5D-05 - Assessing the appropriateness of using EB05 for signal detection in a company’s spontaneous pharmacovigilance database with low case volume in a therapeutic area

Background: The lower bound of the Empiric Bayes Geometric Mean’s 90% credible interval calculated using Multi-item Gamma-Poisson Shrinkage, also known as EB05, is often used in disproportionality analyses to detect unexpected occurrences (i.e. “signals”) of adverse events (AEs) in spontaneous pharmacovigilance reporting systems. However, it is unclear when the use of EB05 would be appropriate for a particular therapeutic area (TA) when the AE case volume is low from the TA. This is relevant when a pharmaceutical company enters into a new therapeutic area.

Objectives: The aim of this study was to assess the appropriateness of using EB05 for signal detection in a low case volume setting.

Methods: Using the US FDA Adverse Event Reporting System (FAERS) database from 2015Q1 to 2024Q2, we investigated the performance of EB05 in detecting AE signals for an FDA approved oncology product (called Drug ABC in this study) with a well described AE profile. To mimic the numbers of drug-event combinations (DECs) in a real-world safety database with a low case volume from a particular TA, we randomly sampled 1.5%, 3%, and 6% of the total DECs in FAERS. We simulated 3 scenarios for each sample by varying the volumes and proportions of DECs for all oncology drugs (up to 13%, which was the proportion of oncology DECs in FAERS), non-oncology drugs (up to 100%), and Drug ABC (up to 50% of all oncology DECs). We explored how EB05 performed and assessed the minimum case volume needed to detect well described adverse drug reactions in the FDA approved label in each scenario. Signal detection for Drug ABC was measured by EB05 ≥2 for the Medical Dictionary for Regulatory Activities preferred terms (PTs) compared to all other drugs. The PTs listed on the FDA label for Drug ABC were defined as true signals.

Results: Key safety warnings on the FDA label for Drug ABC were detected using both the full FAERS and sampled datasets. Varying the volume and proportion of oncology DECs had a negligible impact on the sensitivity and specificity of signal detection. Overall, the sensitivity and specificity in each data sample were consistent with those found in the full FAERS data for all 3 simulated scenarios. For instance, in the 3% case volume scenario (500,000 DECs), the sensitivity was 43% and specificity was 95% when the proportion of DECs for oncology drugs equaled the proportion in FAERS. When the reporting ratio is 2 for a signal, it is necessary to have at least 3,000 DECs for the drug of interest in the database in order to successfully detect the signal (EB05 ≥2).

Conclusions: In the 9 case volume scenarios we explored, EB05 may be used appropriately for oncology signal detection in a company’s safety database having a total case volume as low as 250,000 DECs (6% oncology cases) and as low as 3,000 DECs for a drug of interest given a reporting ratio of 2.