Interobserver agreement for confocal imaging of ampullary lesions: A multicenter single-blinded study

BACKGROUND:: Malignant ampullary lesions can be difficult to classify by endoscopy alone. Probe-based confocal laser endomicroscopy (pCLE) permits in vivo assessment of mucosal structures in the gastrointestinal tracts in the real time. AIM:: The objective of this pilot multicenter study was to assess the interobserver agreement and variance in interpretation of pCLE of ampullary lesions. METHODS:: Twelve pCLE video clips of ampullary lesions were distributed to 6 gastrointestinal specialists at 5 medical centers, blinded to final pathologic results. Six variables were assessed for interobserver agreement using κ statistics. Variables included an epithelial outer border with irregular thickness, dark epithelium without discernable individual cells, heterogenously distributed elongated crypts, reduced number of goblet cells, neovascularization, and final diagnosis. RESULTS:: The overall interobserver agreement for all observers was poor to slight for all variables (κ=0.02, 0.05, -0.01, 0.04, 0.018) except for the first variable with fair degree of agreement (κ=0.27). On the basis of experience, 3 observers were classified as less experienced, whereas 3 were classified as most experienced. Upon stratification, the less experienced observers had poor interobserver agreement for all variables, except 1. The most experienced observers had poor agreement for 2 variables, slight agreement for 3 variables, and fair agreement for the final diagnosis variable. CONCLUSIONS:: The overall interpersonal agreement on pCLE for ampullary lesions was poor. The interobserver agreement was not substantially improved for experienced raters. Further standardization of pCLE image criteria is needed for ampullary lesions. Standardized training may improve interrater reliability to an acceptable level.