MACHINE-LEARNING MODELS FOR PREDICTING PATIENT SURVIVAL AFTER LIVER TRANSPLANTATION

Abstract

In our work we have built models predicting whether a patient will lose an organ after liver transplantation within a specified time horizon. We have used the observations of bilirubin and creatinine in the whole first year after the transplantation to derive predictors capturing not only their static value but also variability. Our models indeed have predictive power which proves the value of incorporating variability of biochemical measurements and it is the first contribution of our paper. The second one is the selection of the best model for the defined problem. We have identified that full-complexity models, such as random forests and gradient boosting, despite having the best predictive power, lack sufficient interpretability which is important in medicine. We have found that generalized additive models (GAM) provide desired interpretability and their predictive power is closer to the predictions of full-complexity models than to the predictions of simple linear models.