Primary Submission Category: Randomized Studies
Statistical challenges of pragmatic randomized trials with intervention-dependent outcome assessment processes
Authors: Jennifer Bobb, Sungtaek Son, Melissa Anderson, Lynn DeBar, Katharine Bradley,
Presenting Author: Jennifer Bobb*
Pragmatic trials are increasingly being conducted that use real-world data such as electronic health records (EHRs) to define study outcomes. Unlike traditional trials in which outcomes are collected at pre-specified time points, follow-up times from EHRs are irregularly spaced, not controlled by the study team, and may be outcome dependent. In this talk, we explore issues that arise when interventions being studied affect outcome assessment, including frequency and timing of follow-up measures. This work is motivated by the MICARE trial among primary care patients with opioid use disorder and depression, in which the intervention is expected to increase documentation of the depression outcome in the EHR (Patient Health Questionnaire [PHQ] measure of depression). We conducted simulations to examine the performance of common statistical approaches in the setting of intervention-dependent outcome assessment times, including simple approaches that select a single follow-up measure per person (e.g., score closest to 12 months), and longitudinal approaches that use all follow-up data. We additionally clarify which estimands are being estimated when observation times are intervention dependent, including time-point specific and (weighted) time-averaged treatment effects. Our results informed the selection of statistical approach for the MICARE trial and have important implications for the design of trials with intervention-dependent measurement processes.