Accuracy and Inter-observer Agreement Among Endoscopists for Visual Identification of Colorectal Polyps Using Endoscopy Images

Background: It is essential to accurately distinguish small benign hyperplastic colon polyps (HP) from sessile serrated lesions (SSL) or adenomatous polyps (TA) based on endoscopic appearances. Our objective was to determine the accuracy and inter-observer agreements for the endoscopic diagnosis of small polyps.

Methods: High-quality endoscopic images of 30 small HPs, SSLs, and TAs were used randomly to create two-timed PowerPoint slide sets-one with and another one without information on polyp size and location. Seven endoscopists viewed the slides on two separate occasions 90 days apart, identified the polyp type, and graded their confidence level. Overall and polyp-specific accuracies were assessed for the group and individual endoscopists. Chi-square tests and Kappa (κ) statistics were used to compare differences as appropriate.

Results: When polyp size and location were provided, overall accuracy was 67.1% for TAs, 50.0% for SSLs, and 41.4% for HPs; the corresponding accuracies were 60%, 44.3%, and 34.3% when polyp size and location were withheld (p < .001). Inter-observer agreement was moderate for TAs (κ = 0.50) and fair for SSLs (κ = 0.26) and HPs (κ = 0.29); the corresponding inter-observer agreements were 0.44, 0.31, and 0.17 with polyp size and location withheld. Accuracy was not affected by knowledge of polyp size, location, or confidence level. Endoscopists with ≥ 10 years (vs. < 10 years) of colonoscopy experience had marginally higher (56% vs. 40%, p = 0.05) accuracy for SSL diagnosis.

Conclusions: The ability to distinguish between small TAs, SSLs, and HPs on their endoscopic appearance is poor regardless of the endoscopists' knowledge of polyp size and location.