Abstract: FR-PO0007
Transforming Care for Elderly Patients with CKD: Clinical Impact of Artificial Intelligence (AI)-Driven Mortality Risk Stratification
Session Information
- Artificial Intelligence and Digital Health at the Bedside
November 07, 2025 | Location: Exhibit Hall, 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
- Roy, Debajyoti M., Changi General Hospital, Singapore, Singapore
- Tiang Guan Bernard, Koh, Changi General Hospital, Singapore, Singapore
- Sridharan, Srinath, Changi General Hospital, Singapore, Singapore
Background
As the global burden of chronic kidney disease (CKD) in the elderly grows, clinicians face increasing challenges in determining which patients may derive meaningful benefit from dialysis. Functional and cognitive decline following initiation of renal replacement therapy is common, and age alone is a poor prognostic indicator. Existing clinical scores such as the REIN and Wick models offer some guidance but are based on linear assumptions and derived from predominantly Caucasian populations. Artificial intelligence (AI), specifically machine learning (ML), offers a powerful alternative capable of modeling complex, nonlinear relationships and adapting to diverse demographic contexts.
Methods
This study employed random forest (RF) and balanced random forest (BRF) algorithms to predict six-month all-cause mortality in a cohort of 1,606 patients aged ≥65 years initiating haemodialysis at a tertiary care center in Southeast Asia. Model interpretability was assessed using SHapley Additive exPlanations (SHAP), allowing transparent identification of key predictive features and facilitating clinical translation.
Results
The RF model outperformed the BRF model, achieving an area under the receiver operating characteristic curve (ROC-AUC) of 0.83 versus 0.82, Fig 1. Both models showed superior discrimination and calibration compared to conventional scoring systems. SHAP analysis revealed the five most influential predictors of mortality: low serum albumin, higher estimated glomerular filtration rate (eGFR), elevated Charlson Comorbidity Index (CCI), older age, and elevated serum phosphate levels.
Conclusion
Our study highlights the potential of ML-driven tools to improve clinical decision-making in nephrology by providing more precise individualized risk assessment. The model is currently undergoing prospective validation in clinical practice where it supports comprehensive geriatric assessment and shared decision making.
Figure 1. ROC-AUC curves for the models