Correlation Research Of Multivariate Failure Time Data Based On Generalized Estimating Equations

Dependent failure time data is often used in various biomedical studies,and the correlation estimation of failure times is an important problem in the analysis of dependent survival data.Generalized estimating equation approach is widely used in the parameter estimation of semi-parametric marginal Cox model.In this paper,we studied the intermediary variables and working correlation matrices in the existing generalized estimating equation.The correlation coefficient of intermediary variables was proved infinitesimal using moment methods so that the working correlation matrices can not represent the correlation of intermediary variables correctly.An improved generalized estimating equation approach is proposed,and the correlation estimation of failure time data are discussed.Simulation studies show that the improved generalized estimation equation can reduce the number of iterations compared with the original method.The proposed method also reflect the correlation information of intermediary variables.Finally,we apply the improved generalized estimating equation approach to a survival data analysis of kidney infections.