Refinement and Validation of a Clinical-Based Approach to Evaluate Young Febrile Infants

Area of Science:

• Pediatric infectious disease
• Clinical decision support systems
• Machine learning in healthcare

Background:

• Existing predictive models for bacterial infections in febrile infants face limited clinical adoption due to implementation barriers.
• There is a critical need for predictive models that utilize widely accessible demographic, clinical, and urine study data.
• Previous work derived novel machine learning and regression models using these factors.

Purpose of the Study:

• To refine previously derived predictive models for bacterial infections in febrile infants.
• To externally validate these refined models in a separate cohort.
• To assess the models' ability to detect serious bacterial infections, including urinary tract infection, bacteremia, and meningitis.

Main Methods:

• A cross-sectional study of 1419 febrile infants (age 0-90 days) evaluated at a pediatric emergency department (2011-2018).
• Models were re-derived excluding insurance status to minimize bias and then tested on a validation sample.
• Performance was evaluated using area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity.

Main Results:

• The refined machine learning and regression models achieved AUROCs of 0.92 and 0.90, respectively, in the validation sample.
• Sensitivities for detecting bacterial infections were high (98.0% for ML, 96.0% for regression), with specificities around 50-54%.
• Model performance was comparable to the initial derivation study.

Conclusions:

• The refined clinical-based predictive models demonstrate consistent performance in estimating bacterial infection risk in febrile infants.
• These models hold potential for aiding clinical judgment in identifying infants with bacterial infections.
• Future research should focus on prospective testing and strategies to enhance clinical implementation of these models.