Comment on 'Empirical assessment of methods for risk identification in healthcare data: Results from the experiments of the Observational Medical Outcomes Partnership'

摘要

As the U.S. Food and Drag Administration's (FDA) Sentinel System nears fruition [1] - the results of which will be used to inform regulatory decision making - it is critical to understand the performance of the observational methods that underpin the system. To this end, we commend Dr. Ryan and colleagues and the Observational Medical Outcomes Partnership (OMOP) for their ambitious efforts to evaluate analytic methods in the context of what they refer to as 'risk identification' [2]. An active drag safety monitoring system can serve multiple purposes - chief among them are the ability to identify previously unsuspected drag safety issues in a hypothesis-free space (which the FDA currently refers to as ' signal generation') and the ability to strengthen a prior hypothesis by measuring the direction, magnitude, and precision about a putative underlying causal relation (which the FDA currently refers to as 'signal refinement'). Because of the evaluation metric the OMOP used, their results apply to the former but not the latter.