Causal methods for Treatment Effect Estimation and Heterogeneity Assessment Using Real-world Data
Tat Thang Vo, PhD Postdoctoral research fellow Department of Statistics and Data Science The Wharton School University of Pennsylvania
From the analysis of Real-World Data (RWD), one can obtain real-world evidence (RWE) about the usage and potential benefits-risks of a medical treatment, which complements results of traditional clinical trials. RWE studies, however, currently face two important challenges. These include (i) incomplete/inaccurate information on potential confounding variables of the considered treatment-outcome relationship and (ii) substantial heterogeneity in the findings obtained from different RWD studies. In this talk, I will discuss methods to address these challenges. For challenge (i), we develop a novel method called instrumented difference-in-differences, which leverages an exogenous variable that does not have any direct causal impact on the outcome trend over time except via the exposure trend over time, and is not associated with the unmeasured confounders on the trend scale. For challenge (ii), we propose a novel framework for heterogeneity assessment in RWE synthesis. The proposed framework allows one to measure how much of the statistical heterogeneity across RWE studies (e.g. 100%) can be attributed to the difference across studies in case-mix (e.g. 30%), in the distribution of important mediators of the treatment-outcome relationship (e.g. 40%) and residual heterogeneity (e.g. 30%). In both cases, we illustrate the use of the new methods by using RWD obtained from the claim database Optum Clinformatics..
Hybrid format: Zoom link to follow 8 February 2023 1:30pm