Why we should be prioritizing the average treatment effect on the treated over other estimands when evaluating drug and device safety

When a drug or medical device is suspected of having a safety problem, observational studies are often utilized with an active comparator cohort design and covariate balancing using the propensity score. Each covariate balancing method is an estimator for a particular estimand, with each estimand characterizing the target population of interest differently as it relates to the treatment effect. In this article, I argue that characterizing the average treatment effect in the treated population (ATT), has a distinct inferential advantage over estimands characterizing the treatment effect in either the comparator population, entire population, and the overlap population, and as such the ATT should be prioritized. Regulatory guidance offers little direction with respect to estimand selection in observational studies, and a review of recent pharmacoepidemiology studies suggests that the ATT is infrequently used. Guidance is offered with respect to selecting among ATT estimators and identifying contexts where alternative estimands might be more informative. An empirical example is used to illustrate the implementation of the described methods. The implications of adopting the recommendations set forth in this article are considered.