Machine-Learning Prediction Model for Personalized Urinary Tract Infection Care in Children

Urinary tract infections (UTIs) are among the most common and serious infections in children, leading to over 1 million office visits annually in the United States and contributing to substantial healthcare costs. Unlike in adults, febrile UTI (fUTI) in children can be a silent and dangerous condition because it is challenging to diagnose, with the potential risk of progression to severe infection and permanent loss of kidney function. Recurrent UTIs, particularly those associated with vesicoureteral reflux (VUR)—a condition affecting 10 percent of children in which urine flows backward from the bladder to the kidneys—pose a high risk for kidney damage due to inflammation and scarring. VUR alone incurs an estimated healthcare burden of over $100 million annually. VUR is linked to 40-50 percent of fUTI cases and, if untreated, can result in severe long-term complications such as hypertension, proteinuria, and end-stage renal disease. Fortunately, VUR can be successfully treated with either surgical correction or medical management, depending on the child’s circumstances. While early identification and management of VUR are critical, there is considerable debate about the optimal timing for performing the voiding cystourethrogram (VCUG), the gold standard diagnostic procedure for VUR. This has led to inconsistent care, with only 64 percent of children receiving follow-up care after fUTI and variable VCUG practices. This research aims to determine which children would benefit most from early VCUG, ensuring timely diagnosis and management while avoiding unnecessary testing and associated risks.

The specific aims of the research are as follows:

• Assess determinants of machine learning (ML) algorithm implementation for pediatric fUTI care.
• Optimize novel ML algorithm to predict high-risk VUR in children who presented with fUTI.
• Assess prediction algorithm implementation for pediatric fUTI care.

Researchers will conduct interviews with clinicians and families to identify facilitators and barriers to implementing a ML model, using the Consolidated Framework for Implementation Research (CFIR 2.0). Insights from these interviews will guide the design of a pilot implementation in outpatient clinical settings to test the ML model's effectiveness in predicting high-risk fUTIs in children and evaluate implementation outcomes. Concurrently, researchers will optimize and validate the ML prediction model and refine the clinical decision support (CDS) data pipeline.

The researchers aim to address the significant health and economic impact of VUR and recurrent UTIs in children by developing and implementing a ML model to optimize the timing and use of VCUG for diagnosis, standardizing practices, minimizing unnecessary procedures and associated risks, and ensuring timely diagnosis for high-risk children. The study seeks to close critical care gaps, enhance clinical decision making, and improve pediatric UTI management quality and safety. Its significance lies in demonstrating a scalable, proof-of-concept algorithm that can be applied to other conditions, advancing a learning healthcare system that empowers both clinicians and caregivers across diverse settings.