Predicting Kidney Discard Using Machine Learning.

Area of Science:

• Nephrology
• Transplant Surgery
• Data Science

Background:

• Significant numbers of deceased donor kidneys are discarded annually in the US, despite a shortage of available organs.
• In 2018 alone, 3,569 kidneys were discarded, highlighting a critical area for improvement in organ transplantation.

Purpose of the Study:

• To compare the efficacy of various machine learning (ML) techniques against traditional logistic regression in identifying deceased donor kidneys at high risk of discard.
• To evaluate the performance of ML models both at the time of initial match run and after incorporating biopsy and machine perfusion data.

Main Methods:

• A cohort of adult deceased donor kidneys from December 2014 to July 2019 was analyzed.
• Machine learning models including Random Forests (RF), Adaptive Boosting, Neural Networks, Support Vector Machines, and K-nearest Neighbors were compared to Logistic Regression (LR).

Main Results:

• Random Forests (RF) demonstrated superior performance over Logistic Regression (LR) in classifying discarded kidneys, achieving a higher area under the receiver operative curve (AUC) and balanced accuracy.
• Incorporating biopsy and machine perfusion variables further enhanced the predictive accuracy of both LR and RF models, with RF achieving an AUC of 0.904.
• RF correctly classified over 388 more kidneys compared to LR in the test dataset.

Conclusions:

• Machine learning techniques, particularly Random Forests, offer a more accurate method for identifying deceased donor kidneys at risk of discard.
• Implementing advanced ML models can potentially improve organ utilization rates and increase the number of successful kidney transplants.