Interpretable Machine Learning in Kidney Offering: Multiple Outcome Prediction for Accepted Offers

Abstract

The decision to accept an organ offer for transplant, or wait for something potentially better in the future, can be challenging. Especially, clinical decision support tools predicting transplant outcomes are lacking. This project uses interpretable methods to predict both graft failure and patient death using data from previously accepted kidney transplant offers. Precisely, using more than twenty years of transplant outcome data, we train and compare several survival analysis and classification models in both single and multiple risk settings. In addition, we use post hoc interpretability techniques to clinically validate these models. In a single risk setting, neural networks provide comparable results to the Cox proportional hazard model, with 0.71 and 0.81 AUROC for predicting graft failure and patient death at year 10, respectively. Recipient and donor ages, primary renal disease, donor eGFR, donor type, and the number of mismatches at DR locus appear to be important features for transplant outcome prediction. We also extended the neural network approach to multiple outcome prediction, maintaining consistent performances and clinical interpretation. Thus, owing to their good predictive performance and the clinical relevance of their post hoc interpretation, neural networks represent a promising core component in the construction of future decision support systems for transplant offering.