Generalized Extraction and Classification of Span-Level Clinical Phrases

Area of Science:

• Medical Informatics
• Natural Language Processing

Background:

• Electronic health records (EHRs) contain critical patient information primarily in unstructured text.
• Automated text extraction and summarization are crucial for making EHR data accessible and understandable.
• Existing clinical note text extraction methods are often narrowly focused or miss important contextual information by only extracting concepts.

Purpose of the Study:

• To develop and evaluate a novel phrase-level text extraction method for clinical notes.
• To improve the extraction of contextual information from unstructured EHR data.
• To provide concise yet contextually rich extracted clinical elements.

Main Methods:

• A three-stage pipeline approach was employed for text extraction.
• The method utilizes clinical concepts as anchor points to extract categorized phrases.
• Focuses on phrase-level extraction rather than solely concept-level extraction.

Main Results:

• The proposed method demonstrated performance comparable to individual human annotators.
• Successfully extracted categorized phrases with relevant contextual information.
• Achieved concise extractions while preserving essential context.

Conclusions:

• Phrase-level extraction offers a promising approach for enhancing EHR data accessibility.
• The developed pipeline effectively extracts clinically relevant information with high accuracy.
• This method has the potential to improve clinical decision-making by providing summarized, contextualized patient data.