Title: Inferences for the correct classification fractions of a continuous biomarker in trichotomous settings
Authors: Peng Shi - University of Kansas Medical Center (United States) [presenting]
Leonidas Bantis - University of Kansas Medical Center (United States)
Abstract: Hepatocellular carcinoma (HCC) is the most common primary cancer of the liver. As such, there is a strong clinical interest in finding new biomarkers for its early detection. When the disease status is trichotomous, the ROC surface is an appropriate tool for assessing the discriminatory ability of a marker. A popular approach for computing cutoffs for decision making is the Youden index and its recent 3-class generalization. However, this method treats the data in a pairwise fashion and is unable to accommodate biomarker scores from all three groups simultaneously. This may result in inappropriate cutoffs that are of no clinical interest. Methods are proposed for such inferences where the cutoffs are based on the minimized Euclidean distance of the ROC surface from the perfection corner. An inferential framework, both parametric and non-parametric, are provided for the derivation of marginal confidence intervals (CIs) and joint confidence spaces (CSs) for the optimized true class rates. Our approaches were evaluated through extensive simulations and finally illustrated using a real data set that refers to HCC patients.