Prospective Development and Validation of a Malignancy Scoring System During Endobronchial Ultrasound Evaluation of Mediastinal Lymph Nodes for Lung and Esophageal Cancer

Abstract

Background: At the time of endobronchial ultrasound (EBUS) staging, ultrasonographic features can be used to predict mediastinal lymph node (LN) malignancy. Predictive tools have been developed, however they have not gained widespread use due to lack of research demonstrating validity and reliability. We sought to develop a novel predictive tool, the Canada Score, capable of predicting malignancy and potentially guide LN biopsy decision making. Methods: We prospectively analyzed the ultrasonographic features of LNs from patients with NSCLC. Ultrasonographic features were identified by a single experienced endoscopist, this data was used to develop the Canada Score. Pathological specimens were used as the gold standard for determination of malignancy. Videos were then circulated to endoscopists across Canada, who were also asked to identify ultrasonographic features for each LN. Hosmer- Lemeshow test, logistic regression, receiver operator characteristic (ROC) curve, and Gwet’s AC1 analyses were used to test the performance, discriminatory capacity, and inter-rater reliability of the Canada Score. Results: A total of 300 LNs from 140 patients were analyzed by 12 endoscopists across 7 Canadian centres. Backwards elimination was used to create a multivariate model. Hosmer-Lemeshow test and ROC curves indicated the model was well-calibrated (chi2=11.86, p=0.1567) with good discriminatory power (c- statistic= 0.72 ±0.042, 95%CI: 0.64-0.80). Beta-coefficients were used to create a simplified score out of four. Evaluation of the tool showed that LNs scoring 3 or 4 had odds ratios of 15.17 (p<0.0001) and 50.56 (p=0.001), respectively for predicting malignancy. A score of 4/4 was associated with 99.59% specificity and a positive likelihood ratio of 22.78. Inter-rater reliability for a score ≥ 3 was 0.81 ± 0.02 (95%CI: 0.77-0.85). Conclusions: The Canada Score shows excellent performance in identifying malignant LN at the time of EBUS. A cut-off of ≥ 3 has the potential to inform decision-making regarding biopsy or repeat/mediastinoscopy if the initial results are inconclusive.