BACKGROUND: Childhood obesity is an important, potentially modifiable risk factor for a range of concurrent and later morbidities. Despite concerns about recent increases in children's body mass index (BMI), supporting data in Australia (as elsewhere) are scant.
OBJECTIVE: To seek anthropometric evidence of a recent secular increase in BMI in primary school children in Victoria, Australia.
DESIGN: Data from two cross-sectional population-based surveys of primary school children (the Victorian subsample of the 1985 Australian Health and Fitness Survey and the 1997 Health of Young Victorians Study) were compared. Similar stratified random sampling and standardized measurement methods were employed in the two studies. Subjects were all children aged 7–12 y with complete height and weight data. Body mass index (BMI (weight/height2)) was used as the index of relative adiposity. Non-parametric and parametric methods were used to examine the pattern and magnitude of change in BMI over the 12 y interval.
RESULTS: Data for 1421 children (50% male, 68% response) from the 1985 survey and 2277 children (51% male, 75% response) from the 1997 survey were analysed. At all ages, mean height and median weight were greater in 1997 than 1985 for both boys and girls. Median BMI was significantly higher in the 1997 sample for all but 12 y-old girls and for boys aged 7, 8 and 10 y (Mann–Whitney U test). The magnitude of the overall increase in BMI was estimated using analysis of covariance for log-transformed BMI adjusted for exact age, which indicated an increase of 1.03 kg/m2 for boys and 1.04 kg/m2 for girls (both P<0.001). Plots of BMI against BMI percentile clearly showed a pattern of higher BMI at any given percentile, especially at the upper percentiles, for all ages and both genders.
CONCLUSIONS: Primary school children in Victoria have become more obese over the last decade. Increases in BMI are most marked at the heavier end of the distribution. Lesser increases in median and mean BMI (confirmed by both parametric and non-parametric statistical models) may also have major public health implications.
Obesity is associated with numerous adverse health effects in children. Immediate psychosocial effects include social discrimination in childhood and self esteem problems in adolescents,1,2 while physical health problems include poorer pulmonary function,3 advanced growth, hyperlipidaemia, glucose intolerance, hepatic steatosis and cholelithiasis, and a wide range of less common pathologic conditions.2,4 Perhaps even more importantly, obesity in the school years is an important independent risk factor for adult obesity.5 In turn, adult obesity is strongly associated with a range of common chronic diseases including non-insulin dependent diabetes mellitus and coronary heart disease;6 even moderately elevated levels of body mass index (BMI) are associated with increased risk of mortality.7,8
The prevalence of adult obesity in both developed and developing countries is increasing.8 The situation appears to be similar among children.8 Recent reports from the US have shown that the prevalence of overweight has increased in nationally representative samples conducted between 1963 and 1991, with the largest increases occurring in the last 15 y.9,10
For Australian children, anthropometric data enabling the detection of changes over time are extremely limited, and the few published studies are compromised by data quality and sampling issues. Comparisons of three surveys of Queensland school children conducted in 1911, 1950 and 1976 suggested substantial increases in age-specific mean height and mean weight for both girls and boys.11 Using a variant of the conventional BMI (2.4 as the exponent on height instead of 2.0) and predicted percentage body fat, only small changes in BMI were reported for girls and almost no change for boys (unfortunately only mean values were available from the 1911 survey). Data from school children in Perth collected in 1983–1984 were compared with data from New South Wales school children collected in 1970–1972,12 and while children in the later study were slightly taller and heavier, differences in BMI were minimal. More recently, a 1992 survey of Melbourne children of European descent was compared with the 1985 Australian Health and Fitness Survey (AHFS) of children across Australia.13 This paper was the first to present data supporting a recent significant increase in BMI in Australian children, particularly among older girls. The most serious problem in interpreting these reports is the difficulty in drawing comparisons between geographically different populations, and other problems include probable differential sampling biases and a lack of technical detail, particularly in the earlier papers.
In summary, previous attempts to detect changes in simple anthropometry over time among Australian school children have consistently reported that children have increased in both height and weight, but have offered only a suggestion that the recent increase in obesity reported in American children is also occurring in Australia. The purpose of this report was to seek clear evidence of a change in anthropometry over time in two separate but comparable cross-sectional surveys of Australian school children conducted in 1985 and 1997.
Data from the Victorian subsample of children studied in the 1985 Australian Health and Fitness Survey (AHFS) were used as the baseline for the comparison. The AHFS has been described in detail elsewhere.14 Briefly, it was a nationally representative two-stage random sample of Australian schoolchildren aged 7–15 y stratified by school and age. It included measurement of height and weight using well-trained observers and calibrated equipment. Schools were randomly selected with a probability proportional to enrolment, and at each school a sample of 10 children (equal number of each gender) was selected within each age year-band between 7 and 15 y inclusive. Informed consent was obtained from a parent or guardian for each participant. The Victorian component of the survey took place in 13 schools over 3 months from June 1985.
The 1997 Victorian primary school sample was drawn from the Health of Young Victorians Study (HOYVS), an epidemiological study of health and wellbeing in children and adolescents in Victoria. Schools were selected using a two-stage random sampling design. In the first stage, schools were stratified by educational sector (government, Catholic and independent) with a probability proportional to total Victorian primary school enrolments, achieving a final sample of 24 primary schools. In the second stage, one intact class at each year level (preparatory to year 6) was randomly selected at each school, unless the total school population was less than 240, in which case the entire school was sampled. Informed consent was obtained from a parent or guardian for each participant. Height and weight were measured by trained observers using protocols based on the AHFS survey methods. Measurements took place over three months beginning September 1997. Table 1 shows technical details for the two studies.
Subjects for this study were all those with complete height and weight (and thus body mass index, BMI) data for the age range common to both surveys (7–12 y inclusive). BMI, calculated as weight in kilograms divided by the square of the height in meters (weight/height2), was used as the index of relative adiposity. The use of BMI in children has some known limitations.16,21 However it is highly predictive of measures such as skinfold thickness,15 densitometry,16,17 and other laboratory measures of body fat in children18 as well as of later adiposity.19,20 It is also practical for use in large-scale epidemiological surveys, and is now recommended as the field measure of choice in children for public health purposes.21
The project was approved by the Royal Children's Hospital Ethics in Human Research Committee.
Data were analysed by age and by gender, because of the substantial differences in growth patterns at different ages and between boys and girls. Exact age was calculated from the test date and birth date recorded in both samples (except AHFS children aged 7 and 8 y, for whom birth date was not recorded). Where exact age could be calculated, the Mann–Whitney U test for independent samples was used to test the assumption that the median age within each year of life in the two surveys was the same.
Because of the substantial right skew in BMI, non-parametric statistical methods were preferred. The skew was decreased both by log transformation and by application of Cole's ‘LMS’ method.22 However, we chose in the first instance not to rely on any distributional assumptions because the shape of the population distribution as well as measures of central tendency of BMI may have changed over time, with changes at the upper (obese) end of the distribution currently of particular interest. The Mann–Whitney U test was used to compare median values of age, weight and BMI between the two independent samples stratified by year of age and gender, while mean values for the more normally distributed height data were compared with independent sample t-tests.
Graphical methods were then used to compare distributions of BMI between 1985 and 1997. Within each sample, each subject's BMI value was ranked from smallest to largest for each age year-band and gender group separately. These ranks were then converted to percentile values between 0 and 100 for each subject. For each age and gender group, and for all ages combined in each gender group, plots of measured BMI against the percentile value for each observation were prepared. These graphs show clearly the percentiles at which the greatest divergence between the two populations occurs.
In order to estimate the overall magnitude of shift in BMI it was necessary to adjust for age, which required the use of parametric methods. Separate analysis of covariance models were constructed for each gender to test the hypothesis that mean BMI differed between the two surveys after adjustment for exact age. This parametric model used log-transformed BMI and permitted adjustment for exact calendar age; it was used even though there was evidence of some heteroscedasticity, as it is moderately robust to mild departures from distributional assumptions. This analysis was restricted to children aged 9–12 y because exact age was not available from the AHFS sample below age 9 y. Stata (Version 5.0) was used for all analyses.23
The achieved 1997 sample of 2277 children was larger than the 1985 sample of 1421 children, but otherwise the two samples were strikingly similar (Table 1). Response rates were slightly higher from the 1997 survey (75%) than the 1985 survey (68%).
Comparisons of means and standard deviations for height and of medians for age, weight and BMI for the two surveys are presented in Table 2 (females) and Table 3 (males). The younger median ages of the 1997 12 y-old girls and boys (P=0.0001 and P=0.001, respectively, Mann–Whitney U test) reflected the known incomplete sample at this year level, because many of the older 12 y-olds were already at secondary school. Very small but statistically significant differences were also noted for girls aged 9 and 11 y. Although from a clinical perspective these small age differences are unlikely to be important sources of bias, any interpretation of the findings below should take these differences in sample ages into account.
The median weight was greater in the 1997 sample for each age year-band. This difference was statistically significant at all ages except 12 y for girls and 9 y for boys. Similarly, the mean height was greater at every age in the 1997 sample and these differences were statistically significant for girls aged 7, 8 and 11 y and for boys aged 7, 8, 10, 11 and 12 y. As a result of these changes in height and weight, median BMI also increased for both genders in the 1997 sample. These differences were statistically significant for all but 12 y-old girls and for boys aged 7, 8 and 10 y.
Figures 1 and 2 show plots of measured BMI against BMI percentile for girls and boys respectively. The 1997 sample shows a clear pattern of an upward shift—a higher BMI at any given sample percentile value, particularly at the higher percentiles. Similar plots were obtained for each individual age year-band for boys and girls separately (not shown but available from the authors upon request).
Comparing mean BMI between the two surveys for boys and girls separately using analysis of covariance, mean BMI was increased by 1.03 kg/m2 for boys and 1.04 kg/m2 for girls (P<0.0001 for both) in 1997 compared with 1985 after adjustment for exact age.
Slight increases in height and substantial increases in weight, with a consequent increase in BMI, were found in Victorian primary school children from two comparable cross-sectional surveys conducted in 1985 and 1997. Results from the parametric modelling suggested a mean increase of slightly more than 1 unit of BMI over the 12 y interval after adjustment for age in both boys and girls. Plots showed convincingly that the most substantial change in BMI has occurred at the heavier end of the distribution, with relatively little change among children at the lighter end of the distribution. It seems reasonable to conclude that a true increase in BMI has taken place between 1985 and 1997, affecting Victorian primary school children of all ages. These findings are consistent with published reports in American children9,10 and British children,26,27 and with the recent WHO review suggesting reasonable evidence of increasing obesity among children in many developing countries.8
In comparing BMI data from the two samples separated in time, changes in the shape of the underlying distribution as well as changes in measures of central tendency are of interest. An important change in the shape of the distribution (such as an increase in the number of very obese children) could occur in the absence of a change in mean or median. Methods such as analysis using z-scores based on a reference population which rely on parametric assumptions about the underlying distributions of growth data are often reported. Analyses which assume that both sets of data (or some transformation of the data) are distributed in a Gaussian fashion could mask important changes in the distribution of obesity. In cases where both sets of data violate the assumptions in a similar way, it seems likely that both methods will lead to similar conclusions, but it is possible to imagine situations where parametric assumptions could lead to bias. We have therefore preferred to show the raw data and let readers draw their own conclusions.
Stratification by year of age is an effective method to control for any possible confounding due to age differences between the two samples. Although children in the 1997 sample were consistently taller and heavier across age groups and genders, median increases were not statistically significant in about half of the male age strata examined. These stratified analyses controlled for possible confounding from age differences, but the smaller sample sizes in each stratum made them less likely to find a true shift in the population BMI distribution (ie more prone to type 2 error). However, even when median shifts were not statistically significant, there were marked changes in the upper end of the distribution. This pattern is clearly seen in the figures, which show that for any given BMI percentile, the actual BMI value was larger in the more recent survey, particularly at higher BMI centiles. We believe that the figures provide the most convincing evidence of a change over time, particularly with the supporting results of the statistical analyses which confirm a central shift and provide an estimate of its overall magnitude.
The observed differences seem unlikely to result from systematic differences in measurement, differences in sampling, or changes in the age structures of Victorian primary school children. Error from systematic measurement differences was minimized by adopting the AHFS measurement protocols for use in the 1997 HOYVS sample. Although the instruments used were not identical, in both studies they were regularly calibrated, and height and weight are generally accepted as being among the most reliable and valid field anthropometric measures.24
The two surveys used slightly different sampling methods so that it is not possible to totally exclude differential sampling error as a bias in the results reported here, but the achieved samples appear strikingly similar. Differential response bias seems unlikely to explain our findings since the surveys achieved similar response rates using similar methodology. It is perhaps possible that more obese children might have been less enthusiastic about a survey of this type, especially in the later survey given recent media concerns about obesity and the degree of thinness that is currently in vogue. If anything, this would mean that our findings would be an underestimate of the change over time.
There were structural age differences within some of the age groups studied. The absolute magnitude of these differences was very small and the direction of the difference would be expected to bias our findings toward the null hypothesis of no increase in body size, since children in the 1985 survey were older in three of the four statistically significant comparisons. Because the AHFS did not record parent ethnic groupings we cannot assess the impact of possible changes in ethnicity between 1985 and 1997, but recent Australian immigration patterns25 are not such that we believe they would account for the observed increases in BMI.
A strength of this study is the consistent pattern which emerged from the three different analytic approaches (non-parametric analyses, plots and parametric models). We have avoided using any arbitrary BMI or weight cut-off points to define obesity. Although this method has the advantage of simplicity, it gives only a crude measure of change in the population distribution. For example, a large change in the proportion of children classified as obese could arise from a small change in the shape of the right tail of the distribution of BMI or from use of a different reference population, while a small change in the chosen cut-off point could result in a substantially different measure of change over time. As a result, this method is far from ideal. In addition, assessing only changes in the proportions of children classified as overweight or obese ignores smaller increases in BMI affecting very large numbers of children which could have important long-term implications at a population level. Between 1985 and 1997 median BMI increased, which by definition means a change affecting at least half of the population. This median increase was most marked in girls, but (unlike recent American research) was seen in primary school children of all ages, rather than being limited to older children.9
In summary, significant differences in height, weight and in BMI have been found between two comparable surveys conducted 12 y apart. These differences suggest that Victorian primary school children of all ages were taller, heavier and more adipose in 1997 than in 1985. The significance of the differences between the two surveys was confirmed by both parametric and non-parametric statistical models. It seems difficult to avoid the conclusion that Australian children are getting fatter.
Stunkard A, Mendelson M . Obesity and body image. I. Characteristics of disturbances in the body image of some obese persons Am J Psychiat 1967 123: 1296–1300.
Dietz WH . Health consequences of obesity in youth: childhood predictors of adult disease Pediatrics 1998 101: 518–525.
Lazarus R, Colditz G, Berkey CS, Speizer FE . Effects of body fat on ventilatory function in children and adolescents: cross-sectional findings from a random population sample of school-children Pediatr Pulmonol 1997 24: 187–194.
Dietz W, Gross W, Kirkpatrick J . Blount Disease (tibia vara): another skeletal disorder associated with childhood obesity. J Pediatr 1982 101: 735–737.
Whitaker RC, Wright JA, Pepe MS, Seidel KD . Predicting obesity in young adulthood from childhood and parental obesity New Engl J Med 1997 337: 869–873.
DeFronzo R, Ferrannini E . Insulin resistance: A multifaceted syndrome responsible for NIDDM, obesity, hypertension, dyslipidemia and atherosclerotic cardiovascular disease Diabetes Care 1991 14: 173–194.
Manson JE, Willett WC, Stampfer MJ, Colditz GA, Hunter DJ, Hankinson SE, Hennekens CH, Speizer FE . Body weight and mortality among women New Engl J Med 1995 333: 677–685.
World Health Organization Consultation on Obesity Obesity: preventing and managing the global epidemic WHO/NUT/NCD/98.1: Geneva 1998.
Troiano RP, Flegal KM . Overweight children and adolescents: description, epidemiology, and demographics Pediatrics 1998 101: 497–504.
Rosner B, Prineas R, Loggie J, Daniels SR . Percentiles for body mass index in U.S. children 5 to 17 years of age. J Pediatr 1998 132: 211–222.
Dugdale AE, O'Hara V, May G . Changes in body size and fatness of Australian schoolchildren 1911–1976 Austral Paediatr J 1983 19: 14–17.
Hitchcock NE, Maller RA, Gilmour AJ . Body size of young Australians aged five to 16 years Med J Austral 1986 145: 368–372.
Hill M, Nowson C, Tavener M, Higgins R, Loesch D . Relationship of dietary restraint and activity patterns to body mass index in Melbourne primary and secondary school children: a preliminary study Austral J Nutr Diet 1997 54: 118–125.
Australian Council for Health Physical Education and Recreation Australian Health and Fitness Survey 1985 ACHPER Publications: Adelaide 1985.
Dwyer T, Gibbons LE . The Australian schools health and fitness survey: physical fitness related to blood pressure but not lipoproteins Circulation 1994 89: 1539–1544.
Lazarus R, Baur L, Webb K, Blyth F . Adiposity and body mass indices in children: Benn's index and other weight for height indices as measures of relative adiposity Int J Obes Relat Metab Disord 1996 20: 406–412.
Pietrobelli A, Faith MS, Allison DB, Gallagher D, Chiumello G, Heymsfield SB . Body mass index as a measure of adiposity among children and adolescents: a validation study J Pediatr 1998 132: 204–210.
Deurenberg P, Weststrate J, Seidell J . Body mass index as a measure of body fatness: age- and sex-specific prediction formulas Br J Nutr 1991 65: 105–114.
Rolland-Cachera M.F, Cole T.J, Sempe M, Tichet J, Rossignol C, Charraud A . Body mass index variations: centiles from birth to 87 years Eur J Clin Nutr 1991 45: 13–21.
Clarke WR, Lauer RM . Does childhood obesity track into adulthood? Crit Rev Food Sci Nutr 1993 33: 423–430.
Dietz WH, Robinson TN . Use of the body mass index (BMI) as a measure of overweight in children and adolescents [editorial; comment] J Pediatr 1998 132 (2): 191–193.
Cole TJ, Green PJ . Smoothing reference centile curves: the LMS method and penalized likelihood Stat Med 1992 11: 1305–1319.
Stata Corporation Stata Statistical Software: Release 5 Stata Press: College Station, TX 1997.
Willett W . Anthropometric measures and body composition. In: MacMahon B (ed) Nutritional epidemiology, 1st edn. Monographs in Epidemiology and Biostatistics, Vol 15: Oxford University Press: New York 1990, pp 217–244.
Australian Bureau of Statistics Overseas arrivals and departures, Australia Commonwealth of Australia: ABS Catalogue No. 3401.0 1997.
Reilly J, Dorosty A, Emmett P . Prevalence of overweight and obesity in British children: cohort study BMJ 1999 319: 1039.
Hughes JM, Li L, Chinn S, Rona RJ . Trends in growth in England and Scotland, 1972 to 1994 Arch Dis Childhood 1997 76: 182–189.
We would like to gratefully acknowledge the funding support provided by the Victorian Department of Human Services which enabled the Health of Young Victorians Study to be conducted.
About this article
Cite this article
Lazarus, R., Wake, M., Hesketh, K. et al. Change in body mass index in Australian primary school children, 1985–1997. Int J Obes 24, 679–684 (2000). https://doi.org/10.1038/sj.ijo.0801218
- body mass index
- secular trend
Project Spraoi: two year outcomes of a whole school physical activity and nutrition intervention using the RE-AIM framework
Irish Educational Studies (2019)
Exercise and BMI in Overweight and Obese Children and Adolescents: A Systematic Review and Trial Sequential Meta-Analysis
BioMed Research International (2015)
The prevalence of underweight, overweight and obesity in a multiracial group of urban adolescent schoolchildren in the Cape Metropole area of Cape Town
South African Journal of Clinical Nutrition (2014)
Trends in weight, height, BMI and obesity in schoolchildren from Santa Rosa (Argentina), 1990–2005/07
Annals of Human Biology (2013)
Prevalence of obesity among Arab school children in Nazareth, Israel: comparison with national (Jewish) and international data
Pediatric Obesity (2013)