Bayesian joint modeling for causal mediation analysis with a binary outcome and a binary mediator
Geneviève Lefebvre, PhD University of Québec at Montréal
Mediation analysis with a binary outcome is notoriously more challenging than with a continuous outcome. In this talk, we will present a new approach for performing causal mediation with a binary outcome and a binary mediator. Our proposal relies on the Student-t approximation to the Bayesian multivariate regression logistic model introduced by O'Brien and Dunson (Biometrics, 2004). We will explain how this latent multivariate model can be used to estimate the natural direct and indirect effects of an exposure in any measure scale of interest (e.g., odds or risk ratio, risk difference). Our novel mediation approach has several valuable features which, to our knowledge, are not found together in current binary-binary mediation models. The model will be illustrated and compared to two existing approaches for conducting causal mediation analyses with this type of data.