Using Machine Learning to Optimize Kidney Transplants

Abstract

Currently, 100,000 people are waiting on the organ donation list in the United States, 17 of which die each day. This is due to the demand for organs far outweighing their supply. There needs to be objective criteria regardless of race, economic status, or sex for the distribution of organs. The most common types of transplants are kidney transplants. For this reason, the allocation of kidneys must be given to the patients who require the kidneys the most. The operational definition for needing kidneys the most should be the patients who are most likely to progress to chronic kidney disease (CKD). Artificial intelligence can help predict which patients will progress to ESRD and has shown promise in doing so. However, many different types of AI models can be used, and many of them have stark differences in how they operate. Comparing these models can allow researchers to understand which models are most effective for diagnosing CKD.  The models featured in this study were the logistic regression, ridge classifier, and decision tree models. All three models had a mean accuracy of 0.975. The logistic regression model had a mean precision of 0.960, a mean recall of 1.00, and a mean F1 score of .980.  The ridge classifier model and the random forest classifier model both had a mean precision of 1.00, a mean recall of 0.958, and a mean F1 score of 0.979.