Estimand-Based Covariate Adjustment under Nonproportional Hazards with Long-Term Survivors

Abstract

In time-to-event studies, the hazard ratio remains a common measure, but its interpretation becomes problematic under nonproportional hazards, which are increasingly seen in modern trials, especially immunotherapy studies with durable responders. At the same time, recent regulatory guidance encourages the use of prespecified prognostic baseline covariates to improve efficiency in randomized trials. In this talk, I will discuss treatment effect estimands that distinguish long-term and short-term survival benefit. To improve efficiency through covariate adjustment, we derive efficient influence functions for the estimands of interest. In particular, we study a concordance-based tau-process summary, which quantifies the probability that one treatment yields longer event-free survival than the other. This provides an interpretable summary of short-term treatment advantage under nonproportional hazards. Overall, this work offers an interpretable covariate-adjusted framework for survival trials with nonproportional hazards and durable benefit, connecting regulatory guidance on covariate adjustment with estimand-based causal inference.

線上視訊請點選連結