The Problem Of Variable Selection In Weibull Parameter Model And Semi-competitive Risk Model

Survival Analysis has been a hot research area of biostatistics,and how to find out the variables which is of great significance for analyzing the survival s-tatus is one of the most important issues in Survival Analysis.The main content of this paper is divided into two parts.In the first part,we extend the multi-parameter regression(MPR)model with generalizing the continuous covariates of the scale parameters to nonparametric and additive forms,forming a generalized MPR model with nonparametric structure(noted as NPMPR model).Also,a variable selection method for NPMPR model is given,by using Akaike Informa-tion Criterion and Bayesian Information Criterion for stepwise variable selection.The simulation results show that the proposed method works well in selecting covariates with nonlinear structures,and greatly improves the MPR model which is insensitive to the variables of nonlinear structure.When applied to Germany breast cancer data,for empirical analysis,the result of variable selection testified the efficiency of the proposed method.In the second part,we discuss the variable selection method in semi-competing risk model,and selects variables for multiple hazard functions simultaneously with Akaike Information Criterion and Bayesian Information Criterion.The result of simulation shows that the proposed method is also effective in handling the semi-competing risk model with three hazard functions,better than the Bayesian variable selection procedure based on the Deviance Information Criterion(noted as BSV method).Besides,the method is applied to colon cancer data,which has a good performance in variable selection.