Identification of avoidable patients at triage in a Paediatric Emergency Department: a decision support system using...

Area of Science:

• Emergency Medicine
• Health Informatics
• Machine Learning

Background:

• Emergency departments face significant overcrowding challenges.
• A fast-track system for non-urgent pediatric cases is being implemented.
• Optimizing patient flow is crucial for efficient emergency care.

Purpose of the Study:

• To develop an optimized Decision Support System (DSS) for directing patients to a fast-track pathway.
• To evaluate Machine Learning (ML) models for predicting avoidable emergency department visits.
• To balance predictive performance, model complexity, and interpretability.

Main Methods:

• Retrospective study of 507,708 pediatric emergency department visits (2014-2019).
• Trained and tested various ML models using triage information to predict visit avoidability.
• Compared model performance using metrics like Area Under the Curve (AUC).

Main Results:

• 41.6% of pediatric emergency visits were classified as avoidable.
• Most ML models achieved similar predictive performance (AUC 74%-80%), outperforming simple triage rules.
• A pruned decision tree demonstrated comparable results to more complex models.

Conclusions:

• The pruned decision tree offers a low-complexity, interpretable, and effective solution for identifying patients suitable for a fast-track system.
• ML-based DSS can significantly improve patient flow management in pediatric emergency departments.
• This research supports the practical application of machine learning in addressing emergency department overcrowding.