A synthetic estimator for the efficacy of clinical trials with all-or-nothing compliance.

摘要: A critical issue in the analysis of clinical trials is patients' noncompliance to assigned treatments. In context a binary treatment with all or nothing compliance, intent-to-treat straightforward approach estimating effectiveness trial. contrast, there exist 3 commonly used estimators varying statistical properties for efficacy trial, formally known as complier-average causal effect. The instrumental variable estimator may be unbiased but can extremely many settings. treated and per protocol are usually more efficient than estimator, they suffer from selection bias. We propose synthetic that incorporates data-driven manner. linear convex combination variable, protocol, estimators, resembling popular model-averaging literature. However, our nonparametric; thus, it applicable variety outcome types without specific distributional assumptions. also discuss construction using an analytic form derived simple normal mixture distribution. apply trial post-traumatic stress disorder.