Since the introduction of the ICH E9 (R1) estimand addendum, handling potential intercurrent events (ICE) has been a strong motivation for research and discussions on the analysis for the primary estimand. Under the estimand framework, causal inference plays a crucial role in mitigating inference bias in clinical trials, particularly when the covariates are imbalanced between the treatment and control groups, or when there is imperfect or no randomization. Thus, to estimate the treatment effect, we use the hypothetical estimand adjusting for covariates, while continuing to include all patients as randomized. This article develops a time-varying covariate model to obtain an asymptotically unbiased estimate of the treatment effect in clinical trials while adjusting for covariates, including time-varying ones, such as treatment discontinuation and the use of rescue medicine, from the causal estimation approach, specifically by utilizing a class of enhanced doubly robust estimators of a treatment effect estimand developed recently by the authors. Since the estimand here is the would-be treatment effect in the absence of ICE, while utilizing data from all patients as randomized, the construction of eDRE is nontrivial. The asymptotic properties of this new enhanced doubly robust estimator are studied. Simulation studies are conducted to evaluate the performance of the proposed method and compare it with several commonly used existing parametric and naive estimates. The proposed method is then applied to analyze a real clinical trial.
Zhang et al. (Tue,) studied this question.