Abstract

Shared-parameter joint models for longitudinal and time-to-event data provide a powerful framework for analyzing the often complex relationships between repeatedly measured biomarkers and clinical events. JMbayes2 is a highly optimized R package that implements a broad class of extended joint models, going beyond the standard setting of a single continuous longitudinal outcome and one survival endpoint.

This talk showcases the model extensions supported in JMbayes2. We show how flexible association structures can be used to characterize complex relationships between multiple longitudinal trajectories and event risk while producing interpretable and clinically meaningful estimates. We also explain how to incorporate more complex event processes (e.g., competing risks and recurrent events) and accommodate different censoring and truncation mechanisms (e.g., interval censoring and left truncation). Finally, we briefly illustrate how fitted joint models can be used to obtain personalized dynamic predictions.

Throughout these examples, the talk demonstrates how JMbayes2 enables researchers to explore alternative formulations, assess the model fit and which extensions are most appropriate for their application, and compare models without switching between different R packages or learning multiple syntaxes.