Comparing Childhood Weight-for-Age to Body Mass Index in the Prediction of Adolescent Obesity and Chronic Disease Risk Factors

Steven D. Stovitz, M.D.
Department of Family Medicine, University of Minnesota

Approximately one-third of pre-pubescent children in the US may be either overweight or at risk for overweight, as defined by a body mass index (BMI) over the 85th% for age and gender matched norms. Childhood obesity is a clear risk factor for adult obesity and associated co-morbidities. An essential first step toward confronting the problem of childhood obesity is an assessment tool that is both accurate and understandable to clinicians and parents. Currently experts recommend the use of BMI in children. However, BMI may have limitations in children, because, unlike adults, tallness in children can be due to excessive caloric intake, and a surrogate marker for subsequent obesity. High BMI prior to age 6 has been associated with elevated height and subsequent obesity in a number of studies. Young children with elevated BMI have accelerated vertical growth before puberty, with a deceleration in height thereafter. They do not become taller adults. Since height-squared is the denominator for the calculation of BMI, early increases in height may mask the risk of subsequent obesity by causing a "falsely" normal BMI pre-puberty. This phenomenon presents a significant limitation of BMI in the assessment of childhood obesity, as it implies that two children of similar age, gender, and BMI (but different heights and weights) may have different probabilities of subsequent obesity.

The hypothesis of this study is that weight-for-age (WFA) will prove as accurate as BMI in assessing a child's risk for adolescent obesity and associated cardiovascular risk factors. If the hypothesis is correct, we believe that this will be an important advancement in combating the public health crisis of childhood obesity due to the simplicity of WFA in comparison to BMI.


In order to test the hypothesis, we will conduct a secondary data analysis of 2695 participants from the Child and Adolescent Trial for Cardiovascular Health (CATCH study) that had anthropomorphic data measured in 3rd grade, and then were retested 10 years later. The analysis will involve a series of general linear models. Adolescent BMI, skinfolds, blood pressure, total cholesterol, HDL-C, and homocysteine will be regressed separately on pre-pubescent WFA and BMI, and compared. All variables will be examined for non-normality, with appropriate transformations used to normalize distributions prior to analysis.

All models will control for the effects of a number of potential covariates and interactions between these potential covariates and childhood WFA and BMI will be included where indicated. We will compare competing models (i.e., one with childhood BMI as a predictor vs. one with childhood WFA as a predictor) using two selection criteria: Bayesian Information Criteria (BIC) and proportion of variance explained (R2). The model with the BIC closest to zero will be considered the better fitting of the two, and R2 will be used to quantify the difference in variance explained. All models will be estimated using SAS PROC GLM.