AI-Powered Signal Detection in Pharmacovigilance

1. Stakeholders across industry, academia and regulatory bodies gain access to evidence-based, practical guidance on AI-driven signal detection and pharmacovigilance to ensure patient safety. 2. Patients and citizens benefit from earlier and more accurate detection of safety signals, leading to timely risk communication and mitigation. 3. AI-powered algorithms and methods enabling faster, more accurate signal detection at medical concept and syndrome levels. 4. A comprehensive catalogue of data sources suitable for AI-based signal detection, with recommendations and principles for a common data model supporting simultaneous analysis of diverse datasets. 5. AI-powered predictive methods for high-accuracy future risk identification, supporting proactive public health interventions. 6. A recommendations framework for real-world implementation of AI systems—covering technical, practical, ethical, legal and governance aspects—and EMA endorsement via the Qualification Procedure. 7. Guidance on human-in-the-loop and human-on-the-loop AI oversight for optimal performance, transparency and regulatory compliance. 8. Ready-to-use templates, tools and integration pathways for embedding AI models into existing PV infrastructures. 9. Comprehensive training, user guides and educational materials, along with a dissemination and uptake strategy. All outcomes anchored in transparency, trustworthiness and adherence to ethical and legal principles of patient-level data use and proprietary information handling.

Projects must focus exclusively on AI applications for signal detection and risk prediction in pharmacovigilance; AI-driven ICSR management and periodic reporting are out of scope. Proposals must demonstrate use of advanced digital technologies (generative AI, machine learning, predictive analytics) applied to both traditional and emerging data sources. They must pilot algorithms across multiple therapeutic areas and stakeholder use-cases, evaluating performance on criteria of accuracy, reliability/repeatability and trustworthiness. Key data sources to be evaluated include EHRs, spontaneous reporting systems (EudraVigilance, FAERS, VigiBase), social media and genomics. Proposals must develop principles for a common data model enabling concurrent analysis of heterogeneous sources. Model interpretability, data provenance and traceability of AI decisions are mandatory. Ethical and legal compliance for patient-level and proprietary data use, transparency and trustworthiness are central. Finally, projects must include a regulatory qualification strategy and proactive engagement with EMA and national authorities.