Deep Learning Algorithm to Predict Need for Critical Care in Pediatric Emergency Departments

Area of Science:

• Artificial Intelligence in Medicine
• Pediatric Emergency Care
• Health Informatics

Background:

• Emergency department (ED) overcrowding is a significant challenge, particularly for pediatric patients who may be undertriaged.
• Accurate prognosis prediction for pediatric ED patients is crucial but complex.

Purpose of the Study:

• To develop and validate a deep learning (DL) algorithm for predicting critical care and hospitalization needs in pediatric ED patients.
• To compare the DL algorithm's performance against established scoring systems and machine learning models.

Main Methods:

• A retrospective cohort study utilized data from the Korean National Emergency Department Information System (2014-2016).
• Included 2,937,078 pediatric patients, with data split into derivation and testing sets.
• Predictors included demographics, chief complaint, symptom onset, arrival mode, trauma status, and vital signs.

Main Results:

• The DL algorithm achieved an area under the receiver operating characteristics curve (AUC) of 0.908 for critical care prediction, significantly outperforming other methods.
• For hospitalization prediction, the DL algorithm's AUC was 0.782, also demonstrating superior performance.
• The DL algorithm significantly outperformed the pediatric early warning score, conventional triage systems, random forest, and logistic regression models.

Conclusions:

• The developed deep learning algorithm offers superior accuracy in predicting critical care and hospitalization for pediatric ED patients.
• This AI-driven approach holds promise for improving triage and resource allocation in overcrowded pediatric emergency departments.