Interobserver agreement on the endosonographic features of lymph nodes in aerodigestive malignancies

Background: Endoscopic ultrasound (EUS) is used to locally stage aerodigestive malignancies. Endosonographic features are used to predict malignant lymph nodes (MLN). Interobserver agreement on the endosonographic features of MLN has not been described. Aims: To evaluate the interobserver agreement among endosonographers on the EUS features of MLN in aerodigestive malignancies along with the specific feature used to make this distinction. Methods: A total of 760 procedures of suspected LN were reviewed. Cases were selected based on cytology-proven malignant or benign LN involvement by FNA with correspondent photodocumentation. Images of each LN were de-identified and distributed to three expert endosonographers. The experts recorded the LN's echogenicity (hypoechoic or other), shape (round or other), border (sharp or fuzzy) and subjective diagnosis (benign or malignant). The relationship between the endosonographers' subjective as well as the pathological diagnosis and LN's endosonographic features were analyzed using logistic regression analysis. Pair-wise comparison between endoscopist and interobserver agreement (kappa statistics) were performed. Results: Images of 41 malignant and 35 benign LN were evaluated. There was fair agreement on shape, Κ = 0.35 (95% CI 0.2-0.5), and moderate agreement on echogenicity and borders, Κ = 0.46 (95% CI 0.31-0.61) and 0.43 (95% CI 0.27-0.58) respectively. The agreement on malignant LN was good, Κ = 0.65 (95% CI 0.5-0.8). The overall diagnostic predictive accuracy ranged from 70 to 77% among the three endoscopists.Two of the three endoscopists assessed shape as the most predictive feature of malignancy (OR 39.4, 95% CI 3.29-470.96). Conclusion: The inter-observer agreement on the individual lymph node features as obtained by EUS is moderate with good overall agreement on the diagnosis. Round shape was the feature most strongly associated with a diagnosis of MLN.