TY - JOUR
T1 - Model selection for two-sample problems with right-censored data
T2 - An application of Cox model
AU - Chen, Chun Shu
AU - Chang, Yu Mei
N1 - Funding Information:
This work was supported by the National Science Council of Taiwan under Grants NSC 98-2118-M-018-003-MY2 and NSC 99-2118-M-029-003 . The authors thank the editor, the associate editor, and the anonymous referees for helpful comments and suggestions. The authors also thank Prof. Hsin-Cheng Huang, Institute of Statistical Science, Academia Sinica, Taipei, Taiwan, and Prof. Pao-Sheng Shen, Department of Statistics, Tunghai University, Taichung, Taiwan, for helpful discussions on methodology and many constructive suggestions about this article.
PY - 2011/6
Y1 - 2011/6
N2 - For investigating differences between two treatment groups in medical science, selecting a suitable model to capture the underlying survival function for each group with some covariates is an important issue. Many methods, such as stratified Cox model and unstratified Cox model, have been proposed for investigating the problem. However, different models generally perform differently under different circumstances and none dominates the others. In this article, we focus on two sample problems with right-censored data and propose a model selection criterion based on an approximately unbiased estimator of Kullback-Leibler loss, which accounts for estimation uncertainty in estimated survival functions obtained by various candidate models. The effectiveness of the proposed method is justified by some simulation studies and it also applied to an HIV+ data set for illustration.
AB - For investigating differences between two treatment groups in medical science, selecting a suitable model to capture the underlying survival function for each group with some covariates is an important issue. Many methods, such as stratified Cox model and unstratified Cox model, have been proposed for investigating the problem. However, different models generally perform differently under different circumstances and none dominates the others. In this article, we focus on two sample problems with right-censored data and propose a model selection criterion based on an approximately unbiased estimator of Kullback-Leibler loss, which accounts for estimation uncertainty in estimated survival functions obtained by various candidate models. The effectiveness of the proposed method is justified by some simulation studies and it also applied to an HIV+ data set for illustration.
KW - Confidence interval
KW - Data perturbation
KW - Generalized degrees of freedom
KW - Information criterion
KW - Kullback-Leibler loss
KW - Median survival time
UR - http://www.scopus.com/inward/record.url?scp=79651471794&partnerID=8YFLogxK
U2 - 10.1016/j.jspi.2010.12.023
DO - 10.1016/j.jspi.2010.12.023
M3 - 期刊論文
AN - SCOPUS:79651471794
SN - 0378-3758
VL - 141
SP - 2120
EP - 2127
JO - Journal of Statistical Planning and Inference
JF - Journal of Statistical Planning and Inference
IS - 6
ER -
