Pediatric body composition references: what’s missing?

See corresponding article on page 49.

In this era of prevalent obesity, when the term “body mass index” (BMI) has become all too familiar to parents and clinicians alike, it is easy to forget that before the 1990s the primary guide for “desirable weight” in the United States was insurance company tables of weights for height associated with the lowest mortality (1). Although limited in numerous respects, these tables did make an attempt to adjust for body composition. On the basis of elbow breadth, different weight ranges were provided for “small,” “medium,” and “large” frames. Today, the CDC advocates the use of BMI for identification of underweight, overweight, and obesity in adults and children (2). Despite ease of use, BMI percentiles in children are limited indicators of body composition at the individual level: the components of BMI change dramatically during growth, with fat mass (FM)⁴ and fat-free mass (FFM) increasing in different proportions relative to weight depending on sex, pubertal status, race-ethnicity, and other individual factors (3).

Percentage body fat reference data for children have been published with the use of a number of different assessment techniques (4–7). Still, percentage body fat is a proportion that includes FFM in the denominator, and depending on whether FFM is low or high percentage body fat may misclassify individuals. Furthermore, FFM [and its skeletal tissue–free compartment, lean mass (LM)] are key indicators of protein status that are important targets in their own right. Stature-adjusted indexes of FM and LM, comparable to BMI (ie, kg/m²) were suggested ≥20 y ago (8) and allow decomposition of BMI into an FM index (FMI) and an LM index (LMI), but pediatric FMI and LMI reference data have yet to be developed for the United States. With this in mind, Weber et al (9) published the first national pediatric FMI and LMI references in the current issue of the Journal.

The new sex-specific FMI and LMI references are based on NHANES dual-energy X-ray absorptiometry measures collected from 1999 to 2004 and span ages 8–20 y. Their strengths are as follows: 1) the very large sample size (∼9000 participants) allows high accuracy at the extreme percentiles; 2) the LMS (lambda-mu-sigma) approach used to construct the curves is currently considered the most appropriate for childhood growth data, as it addresses their nonnormal distributions; and 3) the data are from a nationally representative study in US children that oversampled black non-Hispanic, white non-Hispanic, and Mexican American children. Along with the references generated by Kelly et al (5) and Ogden et al (7) by using the same data set, a comprehensive picture of pediatric dual-energy X-ray absorptiometry–derived body composition variation by age and sex is now available for the United States.

The title of the article suggests that Weber et al (9) tested the ability of the FMI to identify excess adiposity compared with BMI and percentage body fat, but in fact the authors simply compared the positive predictive values of established BMI cutoffs for overweight and obesity (>85th and 95th percentiles of the CDC BMI charts, respectively) to identify high FMI (>75th percentile of their reference) or high percentage body fat [>75th percentile of the reference published by Ogden et al (7)]. Because there is still no gold-standard definition of excess childhood adiposity (ie, one that most strongly relates to chronic disease risk factors), such an analysis is not particularly informative; definitions of “high adiposity” are not anchored in any functional definition. As similarly shown by others (10), BMI percentile >95th successfully identified children with high adiposity (here, FMI percentile >75th) with a positive predictive value ≥0.95 for each sex- and race-stratified group. Not shown were the negative predictive values that would have helped the reader understand what percentage of nonobese individuals were correctly classified as having a normal FMI. Interestingly, there was strong agreement between high adiposity defined as an FMI >75th percentile or as percentage body fat >75th percentile; only 6% of children were differentially classified by using these 2 definitions, which suggests that the concern over misclassification error for percentage body fat in this sample was not particularly important. There were some differences in mean BMI, LMI, and height z scores between those who were concordant and discordant on the 2 measures, but it would have been more to the point to report whether misclassification using percentage body fat is due to grouping children with low LMI (but normal FMI) with those with high FMI. Overall, however, what is missing here is that our current criterion for obesity in children (BMI >95th percentile) already does a reasonably good job of detecting children with elevated adiposity (10), and so the onus is on body composition researchers to show the utility of decomposing weight into FM and LM components for risk stratification or other purposes. Only then will a comparison of the performance of difference proxies and cutoffs be particularly meaningful.

Non-Hispanic black children were found to have lower FMI and greater LMI than non-Hispanic white children. These race differences in body composition were highlighted and discussed by Weber et al (9), yet no argument was made as to why the present references were constructed using data pooled across all races. Their approach does have the advantage that researchers can standardize data in a way that ranks boys or girls on the basis of absolute fatness or leanness regardless of race, but it has the potential disadvantage that it will systematically indicate low FMI and high LMI in non-Hispanic black children, as one example. What is missing here is evidence that directly answers the question of whether or not race-ethnic–specific cutoffs for elevated adiposity (by using FMI, for instance) are needed to distinguish between children with or without related cardiometabolic risk factors. Once optimal body composition cutoffs are established, then a sensitivity and specificity analysis can be conducted to compare different proxies in boys and girls of different race-ethnic backgrounds. Given that this is currently unknown, both pooled and race-ethnic–specific references would be helpful.

Wells et al (11) recently published FMI and FFM index reference growth curves for the UK pediatric population based on estimates from a 4-component model of body composition, which is often considered the gold standard. The source sample comprised 565 children of predominantly European ancestry from London and the southeast of England. A visual comparison of the 2 FMI percentiles is shown in Figure 1. In both sexes, the FMI distribution is shifted upward in the US reference compared with the UK reference, with the 50th percentile for US children being most similar to the 75th percentile for UK children, probably reflecting the higher rate of childhood obesity in the NHANES sample. Care is needed to determine which reference set would be most appropriate for use with a particular study sample.

FIGURE 1.

Comparison of FMI-for-age reference curves, by sex, for the United States and the United Kingdom. FMI, fat mass index.

In summary, the new FMI and LMI pediatric references of Weber et al (9) are a welcome addition to our array of tools for comparing pediatric body composition estimates against a US representative sample. The use of tissue-specific, stature-normalized indexes overcomes the limitations of childhood BMI and percentage body fat by decomposing body weight into its FM and LM components. Race-stratified values would have been valuable but were not included despite the demonstration of significant large race variation in FMI and LMI. As with other growth references, the pediatric FMI and LMI references are descriptive, not prescriptive, and may not accurately reflect the growth process, given that they are based on cross-sectional data. The references will likely be used at this time primarily by researchers, and incorporation of the LMS data into a program that converts raw data into percentiles and z scores will aid in their use by the research community. What is missing is a discussion about future clinical uses of such references. That discussion will not move very far without evidence on appropriate cutoffs for elevated adiposity that distinguish children with and without concurrent disease risk factors, followed by an analysis that shows that measured FMI and LMI actually improve risk detection over simple metrics like BMI percentile. Nonetheless, the new references are important as part of the continuing effort to refine and personalize our recommendations on body weight in a way that accounts for individual variation in body composition.