Each data marker represents 5 cases; each line off of a data marker
represents an additional 5 cases. Lines of best fit were generated using locally
weighted regression techniques.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Williams J, Wake M, Hesketh K, Maher E, Waters E. Health-Related Quality of Life of Overweight and Obese Children. JAMA. 2005;293(1):70–76. doi:10.1001/jama.293.1.70
Context The negative effects of childhood overweight and obesity on quality
of life (QOL) have been shown in clinical samples but not yet in population-based
Objective To determine relationships between weight and health-related QOL reported
by parent-proxy and child self-report in a population sample of elementary
Design, Setting, and Participants Cross-sectional data collected in 2000 within the Health of Young Victorians
Study, a longitudinal cohort study commenced in 1997. Individuals were recruited
via a random 2-stage sampling design from primary schools in Victoria, Australia.
Of the 1943 children in the original cohort, 1569 (80.8%) were resurveyed
3 years later at a mean age of 10.4 years.
Main Outcome Measures Health-related QOL using the PedsQL 4.0 survey completed by both parent-proxy
and by child self-report. Summary scores for children’s total, physical,
and psychosocial health and subscale scores for emotional, social, and school
functioning were compared by weight category based on International Obesity
Task Force cut points.
Results Of 1456 participants, 1099 (75.5%) children were classified as not overweight;
294 (20.2%) overweight; and 63 (4.3%) obese. Parent-proxy and child self-reported
PedsQL scores decreased with increasing child weight. The parent-proxy total
PedsQL mean (SD) score for children who were not overweight was 83.1 (12.5);
overweight, 80.0 (13.6); and obese, 75.0 (14.5); P<.001.
The respective child self-reported total PedsQL mean (SD) scores were 80.5
(12.2), 79.3 (12.8), and 74.0 (14.2); P<.001.
At the subscale level, child and parent-proxy reported scores were similar,
showing decreases in physical and social functioning for obese children compared
with children who were not overweight (all P<.001).
Decreases in emotional and school functioning scores by weight category were
Conclusion The effects of child overweight and obesity on health-related QOL in
this community-based sample were significant but smaller than in a clinical
sample using the same measure.
In 2003, Schwimmer et al1 reported in JAMA that “severely obese children and adolescents
have lower health-related QOL [quality of life] than children and adolescents
who are healthy and similar QOL as those diagnosed as having cancer.”
Child overweight and obesity are now so endemic that many countries are reporting
prevalences of 25% or higher.2 If the associations
seen in the study by Schwimmer et al remain for overweight and obese children
across the population, then a substantial proportion of children and adolescents
could be experiencing major reductions in health-related QOL due to their
Health-related QOL refers to the subset of QOL directly related to an
individual’s health,3 which as defined
by the World Health Organization includes physical, mental, and social well-being.4 During the last decade, new instruments measuring
health and well-being have been developed that are suitable for epidemiological
studies. In adults,5 adolescents,6 and
children,7 these measures have been shown to
discriminate among population groups known to have different levels of health.
Because it is subjective, health-related QOL should be assessed whenever possible
from the participant’s (ie, the child’s) perspective. Nonetheless,
the parent-proxy’s perspective about the child’s health is also
important because a parent’s perspective is likely to be a strong driver
of health service use.8 However, few studies
have compared parents’ and children’s perspective of how health
varies according to weight and none have been performed in community samples
of elementary school children. This largely reflects the absence of parallel
parent and child measures until recently.6
If childhood overweight and obesity lead to a significant reduction
in health-related functioning, this could have population ramifications beyond
the known complications associated with concurrent and future cardiovascular,
endocrine, and psychosocial morbidity.9-13 To
date, only 2 population-based studies relating health and well-being to child
body mass index (BMI)13 have been published,
both suggesting a less striking relationship with weight status than that
reported by Schwimmer et al.1 Both studied
children of elementary school age and used the parent-reported Child Health
Questionnaire as their measure of a child’s health-related QOL. Parents
of children in the “at risk for overweight” category in the US
study and the comparable “overweight” category in the Australian
study reported similar scores to normal-weight children on nearly all Child
Health Questionnaire scales. For children in the “overweight”
category in the US study and the comparable “obese” category in
the Australian study, parents were more likely to report reduced scores on
some scales, but even for these heaviest groups many domains appeared unaffected.
However, neither study included children’s reports of their own health
because the Child Health Questionnaire does not have a child self-report suitable
for this age group.
The PedsQL is a short survey instrument assessing physical, emotional,
social, and school functioning. Strengths of the PedsQL include the availability
of parallel reports by a parent-proxy and a child and relatively low ceiling
effects, which occur if most participants achieve a near perfect score on
a questionnaire. Using this measure, we hypothesized that PedsQL scores reported
in parallel by elementary school children and parents in a large Australian
community-based sample would decrease with increasing weight. In exploratory
analyses, we examined whether age, sex, and socioeconomic status modify apparent
effects of overweight and obesity. Finally, hypothesizing that major effects
will be limited to more severe degrees of overweight that may not coincide
with current empirical BMI cut points, we explored possible relationships
between health-related QOL and BMI to look for evidence of a threshold effect.
The Health of Young Victorians Study was established in 1997. Sampling
and methods have been reported in detail previously.14 Briefly,
participants were selected from the state of Victoria, Australia (population
4.69 million in 199815), using a stratified
2-stage sampling design based on school education sector (government, Catholic,
or independent) and school class level. For the primary school cohort, 24
schools were randomly selected with a probability proportional to size, and
one class at each year level from each school was then randomly selected.
The baseline response rate for students in grades kindergarten through third
grade (ages 5-8 years) in 1997 was 83.2% (1943 of 2336 identified children).
The achieved sample mirrored Victorian census data for age distribution, sex,
ethnicity (parental county of birth), and proportion of indigenous persons.
This study draws on cross-sectional data from the follow-up conducted
in 2000 when 1569 children (80.8% of the original sample of children in kindergarten
through third grade) were resurveyed when in grades 3 through 6. Each child’s
height and weight were measured and each child and one of his/her parents
completed brief written questionnaires. In 2000, the children’s ages
ranged between 8 and 13 years, but 99.3% were aged 9 to 12 years. Because
of the small number of 8- and 13-year-olds, it was not possible to calculate
meaningful mean PedsQL scores for these age groups, so only the 1456 children
aged 9 to 12 years were retained in the analyses.
The study was approved by the ethics in human research committee of
the Royal Children’s Hospital, the Victorian Department of Education,
the Catholic Education Office, and the Independent Schools Office. A parent-proxy
provided written informed consent.
Children’s height and weight were measured at school by trained
field workers. Investigation of anthropometric reliability found no evidence
of systematic bias for intra- or inter-rater comparisons. Body mass index
was calculated as the weight in kilograms divided by the square of height
in meters. Children were classified into 3 mutually exclusive categories of
not overweight, overweight, or obese according to the sex- and age-specific
cut points developed by the International Obesity Task Force.16 The
cut points were derived from the results of nationally representative surveys
of childhood BMI conducted in 6 different countries. For each survey, centile
curves were drawn that at age 18 years passed through the BMI cut points for
overweight and obesity (25 and 30), and the curves were then averaged to provide
age- and sex-specific cut points from 2 to 18 years. Furthermore, using the
US Centers for Disease Control and Prevention 2000 Growth Chart data,17 BMI was transformed to externally standardized z scores based on sex and age to adjust for right skew
in the BMI distribution and physiological changes that occur in BMI with age.
We used Nutstat software (EPI Info, Centers for Disease Control and Prevention,
Atlanta, Ga) to calculate the BMI transformations.
The PedsQL 4.0 is a validated 23-item questionnaire for children aged
2 to 18 years. It assesses physical, emotional, social, and school functioning,
from which total, physical, and psychosocial health summary scores are derived.
The best possible score on the PedsQL is 100 (range of 0-100). Near-identical
parallel parent-proxy and child self-report versions are available, which
were completed independently by the parents (at home) and the children (at
school) in this study. Recently published US population normative data indicate
high levels of internal consistency for both the self-report and parent-proxy
report in 8- to 12-year-olds.18
Using population census data, the Australian Bureau of Statistics computes
socioeconomic indices.19 These indices summarize
the social and economic conditions of Australia by geographic area. The disadvantage
index (population mean [SD] of 1000 ) was used in this study as a measure
of neighborhood socioeconomic status with each individual assigned to a disadvantage
index score based on the postal area in which he/she lived. The disadvantage
index distribution for the population was then divided into 4 quartiles as
equally as possible, with quartiles 1 and 4 representing the lowest and highest
disadvantage indexes, respectively. As a proxy for personal socioeconomic
status, maternal education was stratified into 3 categories (<11 years,
11 or 12 years with or without a diploma or trade certificate, and >12 years).
The sample size had at least 80% power to detect differences of 5 points
on each of the summary or subscale scores. Male and female differences in
demographic variables were explored using χ2 tests for categorical
variables and t tests to compare means of continuous
variables. Total, physical, and psychosocial summary and subscale scores were
computed for both the parent-proxy and child self-reported PedsQL according
to the manual. One-way analyses of variance were used to test for significant
overall mean differences by weight category groupings. Univariate generalized
linear models were used to determine the estimated marginal means of the PedsQL
scales and subscales adjusting for the child’s age, sex, maternal education,
and disadvantage index as covariates. There were interactions between maternal
education and disadvantage index but no other interactions were significant.
Bonferroni tests were applied to the estimated marginal means to provide pairwise
comparisons between weight categories to determine which (if any) groups most
influenced any significant overall differences found. Finally, to examine
possible threshold and nonlinear relationships that might cut across current
overweight and obese cut points, locally weighted regression techniques were
used to generate lines of best fit for the relationships between BMI z scores and PedsQL total, summary, and subscale scores
that differed significantly by weight category; P <.05 was
considered significant. Data were analyzed using SPSS statistical software
(version 11.5, SPSS Inc, Chicago, Ill).
Demographic and weight characteristics of the sample are shown in Table 1. The sample reflects population characteristics
in terms of sociodemographics, but as noted in previously published data,
children retained in the cohort had a lower mean BMI at baseline than children
lost to follow-up (16.9 vs 17.5; P<.001).20 The mean (SD) age of the child participants was 10.4
(1.1) years (range, 9-12 years), the mean (SD) BMI z score
was 0.50 (0.92), and the prevalences of overweight and obesity were 20.2%
and 4.3%, respectively. A slightly higher proportion of females than males
were classified as overweight or obese (21.3% overweight and 4.9% obese females
vs 19.2% overweight and 3.8% obese males), but this difference was not significant
(P = .33). Males and females did not differ
significantly by height, weight, BMI, age in years, weight category, location
of residence, disadvantage index, or maternal education; therefore, except
where specified, males and females were combined for analyses.
Table 2 shows the mean (SD) scores
for each of the PedsQL scales and subscales by weight category. Parent and
child perceptions were strikingly similar, with obese children having the
lowest summary and subscale scores. Parent- and child-reported physical and
social functioning and total scores decreased significantly with increasing
weight, as did the child self-reported psychosocial summary score (P = .004). Similar, although not statistically significant,
trends were seen for the parent-proxy psychosocial summary score and school
functioning and for parent- and child-reported emotional functioning.
Scores were lowest for children in the obese category. For illustrative
purposes, the published total PedsQL scores from the clinical sample of severely
obese children and adolescents reported by Schwimmer et al1 are
also reproduced in Table 2. Our community sample of obese children reported
PedsQL scores that were between the children who were not overweight and the
severely obese clinical sample.
Mean scores for both parent-proxy and child self-reported total PedsQL
scores across the 3 weight categories were stratified by sex, age, maternal
education, and disadvantage index (Table 3).
PedsQL scores were inversely related to weight category for both sexes according
to responses from both children and parents. Low maternal education was associated
with a significant decrease in total PedsQL score by higher weight category.
Although parents and children reported lower QOL with increasing weight status
in children of highly educated mothers, these decreases were not statistically
significant. Similar trends held across all quartiles of the disadvantage
index, with obese children having lower QOL scores than those who were not
overweight, but the strength of association varied. Parents reported a significant
decrease in total PedsQL score by weight category for all age groups except
the oldest, while only children aged 11 years reported significant decreases
in scores, although decreases were present in all age groups.
Table 4 shows estimated marginal
mean (SD) PedsQL scores after adjustment for age, sex, maternal education,
and disadvantage index. As in the univariate analyses, the decreases in total,
physical, and social PedsQL scores with increasing weight category remained
significant for responses from both children and parent-proxies even after
multivariable adjustment. Pairwise comparisons between weight categories showed
that parents reported a significant decline in total and physical PedsQL with
each category increase. For children, the difference in total and physical
PedsQL scores between not overweight and overweight were not significant;
however, a significant decrease in the social functioning subscale across
each increase in weight category was observed. Parents did not report significant
differences in the psychosocial summary scores across weight categories, while
children reported a significant decrease in the psychosocial summary score
between the not overweight category and the obese category.
Relationships between total, physical, psychosocial, and social PedsQL
scores and BMI z scores are shown in the Figure. The lines of best fit show that all of
these scores decrease with increasing BMI, and in most cases the decline begins
just above the mean BMI.
To ensure that definitions of weight category did not influence our
findings, we reanalyzed all data using Centers for Disease Control and Prevention’s
age- and sex-specific cut points, under which 19.1% of the sample were classified
as at risk of overweight and 11.5% as overweight. However, this did not alter
the conclusions of the study. All statistically significant relationships
remained significant, and all nonsignificant relationships remained nonsignificant.
In a large community-based sample, we demonstrated that children’s
health-related QOL measured by the PedsQL decreases across not overweight,
overweight, and obese 9- to 12-year-old children in Victoria, Australia. The
decrease was small for overweight children but more marked for those who were
obese. These new observations are less dramatic than the much lower scores
reported for children attending tertiary clinics,1 but
are consistent with those observed for adults.21
Because health and QOL are multidimensional, it is feasible that some
dimensions may be more affected by overweight or obesity. Overweight and obese
children differed from children who were not overweight most strongly on physical
and social functioning scores, while emotional and school functioning seemed
relatively unaffected. This study confirms and extends previous research restricted
to reports by parent-proxies.13,21,22 The
child self-report and parent-proxy versions of the PedsQL are nearly identical,
making it possible to show just how closely parents and children agree on
aspects of child functioning across the gradient of BMI. This study has shown
that in contrast to other observed health concerns, children’s own subjective
response to overweight and obesity is not more negative than that of their
parents, at least for the broad domains measured herein. However, given that
there is evidence of less agreement between parent and adolescent reports
of health and well-being, future data collection from this cohort may demonstrate
whether agreement dissipates as children age.6,7 We
have confidence in our results because the sample was drawn from a large,
representative, population-based cohort with high response rates in which
BMI had been accurately measured. In addition, the PedsQL conforms to the
multidimensional conceptual framework of the World Health Organization’s
definition for health, has reliable population norms and good measurement
properties, and has been used widely.
Despite the high response rates, studying the second wave of a longitudinal
study limited our ability to generalize to the whole population of Victorian
children. There is some evidence that heavier children may have been underrepresented
in the 2000 study. If these children also had systematically better health-related
QOL than those not followed up, our findings could have been erroneous. However,
our results showed that heavier children were more likely to report lower
QOL and if these children had remained in the study our findings probably
would have been strengthened. Our study was limited to children aged 8 to
13 years. A study incorporating younger and older children and adolescents
is needed to determine applicability of our findings to other age groups.
Because of its generic nature, the PedsQL allows comparisons of health across
whole populations but it may not tap into particular areas of concern experienced
by overweight and obese children or other aspects of QOL.
Although there were sound reasons for using the PedsQL, debate continues
regarding what constitutes health-related QOL and how it differs from health
and functional status. The PedsQL assesses “difficulties or problems
with . . . ” or ill-being. A consensus is emerging
in the pediatric literature, which concurs with research on adults, that ill-being
is not low well-being, and that the absence of ill-being does not necessarily
equate with high well-being.23 It is possible
that overweight and normal weight children may differ little in ill-being
but could experience a lower level of well-being, which the PedsQL would not
capture. In addition, while we may expect that lower QOL is associated with
lower functioning, functional status measures alone in children might be inadequate
as a proxy for QOL because children may adapt to their current health state
or may not have experienced a healthier state, which could result in high
self-reported QOL scores despite obvious functional limitations.24
Our results indicate that health-related QOL, or functioning, begins
to decline as soon as a child is above average weight, with a gradual steepening
as BMI increases. To determine whether a specific BMI threshold must be reached
to stimulate motivation to seek treatment (and whether this directly relates
to perceived health), it would be necessary to conduct a separate study to
investigate the issue of health care use and the associated clinical implications.
These findings have both positive and negative implications. As for
any chronic condition, a relatively small effect of overweight on children’s
functioning across multiple domains is welcome. However, if neither children
nor parents perceive a health effect, it seems unlikely that they will seek
health care or initiate behavioral change that might lead to a healthier BMI,
and consequently lessen the long-term health risk for the current generation
of children. Our findings may explain why so few parents of overweight children
express concern about their child’s weight, yet with a quarter of all
children now overweight or obese, even a minor reduction in health-related
QOL at an individual level is still likely to have a major effect at a population
Further research is needed to determine if these findings can be replicated
in different age groups and countries, and to investigate the temporal relationships
between overweight and obesity and QOL in children. It is not known whether
the health-related QOL in overweight and obese children decreases in response
to their weight, whether children with lower health-related QOL from the outset
are more likely to become overweight, or whether there is a transactional
relationship mediated by the presence of comorbidities or by factors extrinsic
to the child, such as parental mental health. These are complex questions
that will require careful study.
Corresponding Author: Joanne Williams, PhD,
Centre for Community Child Health, Royal Children’s Hospital, Flemington
Road, Parkville 3052, Australia (firstname.lastname@example.org).
Author Contributions: Dr Williams had full
access to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Wake, Hesketh, Waters.
Acquisition of data: Wake, Hesketh.
Analysis and interpretation of data: Williams,
Drafting of the manuscript: Williams, Wake,
Critical revision of the manuscript for important
intellectual content: Williams, Hesketh, Waters.
Statistical analysis: Williams, Hesketh.
Obtained funding: Wake, Hesketh, Waters.
Administrative, technical, or material support:
Study supervision: Wake, Waters.
Funding/Support: This study was supported by
grants from the National Heart Foundation, Financial Markets for Children,
and Murdoch Childrens Research Institute. Dr Wake is supported by a National
Health and Medical Research Council (NHMRC) Career Development Award; Ms Hesketh
by an NHMRC Public Health Postgraduate Scholarship; Dr Maher by an NHMRC project
grant to develop a measure of quality of life for children with cerebral palsy;
and Dr Waters by a Victorian Health Promotion Foundation Public Health Research
Role of the Sponsors: The National Heart Foundation,
Financial Markets for Children, and Murdoch Childrens Research Institute had
no role in the design and conduct of the study. They also did not provide
input into the conception of the research, data acquisition, analysis or interpretation,
nor in supervision of the researchers.
Acknowledgment: We acknowledge the contribution
of all field workers who conducted the data collection.