Bilgi Kuramı Yaklaşımı ile Bilgisayarlı Tomografik Koroner Anjiyografinin Tanısal Değerinin Değerlendirilmesi

Abstract

Classical performance measures of a test, such as sensitivity, specificity, positive predicted value (PPV) and negative predicted value (NPV), are obtained by comparison of the test results with a gold standard test. This study was aimed to assess the diagnostic performance of coronary computed tomographic angiography (CCTA) by using information theory. Diagnostic performance of CCTA was assessed separately for per patient, per segment, per vessel and per level. Sensitivity and specificity were highest for intermediary artery (IMA) (both equals to 1.0). Specificity was higher than sensitivity in all analysis except for per patient analysis; and NPV was higher than PPV in all analysis. Uncertainty was reduced most for IMA. Relative entropy for positive test results and rule in potentials (Pin) were higher compared to the relative entropy for negative test results and rule out potentials (Pout) in all analysis except for per patient analysis. Relative entropy graphics gave similar results. Area under the curves (AUC) of PPV and NPV gave similar results to AUC of relative entropies. Pretest probabilities at the peak values of relative entropies were very close to those at the bending point of PPV and NPV curves. As various hypothetic scenarios demonstrated that the change in relative entropies, Pin and Pout depending on sensitivity and specificity was not monotonic, these measures might sometimes be misleading when different tests were compared. As PPV and NPV curves were monotonic and easier to interpret, these measures might be considered more useful for routine clinical applications.