Biostatistics Weekly Seminar

Pragmatic randomized trials are key tools for research on the comparative effectiveness of medical interventions. Unlike other randomized trials, pragmatic trials are specifically designed to address real-world questions about options for care and therefore to guide decisions by patients, clinicians and other stakeholders. Therefore, characteristics of a pragmatic trial include typical patients and care settings, clinically relevant comparators, unconcealed assignment to treatment, and follow-up time long enough to study long-term clinical outcomes without having to rely on surrogates. While pragmatic trials are useful to guide decision making, they are also especially vulnerable to post-randomization confounding from incomplete adherence and post-randomization selection bias from loss to follow-up. These sources of bias are common in observational epidemiology, and the use of analytic approaches pioneered for observational studies can improve inference from pragmatic trials. Here we propose causal inference guidelines tailored for the analysis of pragmatic randomized trials using methods from observational research. Importantly, conventional methods to adjust for confounding and selection bias do not generally work for post-randomization variables. In fact, conventional methods such as multivariate outcome regression, stratified analyses, propensity score regression and matching, and others may themselves introduce bias. Therefore, our guidelines are based on so-called g-methods, developed by Robins and collaborators since 1986, which can appropriately adjust for post-randomization biases. Because g-methods require data on post-randomization (that is, time-varying) treatments and covariates, embracing these guidelines will require a revised framework for both the design and conduct of both pragmatic trials and other trials with substantial loss to follow-up or non-adherence.

MRBIII, Room 1220
26 February 2020
1:30pm