Prediction tool for renal adaptation after living kidney donation using interpretable machine learning.

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Assessing post-donation renal outcomes is critical for living kidney donors due to varying life expectancies and comorbidities.
• Predicting renal adaptation is essential for donor safety and long-term kidney health.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting renal adaptation after living kidney donation.
• To create a clinical decision support tool for evaluating donor kidney function post-nephrectomy.

Main Methods:

• Utilized AutoScore, a machine learning-based score generator, to build the prediction model.
• Included 823 living kidney donors from 2009-2020, incorporating pre-donation variables, GFR, and kidney volume data.
• Defined fair and good renal adaptation based on post-donation estimated glomerular filtration rate (eGFR) thresholds.

Main Results:

• The developed model demonstrated good predictive capability for renal adaptation.
• Achieved areas under the ROC curve of 0.846 for fair and 0.626 for good adaptation.
• Precision-recall curves showed strong performance, with AUCs of 0.965 and 0.709, respectively.

Conclusions:

• The prediction tool for post-donation renal adaptation exhibits strong predictive performance.
• An interactive, web-based application was developed to aid clinical decision-making.
• This tool can assist clinicians in managing living kidney donors and optimizing post-donation care.