aceEvaluation of large language models for clinical decision support in prostate cancer: a multidisciplinary Delphi study
Aim: Large language models (LLMs) may help interpret clinical guideline content. However, their performance in guideline-based prostate cancer (PC) scenarios remains insufficiently validated. This study evaluated the agreement between a multidisciplinary expert panel and responses generated by three leading LLMs (Claude, Gemini and ChatGPT) when addressing structured, guideline-derived clinical questions on PC. Materials & methods: A multidisciplinary tumor board created 25 structured questions based on the 2025…
Gemini's answers reached expert consensus on 23 of 25 guideline-based prostate cancer questions (92%) and ChatGPT's on 20 of 25 (80%), when given plain-language versions of the guidelines.
Claude reached expert consensus on only 14 of 25 prostate cancer questions (56%), drew explicit disagreement on three, and all models struggled with complex cases, advanced imaging, genetic testing and controversial topics.
Consensus difficulties clustered in complex cases, advanced imaging, genetic testing and questions with limited data or controversial topics, and performance varied by model, so the findings support only routine situations with expert oversight.
Evidence
- Peer-reviewedJournal of Comparative Effectiveness Research2026-11-04
How should this claim be treated?
Truvace Impact Record TRV-2026-2368, v1: “Evaluation of large language models for clinical decision support in prostate cancer: a multidisciplinary Delphi study.” Truvace, 2026-11-04. /record/TRV-2026-2368 (accessed at citation time). sha256 f5439070ca93ade0…
Calibration history
Every change to this record since certification, in the open. None yet — the reading has held since it entered the record.
Certified into the record
How to verify without trusting this page
Fetch the canonical text of any version from /api/record/TRV-2026-2368 and hash it yourself — for example shasum -a 256 on the saved canonical field. The result must equal content_hash, and each version’s text ends with prev:followed by the prior version’s hash (version 1 chains to 64 zeros). If a single character of any version had been altered since certification, the chain would not reproduce.