Construction and validation of a predictive model for urinary tract infection stones based on machine learning

Objective To construct a machine learning model for predicting the risk of infection stones in patients with urinary calculi before surgery, in order to improve the preoperative management of patients with urinary stones.Methods Patients who were admitted to Xuzhou Central Hospital due to urinary calculi from August 2018 to March 2023 were selected and their clinical data were retrospectively analyzed. Through the caret R package, the patients were randomly divided into training and test sets in a ratio of 3:1. Predictors were screened from the training set by Lasso regression analysis, which were then fitted based on nine machine learning models. The performance of the resultant model. was evaluated according to the receiver operating characteristic-area under curve (ROC-AUC), precision recall-area under curve (PR-AUC), accuracy, precision, F1 score, calibration curve, and clinical decision curve.Results A total of 350 patients were included, including 108 patients with infection stones and 242 without infection stones. Through Lasso regression analysis based on 10-fold cross-validation, 11 predictors were obtained, namely urinary pH value, blood uric acid, urinary nitrite, age, urinary crystallization, lymphocyte, urinary protein, gender, hydronephrosis degree, smoking, and urinary bacterial culture. Accordingly, nine machine learning models were established, where the random forest model had the best efficiency (accuracy: 0.83, F1 score: 0.69, PR-AUC: 0.77, precision: 0.77, ROC-AUC: 0.87, 95% confidence interval (CI): 0.78-0.94). The calibration curve results further indicated that the random forest model showed a good curve fit, with the smallest brier score (BS) of all models at 0.13. According to the clinical decision curve, the net benefit obtained by the random forest model was the largest of all models at the thresholds of 0.38-0.71.Conclusions The random forest model is the most effective machine learning model for predicting infection stones before surgery, in which urinary pH, blood uric acid and urinary nitrite are the three most important indexes.