TimeDepFrail: Time-Dependent Shared Frailty Cox Models in R

TL;DR

TimeDepFrail is an R package that extends shared frailty Cox models to include time-varying frailty terms, enabling more flexible survival analysis of clustered data with changing unobserved heterogeneity.

Contribution

It introduces a novel implementation of time-dependent shared frailty models in R, including posterior estimation and prediction capabilities, with reduced computational complexity.

Findings

Uncovered new insights into dropout dynamics over time.

Demonstrated scalability for large datasets.

Provided practical tools for advanced survival analysis.

Abstract

This paper introduces TimeDepFrail, an R package designed to implement time-varying shared frailty models by extending the traditional shared frailty Cox model to allow the frailty term to evolve across time intervals. These models are particularly suited for survival analysis in clustered data where unobserved heterogeneity changes over time, providing greater flexibility in modeling time-to-event data. The package builds on the piecewise gamma frailty model originally proposed by Paik (1994) and refined by Wintrebert et al. (2004). Our key contributions include the integration of posterior frailty estimation, a reduction in computational complexity, the definition of a prediction framework and the efficient implementation of these models within an R package. As a practical application, we use TimeDepFrail to analyze dropout rates within a university, where high dropout rates are a known issue. By allowing frailty to vary over time, the package uncovers new insights into the unobserved factors influencing dropout. TimeDepFrail simplifies access to advanced time-varying frailty models, providing a practical and scalable alternative to more computationally demanding methods, making it highly applicable for large-scale datasets.