EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Artificial intelligence-driven study selection in systematic reviews of randomized controlled trials, emulated trials and economic evaluation studies using large language models

KI-gestützte Studiena uswahl in systematischen Reviews zu randomisierten kontrollierten Studien, emulierten Studien und gesundheitsökonomischen Evaluationsstudien mit Large Language Models

Panu Looareesuwan, Wanchana Ponthongmak iD, Amarit Tansawet, Cholatid Ratanatharathorn, Habib Ur Rehman Owasi, Gareth J McKay iD, John Attia iD, Takehiko Oami, Ammarin Thakkinstian

PLOS Digital Health · 28.08.2026 · 5 (8) · e0001668

DOI: 10.1371/journal.pdig.0001668 ↗

Auf einen Blick

Fünf systematische Reviews wurden darauf geprüft, wie gut LLMs die Studienauswahl automatisieren. Llama 3.1 und Gemini 2.0 Flash erreichten hohe Trefferquoten bei deutlicher Arbeitsersparnis.

Fragestellung

Wie gut können Large Language Models die Studienauswahl in systematischen Reviews automatisieren?

Methode

Fünf SR-Projekte; Vergleich von Gemini 2.0 Flash, Llama 3.1 und Qwen 2.5 mit menschlicher Auswahl; Bewertung von Recall, NNS und Arbeitsreduktion in Titel/Abstract- und Volltextphase.

Ergebnisse

In der Titel/Abstract-Phase: Llama 3.1 Recall 90,00%-100,00 %, Gemini 2.0 Flash 90,48%-100,00 %, Qwen 2.5 76,67%-88,89 %. In der Volltextphase: Llama 3.1 93,33%-100,00 %, Gemini 2.0 Flash 83,33%-100,00 %, Qwen 2.5 66,67%-89,71 %.

Einschränkungen

Probleme bei Reproduzierbarkeit geschlossener Modelle und eingeschränkte Generalisierbarkeit auf verschiedene SR-Themen wurden genannt.

Schlussfolgerung der Autor:innen

LLMs, besonders Llama 3.1 und Gemini 2.0 Flash, können die Screening-Arbeit in SRs deutlich reduzieren und bei einem Recall-first-Ansatz hohe Trefferquoten halten.

Kernaussagen

  • Llama 3.1 und Gemini 2.0 Flash mit hohem Recall in beiden Screening-Phasen
  • Qwen 2.5 mit größter Arbeitsreduktion, aber niedrigerem Recall
  • Gemini 2.0 Flash war deutlich schneller als Llama 3.1 und Qwen 2.5

Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.

Kurz gefasst

Fachartikel in der Zeitschrift PLOS Digital Health, online erschienen am 28.08.2026. Verfasst von Panu Looareesuwan und 8 weiteren Autor:innen. Beteiligte Einrichtung: Department of Clinical Epidemiology and Biostatistics, Faculty of Medicine Ramathibodi Hospital, Mahidol University, Bangkok, Thailand. Die Arbeit nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Faculty of Medicine Ramathibodi Hospital, Mahidol University, National Research Council of Thailand.

Auf einen BlickAbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Systematic reviews (SRs) are key to evidence-based medicine but are often labor-intensive, especially in the study selection step. This study assessed the use of large language models (LLMs) to automate SR study screening and selection. Five SR projects were included: two published therapeutic SRs (SR1–2), two ongoing emulated-trial SRs (SR3–4), and one economic evaluation SR (SR5). The total number of studies screened for each SR was 3,966, 3,147, 695, 3,096 and 485, respectively, with 20, 24, 46, 32 and 70 eligible studies. Three LLMs—Gemini 2.0 Flash, Llama 3.1, and Qwen 2.5—were evaluated using training sets (five studies), title/abstract datasets, and full-text datasets predicted as relevant. Prompts based on the PICOS framework were iteratively refined using a recall-first strategy. Outputs were compared with human reviewer classifications using recall, number needed to screen (NNS), and percentage reduced workload with 95% confidence intervals. In the title/abstract screening phase, Llama 3.1 and Gemini 2.0 Flash achieved consistently high recall (90.00%-100.00% and 90.48%-100.00%), with workload reduction of 61.92%-97.10% and 65.21%-97.03%, respectively. Qwen 2.5 achieved the highest workload reduction (76.16%-99.19%) but showed the lowest recall (76.67%-88.89%). In the full-text selection phase, Llama 3.1 achieved the highest recall (93.33%-100.00%) with workload reductions of 73.15%-97.41%, but slower processing time (approximately 2.3-3.6 minutes per document). Qwen 2.5 yielded lower recall (66.67%-89.71%), despite the highest workload reduction (80.82%-99.44%) and similarly slow inference times (approximately 3.0-4.2 minutes per document). Gemini 2.0 Flash balanced high recall (83.33%-100.00%) with substantial workload reduction (76.71%-98.91%) and markedly faster inference (approximately 4–8 seconds per document). LLMs—particularly Llama 3.1 and Gemini 2.0 Flash—can substantially reduce SR screening workload while maintaining high recall when guided by a recall-first prompting framework. Remaining challenges include reproducibility in closed-source models and generalizability across diverse SR topics.

Autor:innen und Institutionen

  • Department of Clinical Epidemiology and Biostatistics, Faculty of Medicine Ramathibodi Hospital, Mahidol University, Bangkok, Thailand.
  • Faculty of Medicine, Research Affairs, Chulalongkorn University, Bangkok, Thailand.
  • Centre for Public Health, Queen's University Belfast, Belfast, United Kingdom.
  • Centre for Clinical Epidemiology and Biostatistics, School of Medicine and Public Health, University of Newcastle, Newcastle, NSW, Australia.
  • Department of Emergency and Critical Care Medicine, Chiba University Graduate School of Medicine, Chiba, Japan.

ORCID-Profile: Wanchana Ponthongmak, Gareth J McKay, John Attia

Förderung

  • Faculty of Medicine Ramathibodi Hospital, Mahidol University
  • National Research Council of Thailand

Zitieren

APA

Looareesuwan, P., Ponthongmak, W., Tansawet, A., Ratanatharathorn, C., Owasi, H. U. R., McKay, G. J., Attia, J., Oami, T., & Thakkinstian, A. (2026). Artificial intelligence-driven study selection in systematic reviews of randomized controlled trials, emulated trials and economic evaluation studies using large language models. PLOS Digital Health, 5(8), e0001668. https://doi.org/10.1371/journal.pdig.0001668

Vancouver

Looareesuwan P, Ponthongmak W, Tansawet A, Ratanatharathorn C, Owasi HUR, McKay GJ, et al. Artificial intelligence-driven study selection in systematic reviews of randomized controlled trials, emulated trials and economic evaluation studies using large language models. PLOS Digital Health. 2026;5(8):e0001668. doi:10.1371/journal.pdig.0001668

BibTeX
@article{Looareesuwan2026,
  author = {Looareesuwan, Panu and Ponthongmak, Wanchana and Tansawet, Amarit and Ratanatharathorn, Cholatid and Owasi, Habib Ur Rehman and McKay, Gareth J and Attia, John and Oami, Takehiko and Thakkinstian, Ammarin},
  title = {Artificial intelligence-driven study selection in systematic reviews of randomized controlled trials, emulated trials and economic evaluation studies using large language models},
  journal = {PLOS Digital Health},
  year = {2026},
  volume = {5},
  number = {8},
  pages = {e0001668},
  doi = {10.1371/journal.pdig.0001668}
}
Kostenlos starten →