Artificial intelligence in pharmacovigilance – possibilities and challenges

Abstract

Since the last decade there is an increased interest in the application of artificial intelligence (AI) in pharmacovigilance (PV), which is in a time of great change. The aim of this paper was to emphasize the progress made in this area, but also all constraints that should be addressed before the full potential of AI can be exploited. A comprehensive review of scientific papers published in the last decade was undertaken. For the purpose of this work 15 out of 80 searched papers were selected, representing relevant results of experts in PV field. AI may be useful in PV for 1) the automatic execution of tasks associated with case report processing, 2) the identification of clusters of adverse events (AE) representing symptoms of syndromes, 3) the conduction of pharmacoepidemiological studies, and 4) the prediction and prevention of AEs through specific models using real-world data (1). The technical challenges for AI-based PV are lack of high-quality databases, insufficient human resources, weak AI technology and insufficient support from governments (2). Benefits and possibilities of application of AI in PV are numerous, but its successful implementation requires understanding of the complex interactions among all components and human-computer interface. Additional challenge is that PV professionals are traditionally recruited primarily from clinical disciplines with limited trainings in computational approaches to data analysis. Therefore, the education of PV staff and targeted recruitment of AI specialists should be key activities, both in industry and regulatory agencies, to achieve successful implementation of AI in PV.