Interobserver agreement and accuracy among international experts with probe-based confocal laser endomicroscopy in predicting colorectal neoplasia

Background and study aims: A recently developed probe-based, confocal laser endomicroscopy (pCLE) system provides images of surface colonic epithelium in vivo during any endoscopy. Our objective was to assess interobserver agreement, sensitivity, specificity, and overall accuracy in the diagnosis of neoplasia using pCLE. Patients and methods: 53 patients undergoing surveillance and screening colonoscopies were enrolled. A total of 75 lesions, were detected and all were inspected by pCLE prior to sampling or polypectomy. Intravenous fluorescein was used to optimize tissue contrast. Three pCLE users, blinded to histopathologic and endoscopic findings, reviewed the set of video sequences for crypt architecture, vessel architecture, and colorectal neoplasia diagnosis. Histopathologic diagnosis from the corresponding biopsies was the reference gold standard. Results: Of the 75 colorectal lesions, 50 were neoplastic and 25 non-neoplastic. Interobserver agreement was moderate to good for the classification of neoplasia (kappa 0.55, 78% pairwise agreement), and moderate for vessel architecture (kappa 0.41, 67% pairwise agreement) and crypt architecture (kappa 0.49, 69% pairwise agreement). In distinguishing between neoplastic and non-neoplastic lesions, sensitivity, specificity, and accuracy were 76%, 72% and 75%, respectively. When videos of good or excellent quality only were considered, interobserver agreement for classification of neoplasia was higher (kappa 0.83, 92% pairwise agreement), as were sensitivity (88%), specificity (89%), and accuracy (88%). Conclusion: An international collaboration group had moderate to good interobserver agreement using a pCLE system to predict neoplasia, which is acceptable for this study.