Two-stage recurrent events random effects models

This article delves into the intricacies of two-stage recurrent events random effects models, a crucial tool in understanding the dependence between recurrent events and terminal events. The study highlights the importance of these models in medical research, particularly in trials like the LEADER trial and studies on infections in patients with central venous catheters. The article presents two main models: one where the random effect is fully shared between the terminal event and the recurrent events, and another where the random effect is not fully shared. The estimation procedure relies on a pseudo-likelihood approach, which is derived from the observed intensities given the history of the recurrent events and the terminal event. Simulation studies demonstrate the effectiveness of the estimation procedure, even in the presence of censoring. The article also includes an application to the Taichung Peritoneal Dialysis Study, showcasing the practical utility of the models. The findings suggest that the partly shared random effects model is preferable, indicating considerable heterogeneity even after adjusting for important risk covariates. The article concludes with a discussion on the implications of the models and suggestions for future research.