Combining experimental and non-experimental data to examine treatment effect heterogeneity
Determining “what works for whom” is a key goal in prevention and treatment across a variety of areas, including mental health. Identifying effect moderators-factors that relate to the size of treatment effects-is crucial for delivery of treatment and prevention interventions, but doing so is incredibly difficult using standard study designs. Randomized trials, the gold standard for estimating average effects, are typically under-powered to detect moderation. Large-scale nonexperimental studies may provide another way to examine effect moderation, but can suffer from confounding. This talk will describe recent machine learning and Bayesian methods advances to combine randomized trials and electronic health record (EHR) data to examine effect heterogeneity. We present results from simulation studies comparing a set of recently proposed methods for combining data sources, with the goal of estimating conditional average treatment effects. We also provide initial application of the methods to data from randomized trials and electronic health record data of individuals receiving medication treatment for major depressive disorder.
Pre-reading:
Methods for Integrating Trials and Non-experimental Data to Examine Treatment Effect Heterogeneity
Speaker:
Elizabeth Stuart, Johns Hopkins University