Named Entity Recognition in Prehospital Trauma Care

Area of Science:

• Emergency Medical Services (EMS)
• Natural Language Processing (NLP)
• Clinical Data Abstraction

Background:

• Prehospital Emergency Medical Services (EMS) data collection is crucial for quality improvement.
• Manual abstraction of clinically relevant information from unstructured EMS reports is time-consuming and prone to errors.

Purpose of the Study:

• To evaluate off-the-shelf Natural Language Processing (NLP) solutions for automating the labeling of clinically relevant data within EMS reports.
• To develop and validate a qualitative approach for selecting an optimal ensemble of pretrained NLP systems.

Main Methods:

• A qualitative methodology was employed to select the best ensemble of pretrained NLP systems.
• Word embeddings were utilized to create a feature for testing phrase synonymy.
• The performance of an ensemble of NLP systems was compared against individual systems.

Main Results:

• The developed ensemble of NLP systems demonstrated superior performance in data labeling compared to individual systems.
• The feature using word embeddings effectively tested for phrase synonymy, contributing to ensemble selection.

Conclusions:

• Automated labeling of clinical data from EMS reports using NLP offers significant potential for improving data abstraction processes.
• Ensemble methods in NLP can enhance the accuracy and efficiency of extracting valuable information from prehospital care data.