Interobserver variability in CT and MR staging of lung cancer

Our goal was to assess the interobserver variability in staging non-small cell lung cancer using CT and MRI.

As part of the Radiologic Diagnostic Oncology Group (RDOG) study of lung cancer staging, the CT and MR examinations of 40 patients suspected of having non-small cell bronchogenic carcinoma were blindly interpreted by four expert observers. The primary tumor and lymph node stages in the 40 study subjects were similar to the final proportions reported in the RDOG study. Assessed abnormalities included the presence of a lung nodule, chest wall invasion, mediastinal invasion, bronchial involvement, lymph node metastasis in specific node stations, and T and N classifications. Percent agreement and kappa-values were calculated for each of these determinations.

Depending on the finding assessed and the method of analysis, average agreement rates ranged from 58 to 90% for CT and from 61 to 96% for MRI. Average kappa-values were largely between 0.40 and 0.60 when dichotomous analysis was used; weighted kappa-values were similar. With a single exception, no significant differences were found for kappa-values calculated for CT and MRI.

Although interobserver agreement rates are good for determining T and N classification in patients with lung cancer, variability in image interpretation is frequent, even among experienced observers.