A Study on Predicting the Risk of Postoperative Urinary Tract Infection in Children with Ureteropelvic Junction Obstruction Based on Machine Learning Models

Background Congenital ureteropelvic junction obstruction (UPJO) is a common cause of pediatric hydronephrosis, with postoperative urinary tract infection (UTI) affecting prognosis. This study aims to identify UTI risk factors and develop a machine learning-based predictive model for UPJO patients. Methods A retrospective analysis included 150 children with UPJO undergoing dismembered pyeloplasty (2014–2024). Clinical data were collected, and UTIs were classified as uncomplicated or complicated. The Least Absolute Shrinkage and Selection Operator (LASSO) regression selected key predictors, and the dataset was split (7: 3) for training and validation. Machine learning models (LASSO, random forest, support vector machine, gradient boosting, logistic regression) were compared using the receiver operating characteristic curve (ROC), calibration, and decision curve analysis (DCA). Statistical analyses were performed using Python, SPSS 27.0, and R 4.3.1 software. Results Among the 150 patients, 67 (44.67%) developed postoperative urinary tract infection (UTI), including 44 (29.33%) with simple UTI and 23 (15.33%) with complicated UTI. Multivariate logistic regression analysis identified comorbidities (OR = 3.22, P = 0.008), preoperative abnormal white blood cell (WBC) count (OR = 3.04, P = 0.018), preoperative UTI (OR = 3.93, P = 0.006), preoperative urine nitrite positivity (OR = 17.19, P < 0.001), and presence of calculus (OR = 10.43, P = 0.015) as independent risk factors for postoperative UTI.Least Absolute Shrinkage and Selection Operator (LASSO) regression selected seven key variables: comorbidities, Double J stent indwelling time, presence of calculus, preoperative abnormal WBC count, preoperative UTI, preoperative urinary nitrite positivity, and intraoperative blood loss. The logistic regression-based predictive model demonstrated superior performance in the validation set (AUC = 0.81). A nomogram developed from this model exhibited excellent calibration and clinical net benefit. Conclusions This study successfully identified independent risk factors for postoperative UTI in children with UPJO and developed an efficient logistic regression-based predictive model and nomogram. These tools provide reliable evidence for early identification of high-risk patients and the formulation of personalized intervention strategies