Estimating the Effects of Weight and Weight Change Using a Dynamic Causal Model

Bochen Cao, University of Pennsylvania

Most previous studies examining the association between body weight and mortality do not adequately account for dynamic weight change and time-varying confounders such as disease and smoking. This study attempts to address these issues by applying a dynamic marginal structural model to re-examine the association of body weight and mortality in a nationally representative longitudinal dataset. The results suggest that both initial body weight and subsequent weight change have U-shaped associations with mortality. Time-varying confounders, particularly health conditions, appear to introduce relative large biases on the effects of large weight changes and both low (underweight) and high (obese class II/III, BMI>=35) initial weight status. They appear to have a smaller role in influencing the effects of small weight changes and baseline overweight (25.0-29.9) and class I obese (30.0-34.9). We conclude that better estimates of the body weight and mortality association are obtained by adequately accounting for time-varying confounders.