Background: The 2 main types of colon polyps are adenomas and hyperplastic. Pit patterns on the surface of polyps have been described by using magnification chromoendoscopy, which can help differentiate between polyp types. Narrow band imaging (NBI) is a novel technology that enhances the visualization of surface mucosal and vascular patterns on the polyp surface. Earlier we described, in a pilot study, patterns seen on the polyp surface with NBI that can help differentiate between adenomas and hyperplastic polyps with a high degree of accuracy.
Objective: The aim of this study was to evaluate the interobserver and intraobserver agreement (among endoscopists) for the NBI surface mucosal and vascular patterns and prediction of polyp histology and the accuracy of the investigators to predict polyp histology based on these patterns.
Setting: Kansas City Veterans Affairs Medical Center.
Methods: NBI images of the polyp surface mucosal and vascular patterns obtained in our pilot trial were retrieved. A teaching set of 20 images was selected to educate and demonstrate the polyp patterns to 4 endoscopists. Subsequently, the test set of images was evaluated by the 4 endoscopists for quality, polyp pattern, and prediction of polyp type. Interobserver agreement (k value) was calculated among the 4 assessors for the polyp patterns and predicted histology. By using the final histology as the criterion standard, the accuracy of polyp-type prediction was calculated for each assessor. After a period of 2 months, all polyp images were reevaluated by the assessors (as before), and all findings were recorded in a similar fashion. These results were used for calculation of intraobserver agreement (k value) and the accuracy of the assessors in predicting polyp type.
Results: Photographs of 65 polyps were included in the test set and were evaluated by the 4 assessors. Thirty-eight polyps were adenomatous, and 27 were hyperplastic. The kappa value for the interobserver agreement for polyp surface pattern was 0.57 (moderate) and for prediction of polyp type was 0.63 (substantial). The kappa value for the intraobserver agreement of the 4 assessors for the surface patterns was 0.70, 0.65, 0.60, and 0.79, and for the prediction of polyp type was 0.87, 0.71, 0.61, 0.81. The accuracy to predict polyp type ranged from 80% to 86% for the 4 assessors in the first reading and from 85% to 91% in the second reading, with every assessor showing an improvement in accuracy in the second reading.
Limitations: A single-center study, with a limited number of polyps.
Conclusions: This initial evaluation showed that the NBI polyp patterns described in our pilot study are reproducible, easy to learn, reasonably accurate, and have the potential for use in daily clinical practice for the real-time differentiation of colon polyps.