Critical evaluation of three chest radiograph scores in cystic fibrosis.

BACKGROUND A number of chest radiographic scores have been developed to assess the severity of respiratory disease in cystic fibrosis but critical statistical evaluation has been limited. In particular, the chest radiograph component of the National Institutes of Health (NIH) clinical score has not previously been validated. Three different chest radiograph scores have been compared and the association between them and lung function tests investigated. METHODS The interobserver and intraobserver variation of the Brasfield, NIH chest radiograph, and the Royal Children's Hospital (RCH) chest radiograph score was assessed by three observers--a paediatric radiologist, a junior and a senior respiratory physician--who independently scored, on separate occasions, 62 chest radiographs randomly selected from three age strata of patients ranging from 7 to 18 years. Lung function tests were available for 61 patients obtained within three months of the chest radiograph. Two way analysis of variance was used to estimate components of variation in scores. RESULTS Results were similar for the Brasfield and NIH scores, both of which demonstrated greater precision than the RCH score, but the estimated repeatability of the Brasfield and NIH scores can be expected to differ by up to 20% of the maximum score. The reliabilities (intraclass correlation) are all reasonably high at 0.74, 0.73, and 0.61 for the Brasfield, NIH, and RCH scores, respectively. The estimated correlation between radiographic scores and lung function tests, adjusted for attenuation caused by measurement error, showed a similar correlation for all three scoring methods ranging from 0.55 to 0.78. Correlations were slightly greater with FEV1% than FVC%. These correlations are substantial but not high, indicating that a large proportion of the variability in radiographic scores cannot be explained by lung function measurements. CONCLUSIONS The Brasfield and NIH chest radiograph scores have very similar statistical profiles and can be equally recommended if a chest radiograph score is to be used. The RCH radiographic score appears to be less reliable. The limitations of these scores need to be understood.