The randomized clinical trial is recognized by statisticians as the best tool for establishing the causal effect of a treatment on a clinical outcome. However, even with the best of intentions, a randomized clinical trial may turn out not to measure the causal effect of treatment due to patient noncompliance; e.g. refusal of some study participants to take the treatment to which they were assigned.
In this talk, we will discuss the idea of cause and effect of a treatment intervention in a clinical trial with noncompliance through the use of so-called counterfactuals . Through this perspective, we will argue that the randomized clinical trial is indeed a useful method for measuring the causal effect of an intervention when all study participants do comply with their assigned treatments, and we will elucidate the difficulties that arise in making causal inference when they do not.
Because noncompliance is a routine phenomenon in clinical trials, competing strategies for the analysis of data from these trials have emerged, and choice between these has been subject to considerable controversy. We will introduce and discuss the principles underlying the two main methods of analysis, "intention-to-treat" and "as-treated," and through the use of counterfactuals in a simple example of noncompliance, contrast their respective abilities to make proper causal inference. Other methods that may be more appropriate will also be introduced.
This talk is meant to provide a non-technical introduction to some of the basic considerations underlying the study of causal inference. Future talks will focus on formal definitions of concepts, issues that arise in more complex situations, including observational longitudinal studies, and the use of graphical approaches to inference in these settings.
Return to Biostatistics Working Group