B1379
Title: A novel covariate adjustment strategy for guaranteed efficiency gain in randomized clinical trials
Authors: Marlena Bannick - University of Washington (United States) [presenting]
Ting Ye - University of Washington (United States)
Jun Shao - University of Wisconsin - Madison (United States)
Yanyao Yi - Eli Lilly (United States)
Jingyi Liu - Eli Lilly (United States)
Yu Du - Eli Lilly (United States)
Abstract: In randomized clinical trials, adjusting for baseline covariates has been advocated to improve credibility and efficiency for demonstrating and quantifying treatment effects. The augmented inverse propensity weighted (AIPW) estimator is studied, a general form of covariate adjustment that can incorporate linear, generalized linear, and machine learning models. Theoretical conditions are established under which AIPW estimators have guaranteed efficiency gain and universal applicability under covariate-adaptive randomization. Motivated by these conditions, a covariate adjustment strategy is proposed called joint calibration that ensures both guaranteed efficiency gain and universal applicability are achieved. The utility of joint calibration is demonstrated through simulation and analysis of existing trial data.
