Akademik Veri Yönetim Sistemi
WORLD JOURNAL OF UROLOGY, cilt.44, sa.1, 2026 (SCI-Expanded, Scopus)
Background Mini-percutaneous nephrolithotomy (mini-PCNL) is widely used for the management of renal calculi, and postoperative infectious complications remain a concern. Although the addition of suction mitigates the risk of infectious complications, certain patients still develop sepsis, which can be life-threatening. Accurate early prediction of sepsis could improve patient outcomes and guide perioperative management. This study aimed to develop and validate an artificial intelligence (AI)-based model to predict postoperative sepsis in patients undergoing suction mini-PCNL. Methods Data were collected from the prospective, multicenter STUMPS registry, including 2,172 patients undergoing suction mini-PCNL across 30 centers in 21 countries. Patients were stratified based on the development of postoperative sepsis within 30 days. Clinical, operative, and stone-related variables were used to train an XGBoost machine learning model. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, and SHapley Additive exPlanations analysis for feature importance. Results Postoperative sepsis occurred in 87 patients (4.0%). Patients with sepsis had higher BMI, larger stones, prolonged lithotripsy, and longer operative times. The XGBoost model demonstrated excellent predictive performance (AUC= 0.946; sensitivity 88%, specificity 99%, accuracy 98.4%). SHAP analysis identified lithotripsy time, advanced Guy's stone score, tract closure method, and sheath size as the most influential predictors. Conclusions An AI-based model can reliably predict postoperative sepsis following suction mini-PCNL, highlighting operative complexity, stone burden, and patient physiological factors as key contributors. These findings support the potential of machine learning for early risk stratification and improved perioperative decision-making in clinical practice.