Predicting SIRS after PCNL using machine learning: the joint impact of sarcopenia and staghorn stones.

PURPOSE: To develop and validate machine learning models for predicting systemic inflammatory response syndrome (SIRS) after percutaneous nephrolithotomy (PCNL), to establish a web-based prediction tool, and to investigate the association between sarcopenia and staghorn stones as well as their potential synergistic effect on postoperative SIRS. METHODS: Patients undergoing PCNL between January 2021 and August 2025 at The Third Affiliated Hospital of Sun Yat-sen University were retrospectively analyzed and randomly divided into training and validation sets (7:3). Feature selection was performed using elastic net and Boruta. Six machine learning models were developed and compared, with SHAP used for interpretability. The optimal model was used to build a web-based prediction tool. Associations and interaction effects between sarcopenia and staghorn stones were further assessed. RESULTS: A total of 755 patients were included, with a SIRS incidence of 17.62%. XGBoost achieved the best performance (validation set: AUC = 0.863, accuracy = 0.863, sensitivity = 0.763, specificity = 0.883, F1 score = 0.652). SHAP analysis identified staghorn stones and sarcopenia as the most important predictors. Sarcopenia was positively associated with staghorn stones. A significant synergistic effect on SIRS was observed, confirmed by both multiplicative interaction (OR = 4.229, 95% CI 1.354-14.432, P = 0.016) and additive interaction (RERI = 25.473, AP = 0.854, S = 8.600). CONCLUSION: The XGBoost model provides robust prediction of postoperative SIRS after PCNL, and the web-based tool may assist in risk stratification. Sarcopenia and staghorn stones are positively associated, and their coexistence is linked to a higher risk of SIRS with a potential positive interaction, highlighting the need for individualized perioperative management.