Preprint · 2026 · Open access
Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study
Published 1 January 2026 · DOI 10.17605/osf.io/qdhkj
In brief
“Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study” is a 2026 preprint by Francisco Gonzalez. OpenAlex counts 0 citations as of 19 November 2026. It is open access under a public domain licence: the full text is free to read.
- Citations
- 0
- Authors
- 1
Abstract
Background: Systematic reviews of randomized controlled trials (RCTs) represent the highest-quality evidence for informing therapeutic decisions. Risk-of-bias (RoB) assessment is an essential component of systematic reviews, yet the Cochrane RoB 2 tool — the current recommended standard — is complex, time-consuming (up to 358 minutes per article8), and shows suboptimal inter-rater reliability (as low as −0.15 without prior calibration7). Critically, even experienced Cochrane reviewers frequently misapply the tool, treating lack of blinding as automatically implying high risk of bias and applying arbitrary thresholds for missing outcome data12. Large language models (LLMs) have shown moderate agreement with human reviewers for this task15,16, but whether they meet performance thresholds for specific operational roles — low-cost adjunct reviewer, high-risk triage tool, or low-risk confirmation tool — remains unknown, and no study has directly compared two LLMs or evaluated whether PDF upload and text paste yield equivalent results. Objectives: To evaluate and compare the concordance of Claude Sonnet 4.6 (Anthropic) and ChatGPT 5.4 (OpenAI), each under two input modalities (full-text paste and PDF upload), against consensus-based RoB 2 assessments of Cochrane systematic review authors as the reference standard. Secondary objectives include characterizing diagnostic performance (sensitivity, specificity, likelihood ratios) for the three operational roles, comparing LLMs and modalities directly, assessing intra-LLM test-retest stability, and quantifying time efficiency. Methods: Pre-registered reliability and agreement study following GRRAS guidelines18. We searched the Cochrane Database of Systematic Reviews for SRs using RoB 2 published from January 2020 through 31 May 2026. One RCT will be randomly selected per eligible SR to prevent clustering. Cochrane authors' consensus-based RoB 2 judgments serve as the reference standard. A structured prompt calibration phase (12 RCTs) will precede the main study, with prompt design informed by documented misapplication patterns in Cochrane editorial practice12. Two investigators (FG, FM) will independently execute four LLM assessments per RCT (2 models × 2 modalities), each in duplicate for test-retest stability. Primary outcome is weighted Cohen's κ for overall RoB agreement. Secondary outcomes include domain-level agreement, inter-LLM and inter-modality concordance, sensitivity, specificity, and likelihood ratios for high-risk identification, and assessment time. Target sample: 350 RCTs (approximate 95% CI for κ: ±0.1019). Expected results: Weighted κ with 95% CI for each of the four LLM conditions versus the Cochrane reference, alongside sensitivity, specificity, and likelihood ratios for identifying high-risk RCTs, and mean assessment time per condition. Conclusions: This will be the most methodologically rigorous study to date on LLM performance for Cochrane-standard RoB 2 assessment, and the first direct comparison of two current LLMs under two input modalities using a…
Cite this paper
Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.
APA 7
Gonzalez, F. (2026). Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study. https://doi.org/10.17605/osf.io/qdhkj
MLA 9
Gonzalez, Francisco. “Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study.” 2026. https://doi.org/10.17605/osf.io/qdhkj.
Chicago
Gonzalez, Francisco. 2026. “Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study.” https://doi.org/10.17605/osf.io/qdhkj.
Harvard
Gonzalez, F. (2026) ‘Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study’. doi:10.17605/osf.io/qdhkj.
IEEE
F. Gonzalez, “Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study,” 2026, doi: 10.17605/osf.io/qdhkj.
BibTeX
@misc{gonzalez2026comparing,
title = {{Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study}},
author = {Gonzalez, Francisco},
year = {2026},
doi = {10.17605/osf.io/qdhkj},
url = {https://doi.org/10.17605/osf.io/qdhkj}
}
Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)
Topic and keywords
Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
- risk of bias assessment
- large language models
- randomized controlled trials
- Claude Sonnet
- ChatGPT
- inter-rater reliability
- test-retest stability
- diagnostic performance
Work with this paper
Most-cited papers in Decision Sciences
- The PRISMA 2020 statement: an updated guideline for reporting systematic reviewsMatthew James Page, Joanne E. McKenzie, Patrick M. Bossuyt and 23 more · BMJ · 2021 · 144,919 citations
- Evaluating Structural Equation Models with Unobservable Variables and Measurement ErrorClaes Fornell, David F. Larcker · Journal of Marketing Research · 1981 · 118,336 citations
- Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternativesLi‐tze Hu, Peter M. Bentler · Structural Equation Modeling A Multidisciplinary Journal · 1999 · 117,461 citations
- Common method biases in behavioral research: A critical review of the literature and recommended remedies.Philip M. Podsakoff, SCOTT B. MACKENZIE, Jeong Yeon Lee and 1 more · Journal of Applied Psychology · 2003 · 89,248 citations
- Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statementDavid Moher, Alessandro Liberati, Jennifer Tetzlaff and 2 more · BMJ · 2009 · 85,524 citations
Questions about this paper
How do I cite this paper in APA style?
Gonzalez, F. (2026). Comparing Large Language Models for Risk-of-Bias Assessment in Randomized Controlled Trials Using the Cochrane RoB 2 Tool: Protocol for a Reliability and Agreement Study. https://doi.org/10.17605/osf.io/qdhkj
How many times has this paper been cited?
0 times, according to OpenAlex on 19 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.
Is this paper free to read?
Yes. It is open access: sign in and open the full text from this page.
Who wrote this paper?
Francisco Gonzalez.
What is the DOI of this paper?
10.17605/osf.io/qdhkj, permanently at https://doi.org/10.17605/osf.io/qdhkj. Sign in and open it from this page.
Bibliographic data from OpenAlex (CC0), last checked 19 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.