Background and aims:
Narrow-band imaging (NBI) international colorectal endoscopic (NICE) classification has been validated for differentiating hyperplastic from adenomatous polyps. This classification system was based on NBI technology, leaving uncertainty on its applicability to other systems. The aim of this study was to assess accuracy and reliability of histological predictions for subcentimetric polyps by applying NICE classification to Fujinon spectral Imaging Color Enhancement (FICE) system.
A video-library of 55-subcentimeter histologically verified polyps with FICE was prospectively created including polyps that fulfilled inclusion criteria (morphology, size, histology) in consecutive colonoscopies. Six endoscopists with experience in electronic chromoendoscopy independently reviewed the polyps images, scored the polyps as adenoma or hyperplastic, and assigned a level of confidence to the predictions; 20 videos were reassessed at 6 months. The diagnostic performances of the endoscopists was calculated combined and individually according to the histopathology of the polyps. A mixed-effect logistic regression model in which polyps were considered as random-effects and polyp histology, confidence level, readers as fixed-effect was used. Results were expressed as odds ratios (ORs) with 95% confidence intervals (CIs).
Of the 55 polyps (mean size 4.6 mm), 29 (53%) were adenomas and 26 (47%) hyperplastic. Across all the readers and observations, the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and area under the curve (AUC) were 77%, 75%, 88%, 75%, 77%, and 0.82, respectively. Individual rater accuracy ranged from 66% to 96%, being <90% in 5 out of 6 cases. Overall, 68.5% (226 of 330) of predictions were made with high confidence, although there was high variability (Fleiss kappa, 0.15; 95% CI, 0.08-0.22). Sensitivity, specificity, PPV, NPV, accuracy, and AUC for predictions made with high confidence were 81%, 80.5%, 80%, 77%, 82%, and 0.88, being significantly more accurate as compared with a low-confidence of diagnosis (OR, 2.4; 95% CI, 1.2-4.7). Regarding the performance of the individual NICE-criteria, the odds of adenoma detection were 3.4 (95% CI, 1.8-6.3) and 4.0 (95% CI, 2.1-7.5) by using surface and vessels pattern alone, as compared with the color criterion. Inter- and intra-raters agreement with the NICE was only moderate (inter-:Fleiss kappa, 0.51; 95% CI, 0.44-0.56; intra-: kappa, 0.40; 95% CI, 0.20-0.60).
The application of NICE classification to FICE resulted in suboptimal accuracy and only moderate inter-observer agreement.