In studies of weight and mortality, the construct of reverse causation has come to be used to imply that the exposure-outcome relation is biased by weight loss due to preexisting illness. Observed weight-mortality associations are sometimes thought to result from this bias. Evidence for the occurrence of such bias is weak and inconsistent, suggesting that either the analytical methods used have been inadequate or else illness-related weight loss is not an important source of bias. Deleting

more »

... cipants has been the most frequent approach to control possible bias. As implemented, this can lead to deletion of almost 90% of all deaths in a sample and to deletion of more overweight and obese participants than participants with normal or below normal weight. Because it has not been demonstrated that the procedures used to adjust for reverse causation increase validity or have large or systematic effects on relative risks, it is premature to consider reverse causation as an important cause of bias. Further research would be useful to elucidate the potential effects and importance of reverse causation or illness-related weight loss as a source of bias in the observed associations between weight and mortality in cohort studies. bias (epidemiology); body mass index; body weight; confounding factors (epidemiology); epidemiologic methods; mortality; selection bias Abbreviations: BMI, body mass index; COPD, chronic obstructive pulmonary disease; IRWL, illness-related weight loss. Many analyses have investigated the association of relative weight or body mass index (BMI) with mortality with broadly consistent results. Studies in young and middleaged individuals often show a curvilinear relation with increased mortality risk at both low and high BMI values (1-3). Relative risks of mortality in these studies are usually below 2.0 and often below 1.5. Studies in the elderly often show the highest relative risks at low BMI levels, with little excess risk at higher BMI levels (4, 5). It is sometimes suggested that these observations result from reverse causation, namely, bias caused by preexisting illness and attendant weight loss (6-8). In this commentary, we discuss some unresolved methodological issues regarding the concept of reverse causation and the effects of preexisting illness in studies of weight and mortality and make some brief suggestions for directions in future research. This is not intended as a comprehensive literature review of this complex topic. UNRESOLVED ISSUES Definition of reverse causation in studies of weight and mortality