MedGPT Beats 250 Senior Doctors 88:12 on Complex Lung Cancer Cases, Scoring 85 vs 53

新智元 · wechat · 2026-11-12

A multicenter study with top Chinese hospital thoracic surgery departments tested MedGPT against doctors on 40 previously unpublished complex lung cancer cases, graded by 20 experts under 240 pre-set rules including 40 safety red lines.

250 senior doctors averaged 53.26 without assistance; MedGPT scored 85.29
Head-to-head on 750 treatment plans, the AI won 88:12, with answers 40% shorter
Doctors using MedGPT improved from 53.4 to 66.27 (+24.1%)

Ablations show the gap comes from process, not the base model: stripping the system's workflow dropped scores nearly 24 points; removing just the multidisciplinary consultation module cost 16 points.

The system shatters literature into minimal one-conclusion evidence units (200K structured units for lung cancer alone, 8M+ total), maintains dynamic patient state tables, and runs multi-specialty agents with disagreement preserved, dual standard/frontier pathways, virtual-patient simulation, and five-dimension adversarial review before output. Its moat: the curated evidence base, patient records, and outcome-feedback loops — not the underlying model.

Original post →

MedGPT Beats 250 Senior Doctors 88:12 on Complex Lung Cancer Cases, Scoring 85 vs 53

More from Apps