A particular diffusion model for incomplete longitudinal data: application to the multicenter AIDS cohort study

Longitudinal studies, in which individuals are measured repeatedly in time, are often incomplete.We model continuous-time longitudinal data from the Multicenter AIDS Cohort Study using a diffusion model in which the diffusion parameters are functions of the covariates.These data are jointly modeled with the process of time-to-death due to AIDS.We show that, even for large data sets with a large number of missing variables, a Bayesian analysis is feasible using Gibbs sampling and compare a complete case analysis with a Bayesian treatment of missing values.