In 1993 the World Health Organization (WHO) undertook a comprehensive review of the uses and interpretation of anthropometric references. The review concluded that the NCHS/WHO growth reference, which had been recommended for international use since the late 1970s, did not adequately represent early childhood growth and that new growth curves were necessary. The World Health Assembly endorsed this recommendation in 1994. In response WHO undertook the Multicentre Growth Reference Study (MGRS) between 1997 and 2003 to generate new curves for assessing the growth and development of children the world over.
The MGRS combined a longitudinal follow-up from birth to 24 months and a cross-sectional survey of children aged 18 to 71 months. Primary growth data and related information were gathered from 8440 healthy breastfed infants and young children from widely diverse ethnic backgrounds and cultural settings (Brazil, Ghana, India, Norway, Oman and USA). The MGRS is unique in that it was purposely designed to produce a standard by selecting healthy children living under conditions likely to favour the achievement of their full genetic growth potential. Furthermore, the mothers of the children selected for the construction of the standards engaged in fundamental health-promoting practices, namely breastfeeding and not smoking.
This report presents the second set of WHO Child Growth Standards (i.e. head circumference-for-age, arm circumference-for-age, triceps skinfold-for-age and subscapular skinfold-for-age) and describes the methodical process followed in their development. The first step in this process was a consultative expert review of some 30 growth curve construction methods, including types of distributions and smoothing techniques to identify the best approach to constructing the standards. Next was the selection of a software package flexible enough to allow the comparative testing of the alternative methods used to generate the growth curves. Then the selected approach was applied systematically to search for the best models to fit the data for each indicator.
The Box-Cox-power-exponential (BCPE) method, with curve smoothing by cubic splines was selected for constructing the WHO child growth curves. The BCPE accommodates various kinds of distributions, from normal to skewed or kurtotic. The age-based indicators originating at birth required a power-transformation to stretch the age scale (x-axis) as a preliminary step to fitting the curves. For each set of curves, the search for the best model specification began by examining various combinations of degrees of freedom to fit the median and variance estimator curves. When data had a non-normal distribution, degrees of freedom for parameters to model skewness and kurtosis were added to the initial model and adequacy of fit evaluated. Apart from head circumference-for-age, which followed a normal distribution, the other standards in the second set required the modelling of skewness, but not kurtosis. The diagnostic tools used iteratively to detect possible model misfits and biases in the fitted curves included various tests of local and global goodness of fit, worm plots and residual plots. Patterns of differences between empirical and fitted percentiles were also examined, as were proportions of observed versus expected percentages of children with measurements below selected percentiles.
The sample used for the construction of the second set of growth standards was the same one used for the construction of the first set of standards. The methodology described above was followed to generate ─ for boys and girls ─ percentile and z-score curves for head circumference-for-age (0 to 60 months), arm circumference-for-age (3 to 60 months), triceps skinfold-for-age (3 to 60 months) and subscapular skinfold-for-age (3 to 60 months).
The data of the longitudinal and cross-sectional samples were merged without any adjustments and a single model was fitted to generate one continuous set of curves constituting each sex-specific standard.
Head circumference followed a normal distribution. The data for arm circumference and skinfold thicknesses were skewed, so in specifying the model, the parameter related to skewness was fitted in addition to the median and the coefficient of variation. Results from the final model for girls' subscapular skinfold suggested the need to investigate potential improvements in the curves by modelling kurtosis. Adjustment for kurtosis, however had a negligible impact on the final centiles. Therefore, considering that modelling the fourth parameter would increase complexity in application of the standards and create inconsistency between the sexes, the final curves were generated without adjusting for kurtosis.
The power transformation of age was applied to stretch the age scale for each of the sexes before fitting cubic splines to generate the growth curves. The same power transformation of age was applied to both boys' and girls' head and arm circumferences. For the skinfold thicknesses, boys required a higher power transformation than did girls.
Cubic spline fitting was achieved with variable degrees of freedom for each indicator and sex. For the median curves, different degrees of freedom were required for boys and girls for arm circumference and subscapular skinfold. For the coefficient of variation curves, the degrees of freedom varied between sexes for head circumference and subscapular skinfold. For the indicators that required fitting skewness, all but the subscapular skinfold required different degrees of freedom for the parameter modelling skewness.
Overall, concordance between smoothed curves and empirical centiles was free of bias in both the median range and the tails, indicating that the resulting curves provide an adequate description of the true growth of healthy children.
The method used to construct the WHO standards generally relied on the Box-Cox power exponential distribution and the final selected models simplified to the LMS model. As a result, the computation of percentiles and z-scores for these standards uses formulae based on the LMS method. However, as was done for the construction of the first set of growth standards, a restriction was imposed on all indicators to enable the derivation of percentiles only within the interval corresponding to z-scores between -3 and 3. The underlying reasoning is that percentiles beyond ±3 SD are invariant to changes in equivalent z-scores. The loss accruing to this restriction is small since the inclusion range corresponds to the 0.135th to 99.865th percentiles.
The arm circumference and skinfold thicknesses presented right-skewed distributions. When modelled correctly, right skewness has the effect of making distances between positive z-scores increase progressively the farther away they are from the median, while distances between negative z-scores decrease progressively. The LMS method fits skewed data adequately by using a Box-Cox normal distribution, which follows the empirical data closely. The drawback, however, is that the outer tails of the distribution are highly affected by extreme data points even if only very few. A restricted application of the LMS method was thus used for the construction of the indicators with skewed distributions, limiting the Box-Cox normal distribution to the interval corresponding to z-scores where empirical data were available (i.e. between -3 SD and 3 SD). Beyond these limits, the standard deviation at each age was fixed to the distance between ±2 SD and ±3 SD, respectively. This approach avoids making assumptions about the distribution of data beyond the limits of the observed values.
All four indicators presented in this report are a new addition to the previously available set of indicators in the NCHS/WHO reference. Head circumference-for-age is often used in clinical settings as part of health screening for potential developmental or neurological disabilities in infants and young children. Very small and very large circumferences are both indicative of health or developmental risk. Arm circumference-for-age is used as an alternative indicator of nutritional status when the collection of length/height and weight measurements is difficult, as happens in emergency humanitarian situations due to famine or refugee crises. Triceps and subscapular skinfold measurements assess the thickness of subcutaneous tissue and reflect fatness primarily. The skinfold indicators are thus a useful addition to the battery of growth standards for assessing childhood obesity.
The WHO Child Growth Standards provide a technically robust set of tool that represents the best description of physiological growth for children under five years of age. The standards depict normal early childhood growth under optimal environmental conditions and can be used to assess children everywhere, regardless of ethnicity, socioeconomic status and type of feeding.