Happy the man who has been able to learn the causes of things. -Virgil Clinical experience and a sound knowledge of pathophysiology are the traditional pillars of medicine. In recent years, new statistical methods have strengthened the ability of investigators to extract patterns from complex datasets that would otherwise remain hidden. Research to predict clinical events and the methods needed to support these analyses have evolved in tandem over the last few decades in a classical pas de deux. Indeed, the landscape of modern medicine would hardly be recognizable without such techniques. The recent development of fracture prediction tools is a prime example of one clinical application of powerful statistical methods.(1)When prediction of an event is the sole objective, the existence of an association-whether causal or not-may be sufficient. However, to understand how that association arises, and to intervene appropriately, requires a deeper understanding of the cause and effect relationships. Obesity, defined from elevated body mass index (BMI), has generated much controversy in the osteoporosis literature, variably reported to be protective, neutral, or a risk factor for osteoporosis and related fractures.(2-4) Weight and BMI are both strongly correlated with bone mineral density (BMD) as measured from dual-energy X-ray absorptiometry (DXA) in almost all studies, and it is no accident that weight or BMI are a component of virtually all fracture prediction tools.(1) However, when weight is partitioned into its components of lean mass and fat mass, the associations become more complicated. Most studies find that lean mass is positively associated with higher BMD. (5,6) For fat mass a variety of results have been found, (5,(7)(8)(9)(10)(11)(12) although a recent metaanalysis demonstrates a positive association between fat mass and bone density.(13) Meta-analytic results have also shown that among premenopausal women lean mass exerted a greater effect on femoral neck BMD than fat mass (r ¼ 0.45 versus r ¼ 0.30), whereas in postmenopausal women the effects of lean mass and fat mass were similar (r ¼ 0.33 versus r ¼ 0.31).(13) The presence of large correlations among weight, BMI, lean mass, and fat mass have fueled some of this controversy.(4) For example, if one assumes that lean mass is solely responsible for BMD and that fat mass has no effect, then a model that includes weight and fat mass would actually show the latter as having an adverse (negative) effect on BMD simply because fat is interpreted as "not lean" in the model. Body composition is a major factor impacting accuracy errors in DXA, effects that are amplified in longitudinal studies where weight, body composition, and BMD (not to mention a myriad of other factors) are changing simultaneously. (14)(15)(16) A recent meta-analysis examined the association between BMI and fracture risk in prospective cohorts from 25 countries (398,610 women with an average age of 63 years; 2.2 million person-years follow-up) with a reported 22% prevalence of obesity.(17) ...