Abstract: SA-PO0343
Development and Validation of a Mortality Risk Prediction Model for Patients with Prevalent ESKD on Hemodialysis
Session Information
- Dialysis: Epidemiology and Facility Management
November 08, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 801 Dialysis: Hemodialysis and Frequent Dialysis
Authors
- Liu, Annie, Massachusetts General Hospital, Boston, Massachusetts, United States
- Ouyang, Tianqi, Massachusetts General Hospital, Boston, Massachusetts, United States
- Wulczyn, Kendra E., Massachusetts General Hospital, Boston, Massachusetts, United States
- Tang, Mengyao, Massachusetts General Hospital, Boston, Massachusetts, United States
- Ufere, Nneka, Massachusetts General Hospital, Boston, Massachusetts, United States
- Lakin, Joshua R., Dana-Farber Cancer Institute, Boston, Massachusetts, United States
- Kalim, Sahir, Massachusetts General Hospital, Boston, Massachusetts, United States
Background
While many mortality risk prediction models exist in incident end-stage kidney disease (ESKD), few focus on prevalent ESKD. These individuals often receive intensive patterns of end-of-life care that may be misaligned with their values and goals. Accurate mortality risk estimates could guide anticipatory care planning and treatment decisions. We aimed to develop and validate a model to predict 1-year mortality in patients with prevalent ESKD on hemodialysis.
Methods
Adults with prevalent ESKD (defined as receiving hemodialysis for >1 year) between January 1, 2015 and December 31, 2019 were identified in the U.S. Renal Data System (USRDS), a national mixed claims database. The dataset was randomly split into training (70%) and validation (30%) cohorts. Model development and internal validation were conducted in the training cohort. Predictive performance was assessed in the validation cohort. Thirty-six candidate predictors were considered based on literature and clinical expertise. Variable selection was performed using LASSO regression, followed by Fine-Gray competing risk modeling to account for death with kidney transplantation as a competing event. The final model was chosen based on sequential modeling and performance metrics. Bootstrapping with 200 resamples was used to estimate optimism-adjusted discrimination and calibration.
Results
A total of 366,105 patients met inclusion criteria and were divided into training (n=256,273) and validation (n=109,832) cohorts. The average age was 63.8 years (SD 14); 58.1% men; and 10.4% received a kidney transplant. One-year mortality from baseline occurred in 14.7% of patients. The final model contained five predictors: albumin, age, hemoglobin, hospitalization in the past year, and sex. The model demonstrated strong discrimination (c-statistic 0.86) and calibration (slope 0.994 [95% CI 0.97-1.01], intercept -0.026 [95% CI -0.062-0.011]) with minimal performance gains from additional variables. Internal validation confirmed model robustness and stability.
Conclusion
Commonly-measured clinical characteristics can accurately predict 1-year mortality in patients with prevalent ESKD. This parsimonious model may inform future efforts to support clinicians in identifying high risk hemodialysis patients for timely prognostic communication.
Funding
- Private Foundation Support