Abstract:
|
In clinical trials that are subject to noncompliance we commonly seek to estimate the intention-to-treat (ITT) estimand, which does not measure the causal effect of the treatment received and is sensitive to the level of compliance. An alternative estimand, the complier average causal effect (CACE), refers to the average effect of treatment received in the latent subset of subjects who would comply with either treatment. Under the Rubin Causal Model (RCM), five assumptions are sufficient to identify CACE and enable its consistent estimation from trial data. Taking compliance itself to be a random quantity, we observe that CACE can also vary with the fraction of compliance. We propose a “sixth assumption” that specifies that, at the subject level, the randomized treatment effect on compliance is independent of the total causal effects of both randomized treatment and treatment actually received on outcome, in the superpopulation from which trial samples are drawn. This assumption guarantees robustness of CACE to the compliance fraction. We demonstrate the potential degree of sensitivity in a simulation study and a data analysis from a trial of vitamin A supplementation in children.
|