Receiver operating characteristic curve analysis of the utility of thoracic radiographs for the diagnosis of feline cardiac disease

The results of a retrospective study in which 90 sets of feline thoracic radiographs were examined by six readers were evaluated with the aid of receiver operating characteristic (ROC) curve analysis. There were 51 sets of radiographs of 45 normal cats and 39 sets of radiographs of 33 cats with cardiac disease. The readers were asked to classify each set of radiographs as normal or abnormal, with respect to cardiac disease, into one of five diagnostic cholces that reflected reader confidence. The method of generating an ROC curve is discussed. There was no significant difference in the accuracy of interpretation between the five experienced readers. The mean accuracy of this group was 91%. However, inspection of the individual ROC curves demonstrated a fundamental difference in underlying reader bias (i.e., over-or underreading). As expected, the experienced readers were statistically more accurate than an inexperienced reader. ROC curve methodology has potential applications in veterinary medicine not only for assessing observer performance and reader bias, but also for comparing tests and in clinical decision making