A Machine Learning Prediction Model for Immediate Graft Function After Deceased Donor Kidney Transplantation

Area of Science:

• Nephrology
• Transplantation immunology
• Machine learning in medicine

Background:

• Kidney graft function can decline post-transplantation, necessitating dialysis.
• Machine perfusion may not benefit recipients with excellent immediate graft function (IGF).
• Predicting IGF is crucial for optimizing resource allocation in kidney transplantation.

Purpose of the Study:

• To develop a machine learning model for predicting IGF in deceased donor kidney transplant recipients.
• To identify key variables influencing immediate graft function post-KTx.
• To aid in selecting patients who would benefit from expensive preservation techniques.

Main Methods:

• Utilized data from unsensitized recipients of their first deceased donor KTx (2010-2019).
• Employed machine learning algorithms including XGBoost, LightGBM, and Random Forest.
• Split data into 70% training and 30% testing sets for model validation.

Main Results:

• 21.7% of patients experienced IGF.
• The eXtreme Gradient Boosting model achieved the highest predictive performance (AUC, 0.78).
• Five key predictive variables for IGF were identified.

Conclusions:

• A predictive model for IGF in kidney transplantation is feasible.
• This model can enhance patient selection for costly preservation methods like machine perfusion.
• Improved prediction of IGF can optimize post-transplant care and resource utilization.