Regression Discontinuity Designs in Healthcare: Recent Advances and Challenges

The regression discontinuity design (RDD) is a quasi-experimental design that can be used to measure the causal effect of treatments that are assigned based on a running variable crossing a threshold. In healthcare, RDDs have grown increasingly popular but face key challenges, including discrete running variables, precise cohort definition, and time-to-event outcomes. We provide an overview of existing approaches to regression discontinuity and discuss methods to address each challenge. Drawing on several examples in healthcare, we review current RDD estimation strategies, including sharp and fuzzy RDDs, continuous and discrete running variables, as well as different types of outcomes. In this context, the course will highlight existing methods to adjust for right-censoring in settings with time-to-event outcomes, including inverse probability of censoring weighting and doubly robust scores. We will also introduce nonparametric methods for censoring adjustment that can be paired with existing estimators for RDD. After reviewing statistical concepts, we will discuss applications from recent biomedical literature and present applications from our own work. Finally, we conclude with an interactive coding session using new vignettes and an accompanying software package that unifies existing RDD methods for time-to-event outcomes.