Interobserver Agreement in the Classification of Pancreatic Cysts With Needle-Based Confocal Laser Endomicroscopy

Abstract

Introduction: Needle-based confocal laser endomicroscopy (nCLE) may serve as an accurate and clinically relevant adjunct in classifying pancreatic cysts. The question remains whether clinician interpretation of nCLE images is consistent enough to render nCLE a precise tool. The aim of this study was to determine our institutional interobserver agreement for interpretation of nCLE images of pancreatic cysts. Methods: All pancreatic cyst EUS-FNA cases with nCLE from 3/2014-4/2015 at a large tertiary medical center were included. 3 interventional endoscopists who independently had performed the cases reviewed all nCLE images in a blinded manner and classified each case as serous cystadenoma (SCA), intraductal papillary mucinous neoplasm (IPMN), neuroendocrine tumor (NET), or other. Each clinician designated diagnostic certainty (certain/uncertain) and image quality (poor/fair/good). The primary outcome was interobserver agreement in pancreatic cyst classification, as calculated by the Fleiss kappa in SAS 9.3. Secondary outcomes were accuracy of individual and group diagnosis, as compared to overall cyst clinical impression as previously determined by the performing endoscopist, based upon radiology studies, EUS, nCLE, CEA/amylase, cytology and surgical pathology.Table 1: Interobserver Agreement in Pancreatic Cyst Evaluation with nCLEResults: 14 cases were included. Interobserver agreement was “fair to good” (kappa=0.65, p=0.003). nCLE diagnosis by the performing endoscopist was accurate in 11/14 cases. Blinded accuracy was 43%, 50%, and 57%, respectively. Majority consensus, with at least 2 of 3 clinicians independently achieving the same diagnosis, occurred in 12/14 cases. Consensus accuracy was 58% and increased to 67% in cases of unanimous agreement. There was little correlation between consensus diagnostic accuracy and certainty. In all instances of unanimous agreement, regardless of diagnostic accuracy, there was high level of certainty. Image quality appeared to have limited impact on achieving consensus or the accuracy of consensus diagnosis. Conclusion: Our prior work has shown that nCLE enhances accuracy in classifying pancreatic cysts and is clinically relevant in preventing referral for surgical resection. Interobserver agreement is high demonstrating the reliability of this tool, yet both individual and group diagnostic accuracy fell short of that of the performing endoscopist, indicating that clinical context is vital to nCLE's utility. Clinicians would benefit from greater training with nCLE to improve the yield of this new technology.