Leveraging Machine Learning to Predict Delayed Graft Function Occurrence and Length in Kidney Transplant Recipients.

Area of Science:

• Nephrology
• Transplant Surgery
• Machine Learning in Medicine

Background:

• Delayed graft function (DGF) significantly impacts kidney transplant outcomes.
• Current models lack prediction for DGF duration, creating a clinical gap.
• Perioperative factors are crucial but underutilized for DGF prediction.

Purpose of the Study:

• To develop an ensemble machine learning model for predicting DGF occurrence and duration.
• To address the unmet need for predicting the impact of DGF on patient outcomes.
• To leverage perioperative data for enhanced kidney transplant care.

Main Methods:

• Utilized gradient-boosted decision trees (GBDT) for model development.
• Trained and validated the model on a large cohort (2725 patients) with external validation (284 patients).
• Employed k-fold cross-validation and performance metrics like ROC-AUC and accuracy.

Main Results:

• The DGF prediction model achieved a ROC-AUC of 0.77.
• The DGF duration classification model demonstrated 79.2% accuracy, with external validation at 78%.
• Key predictors identified include acute kidney injury (AKI) and donation after circulatory death (DCD) status.

Conclusions:

• GBDT models effectively predict both the likelihood and duration of DGF.
• This predictive capability fills a critical gap in kidney transplant care.
• The model supports personalized transplant strategies, optimizing perioperative planning and interventions for better patient outcomes.