Abstract: FR-PO1080
Preventing Falls in Patients on Dialysis Through Artificial Intelligence (AI)-Driven Risk Prediction
Session Information
- Health Maintenance, Nutrition, and Metabolism
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Health Maintenance, Nutrition, and Metabolism
- 1500 Health Maintenance, Nutrition, and Metabolism
Authors
- Lama, Suman Kumar, Renal Research Institute, New York, New York, United States
- Bourque, Candi, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Monaghan, Caitlin, Renal Research Institute, New York, New York, United States
- Bellocchio, Francesco, Renal Research Institute, New York, New York, United States
- Lee, Carol, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Willetts, Joanna, Renal Research Institute, New York, New York, United States
- Kwitkowski, Melissa A, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Chatoth, Dinesh K., Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
- Neri, Luca, Renal Research Institute, New York, New York, United States
- Chaudhuri, Sheetal, Renal Research Institute, New York, New York, United States
- Usvyat, Len A., Renal Research Institute, New York, New York, United States
Background
Falls in End Stage Kidney Disease (ESKD) patients are common and may be preventable causes of morbidity when mitigatable risks are identified and addressed. It often results in hospitalization, reduced quality of life, and increased healthcare costs. Identifying high-risk individuals remains a critical challenge in clinical practice. We aimed to develop and evaluate an AI model for predicting the risk of patient falls within a 31-day period.
Methods
We developed an artificial intelligent (AI) model using data from adult ESKD patients receiving incenter hemodialysis (ICHD) between January 2023 and December 2023. The dataset comprised of approximately 1,800 variables, including treatment history, laboratory results, demographics, medication usage, and fall history, incorporating both baseline and derived features. Data were randomly split into 60% for training, 20% for validation, and 20% for testing, ensuring that test data included only unseen patients. An XGBoost binary classification model was trained on the training data to distinguish between patients likely to fall and those not at risk.
Results
The AI model achieved an Area Under the ROC Curve (AUC) of 0.686. The model demonstrated a sensitivity (recall) of 0.65, a specificity of 0.626, and a balanced accuracy of 0.627 in predicting patient falls. Key predictive features as shown in Figure 1 SHAP Value Plot included age, prior fall history, use of nervous system medications, and blood pressure measurements.
Conclusion
The developed AI model shows potential in identifying patients at risk of falling. It can be integrated into clinical practice to provide proactive patient care and implement fall prevention strategies.
Funding
- Commercial Support – Fresenius Medical Care