Machine learning algorithms for predicting stone residue and recurrence after lateral decubitus percutaneous nephrolithotomy

This study used machine learning to create a model predicting stone residue and recurrence after lateral decubitus percutaneous nephrolithotomy (PCNL) for renal and upper ureteral stones. Data from 271 patients who underwent PCNL at Peking University First Hospital-MiYun Hospital (Jan 2015-Aug 2022) were collected. Divided into an 80:20 training (n = 217) and test (n = 54) groups, logistic regression, random forest, and extreme Gradient Boosting (XGBoost) algorithms were used to build models. Model performance was evaluated by accuracy, precision, F1 score, receiver operating characteristic curves, and area under the curve (AUC). AUC is a numerical quantification of the area under the receiver operating characteristic curve, with a value range between 0 and 1, and it is used to objectively evaluate the overall discriminative ability of the model. Models for postoperative stone residue and recurrence were constructed. For stone residue, logistic regression had 72.4% accuracy, AUC of 0.721, F1 score of 0.737; random forest, 78.9% accuracy, AUC of 0.652, F1 score of 0.789; and XGBoost, 86.8% accuracy, AUC of 0.87, F1 score of 0.866. For recurrence, logistic regression had 57% accuracy, AUC of 0.433, F1 score of 0.51; random forest, 65.1% accuracy, AUC of 0.625, F1 score of 0.65; and XGBoost, 72.4% accuracy, AUC of 0.68, F1 score of 0.72. The machine learning-based predictive model for residual and recurrent stones after PCNL can assist urologists in making early treatment decisions.