Pairwise Comparison Estimation of Regression Models under Semi-Competing Risks Data

• Main Authors: Chun-Chieh Wang, 王俊傑
• Other Authors: Jin-Jian Hsieh
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/19275653033608135001

碩士 === 國立中正大學 === 統計科學所 === 97 === In this thesis, we consider regression analysis under semi-competing risks data. In semi-competing risks data, a non-terminal event may be dependent censored by a terminal event. Therefore, the distribution of the non-terminal event time is non-identifiable without extra assumptions. Although the regression problem on non-terminal event have be discussed in many literatures, such as Lin et al. (1996), Peng et al. (2006) and Ding et al.(2009), there are still some shortages in efficiency of estimations or application of models in each method. For example, Ding et al. (2009) considered an estimating function of parameters by log-rank type statistic. When the range of Z is too wide, it will produce heavy artificial censoring to reduce the efficiency. Therefore, we apply pairwise comparison techinique suggested by Peng and Fine (2006) to construct the estimating function which can reduce artificial censoring rate. Furthermore, we prove that the proposed estimator is consistent and asymptotic normal. According to Lin et al. (1996), use the re-sampling method to construct the interval estimation and the variance estimation. Moreover, we introduces a model checking approach to check regression model assumption. In this thesis, we also design several simulation settings to examine our proposed approach and compare it with Ding et al. (2009) and Peng et al. (2006). Finally, we apply the proposed approach to analyze a bone marrow transplant data.