Rich Text Formatted EHR Narratives: A Hidden and Ignored Trove

Area of Science:

• Medical Informatics
• Natural Language Processing
• Data Mining

Background:

• Electronic Health Records (EHRs) contain rich clinical narratives.
• Current data mining often ignores structural information within Rich Text Format (RTF) files.
• RTF is widely used in medical information systems.

Purpose of the Study:

• To develop and evaluate an approach for extracting structured information from EHR narratives using RTF constructs.
• To investigate the prevalence and characteristics of RTF-specific features (tables, sections) in clinical notes.
• To demonstrate the added value of RTF-based features over traditional plain text Natural Language Processing (NLP).

Main Methods:

• Parsing RTF files to identify structural elements like tables and sections.
• Analyzing the occurrence and nature of these RTF constructs across diverse EHR narrative types.
• Comparing the performance of NLP models utilizing RTF features against those using only plain text.

Main Results:

• RTF constructs, particularly tables and sections, are present in various EHR narrative types.
• Parsing RTF enables the extraction of structural information not available in plain text.
• Features derived from RTF constructs significantly enhance the utility of NLP for clinical data mining.

Conclusions:

• Utilizing RTF structural information offers a novel and effective method for mining clinical narratives.
• This approach can unlock richer, more accurate data extraction from EHRs.
• The findings support the integration of RTF parsing into standard clinical data analysis workflows.