In the face of rapidly changing patterns of body weight in many populations, the relationship between relative weight and mortality remains a fundamental unsolved problem in epidemiology. The published literature on this question is contradictory and has led to conflicting public health recommendations. Not only is an optimal BMI as yet undefined, but the shape of the BMI-mortality relationship also remains unclear. Making use of the accumulated wealth of prospective research studies, the plan is to apply a systematic analytic approach to examine the variation in the relationship of BMI to mortality in a series of diverse populations. In the preparatory work leading up to this application, series of limitations in the published literature has been identified and for which remedies are proposed (principally by access to larger data sets and the utilization of newer statistical methods). Specifically, the application suggests that the inability to define a consistent relationship between BMI and mortality is a result, in large measure, of the following problems: 1) the attempt to identify a single pattern for all demographic groups (e.g., gender, age, mean BMI of the group); 2) reliance on statistical methods which have limited precision and flexibility (e.g., grouping in quartiles); 3) conceptual differences over the approach to control for potential confounders; 4) use of sub-group analyses and unusual cohorts comprised of volunteers; 5) absence of a formal structure for hypothesis testing and data analysis; and 6) restricting analyses to a single cohort, or meta-analysis based on published reports. To overcome these difficulties, the application provides the following goals: 1) create a data base including a wide range of cohorts from diverse population groups, including representative national population samples, volunteer cohorts and participants in clinical trials; 2) apply a uniform statistical approach to each data set which will involve original analyses at the individual level; 3) formally test whether a common estimate can be defined across groups; 4) define the shape of the BMI-mortality relationship for each sample; 5) conduct meta-regression analysis to define the attributes of study samples which account for potential heterogeneity in the shape of the relationship or the BMI associated with minimum mortality risk; and 6) define heterogeneity of outcomes by cause of mortality. For these analyses, the largest and most diverse collection of prospective samples yet assembled will be used; this assemblage includes 18 studies with a total of 2.7 million participants.
