Development and validation of a deep-learning-based pediatric early warning system: A single-center study.

Area of Science:

• Pediatric critical care medicine
• Artificial intelligence in healthcare
• Clinical informatics

Background:

• Early detection of clinical deterioration is crucial for improving patient outcomes.
• Rapid response teams (RRTs) and early warning scores are established methods for intervention.
• Existing systems require enhancement for greater accuracy and efficiency.

Purpose of the Study:

• To develop and validate a deep-learning-based pediatric early warning system (pDEWS).
• To assess the performance of pDEWS in predicting critical events in pediatric patients.
• To compare pDEWS against existing early warning scores and predictive models.

Main Methods:

• A retrospective observational cohort study of 50,019 pediatric patients.
• Development of a pDEWS using deep learning techniques.
• Validation by comparing pDEWS with modified pediatric early warning score (PEWS), random forest (RF), and logistic regression (LR) models.

Main Results:

• pDEWS demonstrated superior performance in predicting cardiopulmonary arrest (AUROC 0.923 vs. 0.769 for modified PEWS).
• pDEWS significantly reduced alarm burden (82.0% reduction in MACPD) and improved efficiency (89.5% reduction in NNE).
• pDEWS also showed superior performance in predicting unexpected ward-to-PICU transfers compared to existing methods.

Conclusions:

• The developed pDEWS is a highly effective tool for early detection of critical events in pediatric patients.
• Integration of pDEWS into RRTs can enhance operational efficiency.
• pDEWS has the potential to significantly improve clinical outcomes in pediatric care.