Causal Mediation in Weight Management Trials Using Bayesian Nonparametrics

• Activity Number: 482 - Causal Inference for Dynamic Treatment and Mediation
• Type: Contributed
• Date/Time: Wednesday, August 10, 2022 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Woojung Bae* and Michael Daniels
• Companies: University of Florida and University of Florida
• Keywords: Bayesian nonparametrics; Causal inference; Enriched Dirichlet process mixture model

Abstract

In weight management trials, it is of interest to understand how interventions are effective. In a recent trial, we examine how the number of days calorie goals are met mediates the effect of a behavioral intervention on maintaining 18-month weight loss (after a 4-month weight loss program). To do this, we use a flexible Bayesian nonparametric (BNP) method and assume all confounders of the mediator-outcome relationship are measured. We present results from the analysis of this trial and compare them to simpler parametric and BNP approaches. Simulation studies are also conducted to examine the frequentist operating characteristics of this approach.