CoChat

ChatGPT-4o can serve as the second rater for data extraction in systematic reviews

Open access
Motzfeldt Jensen, Mette, Brix Danielsen, Mathias, Riis, Johannes, Assifuah Kristjansen, Karoline, Andersen, Stig, Okubo, Yoshiro, Jørgensen, Martin Grønbech · 2025 · PLOS ONE

The DOI matches its Crossref record; the references it cites are not checked yet.

Abstract

Background: Systematic reviews provide clarity of a bulk of evidence and support the transfer of knowledge from clinical trials to guidelines. Yet, they are time-consuming. Artificial intelligence (AI), like ChatGPT-4o, may streamline processes of data extraction, but its efficacy requires validation. Objective: This study aims to (1) evaluate the validity of ChatGPT-4o for data extraction compared to human reviewers, and (2) test the reproducibility of ChatGPT-4o’s data extraction. Methods: We conducted a comparative study using papers from an ongoing systematic review on exercise to reduce fall risk. Data extracted by ChatGPT-4o were compared to a reference standard: data extracted by two independent human reviewers. The validity was assessed by categorizing the extracted data into five categories ranging from completely correct to false data. Reproducibility was evaluated by comparing data extracted in two separate sessions using different ChatGPT-4o accounts. Results: ChatGPT-4o extracted a total of 484 data points across 11 papers. The AI’s data extraction was 92.4% accurate (95% CI: 89.5% to 94.5%) and produced false data in 5.2% of cases (95% CI: 3.4% to 7.4%). The reproducibility between the two sessions was high, with an overall agreement of 94.1%. Reproducibility decreased when information was not reported in the papers, with an agreement of 77.2%. Conclusion: Validity and reproducibility of ChatGPT-4o was high for data extraction for systematic reviews. ChatGPT-4o was qualified as a second reviewer for systematic reviews and showed potential for future advancements when summarizing data.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pone.0313401

Cites

Statistics

Figures

Fig 1 Overview of the study design. This study (orange) originates from a source review (blue). Articles included in the source review were used for data extraction with ChatGPT-4o. The ’Consensus on extracted data’ from the source review was considered as the reference standard, against which ChatGPT-4o’s responses were tested for validity and reproducibility (black).
Fig 1 Overview of the study design. This study (orange) originates from a source review (blue). Articles included in the source review were used for data extraction with ChatGPT-4o. The ’Consensus on extracted data’ from the source review was considered as the reference standard, against which ChatGPT-4o’s responses were tested for validity and reproducibility (black).CC BY · PubMed Central
Fig 2 ChatGPT-4o response assessment. Evaluation of ChatGPT-4o responses compared to the reference standard. Response categories included ’Completely correct,’ ’Satisfactory,’ ’Lacking information,’ ’Missing all information,’ and ’False information’.
Fig 2 ChatGPT-4o response assessment. Evaluation of ChatGPT-4o responses compared to the reference standard. Response categories included ’Completely correct,’ ’Satisfactory,’ ’Lacking information,’ ’Missing all information,’ and ’False information’.CC BY · PubMed Central
Fig 3 Stratification of ChatGPT-4o responses by outcome reporting status. The proportion of ChatGPT-4o responses in each response category stratified based on whether the articles reported the outcome or not.
Fig 3 Stratification of ChatGPT-4o responses by outcome reporting status. The proportion of ChatGPT-4o responses in each response category stratified based on whether the articles reported the outcome or not.CC BY · PubMed Central
Fig 4 Proportion of correct and satisfactory ChatGPT-4o responses across four data domains. The proportion of ’Completely correct’ and ’Satisfactory’ responses from ChatGPT-4o compared to the reference standard across four data domains: General information, Population, Intervention and comparator, and Outcome.
Fig 4 Proportion of correct and satisfactory ChatGPT-4o responses across four data domains. The proportion of ’Completely correct’ and ’Satisfactory’ responses from ChatGPT-4o compared to the reference standard across four data domains: General information, Population, Intervention and comparator, and Outcome.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

The citation lists are queued

They are fetched later and appear here then.

References 16 · Cited by 60

Research works printed but not linked: 1

  1. Ildiko-Zsuzsa Szabo, 2022, The Effects of Perturbation-based Balance Training on Daily-life and Laboratory Falls in Community-dwelling: A Systematic Review and Meta-Analysis., PROSPERO

Metadata

Authors
Motzfeldt Jensen, M., Brix Danielsen, M., Riis, J., Assifuah Kristjansen, K., Andersen, S., Okubo, Y., & Jørgensen, M. G.
Affiliations
  • Aalborg University Hospital (DK)
  • UNSW Sydney (AU)
Venue
PLOS ONE
Published
2025-01-07
Volume
20
Issue
1
Pages
e0313401
ISSN
1932-6203 (ISSN-L)
OpenAlex
W4406140367
Cited by
60
References
16
Citations by year
2026: 37 · 2025: 23
FWCI
11.37
Type
journal article
Language
en
Open access
Locations
  • PLoS ONE (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • VBN Forskningsportal (Aalborg Universitet) (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Social Sciences › General Social Sciences › Computational and Text Analysis Methods
Keywords
GPT-4o, artificial intelligence, data extraction, systematic reviews, reproducibility, extraction accuracy, clinical trial reporting
MeSH terms
Systematic Reviews as Topic, Artificial Intelligence, Humans, Reproducibility of Results

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: ChatGPT-4o can serve as the second rater for data extraction in systematic reviews. https://app.cochat.ai/papers/10.1371/journal.pone.0313401
The paper
Motzfeldt Jensen, M., Brix Danielsen, M., Riis, J., Assifuah Kristjansen, K., Andersen, S., Okubo, Y., & Jørgensen, M. G. (2025). ChatGPT-4o can serve as the second rater for data extraction in systematic reviews. PLOS ONE. https://doi.org/10.1371/journal.pone.0313401

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.