Background: When a medical product is suspected of having a safety problem, an active comparator cohort design is often implemented along with covariate balancing using the propensity score.
Objectives: In this presentation I argue that balancing methods that target the Average Treatment Effect on the Treated (ATT) have distinct inferential advantages over alternatives, such as caliper matching, and as such should be prioritized.
Methods: Specifically, the ATT answers the highly relevant question of what would happen if the product with safety concern were taken off the market (in favor of the comparator), as well as generalizing the product’s safety to the population of all treated patients.
Results: Prioritizing one estimand (the ATT) over other estimands has precedents in both RCTs and observational studies, examples of which will be provided. Guidance is offered with respect to selecting among ATT estimators and identifying contexts where alternative estimands might be informative.
Conclusions: Future research should prioritize statistical methods designed to estimate the ATT when addressing a medical product safety concern. Beyond the inferential advantages of using the ATT, a single agreed-upon standard can support greater harmonization of methods in pharmacoepidemiology and medical device epidemiology.
