Conference Program

Validity of Tests for Time-to-Event Endpoints in Studies with Covariate-Adaptive Randomization (303697)

View Presentation

Jun Shao, University of Wisconsin-Madison 
*Ting Ye, University of Washington 
YANYAO YI, Eli Lilly and Company 

Keywords: log-rank test, covariate adjustment, permuted block

It was demonstrated that under complete randomization the score test with the Lin and Wei (1989) sandwich variance is valid under model misspecification, and so is the logrank test. However, complete randomization is rarely used in clinical trials and the validity question remained open for restricted randomization (such as permuted block randomization) and covariate-adaptive randomization. Ye and Shao (2020) demonstrated theoretically that the logrank test/score test is conservative under model misspecification and stratified permuted block design. They also showed that stratified log-rank test preserves the Type I error if all prognostic covariates used in randomization are included in stratification. Additionally, Ye, Shao, and Yi (2022) proposed the covariate-adjusted log-rank test that guarantees efficiency gain over the unadjusted test for stratified randomization.