeding has been shown to be protective of over-
weight31; however, in our sample, the majority of participants
were breast-fed, with 90% breast-feeding in the first month of
life and an average length of breast-feeding of >6 months. The
low proportion of non–breast-feeders may have precluded our
finding that breast-feeding is protective. Activity levels, even
though assessed twice by means of an accelerometer, were also
not related to overweight. Many studies, however, have also
found no relation prospectively between activity levels and
overweight.10This may be due to the difficulty in assessing
activity levels or that such assessment may need to be more
proximal to the assessment of overweight.
Some limitations of the study should be discussed. First,
the sample size is relatively small (n = 150). Second, the level
of parent education is higher than that in the general
population and the percentage of minorities lower, limiting
the generalizability of the findings. Third, some of the critical
assessments, for example, childhood personality attributes,
were obtained byparentalreport,possibly biasingtheresults of
this assessment. Fourth, a substantial number of the original
participants dropped out by 9.5 years of age or provided
incomplete data. Moreover, those who dropped out tended to
be less well educated than those who did not, again potentially
biasing the results.
Although the prevention of obesity may require
alterations to an environment that fosters inactivity and the
consumption of calorie-dense foods, the family is also
a potential arena for prevention. The family provides an
24Agras et al The Journal of Pediatrics?July 2004
environment in which patterns of food intake and activity are
learned early in life. The current study sheds some light on
child and parent attributes and behaviors that may affect food
intake and activity patterns. Parental attitudes toward their
child’s weight may lead to behaviors that increase the risk of
their child becoming overweight. Such behaviors may be
amenable to change. Similarly, children’s behaviors such as
temper tantrums and the behavior difficulties posed by the
highly emotional child, which in turn may lead to overfeeding,
are amenable to parental counseling through the use of well-
documented, effective child treatment procedures.32The
findings of this study suggest that at-risk children and parents
may be identifiable in the first few years of a child’s life,
affording the opportunity for early preventive interventions.
Further research is needed to determine whether the in-
dependent risk factors identified in this study are causal risk
factors. This will require a prospective study in which one or
more of the identified risk factors are experimentally
manipulated to determine whether changing the risk factor
will lower the risk for childhood overweight.
trends in overweight among US children and adolescents. JAMA 2002;288:
2.Stunkard AJ, RHJ, Perdersen NL, McClearn GE. The body-mass index
of twins who have been reared apart. N Engl J Med 1990;322:1483-7.
3.Whitaker RC, Wright JA, Pepe MS, Seidel KD, Dietz WH. Predicting
obesity in young adulthood from childhood and parental obesity. N Engl J
4.Eriksson J, Forsen T, Osmond C, Barker D. Obesity from cradle to
grave. Int J Obes Relat Metab Disord 2003;27:722-7.
5. Strauss RS, Pollack HA. Epidemic increases in childhood overweight.
6. Eid EE. Follow-up study of physical growth of children who have had
excessive weight gain in the first six months of life. BMJ 1970;2:74-6.
7. Stettler N, Zemel BS, Kumanyika S, Stallings VA. Infant weight gain
and childhood overweight status in a multicenter cohort study. Pediatrics
8. Agras WS, Kraemer HC, Berkowitz RI, Korner AF, Hammer LD.
Does a vigorous feeding style influence early development of adiposity?
J Pediatr 1987;110:799-804.
9. Barkeling B, Ekman S, Rossner S. Eating behavior in obese and normal
weight 11-year old children. Int J Obes Relat Metab Disord 1992;16:355-60.
10. Parsons TJ. Childhood predictors of adult obesity: a systematic review.
Int J Obes Relat Metab Disord 1999;23:S1-107.
11. Berkowitz RI, Agras WS, Korner AF, Kraemer HC, Zeanah CH.
Physical activity and adiposity: a longitudinal study from birth to childhood.
J Pediatr 1985;106:734-8.
Ogden CL, Flegal KM, Carroll MD, Johnson CL. Prevalence and
of accelerated weight gain in preschool children. Pediatrics 1995;95:126-30.
13.Dietz WH. Television, obesity, and eating disorders. Adolescent
14. Robinson TN. Reducing children’s television viewing to prevent obesity:
a randomized controlled trial. JAMA 1999;282:1561-7.
15.Hammer LD, Bryson S, Agras WS. Development of feeding practices
during the first 5 years of life. Arch Pediatr Adolesc Med 1999;153:
16. Kramer MS, Barr RG, Leduc DG, Boisjoly C, Pless BI. Maternal
psychological determinants of infant obesity: development and testing of two
new instruments. J Chronic Dis 1983;36:329-35.
17.Stunkard AJ, Messick S. The Three Factor Eating Questionnaire to
measure dietary restraint, disinhibition, and hunger. J Psychosom Res 1985;
18. Garner DM, Olmstead MP, Polivy J. Development and validation of
a multidimensional eating disorder inventory for anorexia nervosa and
bulimia. Int J Eat Disord 1985;2:15-34.
19. DruckerRR,HammerLD,Agras WS,BrysonS. Canmothersinfluence
their child’s eating behavior? J Dev Behav Pediatr 1999;20:88-92.
20.Metcalf BS, Curnow JS, Evans C, Voss LD, Wilkin TJ. Technical
reliability of the CSA activity monitor: the early bird study. Med Sci Sports
21. Rothbart MK. Measurement of temperament in infancy. Child Dev
22.Buri JR. Parental authority questionnaire. J Personality Assess 1991;57:
23.Kraemer HC, Stice E, Kazdin AE, Kupfer D. How do risk factors work
together to produce an outcome? Mediators, moderators, and independent,
overlapping, and proxy risk factors. Am J Psychiatry 2001;158:848-56.
24. Kraemer HC, Wilson GT, Fairburn CG, Agras WS. Mediators and
moderators of treatment effects in randomized clinical trials. Arch Gen
25. Kraemer HC. Evaluating medical tests: objective and qualitative
guidelines. Newbury Park Calif: Sage Publications; 1992.
26.Wells JC, Stanley M, Laidlaw AS, Day JM, Stafford M, Davies PS.
Investigation of the relationship between infant temperament and later body
composition. Int J Obes 1997;21:400-6.
27. Carey WB, Hegvik RL, McDevitt SC. Temperamental factors
associated with rapid weight gain and obesity in middle childhood. J Dev
Behav Pediatr 1988;9:194-8.
28.Johnson SL, Birch LL. Parents’ and children’s adiposity and eating style.
29.Birch LL, Fisher JO. Mothers’ child-feeding practices influence
daughter’s eating and weight. J Am Diet Assoc 2000;100:641-6.
30.Sekine M, Yamagami T, Hamanishi S, Handa K, Tomohiro S,
Nanri S, et al. Parental obesity, lifestyle factors and obesity in preschool
children: Results of the Toyama Birth Cohort Study. J Epidemiol 2001;
31.Toschke AM, Vignerova J, Lhotska L, Oscanova K, Koletzko B, Von
Kries R. Overweight and obesity in 6- to 14-year-old Czech children in 1991:
protective effect of breast-feeding. J Pediatr 2002;141:749-57.
32.Sanders MR, Dadds MR. Behavioral family intervention. Needham
Heights, Mass: Allyn & Bacon; 1993.
Klesges RC, Klesges LM, Eck LH, Shelton ML. A longitudinal analysis
Risk Factors for Childhood Overweight: A Prospective Study
from Birth to 9.5 Years