Using Machine Learning Techniques to Predict Hospital Admission at the Emergency Department

Area of Science:

• Emergency Medicine
• Clinical Decision Support
• Biomarker Analysis

Background:

• Prompt identification of emergency department (ED) patients requiring hospital admission is critical.
• Machine learning (ML) shows potential as a diagnostic aid in healthcare settings.

Purpose of the Study:

• To develop and evaluate ML algorithms for predicting hospital admission in the emergency setting.
• To identify key features that assist in clinical decision-making for ED patient disposition.

Main Methods:

• Assessed performance of ML algorithms using features including serum biomarkers (Urea, Creatinine, LDH, CK, CRP, D-dimer), complete blood count, coagulation tests (aPTT, INR), age, gender, triage disposition, and ambulance utilization.
• Analyzed data from 3,204 emergency department visits.

Main Results:

• Developed ML models demonstrating acceptable performance in predicting hospital admission (F-measure: 0.679-0.708, ROC Area: 0.734-0.774).
• Highlighted advantages such as easy access, availability, binary outcomes, and low cost.
• Suggested a potential shift towards more sophisticated clinical decision-making models.

Conclusions:

• Robust prognostic models using common biomarkers can significantly impact the future of emergency medicine.
• Findings support further validation through pragmatic ED trials to confirm clinical utility.