Gita Kusnadi, Grace Wangge, Arif Perdana
This study characterises expert perspectives on when follow-up activities are most likely to provide actionable value. Our results indicate that follow-up improves understanding when it resolves clinical uncertainty, but is inefficient when pursued without regard to impact, supporting a shift to proportionate, purpose-driven follow-up strategies for consistent and efficient use of resources including criteria for prioritisation and de-prioritisation.
OBJECTIVE: To examine the application of artificial intelligence (AI) in pharmacovigilance across the Asia-Pacific and identify reported methodological implementation challenges.
METHODS: MEDLINE, Scopus, and Google Scholar were searched using terms related to artificial intelligence, computational signal detection, pharmacovigilance, and Asia-Pacific countries. Peer-reviewed original studies published in English were included. PRISMA 2020 guideline was followed.
RESULTS: We included 64 studies in 14 countries primarily focused on 1) Adverse Drug Reaction (ADR) identification, 2) ADR prediction and risk factor modelling, 3) Drug safety, monitoring, and evaluation, 4) Predictive modelling, and 5) Data information management. Machine Learning (ML) techniques were the most commonly applied AI methods in pharmacovigilance, followed by natural language processing, deep learning, neural networks, and symbolic and explainable AI. Disproportionality analysis methods were also commonly used across studies. Some challenges reported were relevant to data quality issues, generalizability, clinical workflow integration, implementation technicalities, and cultural barriers.
CONCLUSION: To overcome the challenges of AI application in the Asia-Pacific, a tiered implementation strategy can be employed through establishing a regional collaboration framework and taking into account disparities in technological maturity across countries.