In previous chapters, we introduced various methods for dealing with missing data. In this chapter, we apply some of these methods to make causal inference in randomized clinical trials with noncompliance and possibly missing outcome data.
A randomized clinical trial is the established gold standard for assessing the causal effect of a new intervention on patient outcomes. The validity of using a simple comparison of outcomes across intervention and control groups to estimate the causal effect of the intervention versus control depends on the assumption that the randomized experiment is implemented perfectly as designed. However, in practice, due to various reasons, many violations of the randomization protocol occur, and one of them is noncompliance, which occurs when the actual treatment a patient takes is different from the initially assigned treatment. Although on the surface it seems that the causal inference of randomized clinical data with noncompliance has nothing in common with the analysis of missing data, it turns out that techniques for dealing with missing data can also be adapted to estimate causal effects with randomized clinical data with noncompliance.
In practice, we encounter two types of noncompliance: (1) all-or-none noncompliance (e.g., either compliant or not) and (2) partial noncompliant. Since most of the available methods in the literature are for all-or-none compliance, ...