Predicting the Stone-Free Status of Percutaneous Nephrolithotomy with the Machine Learning System

Abstract

Purpose: The study aimed to create a machine learning model (MLM) to predict the stone-free status (SFS) of patients undergoing percutaneous nephrolithotomy (PCNL) and compare its performance to the S.T.O.N.E. and Guy’s stone scores. Patients and Methods: This is a retrospective study that included 320 PCNL patients. Pre-operative and post-operative variables were extracted and entered into three MLMs: RFC, SVM, and XGBoost. The methods used to assess the performance of each were mean bootstrap estimate, 10-fold cross-validation, classification report, and AUC. Each model was externally validated and evaluated by mean bootstrap estimate with CI, classification report, and AUC. Results: Out of the 320 patients who underwent PCNL, the SFS was found to be 69.4%. The RFC mean bootstrap estimate was 0.75 and 95% CI: [0.65–0.85], 10-fold cross-validation of 0.744, an accuracy of 0.74, and AUC of 0.761. The XGBoost results were 0.74 [0.63–0.85], 0.759, 0.72, and 0.769, respectively. The SVM results were 0.70 [0.60–0.79], 0.725, 0.74, and 0.751, respectively. The AUC of Guy’s stone score and the S.T.O.N.E. score were 0.666 and 0.71, respectively. The RFC external validation set had a mean bootstrap estimate of 0.87 and 95% CI: [0.81–0.92], an accuracy of 0.70, and an AUC of 0.795, While the XGBoost results were 0.84 [0.78–0.91], 0.74, and 0.84, respectively. The SVM results were 0.86 [0.80–0.91], 0.79, and 0.858, respectively. Conclusion: MLMs can be used with high accuracy in predicting SFS for patients undergoing PCNL. MLMs we utilized predicted the SFS with AUCs superior to those of GSS and S.T.O.N.E scores.