Development and Validation of a Classification System to Identify High-Grade Dysplasia and Esophageal Adenocarcinoma in Barrett's Esophagus Using Narrow-Band Imaging

摘要

BACKGROUND & AIMS: Although several classification systems have been proposed for characterization of Barrett's esophagus (BE) surface patterns based on narrow-band imaging (NBI), none have been widely accepted. The Barrett's International NBI Group (BING) aimed to develop and validate an NBI classification system for identification of dysplasia and cancer in patients with BE. METHODS: The BING working group, composed of NBI experts from the United States, Europe, and Japan, met to develop a validated, consensus-driven NBI classification system for identifying dysplasia and cancer in BE. The group reviewed 60 NBI images of nondysplastic BE, high-grade dysplasia, and esophageal adenocarcinoma to characterize mucosal and vascular patterns visible by NBI; these features were used to develop the BING criteria. We then recruited adult patients undergoing surveillance or endoscopic treatment for BE at 4 institutions in the United States and Europe, obtaining high-quality NBI images and performing histologic analysis of biopsies. Experts individually reviewed 50 NBI images to validate the BING criteria, and then evaluated 120 additional NBI images (not previously viewed) to determine whether the criteria accurately predicted the histology results. RESULTS: The BING criteria identified patients with dysplasia with 85% overall accuracy, 80% sensitivity, 88% specificity, 81% positive predictive value, and 88% negative predictive value. When dysplasia was identified with a high level of confidence, these values were 92%, 91%, 93%, 89%, and 95%, respectively. The overall strength of inter-observer agreement was substantial (kappa = 0.681). CONCLUSIONS: The BING working group developed a simple, internally validated system to identify dysplasia and EAC in patients with BE based on NBI results. When images are assessed with a high degree of confidence, the system can classify BE with >90% accuracy and a high level of inter-observer agreement.