Modeling the Dynamic Association of BMI and Mortality in the Framingham Heart Study

PurposeTo examine and model the dynamic association of BMI and mortality in the Framingham Heart Study (FHS).MethodsBMI is transformed to facilitate modeling curvature associations. Logistic models are used to demonstrate whether different conclusions may be obtained for the same group of individuals under different settings created from FHS. Time-dependent covariates Cox models are used to model the association of BMI and mortality when the proportional hazards assumptions for Cox models are violated.ResultsBoth the measurement time of BMI and the length of follow-up affect the conclusions obtained from logistic models, especially for men. Time-dependent covariates Cox models show that the association between BMI and mortality for men depends on the follow-up time, while that for women depends on the age of BMI measurement.ConclusionThe association of BMI and mortality in FHS is a dynamic system that traditional analyses methods may lead to different conclusions for different study designs. This finding is consistent with the results of several other studies done from different perspectives, suggesting that the dynamic features demonstrated in FHS may apply to other populations. Advanced methods such as time-dependent covariates Cox models may be helpful for future analysis.