An Alternative Treatment Effect Measure for Time-to-Event Oncology Randomized Trials

Background/Objectives: Time-to-event endpoints such as Overall Survival (OS), Progression-Free Survival (PFS), and Event-Free Survival (EFS) are central in phase III oncology trials. Hazard ratios from Cox proportional hazards models and log-rank tests are the standard analytic tools, supplemented by Kaplan–Meier estimates. However, these methods depend on proportional hazards to deliver unbiased estimates of treatment effects and large-sample assumptions, and may perform poorly under heavy censoring or non-proportional hazards. We introduce the univariate martingale residual (UMR) as a new endpoint and summary measure that enables exact inference through randomization testing. Methods: The UMR reflects the difference between observed and expected events at the subject level. Average UMRs per treatment arm provide an absolute measure of excess events. A randomization-based testing framework is used to compare treatment arms and compute exact p-values without proportional hazards or asymptotic assumptions. Performance is assessed through simulations and demonstrated using real oncology trial data. Results: UMRs offered robust and interpretable treatment summaries under heavy censoring, non-proportional hazards, and quasi-complete separation, where Cox-based estimates were unstable or undefined. The exact UMR-based randomization test maintained Type I error control and was competitive or more powerful than the log-rank test when proportional hazards were violated. Conclusions: The UMR provides an intuitive, assumption-free summary of treatment effects and supports exact inference. It represents a practical and robust alternative to hazard-ratio-based methods in phase III oncology trials, especially in complex survival settings.