Machine-learning prediction model for personalized urinary tract infection care in children

PROJECT SUMMARY The overarching goal of Dr. Wang’s proposal is to reduce the known risk of renal injury from febrile urinary tract infection (fUTI) in children by implementing a practical, validated clinical decision support algorithm to promptly identify unsafe anatomy before injury occurs. Dr. Wang’s proposal has identified significant care gaps in the

care of fUTI children. The long-term goal is to contribute to optimal management for fUTI in children through implementation of novel, high-value, self-renewing machine learning (ML) models. The overall objective is to identify those critical elements necessary for the development and implementation of predictive modeling to

identify children who would benefit most from early vs later voiding cystourethrogram (VCUG) in primary care (NOT-HS-22-011). The central hypothesis is that ML models can provide accurate prediction of risky fUTI, and thus assist clinicians/families to choose the best timing for VCUG. The rationale is to offer a scientific roadmap

and pilot new strategies that incorporate and implement prediction models to provide true value-based care (NOT-HS-19-011) and equitable resource utilization for children (NOT-HS-21-015, NOT-HS-21-014). This hypothesis has been formulated based on Dr. Wang’s previous work that demonstrates 1) high variability in

post-UTI VCUG practice patterns; and 2) ML models can serve as a promising basis to reliably identify children with high risk for damaging UTI. Leveraging the data from a large pediatric practice network within Boston Children’s Hospital, the following specific aims are proposed: 1) assess determinants for successful ML

algorithm implementation for pediatric fUTI care, 2) prospectively collect data to optimize and validate novel ML algorithms in fUTI children, 3) pilot prediction of pediatric fUTI algorithm implementation to iteratively testing, implementing, and adapting the algorithm using the principles of implementation and behavioral science to

maximize adoption and sustained implementations. In this proposal, Dr. Wang has assembled a multi- disciplinary mentorship team consisting of experts in, qualitative methods, informatics, infectious disease, machine learning, implementation, and behavioral science to help him achieve his goals and has designed a

comprehensive training plan to acquire necessary expertise. Dr. Wang’s unique background combined with his career development plan, and the rich supporting environment (Boston Children’s Hospital, Harvard system, and MIT) position him well to attain the proposed training goals and specific aims, and eventually lead to his

transition to independent surgeon-scientist. Combining machine-learning technology and real-life implementation to tackle the challenge and change the status quo by translating actionable ML results to the bedside is novel and innovative. The study is significant in that successful implementation of an algorithm for

UTI will be proof-of-concept to catalyze a similar approach to optimal clinical decision support for other conditions. This work has broad impact and can be scaled to other institutions and conditions, facilitating interactive improvements that empower clinicians and caregivers to meet the diverse clinical needs of children.