medical and biological sciences - Collegium Medicum - Uniwersytet ...

64Ana-Maria ŠimundićThe shape of a ROC curve and the area under thecurve (AUC) helps us estimate how high the discriminativepower of a test is. The closer the curve is locatedto upper-left hand corner and the larger the area underthe curve, the better the test is at discriminating betweendiseased and non-diseased. The area under thecurve can have any value between 0 and 1 and it is agood indicator of the reliability of the test. A perfectdiagnostic test has the AUC 1.0. whereas a nondiscriminatingtest has an area 0.5. Generally we can saythat the relation between AUC and diagnostic accuracyapplies as described in Table II.Table II. Relationship between the area under the ROC curveand diagnostic accuracyareadiagnostic accuracy0.9 – 1.0 excellent0.8 - 0.9 very good0.7 - 0.8 good0.6 - 0.7 sufficient0.5 - 0.6 bad< 0.5 test not usefulAUC is a global measure of diagnostic accuracy. Ittells us nothing about individual parameters, such assensitivity and specificity. Out of two tests with identicalor similar AUC, one can have significantly highersensitivity, whereas the other significantly higher specificity.Furthermore, data on AUC state nothing aboutpredicative values and about the contribution of the testto ruling-in and ruling-out a diagnosis. Global measuresare there for general assessment and for comparisonof two or more diagnostic tests. By the comparisonof areas under the two ROC curves we can estimatewhich one of two tests is more suitable for distinguishinghealth from disease or any other two conditions ofinterest. It should be pointed that this comparisonshould not be based on visual nor intuitive evaluation[4]. For this purpose we use statistic tests which evaluatethe statistical significance of estimated differencebetween two AUC, with previously defined level ofstatistical significance (P).DIAGNOSTIC ODDS RATIO (DOR)Diagnostic odds ratio is also one global measure fordiagnostic accuracy, used for general estimation ofdiscriminative power of diagnostic procedures and alsofor the comparison of diagnostic accuracies betweentwo or more diagnostic tests. DOR of a test is the ratioof the odds of positivity in subjects with disease relativeto the odds in subjects without disease [4]. It iscalculated according to the formula: DOR =(TP/FN)/(FP/TN).DOR depends significantly on the sensitivity andspecificity of a test. A test with high specificity andsensitivity with low rate of false positives and falsenegatives has high DOR. With the same sensitivity ofthe test, DOR increases with the increase of the testspecificity. For example, a test with sensitivity > 90%and specificity of 99% has a DOR greater than 500.DOR does not depend on disease prevalence; howeverlike sensitivity and specificity it depends on criteriaused to define disease and its spectrum of pathologicalconditions of the examined group (diseaseseverity, phase, stage, comorbidity etc.).DIAGNOSTIC EFFECTIVENESS (ACCURACY)Another global measure of diagnostic accuracy isso called diagnostic accuracy (effectiveness), expressedas a proportion of correctly classified subjects(TP+TN) among all subjects (TP+TN+FP+FN). Diagnosticaccuracy is affected by the disease prevalence.With the same sensitivity and specificity, diagnosticaccuracy of a particular test increases as the diseaseprevalence decreases. This data, however, should behandled with care. In fact, this does not mean that thetest is better if we apply it in a population with lowdisease prevalence. It only means that in absolutenumber the test gives more correctly classified subjects.This percentage of correctly classified subjectsshould always be weighed considering other measuresof diagnostic accuracy, especially predictive values.Only then a complete assessment of the test contributionand validity could be made.YOUDEN'S INDEXYouden's index is one of the oldest measures fordiagnostic accuracy [6]. It is also a global measure of atest performance, used for the evaluation of overalldiscriminative power of a diagnostic procedure and forcomparison of this test with other tests. Youden's indexis calculated by deducting 1 from the sum of test’ssensitivity and specificity expressed not as percentagebut as a part of a whole number: (sensitivity + specificity)– 1.For a test with poor diagnostic accuracy, Youden'sindex equals 0, and in a perfect test Youden's indexequals 1. Youden's index is not sensitive for differencesin the sensitivity and specificity of the test,