Natural Language Processing Calculated ADR

The adenoma detection rate (ADR) is the best measure of the quality of mucosal inspection during colonoscopy. Calculation of ADR typically requires manual entry of pathology data into endoscopy databases, as systems that automatically link pathology reports to endoscopy databases are typically unavailable.

Natural language processing (NLP) uses computer-based artificial intelligence to extract information from text reports. In the current study, investigators tested the accuracy of NLP in calculating a variety of polyp detection measures across 13 Veterans Administration hospitals, using expert review as the gold standard.

The accuracy of NLP was 99.6% for calculating colorectal cancer detection, 95.0% for advanced adenomas, 94.6% for nonadvanced adenomas, 99.8% for advanced sessile serrated polyps, 99.2% for nonadvanced sessile serrated polyps, and 96.8% for hyperplastic polyps ≥10 mm. The accuracy of NLP for calculating the number of adenomas resected was 90.2%.