Generating guideline-concordant and safe recommendations for diabetic kidney disease management via a hierarchical retrieval-augmented large language model
10 affiliations
- 1Department of Pathology, The First Affiliated Hospital, Fujian Medical University, Fuzhou, China
- 2Department of Pathology, National Regional Medical Center, Binhai Campus of the First Affiliated Hospital, Fujian Medical University, Fuzhou, China
- 3Institute of Molecular Medicine, Renji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China
- 4Department of Neurosurgery, Shanghai Tenth People’s Hospital, School of Medicine, Tongji University, Shanghai, China
- 5School of Basic Medical Sciences, Tsinghua University, Beijing, China
- 6Zhangzhou Affiliated Hospital of Fujian Medical University, Zhangzhou, China
- 7Fujian Medical University Provincial Clinical Medical College, Fuzhou, China
- 8The Second Affiliated Hospital of Fujian Medical University, Quanzhou, China
- 9Fuzong Clinical Medical College of Fujian Medical University, Fuzhou, China
- 10Department of Endocrinology, Sanya Central Hospital, The Third People’s, Hospital of Hainan Province, Sanya, China
Abstract
Managing diabetic kidney disease (DKD) is inherently complex, requiring clinicians to synthesize patient history, fluctuating biomarkers, and evolving treatment guidelines. While large language models (LLMs) show promise in medical decision support, their clinical adoption is hindered by factual inaccuracies and a lack of specific reasoning required for individualized patient management. To address this, we developed a hierarchical multi-agent system that integrates a locally deployed retrieval-augmented generation (RAG) framework with a cloud-based advanced reasoning engine, grounding responses in a curated corpus of clinical guidelines. We conducted a multi-center retrospective validation using 267 patient cases. The system’s performance was evaluated against baseline models through a blinded review by twelve independent physicians across clinical dimensions including accuracy, safety, and factuality. Our evaluation reveals that the RAG-enhanced system significantly outperforms unaugmented models in providing accurate, guideline-compliant recommendations. Notably, it substantially reduced safety-critical errors, particularly in identifying medication contraindications related to renal function stages, while achieving high inter-rater reliability. This study demonstrates that anchoring LLMs with authoritative knowledge effectively mitigates hallucination risks and enhances clinical reliability. The proposed framework functions as a reliable on-demand assistant for DKD management, providing guideline-grounded decision support for primary care providers.
Related
- Towards global clinical practice guidelines for the management of non-specific low back pain in primary care: a review of current guideline recommendations and how they have changed over the last 30 years
- Dyslipidemia management in athletes: A comparative commentary on the 2026 ACC/AHA/Multisociety Guideline and contemporary sports cardiology recommendations.
- PROSPECT guideline for total hip arthroplasty: updated systematic review and procedure‐specific postoperative pain management recommendations
- Receipt of guideline-concordant prenatal care among Medicaid-insured women with disabilities in Texas
- Association of Demographic, Economic, and Geographic Factors With Guideline-Concordant Infection Prophylaxis in Patients With Antineutrophil Cytoplasmic Antibody-Associated Vasculitis.
BibTeX
@article{tao2026generating,
title = {{Generating guideline-concordant and safe recommendations for diabetic kidney disease management via a hierarchical retrieval-augmented large language model}},
author = {Xuan Tao and Lan Tian and Chenhao Fang and Jin-hao Li and Jia-qi Lin and Xiao-jie Liu and Xin-ying Song and Zi-hang Chen and Yi-nan Wu and Yan-hong Lin and Ze-yang Zhu and Min-xia Wang and Wan-xuan Chen and Jian-ya Xu and Hong Chen},
journal = {NPJ Digital Medicine},
year = {2026},
month = jul,
doi = {10.1038/s41746-026-03035-0},
url = {https://doi.org/10.1038/s41746-026-03035-0},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Generating guideline-concordant and safe recommendations for diabetic kidney disease management via a hierarchical retrieval-augmented large language model
AU - Xuan Tao
AU - Lan Tian
AU - Chenhao Fang
AU - Jin-hao Li
AU - Jia-qi Lin
AU - Xiao-jie Liu
AU - Xin-ying Song
AU - Zi-hang Chen
AU - Yi-nan Wu
AU - Yan-hong Lin
AU - Ze-yang Zhu
AU - Min-xia Wang
AU - Wan-xuan Chen
AU - Jian-ya Xu
AU - Hong Chen
PY - 2026
DA - 2026/07/20
JO - NPJ Digital Medicine
DO - 10.1038/s41746-026-03035-0
UR - https://doi.org/10.1038/s41746-026-03035-0
AB - Managing diabetic kidney disease (DKD) is inherently complex, requiring clinicians to synthesize patient history, fluctuating biomarkers, and evolving treatment guidelines. While large language models (LLMs) show promise in medical decision support, their clinical adoption is hindered by factual inaccuracies and a lack of specific reasoning required for individualized patient management. To address this, we developed a hierarchical multi-agent system that integrates a locally deployed retrieval-augmented generation (RAG) framework with a cloud-based advanced reasoning engine, grounding responses in a curated corpus of clinical guidelines. We conducted a multi-center retrospective validation using 267 patient cases. The system’s performance was evaluated against baseline models through a blinded review by twelve independent physicians across clinical dimensions including accuracy, safety, and factuality. Our evaluation reveals that the RAG-enhanced system significantly outperforms unaugmented models in providing accurate, guideline-compliant recommendations. Notably, it substantially reduced safety-critical errors, particularly in identifying medication contraindications related to renal function stages, while achieving high inter-rater reliability. This study demonstrates that anchoring LLMs with authoritative knowledge effectively mitigates hallucination risks and enhances clinical reliability. The proposed framework functions as a reliable on-demand assistant for DKD management, providing guideline-grounded decision support for primary care providers.
SN - 2398-6352
ER -