Abstract: FR-PO0034
Artificial Intelligence (AI)-Enhanced Extraction of Candida auris Testing Results from Health Information Exchange Data in a Dialysis Population
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
- Bergling, Karin, Renal Research Institute, New York, New York, United States
- More, Pablo, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Grate, Crystal, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Usvyat, Len A., Renal Research Institute, New York, New York, United States
- Chatoth, Dinesh K., Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Hamilton, Emel, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Zhang, Hanjie, Renal Research Institute, New York, New York, United States
Background
Candida auris (C. auris) is an emerging multidrug-resistant fungus posing infection prevention challenges, particularly in individuals receiving dialysis with frequent healthcare exposure. Timely identification of test results is critical, but review of Health Information Exchange (HIE) records is labor-intensive due to lengthy, heterogeneous, and predominantly negative records.
Methods
We developed an automated pipeline to extract C. auris testing events from XML-formatted HIE documents in 10 dialysis patients. It used keyword screening, context-based chunking around C. auris mentions, and AI-based extraction using GPT-5-mini (OpenAI) to identify results and associated test and result dates. As testing events may include multiple timestamps (e.g., order, collection, result), post-processing consolidated fragmented extractions by merging near-duplicates (≤1-day test-date difference), resolving discrepant results using predefined rules, and removing duplicates from test/result date misclassification.
Results
Manual review identified 19 testing events with positive or negative results. The pipeline identified 21 events, including all 19 confirmed events and 2 additional duplicate or fragmented extractions related to test/result date mix-up, yielding 100% sensitivity and 90.5% positive predictive value. Among the 19 matched events, 17 were fully concordant for test date, result date, and outcome. Of the remaining two matched events, one had the correct test date, but the result date and outcome could not be verified; the other had the correct outcome, but both dates differed from manual review by 1 day. The two additional events reflected fragmentation or switching of test and result dates, across separate extractions. Fig. 1 shows the AI-extracted event timeline by patient.
Conclusion
An AI-assisted workflow extracted C. auris testing events from HIE data with high sensitivity and promising accuracy. This approach may reduce manual review burden and improve timely identification of multidrug-resistant organism results in nephrology care.
Acknowledgment
Large language models were used as part of the AI-assisted workflow described in this study and to assist with abstract preparation. The authors take full responsibility for the integrity, accuracy, and final content of the abstract.
Timeline of AI-extracted C. auris events by patient (positive/negative events with dates only).
Funding
- Commercial Support – Fresenius Medical Care