Predicting Laboratory Test Ordering in Emergency Departments Using Integrated Structured and Unstructured Electronic...

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Clinical Decision Support

Background:

• Laboratory testing is crucial for emergency department (ED) diagnosis but often overused, increasing healthcare costs.
• Electronic health record (EHR) data offers potential for predictive models to optimize test utilization.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting laboratory test use in ED visits.
• To integrate structured and unstructured EHR data for enhanced predictive accuracy.

Main Methods:

• Analysis of 13,115 adult ED visits from the 2021 National Hospital Ambulatory Medical Care Survey-Emergency Department dataset.
• Utilized structured data (demographics, vital signs, history) and unstructured data (chief complaints, injury descriptions) encoded with Bidirectional Encoder Representations from Transformers (BERT).
• Compared four ML model configurations: structured-only, unstructured-only, combined data, and ensemble, evaluating performance using the area under the receiver operating characteristic curve (AUC).

Main Results:

• The combined model, integrating both structured and unstructured data, achieved the highest predictive performance with an AUC of 0.83.
• The combined model outperformed structured-only (AUC=0.78) and unstructured-only (AUC=0.74) models.
• Key predictors included older age, ambulance arrival, abnormal vital signs, and chronic comorbidities; injury-related visits showed lower testing likelihood.

Conclusions:

• Integrating structured and unstructured EHR data significantly improves prediction of laboratory test use in EDs.
• Findings support developing data-driven clinical decision support tools to enhance diagnostic efficiency.
• This approach can help reduce unnecessary laboratory testing in emergency settings.