The intention-to-treat analysis evaluates the causal effect of treatment assignment in a randomized controlled trial; however, participants do not always adhere to assigned treatment and the intention-to-treat effect may differ from the effect of treatment receipt. Although more recent adherence-based methods assess a well-defined causal effect of receiving treatment, adherence is assumed to be dichotomized as all or none. This approach can lack precision in the real world because adherence is a complex and heterogeneous phenomenon. In this article, we illustrate a simple method that provides estimates of bounds on the causal effect of full adherence to treatment in the presence of partial adherence. We first define three types of partial adherence (delayed, partial-dose, posttreatment). We then use casual diagrams to show that categorizing partial adherence as nonadherence in a sensitivity analysis can lead to a violation of the exclusion restriction principle. Finally, we apply recently published sensitivity analyses related to principal stratification that allow for creating bound estimates around the causal effect of treatment in the presence of partial adherence.
From the aCentre for Clinical Epidemiology, Lady Davis Institute for Medical Research, Jewish General Hospital, McGill University, Montreal, QC, Canada; bDepartment of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, QC, Canada; cDepartment of Pediatrics, McGill University, Montreal, QC, Canada; dThe Research Institute of the McGill University Health Centre, Montreal, QC, Canada; eDepartment of Mathematics and Statistics, McGill University, Montreal, QC, Canada; fFaculty of Pharmacy, Université de Montréal, Montréal, QC, Canada.
Submitted March 19, 2016; accepted October 6, 2017.
Data and Programming Code for Replication: The data are simulated and code is available from the author upon request.
Disclosure: The authors report no conflicts of interest.
Supplemental digital content is available through direct URL citations in the HTML and PDF versions of this article (www.epidem.com).
Correspondence: Ian Shrier, MD, PhD, Centre for Clinical Epidemiology, Jewish General Hospital, 3755 Cote Sainte-Catherine Road, Montreal, QC H3T 1E2, Canada. E-mail: firstname.lastname@example.org.