Bayesian Semiparametric Joint Modeling of Gap-Time Distribution for Multitype Recurrent Events and a Terminal Event

Abstract

In biomedical settings, multitype recurrent events such as stroke and heart failure occur frequently, often concluding with a terminal event such as death. Understanding the links between these recurring and terminal events is fundamental to developing interventions that delay detrimental outcomes. Joint modeling is needed to quantify the dependence between event types and between recurrent events and mortality. We propose a Bayesian semiparametric joint model on the gap-time scale for multitype recurrent events and a terminal event. The model includes a shared frailty that links all recurrent types and the terminal event. Each baseline hazard is assigned a gamma-process prior, while regression and frailty parameters receive standard parametric priors. This ensures flexible baselines and familiar effect measures. The construction gives closed-form expressions for the cumulative hazard and frailty component and connects to Breslow-Aalen type estimators as a special case of our estimator, linking the Bayesian procedure to the classical approach. Computationally, we develop a simple MCMC sampler that avoids large matrix factorizations and scales nearly linearly in sample size. A comprehensive simulation evaluates four criteria: accuracy, prediction, robustness, and computation. There is no exact frequentist version of our specification; for comparison, we fit the same model with an EM algorithm in a frequentist framework. Our model and MCMC algorithm demonstrate superior performance on each criterion. We illustrate the approach with data from the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT), jointly analyzing acute and chronic cardiovascular recurrences and death.