The U.S. Food and Drug Administration (FDA) regulates the approval and safe public use of pharmaceutical products in the United States. The FDA uses postmarket surveillance systems to monitor drugs already on the market; a drug found to be associated with an increased risk of adverse events (ADEs) is subject to a recall or a warning. A flawed postmarket decision-making process can have unintended consequences for patients, create uncertainty among providers and affect their prescribing practices, and subject the FDA to unfavorable public scrutiny. The FDA’s current pharmacovigilance process suffers from several shortcomings (e.g., a high underreporting rate), often resulting in incorrect or untimely decisions. Thus, there is a need for robust, data-driven approaches to support and enhance regulatory decision making in the context of postmarket pharmacovigilance. We propose such an approach that has several appealing features—it employs large, reliable, and relevant longitudinal databases; it uses methods firmly established in literature; and it addresses selection bias and endogeneity concerns. Our approach can be used to both (i) independently validate existing safety concerns relating to a drug, such as those emanating from existing surveillance systems, and (ii) perform a holistic safety assessment by evaluating a drug’s association with other ADEs to which the users may be susceptible. We illustrate the utility of our approach by applying it retrospectively to a highly publicized FDA black box warning (BBW) for rosiglitazone, a diabetes drug. Using comprehensive data from the Veterans Health Administration on more than 320,000 diabetes patients over an eight-year period, we find that the drug was not associated with the two ADEs that led to the BBW, a conclusion that the FDA evidently reached, as it retracted the warning six years after issuing it. We demonstrate the generalizability of our approach by retroactively evaluating two additional warnings, those related to statins and atenolol, which we found to be valid. This paper was accepted by Vishal Gaur, operations management.
Methods for analyzing decision-making: a framework approach
