EUS-guided confocal laser endomicroscopy: prediction of dysplasia in intraductal papillary mucinous neoplasms (with video)

Background and Aims: Previous studies have validated EUS-guided needle-based confocal laser endomicroscopy (nCLE) diagnosis of intraductal papillary mucinous neoplasms (IPMNs). We sought to derive EUS-guided nCLE criteria for differentiating IPMNs with high-grade dysplasia/adenocarcinoma (HGD-Ca) from those with low/intermediate-grade dysplasia (LGD). Methods: We performed a post hoc analysis of consecutive IPMNs with a definitive diagnosis from a prospective study evaluating EUS-guided nCLE in the diagnosis of pancreatic cysts. Three internal endosonographers reviewed all nCLE videos for the patients and identified potential discriminatory EUS-guided nCLE variables to differentiate HGD-Ca from LGD IPMNs (phase 1). Next, an interobserver agreement (IOA) analysis of variables from phase 1 was performed among 6 blinded external nCLE experts (phase 2). Last, 7 blinded nCLE-naive observers underwent training and quantified variables with the highest IOA from phase 2 using dedicated software (phase 3). Results: Among 26 IPMNs (HGD-Ca in 16), the reference standard was surgical histopathology in 24 and cytology confirmation of metastatic liver lesions in 2 patients. EUS-guided nCLE characteristics of increased papillary epithelial "width" and "darkness" were the most sensitive variables (90%; 95% confidence interval [CI], 84%-94% and 91%; 95% CI, 85%-95%, respectively) and accurate (85%; 95% CI, 78%-90% and 84%; 95% CI, 77%-89%, respectively) with substantial (kappa = 0.61; 95% CI, 0.51-0.71) and moderate (kappa = 0.55; 95% CI, 0.45-0.65) IOAs for detecting HGD-Ca, respectively (phase 2). Logistic regression models were fit for the outcome of HGD-Ca as predictor variables (phase 3). For papillary width (cut-off >= 50 mu m), the sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) for detection of HGD-Ca were 87.5% (95% CI, 62%-99%), 100% (95% CI, 69%-100%), and 0.95, respectively. For papillary darkness (cut-off <= 90 pixel intensity), the sensitivity, specificity, and AUC for detection of HGD-Ca were 87.5% (95% CI, 62%-99%), 100% (95% CI, 69%-100%), and 0.90, respectively. Conclusions: In this derivation study, quantification of papillary epithelial width and darkness identified HGD-Ca in IPMNs with high accuracy. These quantifiable variables can be used in multicenter studies for risk stratification of IPMNs.