Constructing a machine learning model for systemic infection after kidney stone surgery based on CT values

This study aims to develop a machine learning model utilizing Computed Tomography (CT) values to predict systemic inflammatory response syndrome (SIRS) after endoscopic surgery for kidney stones. The goal is to identify high-risk patients early and provide valuable guidance for urologists in the early diagnosis and intervention of post-operative urosepsis. This study included 833 patients who underwent retrograde intrarenal surgery (RIRS) or percutaneous nephrolithotomy (PCNL) for kidney stones. Five machine learning algorithms and ten preoperative or intraoperative variables were used to develop a predictive model for SIRS. The SHapley Additive exPlanations (SHAP) method was used to explain the distribution of feature importance in the model's predictions. Among the 833 patients, 126 (15.1%) developed SIRS postoperatively. All five machine learning models demonstrated strong discrimination on the validation set (AUC: 0.690-0.858). The eXtreme Gradient Boosting (XGBoost) model was the best performer [AUC: 0.858; sensitivity: 0.877; specificity: 0.981; accuracy: 0.841; positive predictive value: 0.629; negative predictive value: 0.851]. The characteristic importance of the Machine Learning model (ML model) and SHAP results indicated Hounsfield Unit (HU), Urinary protein, Stone burden, and Serum uric acid as important predictors for the model. A machine learning model utilizing CT values was developed to predict postoperative SIRS in endoscopic kidney stone surgery. The model demonstrates strong predictive performance and can assist in assessing the risk of urosepsis in postoperative patients.