Primary Submission Category: Generalizability/Transportability
Data Fusion under Disparate Outcome Measures
Authors: Harsh Parikh, Elizabeth Stuart, Kara Rudolph,
Presenting Author: Harsh Parikh*
Randomized controlled trials (RCTs) serve as the cornerstone for understanding causal effects, yet extending inferences to target populations presents challenges due to effect heterogeneity and underrepresentation. Discrepancies in distributions across individual characteristics between trial and target populations often impair treatment effect extrapolations, especially for underrepresented groups, leading to inaccuracies in decision-making. Our paper addresses the critical issue of identifying and managing underrepresented subgroups in RCTs, proposing a novel framework for refining target populations to enhance treatment effect generalizability. We introduce an optimization-based approach, Rashomon Set of Optimal Trees (ROOT), to characterize underrepresented groups. By minimizing the variance of the target average treatment effect (TATE) estimate, ROOT optimizes the target subpopulation distribution, ensuring more precise treatment effect estimations. Notably, ROOT generates interpretable characteristics of the underrepresented population, aiding researchers in communicating for targeted recruitment in future trials. Through extensive evaluation using synthetic data experiments with complex structures, our approach demonstrates superior precision enhancement and interpretability compared to alternatives. Applying our methodology to an RCT on medication for Opioid Use Disorder (MOUD), we investigate discrepancies between trial results and real-world effectiveness.