Detection of lung cancer on the chest radiograph: a study on observer performance

Study objectives: to study the validity and observers consistency in the detection of lung cancer on the chest radiograph. Materials and methods: the chest radiographs of 100 clinical cases were interpreted by 14 observers. The radiographs were obtained from 30 patients with initially missed but histo pathologically proven non-small cell lung cancer (NSCLC), 35 patients with other cardiopulmonary diseases and 35 patients with no abnormalities. The observers consisted of ten experienced radiologists, two-experienced chest physicians and two residents in radiology. All observers were unaware of the study design. The validity and observer consistency was determined for each observer. Results: the mean sensitivity and specificity of the ten radiologists were 0.36 and 0.90. For the two chest physicians, the mean sensitivity and specificity were 0.29 and 0.96. For the two residents in radiology, mean sensitivity and specificity were 0.25 and 0.94. The mean interobserver kappa and mean intraobserver kappa for the radiologists were 0.38 and 0.54. For the two chest physicians, the mean interobserver kappa was 0.43, while the intraobserver kappa was 0.59. For the two residents in radiology, mean interobserver kappa was 0.35 and the intraobserver kappa was 0.42. There was no significant relation between the consistency parameters and validity parameters. The interobserver and intraobserver kappa values showed good correlation. Conclusion: the validity of the chest radiograph and observers consistency in the detection of nodular lung cancer varies widely. The level of experience is likely to influence the diagnostic performance. (C) 2001 Elsevier Science Ireland Ltd. All rights reserved.