Predicting kidney allograft survival with explainable machine learning.

Area of Science:

• Nephrology
• Transplant Surgery
• Medical Informatics

Background:

• Kidney allograft survival has improved, but risk factors for graft dysfunction persist.
• Identifying predictors of early graft loss is crucial for optimizing patient outcomes.

Purpose of the Study:

• To evaluate a novel machine learning (ML) method for predicting early kidney allograft loss.
• To identify key variables impacting graft survival and inform clinical decision-making.

Main Methods:

• Retrospective cohort study of 627 kidney transplant patients.
• Development and application of an automated ML algorithm using pre-processed patient data.
• Model evaluation via Area Under the Curve (AUC) and interpretation using SHapley Additive exPlanations (SHAP).

Main Results:

• The ML model achieved an AUC of 0.84, with high specificity (0.89) and precision (0.81).
• Serum creatinine at hospital discharge was the most significant predictor of allograft loss.
• Pre-transplant factors including BMI, patient age, and BK polyomavirus (BKPyV) infection also significantly impacted outcomes.

Conclusions:

• Machine learning effectively identifies critical factors for early kidney allograft loss.
• Key predictors include post-transplant serum creatinine, pre-transplant BMI, age, and BKPyV infection.
• ML tools show promise in assisting clinical decisions for kidney transplant management.