Machine Learning-Based Prediction of Urolithiasis Recurrence Using Patient's Clinical Data, Demography, and CT...

Urolithiasis is the formation of stones within the urinary tract with diverse shapes, sizes, and locations. Earlier diagnosis facilitates treatment and complication prevention. This study aimed to propose a method for predicting urolithiasis recurrence using machine learning.

We used clinical data, demographics, and CT findings from 4,246 patients who visited a clinic once or multiple times within three years. The method comprises three phases: data engineering and preprocessing, machine learning model development, and performance evaluation. Six machine learning classifiers were evaluated using standard performance metrics, ROC analysis, calibration analysis, and decision curve analysis.

Across 10 independent repeats with a train/test split, the best‑performing classifier was random forest, with area under the ROC curve (AUC), sensitivity, and positive predictive value of 0.64, 0.87, and 0.84, respectively. Using a 10‑fold cross‑validation strategy, random forest again performed best, with AUC, sensitivity, and positive predictive value of 0.63, 0.90, and 0.83, respectively. A Brier score of 0.18 indicated comparatively better calibration.

This study presents a practical machine learning application for predicting urolithiasis recurrence with clinically acceptable accuracy compared with traditional scoring systems. Six predictive models were assessed using multiple metrics to select the optimal classifier.