The DOI matches its Crossref record; 85 of the 85 references it cites were found in a registry or an index; 35 of the 51 sources its claims cite were checked for support.
Automation of Systematic Reviews with Large Language Models
Open accessAbstract
Abstract Importance Systematic reviews (SRs) inform evidence-based decision making. Yet, many take over a year to complete, are labor intensive, prone to human error, and face reproducibility challenges; thus limiting access to timely and reliable information. Objective To validate a large language model (LLM)-based workflow (otto-SR) to automate three of the most labour intensive tasks in performing SR’s: article screening, data extraction, and risk of bias assessment; and to assess its feasibility in rapidly updating existing reviews. Design, setting, and participants We conducted a validation study in four phases, with direct benchmarking against graduate-level human researchers in phases 1 and 2. Phase 1: article screening performance was measured across 32,357 citations from 5 systematic reviews. The reference standard consisted of the original reviews’ screening decisions after full-text screening. Phase 2: data extraction performance was measured across 4,495 data points from 495 studies in 7 reviews. Phase 3: risk of bias assessment (ROB2, Newcastle-Ottawa, QUADAS2) performance was measured across 345 studies from 12 reviews. Reference standards for Phase 2 and Phase 3 were created after blinded adjudication of the original review extraction and RoB assessments. Phase 4: otto-SR was used to reproduce and update the primary analysis from an issue of Cochrane reviews (n=12 reviews, 146,276 citations), with analytical comparisons to the original meta-analyzed findings. All discrepancies underwent dual human review. Results otto-SR showed high performance in phase 1 article screening ( otto-SR : 96.7% sensitivity, 97.9% specificity; human: 81.7% sensitivity, 98.1% specificity) and phase 2 data extraction ( otto-SR : 93.1% accuracy; human: 79.7% accuracy). In phase 3, otto-SR demonstrated high interrater reliability for risk of bias judgements (ROB2 0.98, Newcastle-Ottawa 0.95, QUADAS2 0.74; Gwet AC2). In phase 4, otto-SR , reproduced and updated the primary analysis from an issue of Cochrane reviews. Across Cochrane reviews, otto-SR incorrectly excluded a median of 0 studies (IQR 0 to 0.25), and found nearly twice as many eligible studies compared to the original authors (n= 114 vs. 64). Meta-analyses based on otto-SR generated screening and extraction outputs, subsequently verified through dual human review, yielded newly statistically significant effect estimates in 2 reviews and negated significance in 1 review. Conclusions and relevance LLMs have high performance in article screening, data extraction, and risk of bias assessments. They can rapidly reproduce and update existing systematic reviews, laying the foundation for automated, scalable, and reliable evidence synthesis.
Overview
CoChat Summary
## TL;DR
This paper presents otto-SR, an end-to-end LLM workflow for automating systematic review screening, data extraction, and update/reproduction tasks. Across multiple benchmarks and a 12-review Cochrane reproducibility exercise, the system outperformed human workflows on screening and extraction and reproduced/updated reviews in two days.
## Key findings
- On screening across five reviews (32,357 citations), otto-SR achieved 96.7% sensitivity and 97.9% specificity in abstract screening, versus human dual screening at 81.7% sensitivity and 98.1% specificity; in full-text screening, otto-SR reached 97.1% sensitivity and 98.7% specificity, while dual human reviewers were at 81.3% sensitivity and 96.8% specificity in the sampled comparison.
- On data extraction across seven reviews (4,559 data points, 495 studies), otto-SR achieved 93.1% accuracy, compared with 79.7% for dual human reviewers and 74.8% for Elicit.
- In the Cochrane reproducibility assessment, otto-SR reproduced and updated 12 reviews from the April 2024 issue in under two days; across these reviews it incorrectly excluded a median of 0 studies (IQR 0 to 0.25) and found a median of 2.0 eligible studies (IQR 1 to 6.5) likely missed by the original authors.
- In the updated/meta-analytic comparisons, otto-SR generated newly statistically significant conclusions in 2 reviews and negated significance in 1 review.
## Methods
- The study built otto-SR as an end-to-end workflow using GPT-4.1 for abstract/full-text screening, Gemini 2.0 Flash for PDF-to-Markdown conversion, and o3-mini-high for data extraction.
- Screening benchmarks used datasets from BenchSR and related sources; the authors evaluated otto-SR on the full citation sets (32,357 citations across five reviews) and compared it with dual human reviewers and Elicit against original author final inclusion decisions.
- Data extraction benchmarks used seven reviews (including SeroTracker, PA-Outcomes, PA-Testing, Sepsis, CKD, Process, and Psyc-meds), with performance assessed using an LLM-as-a-judge framework and blinded human adjudication to create corrected ground truth.
- For the Cochrane reproduction, the authors randomly selected the April 2024 issue, excluded 2 of 14 reviews for missing public data or irreproducible search strategy, and reproduced searches to May 8, 2025; analyses focused on each review’s primary analytical comparison and used meta-analysis in R with the metafor package.
## Limitations
- The paper states that generalizability to other clinical questions and qualitative reviews remains to be tested.
- It states that the LLM parser may have incorrectly extracted information from PDF articles, and that the workflow was limited to main-text content rather than supplementary tables or figures.
- It notes that the LLM-as-a-judge framework may introduce occasional errors, and that when otto-SR and original authors agreed, the authors assumed the value was correct without further adjudication; evident from the text, this could miss shared systematic errors.
Model-written summary of a shortened version of the paper’s full text (it is longer than we read). Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
6 Valid · 9 Not valid · 5 Not checked · 207 Own results and interpretations
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“However, given the known variability in dual human data extraction accuracy (reported rates: 65.8-85.5%) 14-19 , original author-extracted values were not treated as a definitive gold standard”
What the sources say
- Could not verifycites [14] sourceExperimental — accuracy not yet measured
The Data extraction and quality assessment section (page unavailable) says one reviewer extracted data and a second verified them, but gives no accuracy rate for dual human extraction. Searches for the reported 65.8–85.5% range and extraction accuracy found no supporting passage. The decision not to treat original author-extracted values as a gold standard appears to be the citing authors’ own methodological decision. (LLM, conf=0.73)
- Could not verifycites [15] sourceExperimental — accuracy not yet measured
The Data analysis section (p. 3) says that one reviewer extracted data and a second verified them, but it gives no accuracy rate for human data extraction. Searches for the reported 65.8–85.5% range and for treatment of author-extracted values as a gold standard found no supporting passage in our copy. The decision not to treat those values as a gold standard appears to be the citing authors’ own methodological choice. (LLM, conf=0.74)
- Could not verifycites [16] 2024, Time to Benefit of Surgery vs Targeted Medical Therapy for Patients With Primary Aldosteronism: A Meta-analysis, J Clin Endocrinol Metab sourceExperimental — accuracy not yet measured
The Data Extraction and Study Quality Assessment section says two reviewers extracted data in duplicate, but it does not report extraction accuracy. Searches for the 65.8–85.5% rates and for reviewer agreement found no passage addressing those figures. The decision not to treat author-extracted values as a gold standard is the citing authors’ own conclusion, not a finding attributed to this source. (LLM, conf=0.77)
- Could not verifycites [17] sourceExperimental — accuracy not yet measured
The Methods section (Data Extraction and Quality Assessment; page unavailable) says data extraction was performed in duplicate, but gives no accuracy rate. Searches for 65.8–85.5%, extraction accuracy, and whether author-extracted values were treated as a gold standard found no passage addressing those claims. The Discussion’s reference to a gold standard concerns diagnosing primary aldosteronism, not data extraction. (LLM, conf=0.74)
- Not matchedcites [18] Mascarenhas D , Weisz D , Jasani B , Persad N , Main E. Premedication for rapid sequence intubation in neonates - a network meta-analysis. PROSPERO 2022 CRD42022384259 [Internet]. PROSPERO Available from: https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42022384259
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Could not verifycites [19] 2023, A real-world evaluation of the implementation of NLP technology in abstract screening of a systematic review, Res Synth Methods sourceExperimental — accuracy not yet measured
Section 4.1 (Limitations; page unavailable) says there is no definitive gold standard for deciding whether to include abstracts, but it does not address author-extracted data values. Searches for dual human data-extraction accuracy, the 65.8–85.5% range, and original author-extracted values found no supporting passage in our copy. (LLM, conf=0.76)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Delays in completing SRs can have major consequences for evidence-based practice, including prolonged use of ineffective or harmful treatments initially supported by less rigorous evidence 3 .”
What the sources say
- Partly supportedcites [3] Michelson M , Reuter K . The significant cost of systematic reviews and meta-analyses: A call for greater involvement of machine learning to assess the promise of clinical trials. Contemp Clin Trials Commun 2019;16:100443. sourceExperimental — accuracy not yet measured
The Introduction (page not available) says systematic reviews are time-consuming and can evaluate evidence about treatment efficacy. The Conclusion (page not available) says their cost may impede consistent use to assess clinical trials. Neither passage says that delays prolong the use of ineffective or harmful treatments initially supported by less rigorous evidence. (LLM, conf=0.82)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“We previously demonstrated that LLMs can achieve high screening performance 6 .”
What the sources say
- Partly supportedcites [6] 2024, How to optimize the systematic review process using AI tools, JCPP Adv sourceExperimental — accuracy not yet measured
The Introduction to AI says LLM-based tools can assist with systematic-review screening, but it does not report a demonstration of high screening performance. In Methods & results, the paper says the accuracy of two screening tools has not yet been validated; its example of validated AI screening does not establish high performance for LLMs. (LLM, conf=0.83)
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“While several tools have been developed to accelerate SRs 4,5 , none are capable of full automation with human-level accuracy.”
What the sources say
- Not matchedcites [4] 2011, Medical reversal: why we must raise the bar before adopting new technologies, Yale J Biol Med
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Could not verifycites [5] sourceExperimental — accuracy not yet measured
The paper describes techniques for conducting cumulative meta-analyses (Introduction, p. 1) and says they can simplify digesting clinical-trial literature (Future Applications, p. 6). Neither passage assesses tools for accelerating systematic reviews or their capacity for full automation with human-level accuracy. Searches for those terms and for automated review software found no passage addressing that claim. (LLM, conf=0.73)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, we addressed this by performing randomized and blinded adjudication, adopting best practices seen in radiology, ophthalmology, and clinical trials 28-31 .”
What the sources say
- Partly supportedcites [28] 2013, A case study of binary outcome data extraction across three systematic reviews of hip arthroplasty: errors and differences of selection, BMC Res Notes sourceExperimental — accuracy not yet measured
The Discussion recommends double data extraction or independent verification to reduce errors and reveal data-selection choices. It does not describe randomized, blinded adjudication or identify that procedure as a best practice in radiology, ophthalmology, or clinical trials. The citing authors’ statement that they performed adjudication is their own account, not a finding this source can verify. (LLM, conf=0.85)
- Partly supportedcites [29] Tang L , Wang R , Doi SAR , Furuya-Kanamori L , Lin L , Qin Z , et al. Double data extraction was insufficient for minimizing errors in evidence synthesis: a randomized controlled trial [Internet]. 2023 [cited 2025 Jun 8];Available from: http://medrxiv.org/lookup/doi/10.1101/2023.10.16.23297056 sourceExperimental — accuracy not yet measured
In Trial design and setting (page unavailable), the paper describes randomly pairing participants to cross-check extracted data and resolve disagreements by consensus; it also says investigators were blinded to the trial’s randomization sequence. It does not describe randomized, blinded adjudication of original authors’ extraction decisions. The cited procedure therefore supports only part of the attributed claim. (LLM, conf=0.82)
- Supportedcites [30] Xu C , Yu T , Furuya-Kanamori L , Lin L , Zorzela L , Zhou X , et al. Validity of data extraction in evidence synthesis practice of adverse events: reproducibility study. BMJ 2022;e069155. source
The Results report data extraction errors in the systematic reviews examined, supporting the concern about inaccurate original extraction decisions. The randomized, blinded adjudication is described as the citing authors’ own response, not as a procedure attributed to this source. (LLM, conf=0.88)
- Partly supportedcites [31] sourceExperimental — accuracy not yet measured
The Introduction and §2.3 describe adjudication of data-extraction disagreements and randomization of abstractor pairs to different approaches. But §2.4 states that the abstractors were not masked, so this paper does not support describing its adjudication as blinded. The cited paper also does not establish the citing authors’ own procedures. (LLM, conf=0.88)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
122 more claims · 7 with their cited sources checked
Create a free account to see all 227 claims and every source verdict.
- Probably not a substantial claim. 71
- It may state two separate claims as one. 39
- The quote only partly states this claim. 20
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ 5 An agentic workflow of LLMs can rapidly reproduce and update reviews
“After filtering our results to align with the original search cutoffs, we identified 54 additional eligible studies through otto-SR (median 2, IQR: 1 to 6.25 per review) that were likely missed in the original Cochrane reviews (Methods).”
§ 5 An agentic workflow of LLMs can rapidly reproduce and update reviews
“Updating the search to May 8, 2025 identified another 14 new eligible studies (total n = 64, median 2.5, IQR 1 to 7.25 per review) ( Extended Data Table 8 ).”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4 LLMs achieve state-of-the-art SR data extraction performance
“We evaluated the performance of the otto-SR extraction agent and Elicit in data extraction across seven reviews (n=4,559 data points, 495 studies) ( Fig. 3A , Extended Data Table 5 ).”
§ 4 LLMs achieve state-of-the-art SR data extraction performance
“Figure 3: otto-SR extraction agent (o3-mini-high) performance on systematic review data extraction. A. Bar graph displaying data extraction accuracy of the otto-SR extraction agent (green) (4,459 data points), Elicit (teal) (4,459 data points), and dual human reviewers (grey) (1,453 data points) across 7 different systematic reviews.”
Citations
Cites
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · BMJ · doi:10.1136/bmj.n71 · Cited by 108,283
Self-citationRoB 2: a revised tool for assessing risk of bias in randomised trials
Sterne, J. A. C., Savović, J., Page, M. J., Elbers, R. G., Blencowe, N. S., Boutron, I., Cates, C. J., Cheng, H.-Y., Corbett, M. S., Eldridge, S. M., et al. (2019) · BMJ · doi:10.1136/bmj.l4898 · Cited by 34,787
Cumpston, M., Li, T., Page, M. J., Chandler, J., Welch, V. A., Higgins, J. P., & Thomas, J. (2019) · Cochrane Database of Systematic Reviews · doi:10.1002/14651858.ed000142 · Cited by 7,218
Probable Inference, the Law of Succession, and Statistical Inference
Wilson, E. B. (1927) · Journal of the American Statistical Association · doi:10.1080/01621459.1927.10502953 · Cited by 4,476
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2021.03.001 · Cited by 4,265
Self-citationJudging LLM-as-a-Judge with MT-Bench and Chatbot Arena
Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., et al. (2023) · arXiv · doi:10.48550/arxiv.2306.05685 · Cited by 1,888
The Mass Production of Redundant, Misleading, and Conflicted Systematic Reviews and Meta‐analyses
IOANNIDIS, J. P. A. (2016) · The Milbank Quarterly · doi:10.1111/1468-0009.12210 · Cited by 1,483
Meta-Analysis in Clinical Research
L'ABBÉ, K. A., DETSKY, A. S., & O'ROURKE, K. (1987) · Annals of Internal Medicine · doi:10.7326/0003-4819-107-2-224 · Cited by 1,267
Self-citationCumulative Meta-Analysis of Therapeutic Trials for Myocardial Infarction
Lau, J., Antman, E. M., Jimenez-Silva, J., Kupelnick, B., Mosteller, F., & Chalmers, T. C. (1992) · New England Journal of Medicine · doi:10.1056/nejm199207233270406 · Cited by 1,140
Borah, R., Brown, A. W., Capers, P. L., & Kaiser, K. A. (2017) · BMJ Open · doi:10.1136/bmjopen-2016-012545 · Cited by 737
Research works printed but not linked: 1
- 10.1059/0003-4819-155-8-201110180-00009 · doi:10.1059/0003-4819-155-8-201110180-00009
Cited by
Quaquebeke, N. V., Tonidandel, S., & Banks, G. C. (2025) · The Leadership Quarterly · doi:10.1016/j.leaqua.2025.101895 · Cited by 57
Gartlehner, G., Nußbaumer-Streit, B., Hamel, C., Garritty, C., Griebler, U., King, V. J., Devane, D., & Kamel, C. (2025) · Cochrane Evidence Synthesis and Methods · doi:10.1002/cesm.70063 · Cited by 14
Global thermal tolerance compilation for freshwater invertebrates and fish
Bayat, H. S., He, F., Madariaga, G. M., Escobar-Sierra, C., Prati, S., Peters, K., Jupke, J. F., Spaak, J. W., Manfrin, A., Juvigny‐Khenafou, N. P. D., et al. (2025) · Scientific Data · doi:10.1038/s41597-025-05832-w · Cited by 16
“Advances in automated data extraction 122–125 foretell the automation of data synthesis research, which can enhance efforts to update synthetic work amidst ever-increasing output.”
Benefits and Risks of Using AI Agents in Research
Hosseini, M., Murad, M., & Resnik, D. B. (2026) · The Hastings Center Report · doi:10.1002/hast.70025 · Cited by 14
Nguyen, D., Najafabadi, B. T., Lin, L., Kwong, J. C. C., Miller, E., Zlotta, A. R., Fleshner, N. E., Hird, A. E., Lofters, A., Wallis, C. J. D., et al. (2026) · European Urology · doi:10.1016/j.eururo.2026.04.005 · Cited by 7
Metadata
- Authors
- Cao, C., Arora, R., Cento, P., Budak, A., Manta, K., Farahani, E., Cecere, M., Selemon, A., Sang, J., Gong, L. X., Kloosterman, R., Jiang, S., Saleh, R., Margalik, D., Lin, J., Jomy, J., Xie, J., Chen, D., Gorla, J., Lee, S., Zhang, K., Kuang, J., Ware, H., Whelan, M., Teja, B., Leung, A. A., Arora, R. K., Pillay, J., Hartling, L., Noetel, M., Emerson, D. B., Detsky, A. S., Tricco, A. C., Church, G. M., Moher, D., & Bobrovitz, N.
- Affiliations
- University of Toronto (CA)
- Harvard University (US)
- Independent researcher
- McGill University (CA)
- University of British Columbia (CA)
- McGovern Institute for Brain Research, MIT, Cambridge, MA, USA
- University of Waterloo (CA)
- Mount Sinai Hospital (CA)
- Mount Sinai Hospital (US)
- St. Michael's Hospital (CA)
- University of Calgary (CA)
- University of Alberta (CA)
- The University of Queensland (AU)
- University of Ottawa (CA)
- Corresponding author
- Cao, Christian
- ORCID
- Cao, Christian
- Arora, Rohit
- Cento, Paul
- Manta, Katherine
- Farahani, Elina
- Cecere, Matthew
- Selemon, Anabel
- Gong, Ling Xi
- Kloosterman, Robert
- Saleh, Richard
- Margalik, Denis
- Lin, James
- Jomy, Jane
- Xie, Jerry
- Chen, David
- Gorla, Jaswanth
- Lee, Sylvia
- Zhang, Kelvin
- Ware, Harriet
- Whelan, Mairead
- Teja, Bijan
- Leung, Alexander A.
- Arora, Rahul K.
- Noetel, Michael
- Emerson, David B.
- Church, George M.
- Moher, David
- Bobrovitz, Niklas
- Venue
- medRxiv
- Publisher
- openRxiv
- Published
- 2025-06-13
- OpenAlex
- W4411259709
- Cited by
- 33
- References
- 35
- Citations by year
- 2026: 26 · 2025: 7
- Type
- posted content
- Language
- en
- Open access
- green
- doi.org
- Licence
- Locations
- medRxiv (acceptedVersion, cc-by)
- 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
- systematic reviews, large language models, article screening, data extraction, risk of bias assessment, review updating, reproducibility, AI-assisted research, Newcastle-Ottawa Scale, QUADAS-2, inter-rater reliability, meta-analysis reproducibility, evidence-based decision making, human-AI collaboration
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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: Automation of Systematic Reviews with Large Language Models. https://app.cochat.ai/papers/10.1101/2025.06.13.25329541
- The paper
- Cao, C., Arora, R., Cento, P., Budak, A., Manta, K., Farahani, E., Cecere, M., Selemon, A., Sang, J., Gong, L. X., Kloosterman, R., Jiang, S., Saleh, R., Margalik, D., Lin, J., Jomy, J., Xie, J., Chen, D., Gorla, J., . . . Bobrovitz, N. (2025). Automation of Systematic Reviews with Large Language Models. medRxiv. https://doi.org/10.1101/2025.06.13.25329541
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.