Bayesian Applications to Longitudinal Analysis on Medical Data with Discrete Outcomes

Many prediction studies of medical research lead to discrete longitudinal data with repeated measurement and categorical outcomes. Therefore the traditional likelihood-based methods for continuous outcome measures are no longer suitable. With the development of modern computing technologies and improved scope for estimation via iterative sampling methods, Bayesian analysis is becoming increasingly popular among biostatisticians. Markov chain Monte Carlo (MCMC), for the implementation of Bayesian methods has rendered the implementation of complex Bayesian models a reality. In addition, the availability of software like WinBUGS has made the utilization of MCMC straightforward. In this study, we developed a full Bayesian version of generalized linear models for binary longitudinal data and applied it to a longitudinal prediction study of Alzheimer's disease conducted at New York University School of Medicine