Curve smoothing and transformations in the development of growth curves

Abstract

Growth charts such as those published by the National Center for Health Statistics (NCHS) consist of a set of smoothed percentile curves showing the distribution of different aspects of body size for infants, children, and adolescents. The original NCHS growth charts are currently being revised to incorporate additional national data, to include growth curves for body mass index (BMI, in kg/m2), to reduce or eliminate discontinuities, and to use new and improved methods of smoothing. Methods of curve smoothing in the development of growth curves are briefly discussed in the context of this revision by using examples based on the provisional data set for revision of the growth curves. Among the factors that may affect the construction of the revised growth curves are sample sizes, sampling weights, and secular trends. The use of BMI or other weight-height indexes in growth curves presents some new issues because these transformations of weight and height data do not increase monotonically with age. Some of the advantages and disadvantages of several different approaches to creating smoothed percentiles and standardized scores are discussed briefly. These effects need to be considered in the broader context of constructing growth curves. No single method of developing smoothed curves is clearly the best for all purposes, and estimates of overweight and underweight may be sensitive to the method chosen. Alternative approaches to constructing smoothed curves by using weight-height functions other than BMI might warrant further exploration.