Background: A regulatory agency or clinician is interested in the total causal effects of recommending an intervention, which is provided by the intention-to-treat (ITT) analysis. Differing adherence rates will create heterogeneous results in a meta-analysis, which may be considered part of the treatment effect. Alternatively, the individual patient is only interested in the total causal effect of taking versus not taking treatment (complier average causal effect)—how the average population causal effect changes when others decide not to adhere to treatment is of little interest. As these analyses become more frequent, meta-analyses of these patient-focused estimates will become more common. It is essential that meta-analysts understand the fundamental assumptions and limitations of these approaches.

Objectives: To explain why and how non-ITT, patient-oriented treatment causal effects (complier average causal effects), and their underlying assumptions. This presentation will review and explain the different analytical methods currently being used to determine these causal effects with respect to (1) the different questions being addressed by different methods, and 2) underlying assumptions of each method. The results of the different analyses will be compared across two published RCTs to illustrate the strengths and weaknesses of different approaches.