Development of an integrated multidimensional nomogram for predicting stone-free status after retrograde intrarenal surgery: a machine learning-based approach

BACKGROUND: This study aimed to develop and validate a novel preoperative nomogram to predict stone-free status (SFS) in patients undergoing retrograde intrarenal surgery (RIRS) for kidney stones. METHODS: A total of 312 patients who underwent RIRS were retrospectively analyzed. Robust predictors were selected using the intersection of the least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm. Eight machine-learning algorithms were trained and compared. The optimal model was determined based on the receiver operating characteristic curve (ROC), Decision-curve analysis (DCA), and other evaluation indicators, with 5-fold cross-validation performed for internal validation. The best-performing algorithm (LogisticTEST) was further interpreted using Shapley Additive Explanations (SHAP) and visualized through a nomogram. Internal validation was performed with 1,000 bootstrap resamples. RESULTS: Seven key predictors were identified, including stone volume, mean stone density, calyx pelvic height, inferior pole stone, laser mode, Operative Team Proficiency Score, and Number of stones. Among all models, LogisticTEST demonstrated the best discrimination (AUC=0.965) and the greatest clinical net benefit on DCA. SHAP analysis reveals that laser mode is the most critical predictor influencing stone-free status, and the predictive capability of SFS for each patient can be well clinically explained through SHAP. The logistic-based nomogram exhibited strong calibration and interpretability, providing an intuitive clinical decision-support tool. CONCLUSIONS: A machine learning-driven logistic model incorporating laser mode and team experience accurately predicted postoperative SFR following RIRS. Future multi-centre external validation and prospective studies are required to validate our findings.