(A-270) A Collaboration to Explore and Compare Existing Tools for Retrieving Pregnancy Reports from VigiBase

Background: Post-marketing surveillance is essential for complementing the safety profiles of medicinal products, especially for populations generally excluded from clinical trials, such as pregnant individuals. The lack of a standardized pregnancy indicator in the electronic transmission of individual case safety reports (ICSRs) complicates the retrieval of pregnancy-related reports, which may be important for targeted signal detection or when accounting for confounders. Recent initiatives have developed rule-based algorithms to systematically retrieve reports concerning pregnancy, each with different scopes and rationales, tailored for different databases.

Objectives: This study was conducted as a collaboration among the authors of three recently developed pregnancy algorithms to compare them and make them openly available in an R package, empowering professionals working with pregnancy-related research questions to select and use the most appropriate algorithm for their specific needs.

Methods: We first harmonized the inclusion and exclusion criteria of the three pregnancy algorithms, originally developed for VigiBase (A), EudraVigilance (B), and FAERS (C), then implemented each of the three algorithms in VigiBase. Second, we compared the key characteristics of the ICSRs exclusively retrieved by each algorithm.

Results: The pregnancy algorithms A, B, and C retrieved 446 957, 279 515, and 235 653 reports, respectively. Reports exclusively retrieved by each algorithm (142 324, 3 248, and 994 reports, respectively) reflected their unique features. Specifically, algorithm A excluded paternal exposure, algorithm B excluded normal pregnancy and ineffective contraception, and algorithm C had no age restriction.

Conclusions: The differences in phenotyping can be attributed to the varying scopes and desired sensitivity and specificity of each algorithm. Understanding these commonalities and differences is crucial for selecting and interpreting the most appropriate algorithm for specific use cases.