Abstract: PUB056
Reimagining Clinical Nephrology Through Retrieval-Augmented Generation: Systematic Review of Artificial Intelligence (AI) Applications
Session Information
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Vasudev, Ishita, BronxCare Health System, New York, New York, United States
- Bhardwaj, Akinchan, Jawaharlal Institute of Postgraduate Medical Education and Research, Puducherry, PY, India
Background
Retrieval-Augmented Generation (RAG) combines large language models (LLMs) with external knowledge sources to enhance the accuracy of results and reduce errors. In nephrology, RAG-based systems are being explored for clinical decision support, patient education, and research synthesis, prompting a systematic review of their applications and impact
Methods
We conducted a systematic review adhering to PRISMA guidelines, using PubMed, Google Scholar, Scopus and Embase (2015–2025) for publications on applications of RAG in nephrology, limiting our search to English-language human studies and excluding preprints. Eligible publications evaluated RAG-augmented LLMs in nephrology clinical care, patient education and research. Data on use cases and outcomes were extracted for qualitative synthesis
Results
RAG has demonstrated utility across multiple nephrology applications. For clinical decision support, large language models (LLM) demonstrated approximately 99% accuracy in identifying the need for nephrology consultation and 96–99% accuracy in determining the appropriate subspecialty referral. Retrieval assistance also improved diagnostic reasoning in complex cases. In patient education, ChatGPT-4, augmented with RAG, answered kidney disease questions with 96–98% accuracy, ensuring the delivery of reliable information. RAG-enabled guideline-aligned recommendations as well: a CKD chatbot retrieving KDIGO content provided answers consistent with current guidelines. Additionally, the use of RAG in summarizing literature and assisting academic writing suggests its potential to expedite evidence synthesis in nephrology
Conclusion
RAG-augmented AI shows considerable promise in nephrology by providing accurate, up-to-date, and guideline-aligned information for clinical care, patient education, and research. However, rigorous real-world validation is needed, and challenges around data privacy, bias, and workflow integration must be addressed with proper human oversight. With these precautions, RAG tools could become valuable adjuncts to nephrology practice, research and education