Agentic AI-based Framework for Mitigating Premature Diagnostic Handoff and Silent Hallucination in Healthcare Applications

This article presents a multi-agent framework designed to mitigate premature diagnostic handoff and silent clinical hallucinations in healthcare applications, utilizing a neuro-symbolic state-tracking gate and an epistemic uncertainty quantification gate. The framework, evaluated with the llama-3.1-70b-instruct model on 150 test cases, achieved a diagnostic precision of 49.3%, improving by 11.3 percentage points over a baseline. This work is significant for practitioners as it enhances the reliability of LLMs in medical reasoning by enforcing structured data collection and reducing uncertainty in diagnostic outputs.