Near-focus narrow-band imaging classification of villous atrophy in suspected celiac disease: development and international validation.

BACKGROUND AND AIMS: There are no agreed-on endoscopic signs for the diagnosis of villous atrophy (VA) in celiac disease (CD), necessitating biopsy sampling for diagnosis. Here we evaluated the role of near-focus narrow-band imaging (NF-NBI) for the assessment of villous architecture in suspected CD with the development and further validation of a novel NF-NBI classification. METHODS: Patients with a clinical indication for duodenal biopsy sampling were prospectively recruited. Six paired NF white-light endoscopy (NF-WLE) and NF-NBI images with matched duodenal biopsy sampling including the bulb were obtained from each patient. Histopathology grading used the Marsh-Oberhuber classification. A modified Delphi process was performed on 498 images and video recordings by 3 endoscopists to define NF-NBI classifiers, resulting in a 3-descriptor classification: villous shape, vascularity, and crypt phenotype. Thirteen blinded endoscopists (5 expert, 8 nonexpert) then undertook a short training module on the proposed classification and evaluated paired NF-WLE-NF-NBI images. RESULTS: One hundred consecutive patients were enrolled (97 completed the study; 66 women; mean age, 51.2 ± 17.3 years). Thirteen endoscopists evaluated 50 paired NF-WLE and NF-NBI images each (24 biopsy-proven VAs). Interobserver agreement among all validators for the diagnosis of villous morphology using the NF-NBI classification was substantial (κ = .71) and moderate (κ = .46) with NF-WLE. Substantial agreement was observed between all 3 NF-NBI classification descriptors and histology (weighted κ = 0.72-.75) compared with NF-WLE to histology (κ = .34). A higher degree of confidence using NF-NBI was observed when assessing the duodenal bulb. CONCLUSIONS: We developed and validated a novel NF-NBI classification to reliably diagnose VA in suspected CD. There was utility for expert and nonexpert endoscopists alike, using readily available equipment and requiring minimal training. (Clinical trial registration number: NCT04349904.).