ROC (receiver operating characteristic) comparison of methods used in the calculation of area under curve

Accurate and early diagnosis in preventive medicine or clinical studies in the field of health will ensure the control of all possible interactions and results, especially the social course and treatment practices of the relevant disease. A diagnostic test is the measurement or calculation results obtained by the use of evaluation setups or original tools used to determine the presence/absence of the suspected disease. When a new diagnostic test is proposed, it will be used in case the purpose of researching its effect on human health is risk-free, economically viable, sufficiently accurate and stable with lean applications. Receiver Operating Characteristic (ROC) curve analysis is used for diagnostic test performance determination and comparisons. The area under the ROC curve (Area Under Curve - AUC) is considered the best indicator of the accuracy level of the diagnostic test. In this study, since the ROC curve analysis can vary depending on the sample and the distribution properties (parametric/nonparametric) of the variable, the most appropriate analysis approaches and their properties will be discussed according to the data in hand. The differences between the diagnostic test results of patients and healthy individuals were examined according to the gold standard in continuous results criteria, based on the areas under the curve (AUC), which are the result findings of the calculation methods used in the normal distribution (parametric) and otherwise (nonparametric). In addition, as a diagnostic test is a continuous criterion, the effects of converting it into binary results on the properties of the areas under the curve were also tried to be exemplified. As a result of the evaluations made on the sample data set, it was concluded that the parametric calculation methods of the normal distribution diagnostic test results were more appropriate (unbiased). It would be more appropriate to use a nonparametric method (Mann Whitney) in data with large variance and/or skew distributed data. The transformation of the current continuous variable into a dichotomous form and the selection of the cut-off point (high / low) will also affect the calculation results significantly due to the loss of information caused by this method. In this study, the differences observed between the Binormal, Binomial exact_De Long and Nonparametric (Mann-Whitney) analysis methods were not significant, although they do differ at certain levels.