Investigating accuracy of biomarker involving a parametric approach of proportional hazard skewed normal model

Abstract

Time-dependent receiver operating characteristic (ROC) curve is useful to measure the accuracy performance over time. In this paper, we have shown how to determine the accuracy trend using proportional hazard model with continuous skewed normal biomarker and skewed normal time-to-event. Bayesian inference and adaptive multivariate integration over hypercubes are used respectively for parameter estimation and solving the sensitivity and specificity of the time-dependent ROC. The simulation study and application on real data suggest that it is possible to predict the accuracy measurement over time by changing the estimated association parameter between the biomarker and time-to-event data. In addition, studies on the impact of sample size on the ROC curve shows an advantage of this parametric method over conventional nonparametric.