Department of Biostatistics Seminar/Workshop Series

Joint Modeling Compliance and Outcome for Causal Analysis in Longitudinal Studies

Xin (Cindy) Gao, MS

Department of Biostatistics, University of Michigan, Ann Arbor

Tuesday, May 10 1:00pm-2:00pm, Vanderbilt Children's Theater

Randomized studies are well accepted as a standard method to estimate the effect of treatment because it removes both observed and unobserved confounding. However, since patients can choose whether or not to comply with their assigned treatment in many circumstances, non-compliance behavior is common in the randomized studies. In the recent years, causal modeling has been regarded as a promising approach to provide valid estimate of effect of treatment for randomized studies with non-compliance behavior.

Motivated by this, we proposed a Markov compliance and outcome model to jointly model longitudinal measurements of compliance and outcome for randomized studies in presence of non-compliance. In the proposed causal model, we used the potential outcome framework to define pre-randomization principal strata determined by the joint distribution of compliance under treatment and control arms and estimate the effect of treatment within each principal stratum. Besides the causal effects of the treatment, our proposed model could estimate the impact of the causal effect of treatment at the end of follow up interval t on the compliance in the follow up interval t+1. We applied the proposed causal model in a longitudinal mental health study with psychiatric data, and utilized Bayesian methods with Markov chain Monte Carlo algorithm for data analysis.