극단값 분포를 이용한 AFT 변량효과 모델링 접근법

Abstract

Cox's proportional hazards regression model is a semi-parametric method which has often used in the analysis of survival data. However, if the assumption of proportional hazards is not satisfied, the interpretation of the estimates can be difficult. Thus, accelerated failure time (AFT) regression model which usually assumes the distribution of survival time has been used. The AFT model is a linear model that is easier to interpret than the proportional hazards model. In this thesis, we propose a method of estimating AFT random-effect model with the extreme value distribution for analyzing multivariate (or clustered) survival data. For the inference we use a h-likelihood approach and derive the estimation procedure. The proposed method is illustrated with two real data sets, i.e. kidney infection data and multi-center bladder cancer trial data.

keywords: AFT model, Extreme value distribution, H-likelihood, Random effects, Multivariate survival data