Primary Submission Category: Sensitivity Analysis
Inferring Comprehensive Cohort Causal Effects in the Presence of Unmeasured Confounders and Missing Outcomes
Authors: Shiyao Xu, Razieh Nabi, Daniel Scharfstein,
Presenting Author: Shiyao Xu*
Randomized control trials (RCTs) are considered the gold standard approach for estimating causal effect. However, RCTs often enroll patients who are not representative of a broader population. To address these limitations, we consider the comprehensive cohort study (CCS) design, where clinically eligible patients are first asked to enroll in an RCT, and if they decline, are asked to participant in a parallel observational study. Data on baseline covariates, treatments and outcomes are collected on all patients. In this paper, we present a methodological framework for estimating the comprehensive cohort causal effect (CCCE) – the difference in mean potential outcomes had all patients in the CCS received treatment A vs. treatment B, in the presence of unmeasured confounding in the observational arm (handled via sensitivity analysis) and outcome missingness (assumed to be missing at random). We apply our methods to the TOIB study, a CCS to determine the effect of topical versus oral non-steroidal anti-inflammatory drugs (NSAIDs) in managing knee pain among older adults with chronic knee pain. We also conduct a simulation to evaluate the performance of our approach.