Natural language processing accurately categorizes findings from colonoscopy and pathology reports

Area of Science:

• Gastroenterology
• Medical Informatics
• Natural Language Processing

Background:

• Limited understanding of natural language processing (NLP) capabilities for extracting data from free-text gastroenterology reports for secondary use.
• Need for automated methods to analyze clinical data in gastroenterology.

Purpose of the Study:

• To assess the accuracy of an open-source NLP engine in extracting clinically relevant concepts from linked colonoscopy and pathology reports.
• To determine NLP's potential for quantifying quality metrics in gastroenterology.

Main Methods:

• Random selection of 500 linked colonoscopy and pathology reports from 10,798 nonsurveillance colonoscopies.
• Training and testing an NLP system using gastroenterologist annotations as the reference standard.
• Assessing accuracy for highest pathology level, lesion location, adenoma size, and adenoma count.

Main Results:

• The NLP system achieved 98% accuracy in identifying the highest level of pathology.
• Accuracy for lesion location and adenoma size extraction was 97% and 96%, respectively.
• The system demonstrated 84% accuracy in extracting the number of adenomas removed.

Conclusions:

• NLP can accurately extract specific, meaningful concepts from gastroenterology reports with 98% accuracy.
• NLP shows potential as a tool for quantifying specific quality metrics in gastroenterology.
• Automated data extraction from clinical reports can support quality assessment.