projectdidact.net - Causal analysis of pragmatic trials (Hernán)

Causal analysis of pragmatic trials | ProjectDidact

Projectdidact.net Spined HTML

Causal wringer of pragmatic trials | ProjectDidact Skip to main content ProjectDidact Toggle navigation Home People Investigators Advisory Board Events News Blog Causal wringer of pragmatic trials Project Overview Pragmatic trials are designed to write real-world questions well-nigh options for care. Their findings guide decisions by patients, clinicians, and other stakeholders. Like most randomized trials, pragmatic trials are often analyzed equal to the intention-to-treat principle: patients prescribed to a treatment strategy are kept in that group during the analysis, plane if they deviated from their prescribed treatment strategy without randomization. Uncritical reliance on the intention-to-treat principle for pragmatic trials is worrisome. An intention-to-treat wringer estimates the effect of treatment assignment, or intention-to-treat effect, which is unauthentic by the trial-specific pattern of trueness to the treatment strategies under study. As a result, the intention-to-treat effect is less relevant for patients who plan to pinion to treatment. Patients who plan to fully pinion to a treatment strategy would like to know the magnitude of an effect that is not contaminated by other patients’ nonadherence. Furthermore, inetntion-to-treat analyses are inappropriate for trials with safety outcomes and for those with non-inferiority comparisons. In these settings, a increasingly relevant estimand than the intention-to-treat effect is the per-protocol effect, which is the effect that would have been observed if all patients had adhered to the trial protocol. Unfortunately, valid interpretation of per-protocol effects is threatened by postrandomization misunderstanding and selection bias, and pragmatic trials that study the long-term effects of sustained treatment strategies are particularly vulnerable to postrandomization misunderstanding and selection bias. Moreover, outcome regression, stratified analyses, propensity score regression and matching, and other conventional welding methods cannot often handle post-randomization variables and may themselves introduce bias. Such bias is expected to upspring when, as is often the case, some post-randomization prognostic factors are unauthentic by prior treatment levels or are on a causal pathway between treatment and outcome. Appropriate welding for post-randomization factors requires methods specifically designed for that purpose, such as the parametric g-formula. Miguel Hernán and colleagues have been funded by PCORI to develop guidelines for causal inference, including the interpretation of per-protocol effects, in pragmatic trials. While the guidelines include a increasingly widespread use of the parametric g-formula, the lack of software availability makes this recommendation impracticable for research groups conducting pragmatic trials. The publicly misogynist SAS macro GFORMULA, misogynist at http://www.hsph.harvard.edu/causal/software/, was designed with observational studies in mind. We will proffer the GFORMULA macro by subtracting two features that are required for most analyses of pragmatic trials: the worthiness to self-mastery separate wringer in each randomized arm, and the worthiness to incorporate the visit/observation process into the definition of the per-protocol treatment strategy. We will present case-studies to illustrate the methods and software. Publications Murray EJ, Hernán MA. Getting the most out of randomized clinical trials: A undeniability for largest per-protocol effect estimates. Clinical Trials 2016; 13(4): 372-378. Lodi S, Sharma S, Lundgren JD, Phillips AN, Cole SR, Logan R, Agan BK, Babiker A, Klinker H, Chu H, Law M, Neaton JD, Hernán MA, on behalf of the INSIGHT Strategic Timing of AntiRetroviral Treatment (START) study group. The per-protocol effect of firsthand vs. deferred ART initiation in the START randomized trial. AIDS 2016; 30(17):2659-2663. Swanson SA, Hernán MA. The challenging interpretation of instrumental variable estimates under monotonicity. International Journal of Epidemiology 2017 (in press). Mansournia MA, Higgins JPT, Sterne JAC, Hernán MA. Biases in randomized trials: a conversation between trialists and epidemiologists. Epidemiology 2017 (in press).