Machine Learning-Based Integration of Multiple Indicators to Predict Ureteral Access Sheath Placement Failure During Flexible Ureteroscopic Lithotripsy

Background Ureteral Access Sheath (UAS) placement failure during flexible ureteroscopic lithotripsy (FURL) is a critical challenge that leads to prolonged operative time, increased complications, and unexpected surgical termination. This study aimed to develop and validate a machine learning model integrating multiple indicators to predict the risk of UAS placement failure in FURL and to identify its key predictive factors. Methods This retrospective study collected clinical data from 1,153 patients with upper urinary tract calculus who underwent FURL. The data encompassed patient clinical history, laboratory findings, and CT imaging parameters, and innovatively incorporated anesthesia type and various systemic inflammatory markers. We developed and compared nine machine learning models, including Random Forest (RF) and XGBoost. The SHapley Additive exPlanations (SHAP) algorithm was employed to interpret the best-performing model by quantifying the contribution of each feature to the model's predictions. Finally, the model was deployed as a publicly accessible online web application. Results A total of 651 patients were ultimately included, of whom 196 experienced UAS placement failure. Among all models, the Random Forest model demonstrated the best performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.845. SHAP analysis revealed the top five key predictors for UAS placement failure. A decrease in mid-ureteral diameter and distal ureteral diameter, an increase in neutrophils, an increase in the widest part diameter of the renal parenchyma, and a decrease in the calculus long diameter were all found to significantly increase the risk of UAS placement failure. Conclusions This study successfully developed a high-precision machine learning model for predicting UAS placement failure. It is the first to integrate multiple clinical and imaging indicators using a machine learning approach and to analyze feature importance with the SHAP algorithm, offering new perspectives on the underlying mechanisms. The developed online prediction tool can assist clinicians in performing individualized preoperative risk assessments, thereby helping to optimize surgical strategies and reduce the risk of unnecessary surgical trauma and failure. This work represents a significant step in the translation of big data into clinical precision decision-making.