Investigation of diagnostic test performance using information theory

For the application of this thesis, ASO values of 68 subjects who applied to Istanbul Mehmet Akif Ersoy Thoracic and Cardiovascular Surgery Training and Research Hospital for the diagnosis of rheumatic disorder were used. ASO is a value which is used to learn whether the patients have group A beta-hemolytic streptococcal infection which causes these diseases.ASO values were evaluated according to Turbidimetric methods of two different firms. Since the names of the firms were kept secret, these methods were called as I. Turbidimetric method and II. Turbidimetric method. Both ROC and Information Theory Analyses were applied to the results. Therefore, both firms' Turbidimetric method diagnostic test performances were evaluated and which diagnostic test had better performance was determined.The disease diagnosis is considered among the most important parts of the treatment process. The aim of this study is to demonstrate how basic concepts in Information Theory and in Receiver Operating Characteristics (ROC) apply to the problem of quantifying diagnostic test performance.Before evaluation process, Receiver Operating Characteristic (ROC) analysis and Information Theory are described. The role of ROC analysis and Information Theory analysis in the medicine sector are presented in the context of diagnostic tests performance. Moreover, the notations of ROC analysis and Information Theory are introduced. In ROC analysis, it is indicated how to create ROC curves and their properties are explained in detail. After ROC curves are drawn, Area Under the Curve (AUC) concept which is one of the diagnostic test performance measures is evaluated. In Information Theory analysis, the concepts of entropy and conditional entropy are introduced. Afterwards by using these two concepts, Mutual Information value which is one of the diagnostic test performance measures is evaluated. At the final stage, it is indicated which diagnostic test has the best performance based on the mentioned measures.