Abstract: FR-PO0014
Advancing Prognostic Models for Competing Risks in Kidney Transplantation Outcomes: Statistical and Machine Learning Approaches
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
Author
- Awoke, Getahun Mulugeta, Debre Berhan University, Debre Birhan, Amhara , Ethiopia
Background
Modeling competing risks is important in renal transplantation because it allows for the interplay and dynamics of adverse transplant outcomes, such as graft failure and death with a functioning graft. Accurate prediction of these risks would be crucial for identifying high-risk patients and optimizing clinical management strategies. The goal of this study was to develop clinical prognostic models for the prediction of these competing risks to improve clinical decision-making for transplant recipients.
Methods
Data were obtained from a retrospective cohort of renal transplant recipients at the Ethiopian National Kidney Transplant Center from 2015 to 2022. Various statistical models (e.g., cause-specific hazard and Fine-Gray sub-distribution hazard models) and machine learning models (e.g., random survival forest for competing risk (RSFCR) and Cox Boost for competing risk (CB-CR) models) have been used to analyze competing transplant outcomes.
Results
The cumulative incidence of graft failure was 7.4% and death with a functioning graft was 7.2%. An important finding includes chronic rejection, blood urea nitrogen (BUN), post-transplant admissions, acute rejection, and post-transplant fluid overload as significant predictors of renal graft failure. In contrast, major predictors of death with a functioning graft are cardiovascular complications, tacrolimus metabolism rate, chronic rejection, and frequent admissions. The random survival forest for competing risks had the best prediction accuracy.
Conclusion
This study emphasizes the importance of accurate prediction of competing risks of graft failure and death in renal transplant recipients. However, statistical models have value in interpretability, while machine learning models perform well in predictive performance. The application of these findings in clinical practice may allow clinicians to individualize monitoring and therapeutic strategies for transplant recipients.
Funding
- Government Support – Non-U.S.