Natural Language Processing for Assessing Quality Indicators in Free-Text Colonoscopy and Pathology Reports:...

Area of Science:

• Medical Informatics
• Gastroenterology
• Natural Language Processing

Background:

• Manual extraction of colonoscopy quality indicators is labor-intensive.
• Natural Language Processing (NLP) offers automated information extraction from clinical reports.
• NLP can enhance quality control and patient management in healthcare.

Purpose of the Study:

• Develop and evaluate an NLP pipeline for analyzing colonoscopy and pathology reports.
• Assess the pipeline's ability to automatically determine adenoma detection rate (ADR), sessile serrated lesion detection rate (SDR), and surveillance intervals.
• Utilize NLP for large-scale analysis of colonoscopy quality indicators.

Main Methods:

• Developed an NLP tool using 2000 colonoscopy and 1425 pathology reports.
• Tested the NLP system on 1000 colonoscopy reports, comparing its performance to 5 human annotators.
• Analyzed data from 54,562 colonoscopies (2010-2019) using the NLP pipeline.

Main Results:

• The NLP pipeline achieved high accuracy (0.98-1.00) in identifying polyp subtypes, locations, and counts.
• NLP performance for assessing ADR, SDR, and surveillance intervals was comparable to clinical experts.
• Analysis revealed significant individual variations in colonoscopy quality indicators among endoscopists over a 10-year period.

Conclusions:

• The NLP pipeline accurately extracts data from colonoscopy and pathology reports, proving clinically effective for assessing ADR, SDR, and surveillance intervals.
• Automated analysis and feedback on quality indicators can motivate endoscopists to improve performance.
• This NLP system supports enhanced clinical decision-making in colorectal cancer screening programs.