Medical AI Communication Gap Benchmark Explorer

Investigating the ~60% diagnostic accuracy drop when patients communicate symptoms vs structured physicians (RCT Analysis)

RCT BENCHMARK: GPT-4o · LLAMA 3 · COMMAND R+
Select Clinical Case Preset Specialty: Cardiology
Input Framing & Prompt Intervention:
Physician-Structured Intake Transcript HIGH STRUCTURE
62yo M with hx of DM2 presenting with acute onset retrosternal pressure radiating to left jaw, diaphoresis, dyspnea on exertion. Denies pleuritic pain, fever, or prior CAD.
92.0%
GPT-4o Diagnostic Acc.
95.0%
Appropriate Triage
0.12
Ambiguity Index
Diagnostic Pathway Divergence Top LLM Diagnostic Hypotheses
Clinical Pathway: Direct identification of ischemic etiology; prioritized immediate ECG and troponin triage.
2024 Model Benchmark Performance Matrix Randomized Control Study Results
Foundation Model Physician Intake Acc. Patient Narrative Acc. Accuracy Degradation Triage Drop Mitigated Layer Acc.
-59.8%
Overall Diagnostic Drop
-11.9%
Overall Triage Degradation
+34.2%
Prompt Layer Recovery Gain
Enjoy this tool? Build your own with Super