I understand the need to research within the field of body composition. There are important applications of quantifying either the lean or fat body mass in sport, athletic performance, growth and development, aging, health and disease. My early career mentor at UNM, Vivian Heyward, taught me a lot about this topic and field of research. However, it has been a while since I was involved in research based on the validation of different methods to an established criterion method. As such, in preparation for this web page, I did some current reading to make sure my interpretations of the field were not outdated.
My critical interests in this topic are based on the fact that there is no clear criterion method in body composition. Despite this, the majority of research on validation is based on multiple regression, and therefore prediction and explanation of variance. While regression is robust in its purpose of quantifying predictive accuracy, such prediction is only as good as the criterion method is valid. In body composition, this latter issue cannot be addressed unless we go back to the pioneering research methods involved in this topic and partition recently deceased humans to quantify water, protein, mineral and fat, piece by body piece. But of course, this pioneering research did not acquire bodies representative, or in sufficient quantity, of different ethnicity, ages, training states, health conditions, sex, or physique. Thus, we have indirect methods that estimate multiple components that rise to criterion status. At one time, skin-folds ruled this field. Then it was underwater weighing, then underwater weighing with residual volume measurement, then air displacement plethysmography (Bod Pod), and now it appears that dual-energy X-ray absorptiometry (DXA) is the recognized criterion based on the body water and body mineral attributes of this method in allowing a multiple (3) compartment assessment of body composition.
Here is the problem, DXA is not a great criterion because there are numerous assumptions in the method (see Lohman and Chen). As such, DXA has inherent measurement error. All methods of measurement have error, but when there are aspects of the method that lie far from pure empiricism, the method moves further from true criterion status. The best statistical approach to follow in recognizing error of the criterion is the limits to agreement method of Bland and Altman. However, the fact remains that a mean of two methods is still a crude approach at deciphering a criterion measurement.
When considering these facts, is there really a best method of body composition analysis? Furthermore, in science or professional practice, the utility of a body composition method surely must also consider cost (purchase and maintenance), ease of use, scope for user error as well as measurement error, pre-test and during test requirements, measurement time, etc. When does estimated measurement error, relative to the unknown error of a criterion method, become small enough? What methods fall into these ranges? I think these are the questions that should be raised and answered for different applications of body composition measurement.