A Methodological Approach to Validate Pneumonia Encounters from Radiology Reports Using Natural Language Processing.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Pulmonology

Background:

• Pneumonia diagnosis relies on radiologist findings in unstructured electronic health records.
• Current diagnostic methods lack efficient structured data extraction for pneumonia features.

Purpose of the Study:

• To develop a validated methodological approach for pneumonia diagnosis using natural language processing (NLP).
• To convert unstructured radiology report data into structured formats for accurate pneumonia classification.

Main Methods:

• A pneumonia-specific NLP pipeline was created using Clinical Text Analysis and Knowledge Extraction System (cTAKES).
• A lexicon of pneumonia features was developed to parse radiology reports.
• Classification rules were applied to categorize pneumonia episodes as positive, negative, or requiring manual review.

Main Results:

• The NLP pipeline processed 81,707 chest X-ray reports for 91,998 pneumonia episodes.
• It classified 33% as Pneumonia-positive, 19% as Pneumonia-negative, and 48% required manual review.
• Performance metrics showed 76.3% accuracy, 88% sensitivity, and 75% specificity.

Conclusions:

• The developed NLP pipeline demonstrates good performance for validating pneumonia diagnoses.
• This approach is comparable to existing pneumonia-specific NLP systems.