Abstract: TH-OR27
Performance and Global Integration in Clinical Practice of a Fistula Failure Risk-Estimate Tool Based on Artificial Intelligence (AI)
Session Information
- Augmented Intelligence and Digital Health Advances
October 24, 2024 | Location: Room 2, Convention Center
Abstract Time: 05:50 PM - 06:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Ion Titapiccolo, Jasmine, Fresenius Medical Care AG, Vaiano Cremasco, Italy
- Nikam, Milind, Fresenius Medical Care AG, Singapore, Singapore
- Stuard, Stefano, Fresenius Medical Care AG, Bad Homburg, Hessen, Germany
- Hymes, Jeffrey L., Fresenius Medical Care AG, Boston, Massachusetts, United States
- Larkin, John W., Fresenius Medical Care AG, Boston, Massachusetts, United States
- Lama, Suman Kumar, Fresenius Medical Care AG, Boston, Massachusetts, United States
- Chaudhuri, Sheetal, Fresenius Medical Care AG, Boston, Massachusetts, United States
- Bellocchio, Francesco, Fresenius Medical Care AG, Vaiano Cremasco, Italy
- Yeung, Julianna, Fresenius Medical Care AG, Singapore, Singapore
- Usvyat, Len A., Fresenius Medical Care AG, Boston, Massachusetts, United States
- Neri, Luca, Fresenius Medical Care AG, Vaiano Cremasco, Italy
Background
Accurate prediction of fistula failure risk (FFR) in hemodialysis patients is critical for timely intervention and improved clinical outcomes. This study evaluates the performance of an AI-based FFR estimate tool in clinical practice. The data, collected over 12 months (February 2023-January 2024), encompass multiple countries, highlighting variations in clinical outcomes.
Methods
Data were collected from hemodialysis centers in seven countries (Australia, Czech Republic, Italy, Portugal, Singapore, Slovakia, Spain), totaling 83,126 records. The AI risk estimation tool is integrated in the clinical system EuCliD® (Fresenius Medical Care) and provides nephrologists with a monthly estimate of FFR over the next 90 days. Predictive accuracy in comparison with actual failure incidence was assess using AUC of ROC curve. Fistula failures include angioplasty (with/without stent), thrombectomy, switch to a new dialysis access, temporarily unusable fistula, and hospitalization due to fistula complication.
Results
The AI tool exhibited a varied performance across countries, with the global AUC being 78.1%. The global failure incidence percentage was 6.2%. Failure percentages for each of the four considered categories are reported in Table 1. Differences in the habit of reporting data in the clinical system can be highlighted from the table. Risk metrics also varied across countries.
Conclusion
The AI-based FFR estimate tool demonstrated robust predictive performance across diverse clinical settings. The depicted variability shows the importance of localized adaptation of AI tools.
Table 1. Overview of the Fistula Failure Risk (FFR) tool performance across countries. FFR is reported in terms of mean±std. Total sum of failure causes can be higher than 100% because the four causes of failure are not mutually exclusive.
Funding
- Commercial Support – Fresenius Medical Care