高维AFT模型的正则化变量选择

Regularized Variable Selection for High Dimensional Survival Data with AFT Model

针对生存分析中建立生存模型时,如何处理生存数据中特有的数据类型——删失数据,降低高维协变量的维数,更好地识别出真正具有预测性的因子,建立准确的生存模型的问题,提出用STUTE’s加权最小二乘法和删失限制以及LASSO正则化相结合的方法来对AFT模型进行估计。首先,提出STUTE’s加权最小二乘法和删失限制相结合的方法对生存数据中的删失数据进行处理;其次,提出了LASSO的一个新的实现算法进行模型的变量选择,降低模型中协变量的维数,精简模型;最后,通过仿真分析得到提出的新估计方法较已有的LASSO旧算法以及其他的变量选择方法,VSURF算法更能找出“真”因子,建立准确的生存模型。

According to the problem in how to deal with the peculiar data types in the survival data,censoring data,and how to reduce the dimension of high-dimensional covariates and identify the truly predictive factors to build an accurate survival model when building a survival model in survival analysis,this paper proposes to estimate the AFT model by combining STUTE’s weighted least square method,censoring constraints and LASSO regularization method.Firstly,the combination of STUTE’s weighted least square method and censoring restraints is proposed to process the censoring data in survival data.Secondly,a new implementation algorithm of LASSO is put forward to select the variables of the model,reduce the dimension of the covariates in the model and simplify the model.Finally,through the simulation analysis,compared with the existing LASSO algorithm and the VSURF algorithm,the new estimation method can find the“real”factor and set up survival model more accurately.