KTOP

KTOP predicts patient mortality, graft loss, and death with a functioning graft after kidney transplantation from brain-dead deceased donors to inform organ allocation and recipient risk assessment.

Key Features:

• Data-Driven Predictions: Uses a dataset of 32,958 transplants from the Eurotransplant kidney allocation system and the Eurotransplant Senior Program across eight European countries between January 2006 and May 2018.
• Risk Factor Analysis: Employs Cox proportional-hazards models for recipient mortality and proportional subdistribution hazard regression models for graft loss and death with a functioning graft (DWFG).
• Recipient-Donor Models: Constructs prediction models using recipient-only and combined recipient-donor characteristics.
• Mortality Predictors: Identifies recipient diabetes (HR 10.73), retransplantation (HR 3.08 per transplant), and recipient age (HR 1.08) as leading predictors of mortality.
• DWFG Predictors: Notes that similar recipient-related factors influence death with a functioning graft (DWFG).
• Graft Loss Predictors: Associates graft loss with recipient diabetes (SHR 1.32), increased donor age (SHR 1.02), and prolonged cold ischemia time (SHR 1.02).
• Validation and Sensitivity Analyses: Applies leave-one-country-out validation, calibration plots, and time-specific area under the receiver operating characteristic curve (AUC) with up to a 10-year AUC of 0.81.

Scientific Applications:

• Predicting Transplant Outcomes: Estimates individual risks for mortality, graft loss, and DWFG to quantify expected transplant outcomes.
• Organ Allocation and Recipient Selection: Provides risk estimates to inform organ allocation strategies and recipient selection.
• Policy and Resource Planning: Supplies quantitative outcome predictions to support transplantation network policy-making and resource management.
• Pre- and Post-Donor Risk Estimation: Generates individual risk estimates both before and after a donor has been identified.

Methodology:

Models were developed using Cox proportional-hazards regression and proportional subdistribution hazard regression, with recipient-only and combined recipient-donor covariates, and validated via leave-one-country-out analyses, calibration plots, and time-specific AUC calculations (up to 10-year AUC 0.81).