Assessing Deceased-Donor Kidneys Through Posttransplant Survival Prediction Algorithms.

Area of Science:

• Nephrology
• Transplantation Science
• Biostatistics

Background:

• The Kidney Donor Risk Index (KDRI) is a standard tool for deceased-donor kidney quality assessment and allocation in the U.S.
• Current KDRI models exhibit modest predictive accuracy and calibration issues, particularly after recent revisions.
• There is a need for improved prediction models for posttransplant allograft survival.

Purpose of the Study:

• To evaluate novel approaches for predicting posttransplant kidney allograft survival.
• To compare machine-learning and traditional statistical models using various predictor combinations.
• To assess model discrimination and calibration for allograft survival and delayed graft function.

Main Methods:

• Retrospective cohort study of 75,867 adult kidney recipients using Organ Procurement and Transplantation Network data (2007-2021).
• Compared machine-learning and traditional models (proportional hazards, logistic regression).
• Incorporated donor demographic/clinical variables, donor longitudinal lab data, and recipient clinical variables.

Main Results:

• Machine-learning models and inclusion of donor longitudinal lab data did not improve prediction discrimination.
• A proportional hazards model incorporating recipient variables (Kidney Allograft Survival Index) improved discrimination (AUC 0.68) and calibration for allograft survival.
• A logistic regression model with recipient variables showed acceptable discrimination (AUC 0.75) for delayed graft function.

Conclusions:

• Recipient characteristics are crucial for enhancing kidney allograft survival prediction models.
• Improved models with better discrimination and calibration can potentially optimize kidney allocation.
• Further research, including external validation, is warranted.