Investigation of Diagnostic Test Performance Using Receiver Operating Characteristic and Fundamental Concepts of Information Theory

Abstract

Aims: Receive Roperating Characteristic (ROC) curveanda fundamental concept of information theory is directly applicable to evaluation of diagnostic test performance. In this study,the performance of the two diagnosis tests on the field of rheumatic disorder is analyzed byusing receiver operating characteristic and fundamental concepts of information theory. The aims of this study to investigate which diagnosis  tests  has better performance and to demonstrate which test can be an alternative to gold standard test by carrying out ROC and fundamental concepts of information theory. Study Design: ROC analysis and application of fundamentals concept of information theory (entropy, conditional entropy, mutual information). Place and Duration of Study: Department of Statistics, between July 2012 and July 2013. Methodology: Anti-Streptolysin O (ASO) is a value which is used to learn whether the patients have group “A” beta-hemolytic streptococcal infection which causes rheumatic disorder diseases. ASO values of 68 subjects who applied to Thoracic and Cardiovascular Surgery Training and Research Hospital in Istanbul for the diagnosis of rheumatic disorder were used. ASO values were evaluated according toTurbidimetric tests of two different firms. These tests were called as I. Turbidimetric test and II. Turbidimetric test. Both ROC and Information Theory analyses were applied to the data. Therefore, bothfirms’ Turbidimetric test diagnostic performances were evaluated and which diagnostic test had better performance was determined. Results: According to ROC curve results, Area Under curve(AUC) is calculated 0.98 for I. Turbidimetric test and 0.90 for II. Turbidimetric test. On account of information theory analysis; the entropy value is the same but mutual information values are different. According to the result of mutual information, I. Turbidimetric test providesmore diagnostic information than II. Turbidimetric test. Therefore I. Turbidimetric test dominates II. Turbidimetric test. Based on these results, it can be verified that mutual information value is parallel to AUC value. Another result is found for threshold values of tests. According to results an alternative threshold values for tests can be obtained by using mutual information. Conclusion: TheTurbidimetric tests’ performances are examined using ROC and information theory. With regard to ASO values, it is concluded that I. Turbidimetric test is more likely to show the similarity to Nefelometric test in comparison with II. Turbidimetric test. Using I. Turbidimetric test has financial benefits to clinicians, since it is less expensive in contrast with Nefelometric test.