Machine Learning Tools for Predicting Pediatric Urinary Tract Infections Caused by ESBL-producing Bacteria

Area of Science:

• Pediatric infectious diseases
• Medical informatics
• Machine learning in healthcare

Background:

• Increasing global prevalence of pediatric urinary tract infections (UTIs) caused by extended-spectrum β-lactamases (ESBL)-producing bacteria.
• ESBL-UTIs necessitate specialized antibiotic treatment, often leading to delayed empirical therapy, increased ICU admissions, morbidity, and prolonged hospital stays.
• Predicting ESBL-UTIs is challenging but crucial for timely and appropriate patient management.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting pediatric UTIs caused by ESBL-producing bacteria.
• To assist pediatricians in identifying children at higher risk for ESBL-positive UTIs.
• To enable earlier initiation of appropriate empiric antibiotic therapy for ESBL-UTIs.

Main Methods:

• Retrospective analysis of electronic medical records for pediatric patients (1 month to 18 years) with confirmed UTIs from January 2010 to August 2020.
• Data extraction included demographics, clinical information, and laboratory results.
• Development of five ML models using available patient data at UTI presentation for predicting ESBL-positive bacterial infections.

Main Results:

• The study analyzed 35,830 pediatric UTI events.
• Factors significantly associated with ESBL-positive UTIs included age, sex, socioeconomic status, infection site, prior antibiotic use, previous ESBL-UTI history, and specific uropathogen.
• The developed ML models demonstrated a high negative predictive value (~0.98), indicating strong performance in ruling out ESBL-positive UTIs.

Conclusions:

• Machine learning models utilizing data available at UTI presentation can aid clinicians in assessing the probability of ESBL-producing bacterial UTIs in children.
• These models show promise in supporting clinical decision-making for empiric antibiotic selection.
• Further prospective studies are needed to refine model performance and evaluate their impact on clinical outcomes.