Evidence mapPaperPMID 40119675Full record

ArticleJournal of the American Medical Informatics Association : JAMIA2025

High-performance automated abstract screening with large language model ensembles.

Rohan Sanghera, Arun James Thirunavukarasu, Marc El Khoury, Jessica O'Logbon, Yuqing Chen, Archie Watt, Mustafa Mahmood, Hamid Butt, George Nishimura, Andrew A S Soltan

Abstract readValidation Study
In one paragraph

Article in Journal of the American Medical Informatics Association : JAMIA, 2025. The graph could read no effect estimate from its abstract, so it casts no vote on the map. Cited by 14 papers, 1 of them a synthesis that pooled it.

0numbers the graph read from it
0cells of the map it votes in
14citing papers in PubMed, 1 pooled it
field-weighted citation impact
1 · What the graph read from it

What it found

Each row is one number read from the abstract, on the scale the paper reported it, with its interval. Left of the dashed line favours the treatment, right favours the comparator. Under each row is the sentence it came from. New to these charts? A ten-minute tutorial.

The abstract states no effect estimate the extractor could read, or names no intervention and outcome on the map, so this paper lights no cell and moves no belief. It is still indexed, cited and linked below.

2 · The registry

The trial behind it

Trials whose registry record cites this paper, or whose number appears in the abstract. A trial that started after this paper was published is citing it as background, not reporting it.

Neither the registry nor the abstract names a trial number. If this is a trial report, that itself is worth knowing.

3 · Its place in the literature

Who cites it

14 citing papers in PubMed, 1 synthesis or guideline pooled it.

  1. Pooled it
  2. Article
  3. Article
  4. Article
  5. Article
  6. Article
  7. Article
  8. Article
  9. Article
  10. Article
  11. Review
  12. Article
  13. Article
  14. Article
4 · The record

Corrections and comments

PubMed lists nothing against this paper. Absence here is not a guarantee, only a check that was made.

5 · Who and what money

Authors and funding

10 authors.

Rohan SangheraOxford University Hospitals NHS Foundation Trust, Oxford OX3 9DU, United Kingdom.
Arun James ThirunavukarasuOxford University Hospitals NHS Foundation Trust, Oxford OX3 9DU, United Kingdom.ORCID 0000-0001-8968-4768
Marc El KhourySchool of Clinical Medicine, University of Cambridge, Cambridge CB2 0SP, United Kingdom.
Jessica O'LogbonGKT School of Medical Education, King's College London, London WC2R 2LS, United Kingdom.
Yuqing ChenSchool of Clinical Medicine, University of Cambridge, Cambridge CB2 0SP, United Kingdom.
Archie WattOxford Medical School, Medical Sciences Division, University of Oxford, Oxford OX3 9DU, United Kingdom.
Mustafa MahmoodUCL Medical School, University College London, London WC1E 6DE, United Kingdom.
Hamid ButtSchool of Clinical Medicine, University of Cambridge, Cambridge CB2 0SP, United Kingdom.
George NishimuraSchool of Clinical Medicine, University of Cambridge, Cambridge CB2 0SP, United Kingdom.
Andrew A S SoltanOxford University Hospitals NHS Foundation Trust, Oxford OX3 9DU, United Kingdom.ORCID 0000-0003-2391-5361

Funding

Department of Health and Social CareHealthSense Research FundMicrosoft Research Accelerating FoundationMicrosoft Research Accelerating Foundation Models ResearchNational Institute for Health and Care ResearchNHSNHS Foundation Trust
6 · The paper itself

Abstract

objectivescreening is a labor-intensive component of systematic review involving repetitive application of inclusion and exclusion criteria on a large volume of studies. We aimed to validate large language models (LLMs) used to automate abstract screening. MATERIALS AND

methodsLLMs (GPT-3.5 Turbo, GPT-4 Turbo, GPT-4o, Llama 3 70B, Gemini 1.5 Pro, and Claude Sonnet 3.5) were trialed across 23 Cochrane Library systematic reviews to evaluate their accuracy in zero-shot binary classification for abstract screening. Initial evaluation on a balanced development dataset (n = 800) identified optimal prompting strategies, and the best performing LLM-prompt combinations were then validated on a comprehensive dataset of replicated search results (n = 119 695).

resultsOn the development dataset, LLMs exhibited superior performance to human researchers in terms of sensitivity (LLMmax = 1.000, humanmax = 0.775), precision (LLMmax = 0.927, humanmax = 0.911), and balanced accuracy (LLMmax = 0.904, humanmax = 0.865). When evaluated on the comprehensive dataset, the best performing LLM-prompt combinations exhibited consistent sensitivity (range 0.756-1.000) but diminished precision (range 0.004-0.096) due to class imbalance. In addition, 66 LLM-human and LLM-LLM ensembles exhibited perfect sensitivity with a maximal precision of 0.458 with the development dataset, decreasing to 0.1450 over the comprehensive dataset; but conferring workload reductions ranging between 37.55% and 99.11%. DISCUSSION: Automated abstract screening can reduce the screening workload in systematic review while maintaining quality. Performance variation between reviews highlights the importance of domain-specific validation before autonomous deployment. LLM-human ensembles can achieve similar benefits while maintaining human oversight over all records.

conclusionLLMs may reduce the human labor cost of systematic review with maintained or improved accuracy, thereby increasing the efficiency and quality of evidence synthesis.

Indexed as

Abstracting and IndexingNatural Language ProcessingSystematic Reviews as TopicHumansLarge Language Modelsabstract screeningartificial intelligenceevidence synthesisfoundation modellarge language modelsystematic review

Identifiers

PMID40119675
PMCPMC12012331

What Socratic holds

Textmetadata
LicenceCC BY
Read underepoch 390

Registered trials

None linked

Read under generation 80e0d062 · epoch 390. Bibliography from PubMed, PubMed Central and OpenAlex; grants from NIH RePORTER; trial links from ClinicalTrials.gov; estimates, votes and beliefs from the Socratic graph.