Use of Machine Learning to Predict Hospital Admission for EMS-Treated Infants After a Suspected BRUE

Area of Science:

• Pediatric Emergency Medicine
• Health Informatics
• Clinical Decision Support Systems

Background:

• Brief Resolved Unexplained Events (BRUE) are common in infants presenting to emergency medical services (EMS).
• Predicting hospital admission for infants with suspected BRUE is crucial for appropriate resource allocation and patient management.
• Current prediction methods may benefit from advanced analytical approaches.

Purpose of the Study:

• To evaluate the efficacy of various machine learning (ML) classification algorithms in predicting hospital admission for infants with suspected BRUE.
• To identify key clinical variables that influence the prediction of hospital admission in this population.
• To compare the performance of ML models against traditional statistical models.

Main Methods:

• Utilized data from a pediatric care system for infants with suspected BRUE transported by EMS (July 2017-February 2021).
• Applied a random forest model to identify significant predictors of hospital admission.
• Trained and evaluated multiple ML models (e.g., Support Vector Machine, Extreme Gradient Boosting, Logistic Regression) and a statistical model using selected variables.
• Assessed model performance using metrics including Area Under the Receiver Operator Curve (AUROC).

Main Results:

• A total of 508 infants were analyzed; 59% were admitted, and 15% required critical care.
• Key predictors for admission included infant age, bystander interventions, past medical history, and examination findings.
• The Support Vector Machine model achieved the highest AUROC of 0.85, demonstrating strong predictive capability.
• Other ML models (Extreme Gradient Boosting, Random Forest, Logistic Regression) showed comparable performance (AUROC 0.83-0.84).

Conclusions:

• Machine learning models exhibit robust predictive performance for hospital admission in infants with suspected BRUE treated by EMS.
• ML and statistical models demonstrate similar predictive accuracy.
• These findings support the integration of ML tools into clinical workflows for BRUE management.