Assessing Quantile Prediction with Censored Quantile Regression Models

• Activity Number: 530
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Ruosha Li*+ and Limin Peng
• Companies: University of Pittsburgh and Emory University
• Keywords: Censored data ; quantile regression ; mis-specification ; prediction ; distributional properties

Abstract

In biomedical studies, quantiles of survival outcomes are often of interest and provide comprehensive insights into the disease progression. Quantile regression model serves a flexible tool for modeling the survival outcomes, and facilitates straightforward prediction of the quantile survival times. In this work, we propose a meaningful measure that summarizes the model performances in terms of quantile prediction. The proposed measure extends the absolute error to account for censoring and different quantiles. We construct an estimator of the proposed measure without assuming that the model is correctly specified, and establish its asymptotic properties via empirical process theory. In addition, we develop consistent variance estimators that properly account for the non-smoothness of the estimating equations. Extensive numerical studies demonstrate satisfactory performances of the proposed methods under realistic sample sizes.