Early detection of positive urine culture in patients with urolithiasis: a machine learning model with dynamic online nomogram.

BACKGROUND

Positive urine cultures are common in urinary stone patients, yet tools for early infection prediction are limited. To address this gap, a user-friendly, dynamic online nomogram was developed to predict the incidence of positive urine cultures in patients with urolithiasis.

METHODS

A retrospective study was conducted with 3,641 patients with urinary stones at the Second Hospital of Tianjin Medical University. The cohort was split into training and validation sets. Key variables were identified using Least Absolute Shrinkage and Selection Operator (LASSO) regression, while Random Forest and SHapley Additive exPlanations (SHAP) methods were applied to assess their importance. Online nomograms were developed and evaluated for performance through metrics such as area under the curve (AUC), calibration curve, decision curve analysis (DCA), probability density function (PDF), and clinical utility curve (CUC).

RESULTS

Multivariate logistic analysis identified four significant predictors-bacteria (BACT), C-reactive protein (CRP), nitrite, and leukocyte esterase (LEU)-which were integrated into the nomogram. The AUC values for the overall, training, and validation sets were 90.53, 91.22, and 89.06%, respectively. Calibration curves confirmed the nomogram's accuracy, and DCA demonstrated its superior performance over individual metrics. The PDF/CUC method revealed a threshold of 0.168, which effectively distinguished 88.54% of negatives from 78.70% of positives.

CONCLUSIONS

This dynamic online nomogram accurately predicts positive urine cultures in patients with urolithiasis, helping clinicians identify high-risk individuals , optimize antibiotic use, and improve patient outcomes. Further validation and biomarker exploration are needed to enhance its generalizability.