Heterogeneous Recurrent Event Analysis Based on Latent Classes

• Activity Number: 401 - Real-World Survival Data with Multiple Events: Challenges, Opportunities, and Recent Advancements
• Type: Invited
• Date/Time: Wednesday, August 10, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Lifetime Data Science Section
• Author(s): Wei Zhao and Limin Peng* and John Hanfelt
• Companies: Emory University and Emory University and Emory University
• Keywords: Recurrent events; Multiplicative intensity model; Latent Class

Abstract

Recurrent events data frequently arise in chronic disease studies, providing rich information to help uncover clinically relevant heterogeneity underlying chronic disease progression, which can be naturally captured by the concept of latent class. In this work, we investigate a flexible latent class semiparametric multiplicative modeling for recurrent events data. Our model allows for nonparametric baseline intensity function and covariate effects to vary across different latent classes. Utilizing the special characteristics of multiplicative intensity modeling, we derive an iterative estimation procedure that can be stably and efficiently implemented based on existing computational routines without involving smoothing. We also establish asymptotic properties of the resulting estimators, which are greatly complicated by allowing for class-specific nonparametric baseline intensity functions. Results from our numerical studies suggest that applying the proposed latent class recurrent event model can lead to much improved performance in predicting recurrent event trajectories as compared to traditional methods.