Comparison of three-dimensional ROC analysis methods used for three ordered diagnostic groups

Three-dimensional ROC analysis is used to evaluate the performance of a diagnostic test when the disease status is three-class. In three-dimensional ROC analysis, the performance of the diagnostic test is evaluated with the ROC surface and the volume under the surface (volume below the ROC surface-VUS). In this thesis, three-class ROC analysis methods used to evaluate the diagnostic performance of a continuous diagnostic test to be used to classify patients into three ordered disease class that is frequently encountered in health studies were examined and the performances of the estimators used for VUS were compared with the simulation study under different scenarios. According to the results obtained from the simulation study, it was observed that the parametric method used in the estimation of VUS did not provide a significant advantage over the non-parametric method for scenarios where all three groups had a normal distribution and the diagnostic test had a medium, high and very high diagnostic performance. In addition, in the case where it comes from the normal distribution in its three groups and gives a very high VUS value; It was observed that the difference between parametric and non-parametric methods disappeared completely, and the sample size had no effect on the performance of the methods in terms of AQC and bias. For a three-class diagnostic test with moderate performance, where the diagnostic test of the healthy group is normal and the diagnostic test of the patient group is skewed to the right, the non-parametric method gives better results than the parametric method in all sample sizes, and with the increase in the sample size, the non-parametric method is better than the parametric method superiority has increased.