Abstract: FR-PO0053
Artificial Intelligence (AI)-Assisted Generation and Multilingual Translation of Nephrology Patient Education Materials
Session Information
- Artificial Intelligence: Clinical Prediction, Decision Support, and Digital Kidney Care
October 23, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Bizer, Benjamin, Mayo Clinic Minnesota, Rochester, Minnesota, United States
- Arriola Montenegro, Jose J., Mayo Clinic Minnesota, Rochester, Minnesota, United States
- Thongprayoon, Charat, Mayo Clinic Minnesota, Rochester, Minnesota, United States
- Craici, Iasmina, Mayo Clinic Minnesota, Rochester, Minnesota, United States
- Cheungpasitporn, Wisit, Mayo Clinic Minnesota, Rochester, Minnesota, United States
Background
Patient-facing educational materials are an essential component of medical care, yet many exceed recommended readability levels and are unavailable in patients' native languages. Large language models may improve accessibility by generating structured multilingual education directly from clinical documentation. We aimed to evaluate the feasibility of AI-assisted generation and translation of nephrology patient education materials.
Methods
We developed a structured prompting workflow using OpenAI's GPT-5.3 to generate patient education materials from outpatient nephrology provider notes. Twenty-five clinical notes representing five common nephrology presentations were included: hypertension, chronic kidney disease, acute kidney injury, nephrolithiasis, and glomerulonephritis. The model generated patient instructions using predefined disease-specific templates while preserving formatting and clinical content. Outputs were subsequently translated into Spanish, Arabic, German and Hungarian. English-language outputs were evaluated for template adherence, formatting consistency, and content fidelity. Translated outputs were assessed for translational accuracy and structural preservation.
Results
The AI model demonstrated excellent adherence to template instructions overall. The correct template was identified in all cases, including those requiring the integration of multiple templates. Errors were identified in six cases; none were classified as critical. Minor errors primarily involved formatting issues, including font inconsistencies and inappropriate line breaks. Output reproducibility was very strong, with repeated prompts consistently generating identical responses. Translation performance was weakest for Hungarian. The most frequently identified translation issue was the use of non-idiomatic or uncommon terminology.
Conclusion
This proof-of-concept study demonstrates the feasibility of AI-assisted generation and multilingual translation of structured nephrology patient education materials from clinical documentation. Such workflows have the potential to improve accessibility of patient education and warrant further evaluation in real-world clinical settings.