Multivariate survival data under bivariate frailty: An estimating equation approach

A modified frailty model is developed to improve the computing efficiency of the bivariate frailty model proposed by Xue and Brookmeyer (1996, Lifetime Data Analysis 2, 277-290) for the analysis of multivariate survival data. Originally, the frailty was modeled parametrically and a modified EM approach was used to estimate the parameters of interest, however, with intensive computations. The modified frailty model formulates a Poisson regression model and applies quasi-likelihood estimating equations to estimate the parameters of interest. This procedure not only significantly reduces the computation but also avoids using a parametric assumption for the frailty distribution. Simulation studies show the estimators perform well. The method is also applied to a mental health care dataset.