Establishing a standard definition for child overweight and obesity worldwide: international survey

Subjects

We obtained data on body mass index for children from six large nationally representative cross sectional surveys on growth from Brazil, Great Britain, Hong Kong, the Netherlands, Singapore, and the United States (table 1). Each survey had over 10 000 subjects, with ages ranging from 6-18 years, and quality control measures to minimise measurement error. Four of the datasets were based on single samples whereas the British and American data consisted of pooled samples collected over a period of time. We omitted the most recent survey data from the United States (1988-94) because we preferred to use data predating the recent increase in prevalence of obesity.¹⁹ In practice this decision made virtually no difference to the final cut off points.

Centile curves

Centile curves for body mass index were constructed for each dataset by sex using the LMS method,¹⁵ which summarises the data in terms of three smooth age specific curves called L (lambda), M (mu), and S (sigma). The M and S curves correspond to the median and coefficient of variation of body mass index at each age whereas the L curve allows for the substantial age dependent skewness in the distribution of body mass index. The values for L, M, and S can be tabulated for a series of ages. The Brazilian and US surveys (table 1) used a weighted sampling design, and their data were analysed accordingly.

The assumption underlying the LMS method is that after Box-Cox power transformation the data at each age are normally distributed. The points on each centile curve are defined in terms of the formula¹⁵:

where L, M, and S are values of the fitted curves at each age, and z indicates the z score for the required centile, for example, z=1.33 for the 91st centile. Figure 1 shows centiles for body mass index by sex based on the British reference,⁹ with seven centiles spaced two thirds of a z score apart—that is, z=−2, −1.33, −0.67, 0, +0.67, +1.33, and +2.

Figure 1 also shows body mass index values of 25 and 30 kg/m² at age 18; 25 kg/m² is just below the 91st centile in both sexes, whereas 30 kg/m² is above the 98th centile. The body mass index (BMI) values can be converted to exact z scores from the L, M, and S values at age 18, with the formula²:

The body mass index of 25 kg/m² at age 18 is z score +1.19 in females, corresponding to the 88th centile, and +1.30 in males, on the 90th centile. Therefore the prevalence of overweight at age 18 is 10-12%. A body mass index of 30 kg/m² at age 18 is on the 99th centile in both sexes, an obesity prevalence of about 1%.

Each z score substituted into equation 1 provides the formula for an extra centile curve passing through the specified point (dotted line in fig 1). Each centile curve defines cut off points through childhood that correspond in prevalence of overweight or obesity to that of the adult cut off point—the curve joins up points where the prevalence matches that seen at age 18.

This process is repeated for all six datasets, by sex. Superimposing their curves leads to a cluster of centile curves that all pass through the adult cut off point yet represent a wide range of overweight and obesity. The hypothesis is that the relation between cut off point and prevalence at different ages gives the same curve shape irrespective of country or obesity. If sufficiently similar the curves can be averaged to provide a single smooth curve passing through the adult cut off point. The curve is representative of all the datasets involved but is unrelated to their obesity—the cut off point is effectively independent of the spectrum of obesity in the reference data.

Agreement of the centile curves

The major uncertainty with our approach, and the test of its validity, is the extent to which the centile curves for the datasets are of the same shape. Figures 3 and 4 show that although the agreement is reasonable it is not perfect. If it were perfect—that is, all the curves were superimposed—the reference cut off points applied to a given dataset would give the same prevalence for obesity at all ages, which could be predicted from the prevalence at age 18. So the different shapes in figures 3 and 4 show to what extent the age specific prevalence deviates from the age 18 prevalence within datasets.

We did consider six other datasets for our analysis (Canada, France, Japan, Russia, Sweden, and Venezuela) but we excluded them because they were either too small or nationally unrepresentative. Their centile curves for overweight in figure 7 are similar to those in figures 3 and 4. (Data for Japan and girls in Sweden and Venezuela are omitted as they do not extend to age 18). Singapore and Canada are clear outliers during puberty, whereas Russia stands out earlier in childhood. The median curves for Japan and Hong Kong are similar in shape (not shown), suggesting that Singapore is atypical of Asia.

Nothing obvious explains Singapore's unusual pattern of overweight in puberty. Omitting it from the averaged country curves would lower the cut off points for both sexes by up to 0.4 body mass index units or 0.14 z score units at age 11-12. This compares to a range of three units between the lowest and highest curves at this age. Therefore, even though Singapore looks different from the other countries, its impact on the cut off points is only modest. Because there is no a priori reason to exclude Singapore, and because so little is known about growth patterns across countries, we have chosen to retain it in the reference population.

Extending the dataset

We recognise that the reference population made up of these countries is less than ideal. It probably reflects Western populations adequately but lacks representation from other parts of the world. The Hong Kong sample may, however, be fairly representative of the Chinese, and the Brazilian and US datasets include many subjects of African descent. Although additional datasets from Africa and Asia would be helpful, our stringent inclusion criteria of a large sample, national representativeness, minimum age range 6-18 years, and data quality control, mean that further datasets are unlikely to emerge from these continents in the foreseeable future. To our knowledge no other available surveys satisfy the criteria. It is not realistic to wait for them because there is an urgent need for international cut off points now. Also, our methodology aims to adjust for differences in overweight between countries, so it could be argued that adding other countries to the reference set would make little difference to the cut off points. None the less, further research is needed to explore patterns of body mass index in children in Africa and Asia.

Puberty

The body mass index curves in figure 6 show a fairly linear pattern for males but a higher and more concave shape for females. This sex difference can also be seen in the individual curves of figures 2 to 4 reflecting earlier puberty in females. The sensitivity of the curve's shape to the timing of puberty may affect the performance of the cut off points in countries where puberty is appreciably delayed,²¹ although delays of less than two years are unlikely to make much difference.

Use of cut off points

The cut off points in table 4 are tabulated at exact half year ages and for clinical use need to be linearly interpolated to the subject's age. For epidemiological use, with age groups of one year width, the cut off point at the mid year value (for example, at age 7.5 for the 7.0-8.0 age group) will give an essentially unbiased estimate of the prevalence.

The centiles for obesity involve more extrapolation than the centiles for overweight, which may explain the greater variability across datasets in figure 4 compared with figure 3. For this reason the obesity cut off points in figure 6 are fairly imprecise and are likely to be less useful than the cut off points for overweight.

The approximate prevalence values for overweight and obesity in tables 2 and 3 are calculated as the tail areas of the body mass index distribution in each sample at age 18, as estimated by the LMS method. This assumes that the distribution is normal after adjusting for skewness, which is inevitably only an approximation. In the British data there was slight kurtosis (heavy tails) in the distribution of body mass index,¹⁵ with 2.8% of the sample rather than the 2.3% expected exceeding a z score of 2. Therefore the true prevalences for the other datasets here may differ slightly from the values quoted.

The principle used to obtain cut off points for overweight and obesity in children could also provide a cut off point for underweight in children, based on the World Health Organisation's cut off point of a body mass index of 18.5 kg/m² for adult underweight. A body mass index of 18.5 kg/m² in a young adult is, however, equivalent to the British 12th centile,⁹ an unacceptably high prevalence of child underweight. A possible alternative would be a cut off point of a body mass index of 17 kg/m², on the British second centile at age 18.⁹ Although substantial data link cut off points of 25 and 30 kg/m² to morbidity in adults²² and the corresponding centile cut off points are associated with morbidity in children,²³ the health effects of cut off points corresponding to a body mass index below 17 or 18.5 kg/m² have not been studied. These cut off points for underweight need validating as markers of disease risk.

Based on cross sectional data the curves give no information about centile crossing over time—a weakness of most “growth” charts. Longitudinal data are needed to derive correlations of body mass index from one age to another, which then define the likely variability of centile crossing.²⁴,²⁵

Conclusions

Our analysis provides cut off points for body mass index in childhood that are based on international data and linked to the widely accepted adult cut off points of a body mass index of 25 and 30 kg/m². Our approach avoids some of the usual arbitrariness of choosing the reference data and cut off point. Applying the cut off points to the national datasets on which they are based gives a wide range of prevalence estimates at age 18 of 5-18% for overweight and 0.1-4% for obesity. A similar range of estimates is likely to be seen from age 2-18. The cut off points are recommended for use in international comparisons of prevalence of overweight and obesity.