Pascal Geldsetzer, PhD
Assistant Professor of Medicine in the Division of Primary Care and Population Health
Title: Regression Discontinuity in Electronic Health Record Data
Abstract: Regression discontinuity in electronic health record (EHR) data combines the main advantage of randomized controlled trials (causal inference without needing to adjust for confounders) with the large size, low cost, and representativeness of observational studies in routinely collected medical data. Regression discontinuity could be an important tool to help clinical medicine move away from a “one size fits all” approach because, along with the increasing size and availability of EHR data, it would allow for a rigorous examination of how treatment effects vary across highly granular patient subgroups. In addition, given the broad range of health outcomes recorded in EHR data, this design could be used to systematically test for a wide range of unexpected beneficial and adverse health effects of different treatments. I will talk about the broad motivation for this research and discuss examples from some of our ongoing work in this area. If there is time, I will also discuss some of my ongoing research on improving healthcare services for chronic conditions in low- and middle-income country settings.