From manual to automated pharmacovigilance processes and efficient signal management application

ABSTRACT Çetin, R. A. (2024). From Manual to Automated Pharmacovigilance Processes and Effıcient Signal Management Application. Yeditepe University, Institute of Health Sciences, Department of Pharmaceutical Toxicology MSc thesis, İstanbul. Pharmacovigilance (PV) tasks, which are generally time-consuming and labor-intensive activities, play a critical role in ensuring drug safety. This thesis study aims to evaluate the efficiency of automation in selected PV tasks and signal detection. It was conducted in two phases. In the first phase, the study examines the efficiency of automated processes for the PV tasks, Individual Case Safety Report (ICSR) entries and Periodic Benefit-Risk Evaluation Report (PBRER) preparation. A small-scale group of PV employees with different levels of experience participated in this phase. The second phase of the study focuses on identifying drug-adverse event (AE) interaction patterns within the FDA Adverse Event Reporting System (FAERS) data using machine learning (ML) techniques, focusing on 10-year data from January 2012 to December 2021. The dataset included 13,568,552 cases, and six different ML algorithms were applied to detect potential signals. The findings of the first phase suggested that despite the limitation of sample size and not being able to include all possible scenarios for PBRERs and case entries, automated processes improve the time efficiency for PV tasks by providing an average of 38.5% time savings for the ICSR entry task and an average of 26% time savings for PBRER preparation task. In the second phase of the study, XGBoost was selected as the most effective model among the applied ML algorithms, achieving an accuracy of 0.97, an F1 score of 0.71, and an AUC of 0.95. A significant finding was that high-scoring drug-AE combinations demonstrated exponential increases in case counts over the years, indicating their potential as signals. The results suggest that ML methods have potential for accurately identifying drug-AE interactions, contributing to more efficient pharmacovigilance. Keywords: Pharmacovigilance · PBRER · Automation · Signal · Machine Learning