A look on the effects of treatment noncompliance in the multivariate CACE analysis via Bayesian approach

Abstract

Noncompliance is an important issue in the design and conduct of randomized controlled trials (RCTs) – a type of study in which subjects are randomly assigned to either a treatment group receiving some clinical intervention or a control (placebo) group.  Treatment noncompliance is a common issue in RCTs that may plague the randomization settings and may produce treatment effect estimates that are biased.

The Complier Average Causal Effect (CACE) is a methodology that is popular in estimating the impact of an intervention among treatment even when there is noncompliance.  Yue Ma in 2018 introduced the Multivariate CACE (MCACE) analysis and showed that the methodology outperformed the classical CACE methodology via the maximum likelihood estimation (MLE) approach [8].

This paper explores the behavior of the model treatment estimates of the MCACE model via a bayesian estimation (BayesE) approach.  The proposed BayesE methodology explores impact on the treatment effect parameters when varied values of compliance rates  are imposed.  Here,  and a  of 20% implies an 80% noncompliance.  The derived MCACE models are then compared to the derived MCACE models using MLE.

Simulation study shows that as  increases from 20% to 80%, the derived treatment effect estimates of the MCACE model via BayesE gave more precise values than the treatment effect estimates derived via MLE.  Comparison of the two models is based on its corresponding mean square error (MSE) values.