Minimizing Signal Loss and Optimizing Pharmacovigilance in VAERS

Background: The Vaccine Adverse Event Reporting System (VAERS) serves as a critical repository for adverse event (AE) data spanning over three decades, yet its passive reporting nature, outdated infrastructure, data inconsistencies, inability to establish causality, and lack of denominator data for incidence calculations limit its pharmacovigilance utility and necessitate modernization.

Objectives: To evaluate three key limitations of VAERS – data quality and inconsistency, inadequate signal detection with no inherent causality inference, and inability to compute incidence ratesand propose a comprehensive framework for its redesign, incorporating enhanced data cleaning, advanced signal detection, and automated causality assessment.

Methods: An extensive literature review was conducted alongside analysis of the VAERS database, focusing on 2021 data. Data cleaning addressed vaccine lot inconsistencies, free-text extraction, and date standardization. MedDRA Preferred Terms (PTs) were condensed to mitigate coding dispersion and recover obscured signals. Proportional Reporting Ratio (PRR) analyses detected safety signals, followed by causality assessments using the Bradford-Hill Criteria (BHC) to generate a Composite Causality Score. Myocarditis following COVID-19 vaccination (particularly Moderna) was used as a test case for signal detection and automation. A cutting-edge AI tool was evaluated for automated PRR calculation and BHC-based causality ranking. Temporal signals in early-life vaccination (via CAGE_MO variable) were also examined.

Results: Literature confirmed VAERS limitations, including underreporting and data infidelity. Post-cleaning of 2021 data, the PRR for myocarditis was 23.7 (COVID-19 vs. influenza vaccines), with BHC assessment confirming a strong causal link to Moderna COVID-19 vaccines, particularly in male children with elevated incidence based on CDC dose data. Coding dispersion obscured additional signals, while condensation revealed them. The AI tool successfully flagged myocarditis with PRR >2 and calculated a Composite Causality Score of 9.76/10. Analysis of early-life temporal variables highlighted potential for improved signal detection in birth-dose reports.

Conclusions: VAERS requires fundamental upgrades to support inherent disproportionality analyses, causality assessments, and incidence calculations. Integrating automated statistical methods, AI-driven tools, transparent data handling, and higher-granularity variables would enhance data quality, usability, and reliability, transforming VAERS into a more effective public health pharmacovigilance resource.

Keywords: COVID-19, AE, VAERS, PRR, BHC, myocarditis, AI tool