Shape of the BMI-Mortality Association by Cause of Death, Using Generalized Additive Models: NHIS 1986-2002.

Anna Zajacova, University of Michigan

This paper employs generalized additive models (GAMs) to describe the shape of the association between BMI and cause-specific mortality. GAMs are a type of nonparametric models that allow a regression-like analysis without imposing a priori constraints on the functional form of the predictors, and provide a flexible but parsimonious description of the predictors’ effects. We examine the association between BMI and mortality from diabetes, cardiovascular, cerebrovascular, cancer, and external causes. The analyses are based on the National Health Interview Survey 1986-2000 linked to the National Death Index, providing basic mortality information on over 1,000,000 individuals through 2002. The results indicate that the shape of the association between BMI and mortality for each cause depends strongly on the duration of followup with a clearly defined optimal BMI associated with low mortality at short followup duration but a much flatter shape for longer duration. Explanations for these patterns are discussed.