Abstract
A joint model for a longitudinal biomarker and recurrent events is proposed. This general model accommodates the effects of covariates on the biomarker and event processes, the effects of accumulating event occurrences, and effects caused by interventions after each event occurrence. Association between the biomarker and recurrent event processes is captured through a latent class structure, which also serves to handle an underlying heterogeneous population. We use the EM algorithm for maximum likelihood estimation of the model parameters and a penalized likelihood measure to determine the number of latent classes. This joint model is validated by simulation and illustrated with a data set from epileptic seizure study.