Diagnostic accuracy of the Canada Lymph Node Score for predicting malignant mediastinal lymphadenopathy in patients undergoing endobronchial ultrasound-guided transbronchial needle aspiration in a tuberculosis-endemic setting: A cross-sectional study.

BACKGROUND: Endoscopic ultrasound features predicting malignant mediastinal lymphadenopathy have been extensively studied. The Canada Lymph Node Score (CLNS) comprises four endoscopic ultrasound characteristics, at least three of which have been shown to predict malignancy in a Western population. Two of the CLNS variables are commonly described in tuberculous mediastinal lymphadenopathy as well, thus posing a diagnostic challenge. OBJECTIVES: To evaluate the diagnostic accuracy of the CLNS in predicting malignant mediastinal involvement (primary objective), and to study other endoscopic ultrasound variables to formulate a new predictive model (secondary objective). METHODS: In this prospective cross-sectional study from North India, consecutive patients undergoing endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) were included. Nodal endoscopic ultrasound features were recorded in real time before needle aspiration, and the CLNS was determined for each lymph node. The primary outcome was diagnostic accuracy of the CLNS, determined by the area under the curve (AUC). The secondary outcome was the identification of additional ultrasound predictors to develop a new model by logistic regression. RESULTS: A total of 132 nodes from 82 patients were sampled. Characteristics examined were short-axis diameter, central hilar structure, coagulation necrosis sign (CNS), margins, echogenicity, vascular patterns, conglomeration, shape and calcification. The CNS (odds ratio (OR) 4.28, p=0.005), heterogeneous echogenicity (OR 2.45, p=0.036) and round shape (OR 3.02, p=0.03) significantly predicted malignant involvement. The mean (standard deviation) AUC for the CLNS and the new model (comprising the CNS, round shape and heterogeneous echogenicity) was 0.72 (0.04) (95% confidence interval (CI) 0.635 - 0.814) and 0.74 (95% CI 0.635 - 0.844), respectively. CONCLUSION: The CLNS predicts malignant mediastinal involvement at a cut-off score of 3 in a tuberculosis-endemic setting. The proposed new model demonstrated comparable diagnostic performance to that of the CLNS for discriminating malignant from non-malignant nodes. STUDY SYNOPSIS: What the study adds. In a tuberculosis (TB)-endemic setting, the Canada Lymph Node Score (CLNS) maintains reasonable diagnostic accuracy for predicting malignant mediastinal lymph node involvement. A new model, comprising the coagulation necrosis sign, round shape and heterogeneous echotexture, shows equivalent diagnostic performance.Implications of the findings. The CLNS is a clinically valuable tool for differentiating malignant from non-malignant mediastinal nodes in TB-prevalent regions. However, overlapping endoscopic ultrasound features between granulomatous and malignant conditions may reduce its specificity.