Empirical Likelihood Ratio Confidence Interval Estimation of Best Linear Combination of Biomarkers

• Activity Number: 610
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Xiwei Chen*+ and Albert Vexler and Marianthi Markatou
• Companies: and University at Buffalo and University at Buffalo
• Keywords: Receiver operating characteristic curve (ROC) ; Area under the ROC curve ; Empirical likelihood ; Kernel ; Best linear combination

Abstract

To improve the accuracy of diagnostic tests, combinations of biomarkers are considered that maximize the area under the receiver operating characteristic (ROC) curve (AUC). The best linear combination (BLC) of biomarkers' values under the multivariate normality assumption is well-addressed in the literature. This approach may lose efficiency when the data distribution deviates from the normal assumption. We propose a novel smoothed empirical likelihood (EL) approach that incorporates the kernel AUC estimator to construct nonparametric confidence intervals of the BLC-based AUCs. The proposed method has several advantages including the feasibility to use gradient-based methods for fast computation of BLC coefficients and the employment of powerful likelihood methods without specification of underlying data distributions. Simulation results show that the proposed method performs well even when the distribution of biomarkers is skewed, a situation commonly met in practice. We further illustrate the efficiency of the proposed method with the data from a clinical experiment related to atherosclerotic coronary heart disease.