Evaluating the OpenAI’s GPT-3.5 Turbo’s performance in extracting information from scientific articles on diabetic retinopathy
Open accessThe DOI matches its Crossref record; 10 of the 10 references it cites were found in a registry or an index.
Abstract
We aimed to compare the concordance of information extracted and the time taken between a large language model (OpenAI’s GPT-3.5 Turbo via API) against conventional human extraction methods in retrieving information from scientific articles on diabetic retinopathy (DR). The extraction was done using GPT3.5 Turbo as of October 2023. OpenAI’s GPT-3.5 Turbo significantly reduced the time taken for extraction. Concordance was highest at 100% for the extraction of the country of study, 64.7% for significant risk factors of DR, 47.1% for exclusion and inclusion criteria, and lastly 41.2% for odds ratio (OR) and 95% confidence interval (CI). The concordance levels seemed to indicate the complexity associated with each prompt. This suggests that OpenAI’s GPT-3.5 Turbo may be adopted to extract simple information that is easily located in the text, leaving more complex information to be extracted by the researcher. It is crucial to note that the foundation model is constantly improving significantly with new versions being released quickly. Subsequent work can focus on retrieval-augmented generation (RAG), embedding, chunking PDF into useful sections, and prompting to improve the accuracy of extraction.
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.
Integrity signals
Registry
Cites
Statistics
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.
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?
The cited source doesn’t support the claim as written.
In the citing paper
“Moreover, several studies have relied on LLMs to assess their own outputs (12, 14, 21–24), raising concerns about reliability and bias.”
From Performance of large language models and prompt engineering strategies for data extraction in systematic reviews (2026) · cites this paper as [21]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Earlier reports have shown that LLMs can markedly reduce data extraction time compared with manual approaches (17, 21, 24).”
From Performance of large language models and prompt engineering strategies for data extraction in systematic reviews (2026) · cites this paper as [21]
Experimental — accuracy not yet measured
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
Cites
Clinical guidelines: Potential benefits, limitations, and harms of clinical guidelines
Woolf, S. H., Grol, R., Hutchinson, A., Eccles, M., & Grimshaw, J. (1999) · BMJ · doi:10.1136/bmj.318.7182.527 · Cited by 2,548 · cited 2×
“Additionally, misinformed clinical guidelines could potentially encourage ineffective, harmful, or wasteful interventions [9].”
BackgroundMarshall, I. J., & Wallace, B. C. (2019) · Systematic Reviews · doi:10.1186/s13643-019-1074-9 · Cited by 638 · cited 1×
“Systematic reviews require high accuracy in meth-ods, which may be difficult for AI to attain [6].”
BackgroundDetecting Hallucinated Content in Conditional Neural Sequence GenerationPreprint
Zhou, C., Neubig, G., Gu, J., Diab, M. T., Guzmán, F. (., Zettlemoyer, L., & Ghazvininejad, M. (2020) · Findings · doi:10.48550/arxiv.2011.02593 · Cited by 242 · cited 1×
“One of the challenges of accuracy is the occurrence of potential hallucination, where it produces information that may sound plausible but are either factually incorrect or unrelated to the given context [10].”
BackgroundThe problems with systematic reviews: a living systematic review
Uttley, L., Quintana, D. S., Montgomery, P., Carroll, C., Page, M. J., Falzon, L., Sutton, A., & Moher, D. (2023) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2023.01.011 · Cited by 184 · cited 2×
“Consequently, poorly conducted systematic reviews can potentially misinform future clinical guidelines and policies [8].”
BackgroundDetecting Hallucinated Content in Conditional Neural Sequence Generation
Zhou, C., Neubig, G., Gu, J., Diab, M., Guzmán, F., Zettlemoyer, L., & Ghazvininejad, M. (2021) · Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021 · doi:10.18653/v1/2021.findings-acl.120 · Cited by 122
Mancin, S., Sguanci, M. E., Andreoli, D., Soekeland, F., Anastasi, G., Piredda, M., & Marinis, M. G. D. (2023) · MethodsX · doi:10.1016/j.mex.2023.102532 · Cited by 107 · cited 1×
“It is critical for policy and clinical guidelines to be based on the best available evidence [5].”
BackgroundResource use during systematic review production varies widely: a scoping review
Nussbaumer-Streit, B., Ellen, M., Klerings, I., Sfetcu, R., Riva, N., Mahmić-Kaknjo, M., Poulentzas, G., Martinez, P., Baladia, E., Ziganshina, L. E., et al. (2021) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2021.05.019 · Cited by 98 · cited 1×
“Depending on the experience of the researcher and the number of studies selected, conducting a systematic review can take up to 2 years [7].”
BackgroundThe Policy Role of Systematic Reviews: Past, Present and Future
Haddaway, N. R., & Pullin, A. S. (2014) · Springer Science Reviews · doi:10.1007/s40362-014-0023-1 · Cited by 81 · cited 1×
“Systematic reviews are often regarded as the gold standard for evidence synthesis [8], appraising the latest evidence transparently and objectively by employing established standards [2].”
BackgroundA systematic review and meta-analysis on ChatGPT and its utilization in medical and dental research
Bagde, H., Dhopte, A., Alam, M. K., & Basri, R. (2023) · Heliyon · doi:10.1016/j.heliyon.2023.e23050 · Cited by 66 · cited 3×
“The results reported different accuracy ranges of ChatGPT, from 18.3 to 100% [1].”
BackgroundInfluentialEnhancing title and abstract screening for systematic reviews with GPT-3.5 turbo
Gargari, O. K., Mahmoudi, M. H., Hajisafarali, M., & Samiee, R. (2023) · BMJ Evidence-Based Medicine · doi:10.1136/bmjebm-2023-112678 · Cited by 46 · cited 1×
“In that study, the authors reported that precise and detailed prompts yielded more accurate responses [4].”
Background
Cited by
Scherbakov, D., Hubig, N., Jansari, V., Bakumenko, A., & Lenert, L. A. (2024) · arXiv · doi:10.48550/arxiv.2409.04600 · Cited by 156
Scherbakov, D., Hubig, N., Jansari, V., Bakumenko, A., & Lenert, L. A. (2025) · Journal of the American Medical Informatics Association · doi:10.1093/jamia/ocaf063 · Cited by 154
Enhancing medical AI with retrieval-augmented generation: A mini narrative review
Gargari, O. K., & Habibi, G. (2025) · DIGITAL HEALTH · doi:10.1177/20552076251337177 · Cited by 98
Evaluating Large Language Models in Ophthalmology: Systematic Review
Zhang, Z., Zhang, H., Pan, Z., Bi, Z., Wan, Y., Song, X., & Fan, X. (2025) · Journal of Medical Internet Research · doi:10.2196/76947 · Cited by 49
Ferroptosis: a novel mechanism of cell death in ophthalmic conditions
Yang, Y., Lin, Y., Han, Z., Wang, B., Zheng, W., & Wei, L. (2024) · Frontiers in Immunology · doi:10.3389/fimmu.2024.1440309 · Cited by 46
Sercombe, J., Bryant, Z., & Wilson, J. (2025) · JMIR Formative Research · doi:10.2196/68666 · Cited by 16
Rashid, M., Yi, C. S., Sathapanasiri, T., Udayachalerm, S., Boonpattharatthiti, K., Insuk, S., Veettil, S. K., Lai, N. M., Chaiyakunapruk, N., Dhippayom, T., et al. (2025) · Value in Health · doi:10.1016/j.jval.2025.07.001 · Cited by 26
Hilkenmeier, F., Pelzer, M., Stierle, C. M. G., & Fink‐Lamotte, J. (2025) · Social Science Computer Review · doi:10.1177/08944393251404052 · Cited by 22
How to Write Effective Prompts for Screening Biomedical Literature Using Large Language Models
Colangelo, M. T., Guizzardi, S., Meleti, M., Calciolari, E., & Galli, C. (2025) · BioMedInformatics · doi:10.3390/biomedinformatics5010015 · Cited by 14
“5, improving coherence and reducing susceptibility to generating off-topic or erroneous statements [36].”
Creating a biomedical knowledge base by addressing GPT inaccurate responses and benchmarking context
Darnell, S. S., Overall, R. W., Guarracino, A., Colonna, V., Villani, F., Garrison, E. P., Isaac, A., Muli, P., Muriuki, M. F., Kabui, A., et al. (2025) · Open Research Africa · doi:10.12688/openresafrica.15854.1 · Cited by 6
Topic Modeling as a Tool to Identify Research Diversity: A Study Across Dental Disciplines
Colangelo, M. T., Guizzardi, S., & Galli, C. (2024) · Metrics · doi:10.3390/metrics1010003 · Cited by 7
“To get a better representation of the topics, we then relied on OpenAI’s GPT 3.5 turbo to generate labels for the topics from the keywords [34].”
InfluentialRubinstein, M., Grant, S., Griffin, B. A., Pessar, S. C., & Stein, B. D. (2025) · Cochrane Evidence Synthesis and Methods · doi:10.1002/cesm.70031 · Cited by 4
“A comparison of OpenAI's GPT ‐ 3.5 Turbo with human data collection from primary studies found the LLM significantly reduced the time to collect simple information in primary study text [6].”
InfluentialOami, T., Okada, Y., Maeda, K., & Nakada, T.-A. (2026) · Frontiers in Digital Health · doi:10.3389/fdgth.2026.1799623 · Cited by 4
“Earlier reports have shown that LLMs can markedly reduce data extraction time compared with manual approaches (17, 21, 24).”
Shereefdeen, H., Thaivalappil, A., Young, I., & MacKay, M. (2025) · INQUIRY The Journal of Health Care Organization Provision and Financing · doi:10.1177/00469580251399374 · Cited by 3
Rogić, S., Mancarci, B. O., Xu, B., Xiao, A., Yan, C., & Pavlidis, P. (2026) · Database · doi:10.1093/database/baag041 · Cited by 4
“Unlike approaches that rely primarily on string similarity and predefined lexicons, LLMs use contextual cues to disambiguate entities.”
Renedo, D. B., Koo, A. B., Amllay, A., Sujijantarat, N., Fathima, B., Haynes, J. O., Falcone, G. J., Sheth, K. N., Fadel, S. A., Johnson, M. H., et al. (2025) · Journal of NeuroInterventional Surgery · doi:10.1136/jnis-2025-023956 · Cited by 2
Autonomous Infrastructure Healing: A Multi-Agent Kubernetes Recovery Framework
Joyce, J. E., Nagaraj, A. K., Soni, S., Shaik, U. A. N. S., Patil, V., & Tripathi, A. (2025) · Supercomputing India (SCI) · doi:10.1109/sci68648.2025.11333843 · Cited by 2
“CrewAI agents uses OpenAI’s GPT 3.5 Turbo model [21] which is a powerful language generation system.”
InfluentialWang, Z., Fang, Y., & Li, Q. (2026) · Journal of Translational Medicine · doi:10.1186/s12967-026-08211-0 · Cited by 4
“However, more complex information tasks still need to be handled by human researchers [79].”
Hilkenmeier, F., Stoltenberg, M., & Stierle, C. M. G. (2026) · Systematic Reviews · doi:10.1186/s13643-026-03228-4 · Cited by 8
Renedo, D. B., Rivier, C. A., Acosta, J. N., Koo, A. B., Sujijantarat, N., Hebert, R. M., Falcone, G. J., & Matouk, C. C. (2026) · Surgery · doi:10.1016/j.surg.2026.110144 · Cited by 1
Fleurence, R., Qureshi, R., Aggarwal, R., Bian, J., Cheyne, S., Dawoud, D., Faden, R., & Chhatwal, J. (2026) · Value in Health · doi:10.1016/j.jval.2026.06.012 · Cited by 8
Sercombe, J., Bryant, Z., & Wilson, J. (2024) · doi:10.2196/preprints.68666
Darnell, S. S., Overall, R. W., Guarracino, A., Colonna, V., Garrison, E. P., Isaac, A., Lisso, F., Yousefi, S., Ashbrook, D. G., Huang, P.-S., et al. (2024) · Qeios · doi:10.32388/1amker
Mutin, J., Hafizin, M.-L., El-Jammal, V., Morel, C., Leca, L., Bergeron, M., & Bussières, J. (2025) · Le Pharmacien Clinicien · doi:10.1016/j.phacli.2025.07.006
Jansen, T., Liebenow, L. W., Schaller, N.-J., Hattie, J., & Møller, J. (2026) · Educational Psychology Review · doi:10.1007/s10648-026-10136-5 · Cited by 4
Decoding the evolution: artificial intelligence-powered survey of trends in bone regeneration
Colangelo, M. T., & Galli, C. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-32950-0.00010-7
Pavlidis, P., Mancarci, B. O., Maximo, A., Yan, C., & Schwartz, R. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.07.30.741874
“Other previous work with LLMs showed promising results for a range of data annotation and information extraction tasks (Gue et al. 2024; Mahmoudi et al. 2024; Chen et al. 2025; Jensen et al. 2025; Kainer 2025; Poretsky et al. 2025; Riquelme-García et al. 2025; Turner et al. 2025).”
Borty, S. C., Bhowmick, T., Hasan, M. M., Ramos, M. D., & Sakib, M. N. (2026) · International Conference on Intelligent Multimedia, Networking, and Security (IMNS) · doi:10.1109/imns67862.2026.11655253
“5-Turbo [19] model, which combines them with the user query to generate contextually coherent responses.”
Shatskiy, A. S., Deigheidy, E. M., Masyutina, S. E., & Mamalyga, M. L. (2026) · Complex Issues of Cardiovascular Diseases · doi:10.17802/2306-1278-2026-15-4-153-163
Adam, G. P., Davies, M. C., Caputo, E. L., Mai, H. J., Coppola, E. L., Holmer, H. K., Kuhn, E., Balk, E. M., Trikalinos, T. A., Ivlev, I., et al. (2026) · Research Synthesis Methods · doi:10.1017/rsm.2026.10118
Mapping Generative AI, Consumer Behavior, and Digital Trust in Global Marketing
Petroccione, G., & Asady, S. (2026) · Journal of Global Marketing · doi:10.1080/08911762.2026.2747193
Pietrancosta, M., Wang, S., Wu, Y., Brulotte, V., Harutyunyan, R., Hemmerling, T. M., Morisson, L., & Laferrière-Langlois, P. (2026) · Research Square · doi:10.21203/rs.3.rs-10879359/v1
Metadata
- Authors
- Gue, C. C. Y., Rahim, N. D. A., Rojas-Carabali, W., Agrawal, R., RK, P., Abisheganaden, J., & Yip, W. F.
- Affiliations
- National Healthcare Group (SG)
- Nanyang Technological University (SG)
- Corresponding author
- Yip, Wan Fen
- Venue
- Systematic Reviews
- Published
- 2024-05-16
- Volume
- 13
- Issue
- 1
- Pages
- 135-135
- ISSN
- 2046-4053 (ISSN-L)
- PMC
- PMC11097400
- OpenAlex
- W4396959345
- Cited by
- 31
- References
- 9
- Citations by year
- 2026: 14 · 2025: 13 · 2024: 4
- FWCI
- 26.99
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Systematic Reviews (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Social Sciences › General Social Sciences › Computational and Text Analysis Methods
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
- Keywords
- GPT-3.5-turbo, large language models, automatic information extraction, diabetic retinopathy, scientific articles, risk factors, study inclusion and exclusion criteria, prompt engineering, retrieval-augmented generation
- MeSH terms
- Diabetic Retinopathy, Humans, Natural Language Processing, Information Storage and Retrieval, Data Mining
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
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.
