Tencent Health AI Clinical Assistant Diagnostic Support

Problem

When clinical notes, symptoms, and lab results are mixed together, clinicians often need to determine the most likely diagnosis, identify conditions to exclude, find missing information, and recognize urgent findings that require immediate action. This skill organizes that process into a fixed diagnostic reasoning workflow rather than free-form chat.

How It Works

The skill parses input as unstructured clinical text or JSON structured data. It then selects an output path by priority:
- If fast_track_triggers match an urgent condition, it outputs only urgent care guidance and skips routine diagnostic analysis.
- If no current diagnosis exists and the top suspected diagnosis has high confidence, it outputs a suspected diagnosis report.
- If a current diagnosis exists, it outputs differential diagnosis and misdiagnosis/missed diagnosis risk assessment.
- If confidence is medium or low, it forces a missing-information report instead of speculative diagnosis.

The reasoning uses hypothesis-deduction: generate 3-7 initial diagnostic hypotheses, rank common diseases first, and match each hypothesis against supporting, contradictory, and unconfirmed evidence. Every conclusion must pass three validation gates:
1. R01-R08 rule checks, including evidence traceability, contradiction detection, red-flag omission, minimum differential count, and confidence gating.
2. Model cross-validation for diagnosis-evidence consistency, differential completeness, and risk fit.
3. Final completeness check to ensure traceable evidence, safety disclaimer, and template format.

Boundaries

This skill supports clinical decision-making and does not replace a licensed physician's final diagnosis. It does not provide treatment plans, medication adjustments, drug interaction lookup, patient-facing advice, or standardized medical record drafting. All diagnostic evidence must come from the user's input, with no speculation. Low-confidence cases avoid suspected-diagnosis output and instead request missing tests or history. When urgent triggers are hit, it prioritizes immediate actions and time windows over lengthy differentials.