Observer variation in pattern type and extent of disease in fibrosing alveolitis on thin section computed tomography and chest radiography.

In fibrosing alveolitis the pattern type on thin section computed tomography (CT) predicts histological appearances at open lung biopsy and the likelihood of response to treatment. To test the level of inter- and intra-observer variability on CT and chest radiography (CXR), the pattern type and extent of disease were assessed by four observers (two experienced, two inexperienced). A total of 126 CT examinations and 108 concurrent postero-anterior chest radiographs were scored on two occasions, at least 8 weeks apart. A confidence rating was assigned to each observation. Three out of four observers agreed on pattern type in 81% of cases on CT compared with 54% on CXR (kappa coefficient 0.48 and 0.16 for CT and CXR, respectively). Inter-observer variability in categorizing pattern type on CT was lowest in patients with the highest confidence scores (kappa = 0.63). Confident observations were associated with extensive or moderately extensive disease (P < 0.001), and with a predominantly reticular pattern (P < 0.0001). Intra-observer variability for pattern type on CT was less for the experienced observers (kappa = 0.78 and 0.70) than for the inexperienced group (kappa = 0.50 and 0.37). Inter-observer variability for extent of disease was significantly less on CT than on CXR (standard deviations 7.8% and 9.2% respectively, P < 0.001). This study shows that observer variability using a clinical grading system is lower with CT than with chest radiography in fibrosing alveolitis.