Sternfeld et al. Respond to: "Body Composition in Studies of Aging"

Dr. Harris articulates several important issues regarding the study of health effects of body composition in the elderly population (1). First is the hypothesis that the separate elements of body composition (i.e., bone, lean mass, and fat) may have differential relations with specific health outcomes. For instance, low bone density may be a risk factor for fracture, but not for osteoarthritis or atherosclerosis, while a high level of fat may have the opposite relations. The purpose of our study, reported in this issue of the Journal (2), was to test this hypothesis, specifically that a lower level of lean mass, but not a higher level of fat, is a risk factor for impaired physical function. However, our data did not support this hypothesis. More important, the data suggested that separating lean and fat, at least in terms of their relations to functioning, may not advance our understanding of the disablement process as much as consideration of lean mass relative to fat mass. The reason for this is that the more fat mass a person has, the more lean mass he or she has as well; for this reason, both fat mass and lean mass were inversely related to poorer function in unadjusted analyses. Dr. Harris refers to this problem and uses the term “sarcopenic obesity” to describe people with proportionally low lean mass relative to overall size. She hypothesizes that such a relative measure may be more related to functioning in a higher-functioning population than in a frail one. Although the implication of this hypothesis is that there is an absolute threshold of lean mass, below which values are classified as frailty, it does not address the process by which a person reaches that threshold, which may have more to do with the relative amount of lean mass than with the absolute amount. The second issue concerns the limitations of both selfreported and performance-based measures of physical function. Although measurement approaches continue to evolve (3), this problem exists for all studies of physical function in the elderly (4, 5). Because each measure has both advantages and disadvantages, we chose several types of measures rather than only one. The performance-based walking speed and grip strength were selected because each assessed a different physiologic capacity (aerobic capacity and muscular strength, respectively, both of which are important for overall physical functioning). The self-reported measures were selected because, as Dr. Harris points out, physiologic capacity does not always reflect actual functioning. Although Dr. Harris is correct that our composite measure of self-reported functioning included both upperand lowerbody functions, as well as simpler and more complex functions, we do not think that this resulted in misleading findings. The great majority of those defined as limited by the composite measure reported problems with lower-body tasks. The fact that both slower walking speed and selfreported limitation were directly related to fat mass, inversely related to the lean-to-fat ratio, and generally not related to lean mass supports the conclusion that both measures were assessing a similar functional domain. Finally, Dr. Harris suggests that regional measures of lean mass may be more informative in relation to function than are overall measures. In particular, the complex relation between grip strength and lean mass observed in our study might be better understood by examining appendicular lean mass or even arm mass specifically. We agree with this and regret that the method of body composition assessment that we used did not allow for this type of analysis. Although we appreciate Dr. Harris’ comments in general, we respectfully disagree with two specific criticisms of our study. First, it is highly unlikely that our derived, populationspecific regression equations for predicting body composition from the bioelectric impedance data were biased by the stratified sample selection. The measures of fat-free mass, lean mass, and fat mass from these equations correlated highly with those from the equations of Roubenoff et al. (6) (r ≥ 0.99) that were derived from 455 elderly participants in the Framingham cohort. Second, while the use of variables that represent residual values from linear regression may be problematic under certain, specific circumstances, this is, in general, an analytic approach that is defensible both mathematically and intuitively. Furthermore, the interpretation of a residual variable created from multiple independent variables is no more obscure than when there is only a single independent variable; in both cases, the residual variable is