Evidence-to-decision: From exposome data to evidence to action through agentic AI

Area of Science:

• Environmental Health
• Public Health Policy
• Artificial Intelligence in Medicine

Background:

• Public health decisions involve complex trade-offs under uncertainty.
• The GRADE Evidence-to-Decision (EtD) framework aids decision-making but struggles with large-scale data.
• A Human Exposome Project generates vast, complex evidence.

Purpose of the Study:

• To propose agentic artificial intelligence (AI) for operationalizing the EtD framework.
• To enable feasible decision-making for exposome-scale data.
• To ensure AI systems maintain scientific rigor.

Main Methods:

• Agentic AI, comprising autonomous agents, can automate EtD criteria.
• Six agent families map to EtD criteria: evidence extraction, risk-of-bias, uncertainty quantification, causality, cost-outcome analysis, and validation.
• Five governance requirements ensure AI reliability: traceability, versioning, benchmarking, honest uncertainty, and accountability.

Main Results:

• Agentic AI offers a scalable solution for complex public health evidence synthesis.
• The proposed AI framework addresses the data volume and complexity of the Human Exposome Project.
• Specific AI agent families and governance principles are outlined for EtD operationalization.

Conclusions:

• Agentic AI is essential for making exposome-scale public health decisions feasible and rigorous.
• AI must inherit the scientific standards of evidence-based medicine to be trustworthy.
• Implementing these AI systems is critical for advancing public health.