SurvivalGPU: Scalable Survival Analysis for R and Python.
Abstract
Survival analysis is widely used across fields ranging from predictive maintenance to public health to model time-to-event outcomes. The Cox proportional hazards model is the standard approach, with a rich landscape of methodological developments extending its scope and flexibility. One notable extension is the Weighted Cumulative Exposure (WCE) model, which captures the longitudinal impact of time-varying exposures. However, applying these models to large-scale databases such as electronic health records faces significant computational bottlenecks, making iterative procedures like bootstrapping or cross-validation prohibitively expensive. We introduce survivalGPU, an R and Python package leveraging GPU acceleration through PyTorch and KeOps to address these limitations, while preserving compatibility with the reference survival and WCE packages.