A Comprehensive Review: Natural Language Processing and Leveraging Deep Learning Techniques for Adverse Drug Reaction Detection in Pharmacovigilance

Abstract

Adverse drug reactions (ADRs) represent a major global cause of infirmity and fatality rate. The timely and precise identification of ADRs plays a pivotal role in enhancing patient safety and healthcare outcomes. In light of the expanding volume of real-world data sourced from electronic health records (EHRs), social media, and various outlets, the need for effective detection has become increasingly paramount. There are growing opportunities to apply advanced computational techniques like natural language processing (NLP) and deep learning to enhance pharmacovigilance and ADR surveillance. This scoping review comprehensively examines the existing literature on harnessing NLP and deep learning for pharmacovigilance and ADE detection from EHR narratives. Following PRISMA-ScR guidelines, several studies were included after systematic screening. This review highlights the transformative impact of NLP in enabling rapid, routine ADE detection for real-time safety monitoring. However, barriers related to EHR data sharing and complexity in establishing causality persist. Pre-trained NLP models like Clinical BERT shows promise for multi-site ADE detection. Future directions involve hybrid techniques and transfers learning approaches to detect ADEs from evolving clinical terminology. Ultimately, advancing NLP methodologies for EHR-based pharmacovigilance promises to strengthen medication safety practices and improve patient outcomes.