Construction and validation of a machine learning model for predicting urine culture positivity in children with suspected urinary tract infections.

BACKGROUND

Urinary tract infections (UTIs) are common in pediatrics. This study aimed to develop and validate a machine learning model for predicting urine culture positivity in children with suspected UTIs and to characterize regional pathogen and antimicrobial resistance profiles.

METHODS

Clinical and laboratory data were collected from children with suspected UTIs. Variables were selected via univariate analysis, LASSO regularization, and recursive feature elimination. Machine learning models were constructed using selected variables, followed by validation and interpretability analysis. A web application was developed to assist in the early prediction of urine culture positivity.

RESULTS

Infants aged 0-12 months accounted for 61.10% of enrolled children. The overall positive rate of urine culture was 42.17%, higher in boys. Gram-negative bacteria dominated the pathogen spectrum, with the two major strains exhibiting high rates of resistance to first- to third-generation cephalosporins and susceptibility to β-lactam/β-lactamase inhibitor combinations and carbapenems. Seven variables were used to construct seven machine learning models, among which the random forest (RF) model yielded the best performance, with an AUC of 0.814 (95% CI: 0.742-0.875), and sensitivity of 0.770 (95% CI: 0.662-0.865). A web-based calculator was established for individualized prediction.

CONCLUSIONS

We developed an RF-based online tool using seven routine laboratory parameters, demonstrating favorable predictive performance for pediatric urine culture positivity. Furthermore, our mapped local pathogen and antimicrobial resistance profiles help clinicians initiate rational empirical antibiotic therapy.