Author + information
- Mustafa Kılıçkap1,
- Ahmet Ergun Karaağaoğlu2,
- Kerim Esenboğa1,
- Çağlar Uzun3,
- Çetin Atasoy3,
- Deniz Kumbasar1 and
- Çetin Erol1
It is important to know at which pretest probability a diagnostic test gives maximum information. Information theory quantifies the information content of a test, expressed as “bit”, and shows at which pretest probability the test gives maximum information. We assessed the diagnostic value of computerized tomographic coronary angiography (CTA) by using information theory.
Conventional coronary angiography was used as gold standard test, and 50% diameter stenosis was used as cut-off value for severe stenosis. Information content (mutual information), which gives overall test performance, was calculated across the pretest probabilities between 0.01 and 0.99 for CTA and for a perfect test that has a sensitivity and specificity of 0.999999. Ratio of the area under the curves of these values (CTA/perfect) denotes what percent of diagnostic information is obtained compared to a perfect test. Relative entropies give the information of positive and negative test results separately. Analyses were made as per patient (n=69) and per segment (n=1076) basis.
Per segment analysis gives more information than per patient analysis (CTA/perfect values were 33% and 15%, respectively). Maximum information is obtained for intermediary artery (CTA/perfect=91%) and for proximal segments (CTA/perfect =39%). Relative entropy values demonstrate that positive test is more informative in per segment analysis (Figure 1-A), however, negative test is more informative in per patient analysis (Figure I-B). Diagnostic yielding of positive test result increases to its maximum at pretest probability of about 20% in per segment analysis, and that of negative result increases to its maximum at pretest probability of about 80% in per segment and per patient analysis (Figure 1A-B).
Relative entropy graphics give clinically important information in terms of assessing the pretest probability at which diagnostic value of CTA reaches its maximum. This study shows that CTA gives more information if the result is negative in per patient analysis; however, the opposite is true for per segment analysis.