Event

A brief history and adoption curve for incorporating real-world evidence (RWE) into the regulatory pipeline and introduction to the nascent field of External Control Arm (and alternate definitions/configurations, including synthetic control arms) will be presented. In so doing, traditional approaches and their limitations will be explored as well as challenges that arise from “performative propensity scores” and the ubiquitous “table two fallacy.” An overview of artificial intelligence and machine learning approaches will be provided, and then a coherent and novel framework for the best available methods for these types of studies will be introduced, namely the introduction of targeted learning and targeted maximum likelihood estimation (tmle). This elegant marriage of how causal inference and machine learning fulfils the promises of AI in healthcare will be demonstrated--a counterfactual framework that incorporates rich context to make outcomes comparable and effects estimable. Finally, a perspective on a way forward, both sensitive to the challenges and traditions of current best practices and optimistic toward the future of RWE generation, will be offered. 

PREREQUISITE: Must have basic knowledge of confounding, regression, notion of machine learning, basic intro statistics knowledge (probability, type I error, confidence interval). Participants will need to use a laptop with access to the Internet and RStudio installed to participate in the hands-on exercises.

Faculty

Basic Schedule:
4 Hours | Course runs 2 consecutive days, 2 hours each day