Optical Evaluation for Predicting Cancer in Large Nonpedunculated Colorectal Polyps Is Accurate for Flat Lesions.

Background and Aims: The ability of optical evaluation to diagnose submucosal invasive cancer (SMIC) prior to endoscopic resection of large (≥20 mm) nonpedunculated colorectal polyps (LNPCPs) is critical to inform therapeutic decisions. Prior studies suggest that it is insufficiently accurate to detect SMIC. It is unknown whether lesion morphology influences optical evaluation performance.
Methods: LNPCPs ≥20 mm referred for endoscopic resection within a prospective, multicenter, observational cohort were evaluated. Optical evaluation was performed prior to endoscopic resection with the optical prediction of SMIC based on established features (Kudo V pit pattern, depressed morphology, rigidity/fixation, ulceration). Optical evaluation performance outcomes were calculated. Outcomes were reported by dominant morphology: nodular (Paris 0-Is/0-IIa+Is) vs flat (Paris 0-IIa/0-IIb) morphology.
Results: From July 2013 to July 2019, 1583 LNPCPs (median size 35 [interquartile range, 25-50] mm; 855 flat, 728 nodular) were assessed. SMIC was identified in 146 (9.2%; 95% confidence interval [CI], 7.9%-10.8%). Overall sensitivity and specificity were 67.1% (95% CI, 59.2%-74.2%) and 95.1% (95% CI, 93.9%-96.1%), respectively. The overall SMIC miss rate was 3.0% (95% CI, 2.3%-4.0%). Significant differences in sensitivity (90.9% vs 52.7%), specificity (96.3% vs 93.7%), and SMIC miss rate (0.6% vs 5.9%) between flat and nodular LNPCPs were identified (all P < .027). Multiple logistic regression identified size ≥40 mm (odds ratio [OR], 2.0; 95% CI, 1.0-3.8), rectosigmoid location (OR, 2.0; 95% CI, 1.1-3.7), and nodular morphology (OR, 7.2; 95% CI, 2.8-18.9) as predictors of missed SMIC (all P < .039).
Conclusions: Optical evaluation performance is dependent on lesion morphology. In the absence of features suggestive of SMIC, flat lesions can be presumed benign and be managed accordingly.
(Copyright © 2021 AGA Institute. Published by Elsevier Inc. All rights reserved.)