An Innovative Model to Predict Pediatric Emergency Department Return Visits

Area of Science:

• Pediatric Emergency Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Return visits (RVs) to the emergency department (ED) are a key indicator of healthcare quality.
• Predicting early RVs in pediatric populations is crucial for resource allocation and patient care.

Purpose of the Study:

• To develop and compare machine learning models for predicting early readmission risk in pediatric EDs.
• To identify key factors influencing pediatric ED return visits.

Main Methods:

• Retrospective study of pediatric patients (<15 years) with ED visits within 120 hours post-discharge.
• Development and comparison of Artificial Neural Network (ANN) and Classification Tree (CT) predictive models.
• Assessment of model performance using accuracy, sensitivity, and specificity.

Main Results:

• A Classification Tree (CT) model demonstrated superior sensitivity (79.8%) in predicting pediatric ED return visits compared to Artificial Neural Network (ANN) (6.9%).
• While ANN had higher overall accuracy (91.3%), CT's specificity was comparable (97% vs. 98.3%).
• Key predictors for RVs included time of arrival/discharge, triage priority, age, and diagnosis.

Conclusions:

• Classification Tree models offer a promising tool for identifying pediatric patients at high risk of early return visits to the ED.
• These predictive models can support ED staff in implementing strategies to prevent unnecessary readmissions.
• The findings highlight the potential of machine learning in enhancing the quality of emergency care for children.