|
Abstract:
|
In randomized clinical trials, in addition to intention to treat analysis, the treatment as delivered or as received analysis to account for treatment non-adherence also provides important information on the effectiveness of treatment. The principal stratification framework provides inference about causal effects among subpopulations characterized by potential compliance. In this investigation, we propose to use multiple imputation to identify potential compliers and estimate the complier average causal effect (CACE). Two multiple imputation strategies were considered based on the assumptions of exclusion restriction and principal ignorability. Simulation studies were performed to evaluate the proposed method and an analysis of a clustered randomized study of pharmacist intervention on pre-diabetic patients was presented.
|
