Integrating structured and unstructured data for predicting emergency severity: an association and predictive study...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Emergency department (ED) triage is crucial for timely patient care.
• Traditional triage relies on structured data, but unstructured clinical notes offer potential for enhanced predictive modeling.
• This study evaluated combining structured and unstructured data for improved emergency severity prediction.

Purpose of the Study:

• To assess the effectiveness of integrating structured and unstructured data for predicting emergency severity.
• To explore associations between patient characteristics and emergency severity outcomes.
• To compare the performance of machine learning models using different data configurations.

Main Methods:

• Utilized 2021 National Hospital Ambulatory Medical Care Survey (NHAMCS) data for adult ED patients.
• Categorized emergency severity using the Emergency Severity Index (urgent: 1-3, non-urgent: 4-5).
• Processed unstructured data (chief complaints, reasons for visit) with a Bidirectional Encoder Representations from Transformers (BERT) model; applied Logistic Regression, Random Forest, Gradient Boosting, and Extreme Gradient Boosting to structured, unstructured, and combined data.

Main Results:

• The study included 8,716 adult patients; 74.6% were classified as urgent.
• Significant predictors of severity included older age, elevated heart rate, chronic kidney disease, and coronary artery disease.
• Gradient Boosting with combined data achieved the highest performance (AUC: 0.789, accuracy: 0.726, precision: 0.892).

Conclusions:

• Combining structured and unstructured data enhances emergency severity prediction in ED patients.
• Integrating text data into predictive models offers more accurate severity assessments, improving resource allocation and patient outcomes.
• Future research should focus on real-time application and validation in diverse clinical settings.