Enhancing Safety Signaling: Integrating Clinical Trials and Post-Marketing Adverse Event Reports

Problem definition: Adverse drug reactions (ADRs) cause numerous emergency room visits and thousands of deaths annually in the U.S. Although regulators like the Food and Drug Administration (FDA) aim to detect safety issues during clinical trials, these trials are often too small, short, or restrictive to capture rare or delayed ADRs. Consequently, regulators rely on spontaneous adverse event reporting systems (eg, FAERS) to identify potential safety issues—termed safety signals—by detecting side effects that occur disproportionately among patients taking a particular drug. However, these observational data are subject to selection and reporting biases, which can lead to false signals or missed risks. Methodology/results: We propose a Bayesian framework to enhance the hypothesis generation step in safety signaling, offering two methodological advances over standard approaches that rely solely on observational data. First, it integrates pre-approval clinical trial and post-approval observational data. Second, it uses information from common ADRs to estimate and correct bias in rare ADRs. We assess the approach using an analytical model and a large-scale numerical study with real clinical trial and FAERS data. Incorporating clinical trial data substantially reduces signaling costs, and using a common ADR for debiasing yields further gains. This refinement improves decisions for about 1% of ADRs, translating to expected cost savings of 1.5–2.3% (approx. $32–159 million), with performance robust to moderate parameter misspecification. Managerial implications: By using existing regulatory data, our approach supports a more efficient and timely hypothesis generation process in post-approval surveillance.