基于Cox模型的遗传方差分量模型研究及应用
发布时间：2018-11-27 07:03
【摘要】：复杂疾病是环境因素和遗传因素共同作用的结果。为研究复杂疾病病因,遗传流行病学首先需对疾病性状进行家庭聚集性评价。家庭成员间性状不独立是家系资料的特点。对常见数量性状和质量性状,可采用边际模型或广义线性混合效应模型解决家系内相关的问题。但如发病年龄、月经初潮等删失性状,则需要在生存分析的框架下研究,Cox模型是生存分析中应用最广泛的模型之一。经典的遗传方差分量模型将家系中各成员间复杂性状的家庭相关来源分为加性遗传随机效应、显性遗传随机效应和家庭共享环境随机效应,且通常假设随机效应服从多元正态分布。本文在Cox模型的基础上引入遗传和环境随机效应,构造包含多个随机效应的Cox遗传方差分量模型。由于随机效应服从多元正态分布,给模型参数估计时求解高维积分带来极大困难,本课题探讨了应用Laplace近似法和惩罚性似然理论来解决高维积分的困难;在R和S-Plus软件中编制程序实现了核心家系和扩展家系资料的随机效应方差协方差矩阵的设计矩阵,并利用coxme函数实现了Laplace近似的惩罚性偏似然(PPL)参数估计;同时将Cox遗传方差分量模型应用于广东顺德的原发性肝癌核心家系资料和江苏泰兴的原发性肝癌扩展家系资料,利用Laplace近似的PPL参数估计方法建立原发性肝癌发病年龄的Cox遗传方差分量模型,以说明该模型和参数估计方法的可行性。本研究旨在为遗传流行病学研究者提供一种有效的、灵活的用于评价删失性状家庭聚集性的统计分析工具。
[Abstract]:Complex diseases are the result of both environmental and genetic factors. In order to study the etiology of complex diseases, genetic epidemiology needs to evaluate the family aggregation of disease traits. The character of family members is not independent is the characteristics of family data. For common quantitative and qualitative traits, marginal model or generalized linear mixed effect model can be used to solve the related problems in families. However, such as age of onset, menarche and so on, need to be studied in the framework of survival analysis. Cox model is one of the most widely used models in survival analysis. The classical genetic variance component model classifies the family related sources of complex traits among the family members into additive genetic random effect, dominant genetic random effect and family shared environment random effect. And it is usually assumed that the random effect is from the multivariate normal distribution. In this paper, based on the Cox model, genetic and environmental random effects are introduced to construct the Cox genetic variance component model with multiple random effects. Because the random effect is from the multivariate normal distribution, it is very difficult to solve the high dimensional integral in the parameter estimation of the model. In this paper, the difficulty of applying Laplace approximation and punitive likelihood theory to solve the high dimensional integral is discussed. The design matrix of random effect variance covariance matrix of core family and extended family data is realized by programming in R and S-Plus software, and the punitive partial likelihood (PPL) parameter estimation of Laplace approximation is realized by using coxme function. At the same time, the Cox genetic variance component model was applied to the nuclear family data of primary liver cancer in Shunde, Guangdong Province, and the extended family data of primary liver cancer in Taixing, Jiangsu Province. The Laplace approximate PPL parameter estimation method was used to establish the Cox genetic variance component model of the onset age of primary liver cancer to demonstrate the feasibility of the model and the parameter estimation method. The purpose of this study is to provide an effective and flexible statistical analysis tool for genetic epidemiologists to evaluate the family clustering of censored traits.
【学位授予单位】：广东药学院
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：R363
本文编号：2359788
[Abstract]:Complex diseases are the result of both environmental and genetic factors. In order to study the etiology of complex diseases, genetic epidemiology needs to evaluate the family aggregation of disease traits. The character of family members is not independent is the characteristics of family data. For common quantitative and qualitative traits, marginal model or generalized linear mixed effect model can be used to solve the related problems in families. However, such as age of onset, menarche and so on, need to be studied in the framework of survival analysis. Cox model is one of the most widely used models in survival analysis. The classical genetic variance component model classifies the family related sources of complex traits among the family members into additive genetic random effect, dominant genetic random effect and family shared environment random effect. And it is usually assumed that the random effect is from the multivariate normal distribution. In this paper, based on the Cox model, genetic and environmental random effects are introduced to construct the Cox genetic variance component model with multiple random effects. Because the random effect is from the multivariate normal distribution, it is very difficult to solve the high dimensional integral in the parameter estimation of the model. In this paper, the difficulty of applying Laplace approximation and punitive likelihood theory to solve the high dimensional integral is discussed. The design matrix of random effect variance covariance matrix of core family and extended family data is realized by programming in R and S-Plus software, and the punitive partial likelihood (PPL) parameter estimation of Laplace approximation is realized by using coxme function. At the same time, the Cox genetic variance component model was applied to the nuclear family data of primary liver cancer in Shunde, Guangdong Province, and the extended family data of primary liver cancer in Taixing, Jiangsu Province. The Laplace approximate PPL parameter estimation method was used to establish the Cox genetic variance component model of the onset age of primary liver cancer to demonstrate the feasibility of the model and the parameter estimation method. The purpose of this study is to provide an effective and flexible statistical analysis tool for genetic epidemiologists to evaluate the family clustering of censored traits.
【学位授予单位】：广东药学院
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：R363
【参考文献】
相关博士学位论文 前1条
1 郜艳晖;复杂性状家庭聚集性统计分析方法的研究[D];复旦大学;2004年
，本文编号：2359788
本文链接：https://www.wllwen.com/yixuelunwen/shiyanyixue/2359788.html
最近更新
教材专著
