Development and validation of a machine learning framework for improved resource allocation in the emergency...

Area of Science:

• Emergency Medicine
• Artificial Intelligence
• Health Informatics

Background:

• The Emergency Severity Index (ESI) is widely used for triage in US emergency departments (EDs), but mistriage affects 32.2% of visits.
• Accurate prediction of patient resource needs is crucial for efficient ED operations and patient flow.

Purpose of the Study:

• To develop a machine learning framework for predicting patient resource requirements during emergency department triage.
• To enhance resource allocation and mitigate mistriage in emergency departments.

Main Methods:

• Retrospective analysis of 391,472 ED visits using the MIMIC-IV database.
• Development of 144 machine learning models using Azure AutoML to predict laboratory tests, imaging, and medications.
• Evaluation of model performance using AUROC, F1 score, accuracy, precision, and recall.

Main Results:

• Machine learning models achieved an average AUROC of 0.82 and accuracy of 0.76.
• Chief complaint was identified as a significant predictor across various resource needs.
• The framework demonstrated high accuracy in predicting patient resource utilization.

Conclusions:

• Machine learning models can accurately predict patient resource needs in the ED.
• This predictive capability can significantly improve patient flow and resource allocation in resource-constrained emergency departments.
• The developed framework offers a promising approach to optimize emergency care delivery.