New Agentic AI Framework Boosts Diagnostic Precision in Healthcare

This research introduces an innovative multi-agent AI framework specifically designed to enhance the reliability of medical reasoning in healthcare applications. The framework tackles two critical failure modes prevalent in open-ended conversational AI: premature diagnostic handoff and silent clinical hallucinations that might otherwise go undetected. Instead of relying on a single LLM to make routing decisions, this system employs deterministic orchestration constraints. The framework integrates two key safety mechanisms. First, a neuro-symbolic state-tracking gate ensures the comprehensive collection of patient information by enforcing the OLDCARTS clinical protocol (Onset, Location, Duration, Character, Aggravating/Alleviating factors, Radiation, Timing, and Severity). This gate prevents diagnostic transitions until all required data dimensions are gathered. Second, an epistemic uncertainty quantification (UQ) gate calculates semantic entropy across multiple independent diagnostic samples to identify and block inconsistent or divergent outputs before they reach a patient. Evaluated using simulated patient agents on 150 test cases, the full architecture achieved a 49.3% diagnostic precision, representing a significant 11.3 percentage point improvement over an unconstrained baseline. The study also found a statistically significant correlation between structured information gathering (OLDCARTS completeness) and reduced diagnostic uncertainty, highlighting the framework's effectiveness.