Smart emergency care: a narrative review of predictive machine learning models

Area of Science:

• Emergency Medicine
• Artificial Intelligence
• Data Science

Background:

• The Emergency Department (ED) requires timely patient outcome assessments for optimal care and resource management.
• Machine learning (ML) offers advanced tools for predicting critical ED outcomes.

Purpose of the Study:

• To review ML-based predictive models for ED outcomes (mortality, ICU admission, discharge).
• To identify limitations and future research directions for ML in the ED.

Main Methods:

• Narrative review of ML models for ED outcomes (Jan 2015-Dec 2024).
• Analysis of ML techniques, data sources, and evaluation metrics.
• Screening of 156 studies, with 45 included.

Main Results:

• ML models achieve high predictive accuracy (AUC-ROC 0.75-0.95) for ED outcomes.
• Ensemble methods and neural networks show strong performance.
• Personalized and explainable AI (XAI) improve precision and interpretability.
• Limitations include data heterogeneity and poor generalizability.

Conclusions:

• ML is transforming ED predictive modeling and clinical decision-making.
• Personalized and explainable models enhance trust and usability.
• Further research needed on data quality, standardized metrics, and multi-center validation.