APPLICATION OF RECEIVER-OPERATING-CHARACTERISTIC (ROC) CURVES TO VETERINARY CLINICAL PATHOLOGY

Assessment of diagnostic tests in veterinary clinical pathology usuall y includes analytical performance characteristics and values of sensit ivity and specificity estimated when comparing test results from affec ted and non-affected animals to a cutoff value corresponding to the up per/lower limit of the reference range. This cutoff value is but one o f many to which the test results can be compared and because sensitivi ty and specificity varies as the cutoff value is changed, a complete p icture of the test's clinical usefulness is not presented by only one pair of sensitivity and specificity results. Instead, a complete pictu re of the test's clinical usefulness can be appreciated by the entire spectrum of sensitivity and specificity pairs resulting from continuou sly varying the cutoff value over the entire range of test results. Th e receiver-operating-characteristic (ROC) curve provides a view of thi s whole spectrum of sensitivity/specificity pairs and it allows the in vestigator to examine the test's discriminative power over all possibl e cutoff values. This paper presents methods for preparing and analysi ng ROC curves and illustrates the principles presented by applying ROC -curve analysis to a set of data consisting of protein and L-lactate c oncentrations in peritoneal fluid from horses suffering from either su rgical or medical acute abdomen.