Multicenter validation of a deep-learning-based pediatric early-warning system for prediction of deterioration events

Area of Science:

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

Background:

• Early recognition of patient deterioration is vital for improving clinical outcomes in pediatric care.
• A deep-learning-based pediatric early-warning system (pDEWS) was developed to address this need.
• Validation of the pDEWS's clinical performance was the primary objective.

Purpose of the Study:

• To validate the clinical performance of a novel deep-learning-based pediatric early-warning system (pDEWS).
• To assess the pDEWS's ability to identify critical deterioration events in pediatric patients.

Main Methods:

• A retrospective multicenter cohort study included 28,758 pediatric patients from five academic children's hospitals.
• Electronic medical records were used to evaluate pDEWS performance in predicting in-hospital cardiac arrest (IHCA) and unexpected intensive care unit transfers (UIT).
• pDEWS performance was compared against the modified pediatric early-warning score (PEWS), logistic regression (LR), and random forest (RF) models.

Main Results:

• The pDEWS demonstrated superior performance in predicting deterioration events compared to modified PEWS, LR, and RF models.
• Key performance indicators included a larger area under the receiver operating characteristic curve and fewer false alarms.
• The pDEWS showed a lower mean alarm count per day and required examining fewer cases.

Conclusions:

• The pDEWS significantly outperformed existing models in the early and accurate prediction of pediatric deterioration events.
• These findings highlight the potential of pDEWS as an effective screening tool for optimizing rapid response team operations.
• The system's robust performance was consistent across various clinical situations, patient demographics, and event timing.