Diagnosis analysis in breast cancer and cervical cancer screening

Assessment of the accuracy of diagnostic procedures has been made independent of the diagnostic criteria used by means of Relative Operating Characteristics (ROC) analysis. A ROC curve describes the mutual relationship between the sensitivity and specificity of a diagnostic decision on the basis of various diagnostic criteria. The construction of such ROC curves is made possible if diagnoses are graded into levels of certainty. The curve enables the choice of an operating point with predetermined sensitivity and specificity values for the diagnosis decision. The population-based breast-cancer and cervical cancer screening projects carried out in Utrecht demonstrated an excellent fit between actual data and the calculated ROC curves. Analysis of the accuracy or performance of cytological diagnosis uncovered a problem arising from the similarly graded histopathological reference criteria used to determine the 'truth' of the cytological diagnosis decisions. The proposed solution is a serial calculation of ROC curves, one for each level differentiating between the histopathological categories. The ensuing three-dimensional ROC hill may reveal a summit marking numerically advantageous diagnosis criterion levels for both the test and the disease to be detected, or a depression signalling locally below-standard detection performance.