Skip to content
DraconiticResearch
Search Sources API
190,851 studies31 sourcesUpdated 50 min ago
Peer-reviewedCC BYNature Medicine

Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial

  • Ambrose Agweyu1,2,3
  • Paul Mwaniki2,3
  • Vaishnavi Menon4
  • Robert Korom5
  • Lynda Isaaka2,3
  • Conrad W. Wanyama2,3
  • Jaspret Gill4
  • Sarah Kiptinness5
  • Najib Adan5
  • Mira Emmanuel-Fabula1,6,7
  • Richard D Riley4,8
  • Lucinda Archer4,8
3 more authors
  • Alastair Keith Denniston4,8
  • Xiaoxuan Liu4
  • Bilal Akhter Mateen9,10
10 affiliations
  1. 1Department of Infectious Disease Epidemiology and International Health, London School of Hygiene and Tropical Medicine, London, UK
  2. 2Kenya Medical Research Institute-Wellcome Trust Research Programme, Nairobi, Kenya
  3. 3Keprecon, Nairobi, Kenya
  4. 4Department of Applied Health Sciences, School of Health Sciences, College of Medicine and Health, University of Birmingham, Birmingham, UK
  5. 5Penda Health, Nairobi, Kenya
  6. 6Foundations for Appropriate Technologies in Health, Geneva, Switzerland
  7. 7PATH, London, UK
  8. 8National Institute for Health and Care Research Birmingham Biomedical Research Centre, Birmingham, UK
  9. 9Department of Applied Health Sciences, School of Health Sciences, College of Medicine and Health, University of Birmingham, Birmingham, UK. [email protected]
  10. 10PATH, London, UK. [email protected]
Open at sourcePDFDOICite

Abstract

Rigorous evidence on the performance of large language models (LLMs) in real-world, low-resource clinical settings remains limited. Here we conducted a pragmatic, cluster-randomized trial in 16 primary care facilities in Kenya. Clinical officers were randomized to use the electronic medical record with or without LLM assistance. The primary outcome was an expert-adjudicated composite of treatment failure events experienced within 14 days of enrollment. Between 22 April and 16 July 2025, 9,691 patients were enrolled, overseen by 103 clinical officers (52 in the LLM-assisted arm and 51 in the control arm). Treatment failure occurred in 102/4,693 patients (2.2%) in the intervention arm and 94/4,654 (2.0%) in the control arm (adjusted odds ratio 0.77, 95% confidence interval 0.55 to 1.08, P = 0.13). The primary outcome did not differ significantly between groups. No serious adverse events were judged related to the intervention, and independent review of the adverse events did not identify a safety signal. In this trial, LLM assistance was safe but did not reduce treatment failure within 14 days and any benefit, if present, is probably modest.Pan-African Clinical Trials Registry: 202502499779176.

Related

  1. A pragmatic evaluation of outcome fragility in randomized clinical trials comparing open microdiscectomy and endoscopic-assisted discectomy: A meta-analysis
    Michael D. Brown, Alqasim Elnaggar, Obed Barkus +5
    North American Spine Society journalPeer-reviewedCC BY-NC-NDFull textSurgery & anaesthesiaEurope PMC
  2. A Pragmatic Randomized Trial of an EHR-Integrated Generative AI Chart Summarization Tool for Ambulatory Clinicians
    A. T. Chin, N. Zhu, S. Vangala +5
    medRxivPreprintCC BY-NCFull textEpidemiology & public healthmedRxiv
  3. SURMOUNT‐REAL UK: A Pragmatic Randomized Clinical Trial to Assess the Effectiveness of Tirzepatide in Adults With Obesity
    Martin K. Rutter, Julie Mount, J. Martin Gibson +12
    Obesity (Silver Spring, Md.)Peer-reviewedCC BY 4.0Full textPharmacology & toxicologyPMC
  4. Collaborative Care and Everyday Planning in Care Counselling (CoPlan study): study protocol of a pragmatic randomized clinical trial and protocol changes to quasi-experimental evaluation during recruitment
    Ingrid Hendlmeier, Martina Schäufele, Petra Meuer +3
    TrialsPeer-reviewedCC BY 4.0Full textClinical trials & evidence synthesisCrossref
  5. Reclaiming human clinical reasoning in generative AI-enabled healthcare: a perspective on cognitive agency, skill evolution, and accountability
    Jia Xu, Mengjiao Zhang, Bing Zheng +2
    Frontiers in Public HealthPeer-reviewedCC BY 4.0Full textEpidemiology & public healthCrossrefPMC

Cite

BibTeXRIS
BibTeX
@article{agweyu2026generative,
  title = {{Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial}},
  author = {Ambrose Agweyu and Paul Mwaniki and Vaishnavi Menon and Robert Korom and Lynda Isaaka and Conrad W. Wanyama and Jaspret Gill and Sarah Kiptinness and Najib Adan and Mira Emmanuel-Fabula and Richard D Riley and Lucinda Archer and Alastair Keith Denniston and Xiaoxuan Liu and Bilal Akhter Mateen},
  journal = {Nature Medicine},
  year = {2026},
  month = jun,
  doi = {10.1038/s41591-026-04503-6},
  url = {https://doi.org/10.1038/s41591-026-04503-6},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial
AU  - Ambrose Agweyu
AU  - Paul Mwaniki
AU  - Vaishnavi Menon
AU  - Robert Korom
AU  - Lynda Isaaka
AU  - Conrad W. Wanyama
AU  - Jaspret Gill
AU  - Sarah Kiptinness
AU  - Najib Adan
AU  - Mira Emmanuel-Fabula
AU  - Richard D Riley
AU  - Lucinda Archer
AU  - Alastair Keith Denniston
AU  - Xiaoxuan Liu
AU  - Bilal Akhter Mateen
PY  - 2026
DA  - 2026/06/26
JO  - Nature Medicine
DO  - 10.1038/s41591-026-04503-6
UR  - https://doi.org/10.1038/s41591-026-04503-6
AB  - Rigorous evidence on the performance of large language models (LLMs) in real-world, low-resource clinical settings remains limited. Here we conducted a pragmatic, cluster-randomized trial in 16 primary care facilities in Kenya. Clinical officers were randomized to use the electronic medical record with or without LLM assistance. The primary outcome was an expert-adjudicated composite of treatment failure events experienced within 14 days of enrollment. Between 22 April and 16 July 2025, 9,691 patients were enrolled, overseen by 103 clinical officers (52 in the LLM-assisted arm and 51 in the control arm). Treatment failure occurred in 102/4,693 patients (2.2%) in the intervention arm and 94/4,654 (2.0%) in the control arm (adjusted odds ratio 0.77, 95% confidence interval 0.55 to 1.08, P = 0.13). The primary outcome did not differ significantly between groups. No serious adverse events were judged related to the intervention, and independent review of the adverse events did not identify a safety signal. In this trial, LLM assistance was safe but did not reduce treatment failure within 14 days and any benefit, if present, is probably modest.Pan-African Clinical Trials Registry: 202502499779176.
SN  - 1078-8956
SN  - 1546-170X
SN  - 1744-7933
ER  -