主題:Extended Bayesian Information Criteria in High-dimensional Cox Proportional Hazards Model and Cure Model
主講人:李曉 博士
時間:2013年11月28日(周四)下午15:00
地點:北院卓遠樓305
主辦單位:統(tǒng)計與數(shù)學學院
摘要:Cox proportional hazards model and Cure model are often used to quantify the contributions of various factors to the survival of patients or other subjects. They are widely used for the analysis of survival data from clinical trials and other health studies. It has long been recognized that the genetic factors play vital roles in the survival of patients. Recent advances in biotechnology make it possible to collect information on a massive number of genetic factors from each subject. At the same time, only a small subset of them is likely explanatory. Identifying the most explanatory covariates out of a massive number of candidates is both scientific necessity and a statistical challenge. Regularization procedures are now the standard procedures applied to this type of data set. After a tuning parameter is given, this type of approach fits a model with most covariate effects estimated zero. Thus, it automatically accomplishes the task of selecting the most explanatory covariates. Yet the best choice of the tuning parameter remains an important research topic. In this paper, we investigate the use of extended Bayesian information criterion (EBIC) in the context of variable selection and to determine the level of tuning. We show that this criterion works well under the high dimensional Cox models and prove its selection consistency. We also establish a variable selection procedure based on EBIC for Cure model. The proposed methods are evaluated through Monte-Carlo simulations and application to a breast cancer data set.
李曉博士簡介:
中國科技大學統(tǒng)計學博士,,講師。研究領域為生存分析和變量選擇等,。