A1430
Title: Time-dependent effects in survival analyses
Authors: Martina Mittlboeck - Medical University of Vienna (Austria) [presenting]
Abstract: Special statistical methods have been developed for the analysis of censored time-to-event data. Commonly used approaches, such as the Cox proportional hazards model and the log-rank test, rely on the assumption of proportional hazards. However, treatment comparisons and recommendations may be complicated by non-proportional hazards arising from delayed treatment effects, diminishing benefits over time, or crossing survival curves. Crossing hazards are frequently encountered in stem cell transplantation (SCT) studies, where the risk of severe early adverse effects is balanced against the prospect of improved long-term outcomes. A particular challenge in SCT studies is that group membership is often unknown at baseline and is determined only after a potentially lengthy donor search process. If the search is terminated prematurely, group membership may remain unresolved. A novel method for estimating group membership has been proposed that enables an unbiased comparison of treatment groups. As non-proportional hazards may substantially affect treatment comparisons and clinical decision-making, they should be considered already at the trial planning stage, including study design, sample size determination, trial duration, and the choice of statistical methods. The methodological challenges arising from unknown group membership and non-proportional hazards are illustrated and discussed using a real-world SCT data example.
