Many studies of drug safety and comparative effectiveness contrast outcomes in treated patients against patients exposed to a comparable control drug. Investigators typically match members of the two groups based on possible confounding variables, often via the propensity score. In this work, we propose and evaluate improvements to the first step of the cohort comparison process: selecting the control patient group.
Learning Objective: Attendees will understand the weaknesses of the current cohort study procedure and how to use embeddings to improve cohort selection.
Rachel Melamed (Presenter)
University of Chicago