Large language models detect methodological errors that human peer review misses in clinical trials
- 1University of Southern California
Abstract
Peer review is the principal quality control of published research, but methodological errors can go undetected. In an earlier experiment with errors inserted into three reports of randomized clinical trials, untrained human reviewers achieved a mean major-error detection score of 27%. Here, we tested whether 15 large language models (LLMs), including four open-weight models, detect the same errors. Their major-error detection scores ranged from 53% to 96%. To reduce the risk of memorization, we repeated the test on three clinical trials published in 2026. Scores ranged from 51% to 99%, and from 66% to 87% for the open-weight models. Notably, LLMs still flagged 2% to 29% of the error types in control reports without inserted errors, most often conclusions not supported by the results and missing reasons why screened patients were not enrolled. Across the 12 models, mean review times ranged from 1.4 to 16.3 minutes. Opus 5.5 achieved the highest major-error detection score (99%), while GPT-6.1 Sol and GPT-6 Luna offered cost-effective alternatives with scores of 92% and 87%, respectively. Together, these tests provide a benchmark for specialized review models and suggest that relying on human peer review alone may leave avoidable methodological errors undetected.
Related
- Large language models in rheumatology: applications, methodological quality and clinical translation – a systematic review
- Methodological quality and clinical applicability of long COVID guidelines and scientific statements: a systematic review using AGREE II and AGREE-REX
- Errors, Hallucinations, and Clinical Impact of General-Purpose Multimodal Large Language Models in Histopathology
- Calibrated Reliance on Large Language Models for Personalized Surgical Care: A Critical Narrative Review of Confidence, Clinical Reasoning and Decision Support
- Post-Treatment Surveillance to Detect Recurrent Disease in Esophagogastric Cancer: A Review of Current Practice and Future Perspectives
BibTeX
@misc{brian2026large,
title = {{Large language models detect methodological errors that human peer review misses in clinical trials}},
author = {D. Brian and X. I. Pereira Hernandez and H. Yin and J. Koury and R. Rust},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.25.26366455},
url = {https://www.medrxiv.org/content/10.64898/2026.10.25.26366455v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Large language models detect methodological errors that human peer review misses in clinical trials
AU - D. Brian
AU - X. I. Pereira Hernandez
AU - H. Yin
AU - J. Koury
AU - R. Rust
PY - 2026
DA - 2026/10/25
PB - medRxiv
DO - 10.64898/2026.10.25.26366455
UR - https://www.medrxiv.org/content/10.64898/2026.10.25.26366455v1
AB - Peer review is the principal quality control of published research, but methodological errors can go undetected. In an earlier experiment with errors inserted into three reports of randomized clinical trials, untrained human reviewers achieved a mean major-error detection score of 27%. Here, we tested whether 15 large language models (LLMs), including four open-weight models, detect the same errors. Their major-error detection scores ranged from 53% to 96%. To reduce the risk of memorization, we repeated the test on three clinical trials published in 2026. Scores ranged from 51% to 99%, and from 66% to 87% for the open-weight models. Notably, LLMs still flagged 2% to 29% of the error types in control reports without inserted errors, most often conclusions not supported by the results and missing reasons why screened patients were not enrolled. Across the 12 models, mean review times ranged from 1.4 to 16.3 minutes. Opus 5.5 achieved the highest major-error detection score (99%), while GPT-6.1 Sol and GPT-6 Luna offered cost-effective alternatives with scores of 92% and 87%, respectively. Together, these tests provide a benchmark for specialized review models and suggest that relying on human peer review alone may leave avoidable methodological errors undetected.
ER -