From Traditional Statistics to Artificial Intelligence: Advancing Pediatric UTI Recurrence Prediction in Low-Resource Communities

Urinary tract infections (UTIs) are among the most common bacterial infections in children, with recurrent episodes posing risks for renal scarring and long-term kidney damage. This study aimed to evaluate the utility of artificial intelligence (AI)-based models in predicting pediatric UTI recurrence, especially in low-resource settings.

A retrospective cohort study of 211 pediatric UTI cases was conducted between 2010 and 2025 at a single center in Iraq. Data included demographics, laboratory and imaging findings, and clinical outcomes. Four predictive models were developed: Logistic Regression, Random Forest, XGBoost, and Deep Learning. Models' performance was assessed using ROC-AUC, for accuracy, sensitivity, and specificity. SHapley Additive Explanations (SHAP) were used for interpretability.

The Deep Learning model achieved the highest performance (AUC-ROC: 0.94, accuracy: 90.2%), followed by XGBoost (AUC-ROC: 0.92), and Random Forest (AUC-ROC: 0.89). Logistic Regression performed the lowest (AUC-ROC: 0.78). SHAP analysis identified vesicoureteral reflux (VUR) grade ≥3, renal scarring, female sex, and rural residence as the most influential predictors of recurrence.

This study confirms that AI models significantly outperform traditional statistical methods in predicting recurrent pediatric UTIs. Key risk factors identified through SHAP align with established clinical knowledge, supporting the validity of AI predictions. The study also highlights healthcare disparities, particularly the elevated risk in rural populations. Limitations include its single-center design and lack of external validation.

AI-based predictive models, especially Deep Learning and XGBoost, offer high accuracy and clinical relevance for early risk stratification in pediatric UTIs. Their integration into digital health systems could enhance personalized care and reduce recurrence-related complications.