From black box to clear box: explainable AI for next-gen pharmacovigilance

Area of Science:

• Pharmacovigilance and Artificial Intelligence
• Regulatory Science
• Explainable AI (XAI)

Background:

• Traditional pharmacovigilance faces challenges like underreporting and data issues.
• Artificial Intelligence offers automation and advanced pattern recognition for drug safety.
• Explainable AI (XAI) is essential for regulatory acceptance of AI in pharmacovigilance.

Purpose of the Study:

• To review the international regulatory landscape for AI in pharmacovigilance.
• To propose XAI methods (SHAP, LIME) for building regulatory trust.
• To recommend a framework for AI in pharmacovigilance that ensures transparency and safety.

Main Methods:

• Systematic literature search of regulatory guidelines and academic databases (US, EU, India).
• Analysis of regulatory frameworks including the EU AI Act, FDA GMLP, and India's Responsible AI principles.
• Examination of XAI techniques for their applicability in pharmacovigilance.

Main Results:

• The EU AI Act designates PV algorithms as high-risk, requiring human oversight and transparency.
• Regulatory principles in the US and India also emphasize responsible AI implementation.
• SHAP and LIME are identified as key methods for achieving explainability and auditability.

Conclusions:

• A regulatory-fit XAI framework, integrating human oversight, is recommended for AI in pharmacovigilance.
• This framework aims to enhance safety signal detection and foster international trust.
• Convergence on traceability, governance, and transparency is vital for AI adoption in drug safety.