Subject Area
Statistics
Abstract
Recurrent event data frequently arise in clinical studies where individuals experience repeated, possibly related, events over time. These data are often accompanied by sparse and irregular longitudinal measurements, creating challenges for traditional joint modeling approaches that struggle to account for time-dependent associations and within-subject correlations. We propose FRAILTY (Functional Regression with AutoRegressIve fraiLTY), a novel two-step framework that integrates functional principal component analysis (PACE) with a dynamic frailty model featuring autoregressive structure. FRAILTY accommodates both scalar and functional predictors and captures within-subject dependence across recurrent events. To further extend its utility, we develop a multivariate joint modeling framework that simultaneously characterizes multiple types of recurrent events along with a terminal event, leveraging a shared structured frailty component. This extension enables more flexible modeling of interdependencies between different recurrent processes and the risk of terminal events, particularly in the presence of time-varying covariate effects derived from longitudinal trajectories. Estimation is carried out via an EM algorithm combined with Gaussian quadrature for efficient likelihood maximization under informative censoring. Simulation studies demonstrate that FRAILTY and its multivariate extension outperform existing methods in estimation accuracy, robustness under data sparsity, and predictive performance. Applications to the SPRINT and MSTONE studies illustrate its capacity to uncover clinically relevant patterns in complex event histories. To assess predictive performance, we also develop a weighted concordance index for recurrent event data subject to induced dependent censoring. Using inverse censoring probability weighting, our estimator reduces bias and mean squared error compared to existing C-index estimators. Application to the MSTONE dataset illustrates its utility in real-world predictive evaluation.
Degree Date
Summer 8-5-2025
Document Type
Thesis
Degree Name
Ph.D.
Department
Department of Statistics and Data Science
Advisor
Yu-Lun Liu
Number of Pages
118
Format
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Recommended Citation
Kong, Zifang, "Statistical Methods for Joint Outcome Modeling and Dynamic Assessment of Recurrent Events" (2025). Statistical Science Theses and Dissertations. 53.
https://scholar.smu.edu/hum_sci_statisticalscience_etds/53
Included in
Applied Statistics Commons, Biostatistics Commons, Statistical Models Commons, Survival Analysis Commons
