Interobserver Variation in the Computed Tomographic Evaluation of Mediastinal Lymph Node Size in Patients With Potentially Resectable Lung Cancer

OBJECTIVE: To measure the reliability of the assessment of mediastinal lymph node size in computed tomographic (CT) scans of the thorax. DESIGN: Observer agreement study in which radiologists, blinded to one anothers' interpretation, were randomized to read 30 scans each. POPULATION: Sixty scans from patients with apparently operable non-small cell carcinoma of the lung were read by radiologists responsible for clinical interpretation (clinical radiologists) and four radiologists with a special interest in thoracic CT (study radiologists). MEASUREMENTS: Radiologists measured the size of left and right superior mediastinal nodes, aortic nodes, and the subcarinal nodes and, on the basis of whether any nodes accessible to mediastinoscopy were greater than 1 cm, recommended whether mediastinoscopy be undertaken. Agreement was quantified using kappa, a measure of chance-corrected agreement. RESULTS: Among all radiologists, agreement on whether there were any nodes larger than 1 cm for right superior mediastinal nodes was 0.68; for left superior mediastinal nodes it was 0.28; for aortic pulmonary nodes it was 0.62; for subcarinal nodes it was 0.58; and for any node greater than 1 cm and accessible to mediastinoscopy it was 0.61. The agreement was very similar when the analysis was restricted to the study radiologists. CONCLUSION: Although the good level of overall agreement suggests that CT provides useful information in the evaluation of mediastinal lymph node size, the disagreement was sufficient that it likely contributes to suboptimal sensitivity and specificity of CT in detecting tumor spread reported in previous studies.