Abstract / Summary
This study sought to develop and externally validate an interpretable machine learning–based model to predict the risk of urinary tract infection (UTI) following minimally invasive surgery (MIS) in patients with upper urinary tract calculi. Clinical data from patients suffering upper urinary tract calculi who received MIS in the Department of Urology at our institution in 2024 were retrospectively collected as the development cohort. Data from another tertiary center were collected for external validation. Univariate analysis focused on identifying variables significantly associated with postoperative UTI, alongside feature selection with least absolute shrinkage and selection operator (LASSO) regression. A total of eight machine learning algorithms—logistic regression, K-nearest neighbors (KNN), support vector machine (SVM), Gaussian naive Bayes (GNB), elastic net, extreme gradient boosting (XGBoost), random forest, and Multilayer Perceptron (MLP)—were used to build prediction models. The performance of the models was quantified via receiver operating characteristic (ROC) curves. SHapley Additive exPlanation (SHAP) was adopted to more clearly explain the best-performance model. Among the 546 patients with 65 clinical variables included in the study, 62 developed postoperative UTI, yielding an infection rate of 11.35%. LASSO regression identified seven key predictors: urine specific gravity, platelet-to-albumin ratio (PAR), C-reactive protein (CRP), planned surgical approach, urinary white blood cell count, nitrite, and leukocyte esterase. Among all models, XGBoost showed the best discriminative performance, yielding an area under the curve (AUC) of 0.716 in the external validation cohort. An interpretable machine learning model based on readily available clinical parameters showed moderate discrimination in the external validation cohort and may assist in identifying patients at increased risk of postoperative UTI. Further prospective evaluation is required before clinical implementation.