27.10.2025 Neue Publikation: Modeling dependent censoring in time-to-event data using boosting copula regression

This paper presents a boosting copula regression framework for modeling dependent censoring in survival analysis. In many real-world studies, censoring is not independent of the event of interest, for example, when patients withdraw from a clinical trial due to a worsening health condition. In such cases, classical methods like the Cox model may lead to biased conclusions. To address this challenge, we jointly model event times T and censoring times CC using a parametric copula, which allows for a flexible and realistic representation of their joint dependence. Rather than assuming the marginal distributions are known, all distribution parameters—including the copula parameter—are estimated simultaneously as functions of (potentially different) covariates. The use of model-based boosting enables data-driven variable selection and makes estimation feasible in high-dimensional settings where traditional approaches reach their limits. We evaluate the proposed method in an extensive simulation study and demonstrate its practical usefulness in an observational colon cancer study, where we gain additional insight into dependencies between survival and censoring times that would remain hidden under standard independent-censoring assumptions. Details of the paper can be found here.