Machine learning-based screening of characteristic factors for urinary tract infection following ureteral stone surgery and construction and validation of risk prediction models

Ureteroscopic lithotripsy has emerged as the cornerstone treatment modality for ureteral stones due to its exceptional success rates and minimal complication profiles. Nevertheless, postoperative urinary tract infection (UTI) remains a prevalent and serious complication that can result in prolonged hospitalization, increased healthcare expenditure, and compromised patient outcomes. While existing studies have explored UTI risk factors, specific analyses targeting post-ureteroscopic lithotripsy UTI remain insufficient, and efficient, precise individualized prediction tools are lacking. To utilize multiple machine learning methodologies for screening key risk factors associated with UTI following ureteroscopic lithotripsy, combined with multivariable logistic regression to construct and validate prediction models, aiming to provide clinical reference for risk assessment and individualized management. This single-center retrospective cohort study included 533 patients who underwent ureteroscopic lithotripsy between January 2019 and January 2023, randomly divided into training (n = 373) and validation (n = 160) groups at a 7:3 ratio. LASSO regression, support vector machine, and random forest methods were employed to screen UTI-related characteristic factors. Univariate and multivariable logistic regression analyses were performed in the training group to construct a nomogram model based on significant independent risk factors. Model performance was evaluated through receiver operating characteristic curves, calibration curves, and decision curve analysis for both internal and external validation. Machine learning screening identified 16 potential risk factors, with multivariable analysis determining stone size, operative time, white blood cell count, C-reactive protein, diabetes history, positive preoperative urine culture, and American Society of Anesthesiologists score as independent risk factors. The nomogram model constructed based on these factors demonstrated an area under the curve of 0.802 in the training group and 0.746 in the validation group, both showing excellent discriminative ability. Calibration curves revealed good fit between predicted probabilities and actual outcomes (mean absolute error: 0.027 in training group, 0.021 in validation group). Decision curve analysis indicated that the model provided net clinical benefit across risk thresholds ranging from 0% to 99%. The nomogram model constructed based on multivariable logistic regression, incorporating key risk factors identified through machine learning screening, effectively predicts UTI risk following ureteroscopic lithotripsy with satisfactory accuracy and clinical applicability, providing valuable reference for perioperative management and individualized prevention strategies.