Personalized Prognostication of Kidney Transplant Candidates on the Waiting List: A Machine Learning Approach

Abstract

Abstract BackgroundKidney transplantation (KT) remains the best available treatment for end-stage kidney disease. Currently, there is an unmet need for personalized clinical decision aids that help clinicians and KT candidates on the waitlist (WL) make informed decisions. To this end, we took a machine learning (ML)-based approach by using random survival forest with competing risks (RSF-CR) to develop a personalized prognostication system to predict the patient outcomes.MethodsWe obtained data from the Scientific Registry of Transplant Recipients (n=65,583) and developed RSF-CR models to predict the cumulative incidence functions (CIFs) of receipt of KT (CIF KT ) and WL removal due to death or deteriorating condition (CIF death ) at one year post-listing. Hyperparameter tuning was performed with the integrated Brier score of out-of-bag predicted CIFs as the performance metrics. Feature selection was performed based on variable importance computed by the optimized model. Final model performance was evaluated on a hold-out validation set via receiver operating characteristics (ROC) analysis, decision curve analysis, and additional performance measures (e.g. recall, precision, and F1).ResultsThe selected model revealed 16 and 18 prognostic factors for the prediction of CIF KT and CIF death respectively. The model also demonstrated good prognostic ability with an area under the ROC curve and an integrated Brier score of 73.4 [72.7, 74.0] and 16.1 [0.0, 40.5] for CIF KT and 75.0 [73.7, 76.4] and 4.6 [0.0, 11.4] for CIF death on a validation dataset. Dichotomization at selected threshold probabilities yielded a precision, recall, and F1 of 0.927, 0.824, and 0.873 for CIF KT and 0.975, 0.785, and 0.869 for CIF death. ConclusionsRSF-CR revealed new predictors of the potential outcomes at one year post-listing for KT candidates. While the class imbalance problem was observed, adjustment of the threshold probability allowed the model to become a binary classifier with outstanding performance. This work demonstrated the value of using RSF-CR in the identification of prognostic factors and personalized prediction of the outcomes for KT candidates on WL. More research is warranted to fully unlock the potential of ML in providing personalized medicine.

Ähnliche Arbeiten