Predicting vesicoureteral reflux outcomes using artificial intelligence: A critical appraisal using APPRAISE-AI.

Abstract

Vesicoureteral reflux (VUR) is a common congenital urinary tract anomaly in children, associated with recurrent urinary tract infections (UTIs) and long-term sequelae such as renal scarring and chronic kidney disease. Artificial intelligence (AI) has recently emerged as a promising approach for improving VUR diagnosis, prognosis, and treatment stratification. This narrative review identified studies from the AI-PEDURO repository, a living database of AI applications in pediatric urology, most recently updated in June 2024. Eligible studies employed machine learning methods to predict clinically relevant outcomes of VUR or UTI. Seventeen studies met the inclusion criteria, with common applications including VUR grading from voiding cystourethrograms, prediction of UTI recurrence, spontaneous resolution of VUR, and outcomes following antibiotic prophylaxis or endoscopic injection therapy. Neural networks, tree-based algorithms, and support vector machines were the most frequently used approaches. Using the APPRAISE-AI tool, the median overall study quality was moderate, with strengths in clinical relevance and reporting quality, but persistent weaknesses in methodological conduct, robustness, and reproducibility. AI applications in VUR demonstrate strong potential to enhance diagnostic accuracy, personalize treatment, and predict outcomes; however, most published models remain of low to moderate quality. Adoption of standardized reporting frameworks and multi-institutional collaboration will be essential for improving rigour and accelerating clinical translation.