Presentation given at the ICSA Applied Statistics Symposium.

Program

An essential aspect of medical diagnostic testing using biomarkers is to find an optimal cut-point that categorizes a patient as diseased or healthy. This aspect can be extended to the diseases which can be classified into more than two classes. For diseases with general k (k>2) classes, well-established measures include hypervolume under the manifold and the generalized Youden Index. Another two diagnostic accuracy measures, maximum absolute determinant (MADET) and Kullback-Leibler divergence measure (KL) are recently proposed.This research proposes a new measure of diagnostic accuracy based on concordance and discordance (CD) for diseases with k (k>2) classes and uses it as a cut-points selection criterion. The CD measure utilizes all the classification information and provides more balanced class probabilities. Power studies and simulations show that the optimal cut-points selected with CD measure may be more accurate for early-stage detection in some scenarios compared with other available measures. As well, an example of an actual dataset from the medical field will be provided using the proposed CD measure.