Bayesian joint modeling of longitudinal data with dropout
by
Sanjib Basu
Northern Illinois Unniversity

Incomplete data are common in longitudinal studies. We propose a Bayesian model for non-ignorable dropout where the dropout process is modeled jointly with the longitudinal observation process. As an application, we consider data from a crossover trial which is a popular design in early phases of clinical trials and in bioavailability and bioequivalence studies. The second application considers a longitudinal clinical trial that involves multiple longitudinal endpoints and competing risks of dropout from the study. We discuss parametric and semiparametric models involving longitudinal and survival components and develop Bayesian model comparison to compare different dropout models.

Date received: March 16, 2008