Post-dialysis blood urea nitrogen value prediction using machine learning

Abstract

【Introduction】Optimal dialysis dose is an important factor in the survival of patients undergoing hemodialysis；however, methods for accurate pre-dialysis prediction of the dialysis dose are yet to be established. In this study, we developed a machine-learning model for the prediction of post-dialysis blood urea nitrogen (BUN) values based on pre-dialysis patient characteristics and dialysis parameters, and evaluated its accuracy and usefulness. Furthermore, we computed the interpretability of the machine-learning model using the SHapley Additive exPlanations (SHAP) method. 【Method】We collected blood draw data, characteristics, and dialysis data of patients who underwent hemodialysis or hemodiafiltration between May 2020 and May 2022 at Sun Clinic. We randomly divided the data and used 70％ of the patient data to train the machine-learning model for post-dialysis BUN value prediction by the XGBoost algorithm, and used the remaining 30％ of the patient data to test the accuracy of the model. Model performance was evaluated using the coefficient of determination (R²). Feature importance was computed using the SHAP method. 【Results】We included a total of 246 patients and 935 dialysis sessions. Machine-learning model prediction showed R²＝0.87. Features that had the greatest impact on the prediction were the pre-dialysis BUN value, pre-dialysis body weight, and blood flow rate. 【Conclusions】The machine-learning model developed in this study can predict the post-dialysis BUN value with high-level accuracy, and can be used to estimate single-pool Kt/V using the predicted post-dialysis BUN value.