Interobserver variability in the detection of mediastinal and hilar lymph nodes on CT in children with suspected pulmonary tuberculosis

Abstract

Background: The presence of mediastinal or hilar adenopathy is critical for the diagnosis of pulmonary TB. Interobserver variability in the detection of lymphadenopathy on CT in children affects the usefulness of CT as a gold standard.

Objective: To determine the interobserver variability for the detection of hilar and mediastinal adenopathy on CT in children.

Materials and methods: One hundred children with clinically suspected pulmonary TB were prospectively recruited for CT scanning of the chest. Four observers reviewed the scans independently for the presence of lymphadenopathy at predetermined sites. Overall Kappa statistic was determined for each recognised site of mediastinal and hilar lymphadenopathy.

Results: Kappa statistics showed that observers only agreed moderately in their detection of lymphadenopathy. The site of best agreement was the right hilum, followed by the subcarinal, right paratracheal and precarinal locations. Observers differed most at the anterior mediastinum and left hilum. The best Kappa statistic was for the overall presence of lymphadenopathy taking all sites into account.

Conclusions: Imaging techniques that are considered the gold standard for particular diseases must be validated pathologically, and if this is not possible, interobserver variability should be evaluated. CT is considered the gold standard for detecting lymphadenopathy, but we have shown only moderate agreement between readers. Readers had difficulty in distinguishing lymphadenopathy from normal thymus and were unable to distinguish normal from pathological nodes without a predetermined size threshold for abnormality. The right hilum and the sites around the carina are the most reliable for the reported presence of lymphadenopathy.