Statistical Inferences for the ROC Curve Analysis of the optimal Marker


Tzu-Heng Chiou
2009-09-11  12:30 - 14:30
Room 405, Mathematics Research Center Building (ori. New Math. Bldg.)

The receiver operating characteristic (ROC) curve of the likelihood ratio L(Y)=f_{1}(Y)/f_{0}(Y) has been shown to be highest among all transformations, where Y represents the marker, D denotes the disease status with D=1 indicating disease and D=0 as no disease, and f_{D}(y) is the conditional probability density function of Y. In this study, we developed the inferences based on a non-parametric estimator of L(Y) for the related accuracy measures, and extended the proposed procedures to the optimal composite of multiple markers. The applicability of our methods are further demonstrated through a class of simulations and the analyses of two data from the studies of diabetes and liver disorders. (This is a joint work with Chin-Tsang Chiang)