Analysis of the diagnostic performance in breast cancer screening by relative operating characteristics

In medical usage, relative operating characteristics (ROC) refer to the symptoms, signs, and laboratory values used for making a diagnosis. Diagnosis analysis by ROC measures diagnosis performance independently of the diagnostic criteria used, thus revealing the arbitrary character of diagnosis. In order to carry out ROC analysis, the diagnosis has to be graded in levels of certainty of the presence of disease (or suspicion of a lesion). When both the graded diagnosis decisions and the ultimate proof of presence or absence of disease are available, ROC analysis can easily be carried out. Analysis of 50,102 successive examinations of 19,400 women in the Utrecht breast cancer screening projects presents a remarkably good fit of the observed data with the calculated values. In the medical context even small deviations of the actual sensitivity and specificity values from the ROC curve values must be considered as signs of differences in the diagnosis procedure and should lead to further analysis of the diagnostic criteria and the examined population. Consequently the separate effect of diagnosis impairment by the mammographic risk patterns (Wolfe) on diagnostic accuracy could be isolated. The effect of age on the detection capability of breast cancer by mammography appeared to operate in narrow age groups.