Risk-based Heterogeneity of Treatment Effect in Trials and Implications for Surveillance of Clinical Effectiveness Using Regression Discontinuity Designs

To the Editor:

The Perspective by Iwashyna and colleagues highlights the importance of stratifying clinical trial participants by level of risk to assess effect heterogeneity (1). By comparing efficacy across values of a risk score, an “efficacy threshold” can be identified at which patients could be provided (or not provided) the intervention. A critical question, however, is how to assess the real-world effectiveness of an efficacy threshold identified in such a trial. The quasiexperimental regression discontinuity offers an approach to evaluate real-world effectiveness of interventions found efficacious in subgroups of trial participants. Regression discontinuity analysis leverages situations in which interventions are instituted at a threshold of a continuously measured variable (2–4). Noise in measurements of that assignment variable quasirandomly assigns patients to be above or below the threshold and thus to intervention or control. An efficacy threshold that was previously identified in a trial designed according to Iwashyna and colleagues’ specifications could be later tested for effectiveness in real-world settings—and across different patient populations and care delivery systems. Investigators could use observational regression discontinuity methods for active surveillance of effectiveness by evaluating changes in the relationship between predicted risk and observed outcomes at the risk threshold for efficacy suggested by a trial. Health care systems may further use regression discontinuity analyses to explore whether risk thresholds identified in clinical trials optimally identify thresholds of efficacy among their patient populations, allowing titration of optimal risk cutoffs at individual hospitals. We encourage further investigation of methods to pair risk-based analysis for heterogeneity of treatment effects in clinical trials with post-trial, continuous surveillance and optimization of real-world effectiveness through quasiexperimental approaches.