Abstract:
|
Noncompliance to assigned treatments is a common challenge in the analysis and interpretation of randomized trials. The complier average causal effect (CACE) estimation approach provides a useful tool for addressing noncompliance problems, where CACE is the average difference in potential outcomes for the response in a subpopulation of subjects who comply with their assigned treatment. In this article, we present an analytic approach to estimating the causal effect with noncompliance in a meta-analytic setting. Study-specific random effects account for between-study heterogeneity are included to estimate CACE under Bayesian hierarchical models. Using model selection techniques, we explore a range of random effects models to identify a best-fitting model. Finally, we conduct simulations to illustrate the importance of including appropriate random effects and the impact of overfitting and underfitting on model performance. The results are illustrated through reanalyzing a multicenter large field trial -- the multiple risk factor intervention trial (MRFIT), where noncompliance of smoking rates vary across centers.
|