An auditable and source-verified framework for clinical AI decision support: integrating retrieval-augmented...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Trustworthy AI

Background:

• Clinical decision support (CDS) systems using artificial intelligence (AI) show potential but face limited adoption due to transparency and accountability concerns.
• Generative and data-driven AI often lack clear explanations for their recommendations, hindering clinical trust and verification.
• Existing AI in healthcare struggles with demonstrating the evidentiary basis for its outputs.

Purpose of the Study:

• To present a conceptual framework for auditable and source-verified AI-based clinical decision support.
• To enhance transparency, verifiability, and accountability in AI-driven healthcare recommendations.
• To improve clinician trust and regulatory readiness for AI in clinical settings.

Main Methods:

• Integration of a curated medical knowledge base with provenance metadata.
• Utilizing a retrieval-augmented reasoning (RAG) engine to link recommendations to guidelines and peer-reviewed sources.
• Implementation of a tamper-evident audit logging mechanism for recording system processes and evidence.

Main Results:

• The article presents a synthesized system design, not a new algorithm or prototype.
• The proposed framework aims to improve traceability of AI recommendations by linking them to verifiable sources.
• It addresses key challenges including knowledge governance, citation fidelity, bias, usability, privacy, and regulatory alignment.

Conclusions:

• The conceptual framework provides a pathway for developing trustworthy, auditable AI clinical decision support.
• Addressing feasibility challenges is crucial for successful implementation and validation.
• A staged evaluation roadmap involving simulations and user research is proposed for future empirical validation.