Estimating parameters of a frailty semi-competing model with measurement errors in covariates
Lianfen Qian, Department of Mathematical Sciences, Florida Atlantic University, USA
In lifetime data analysis, it is common to observe multiple endpoints of risks. In this paper, we consider a shared frailty semi-competing model with measurement errors in covariates for cluster data with two semi-competing risks. Under the assumptions of shared Gamma frailty within each cluster and Weibull baseline hazards, we propose a corrected maximum likelihood estimation for covariate effects and Bayes estimation for the frailties. We derive the theoretical formulas for EM algorithm which is utilized for numerical optimization. To evaluate the finite sample performance of this method, we conduct the simulation studies which show that the proposed method works better than the Bayes estimation with MCMC algorithm. Moreover, the proposed method is robust to model mis-specification in terms of with or without measurement errors. For illustration purpose, we apply the proposed method to the monoclonal gammopathy of undetermined significance data. The results show that age is significant for all three baseline hazards, while the size of the monoclonal protein spike at diagnosis is significant only for the hazard from healthy to plasma cell malignancy. This is joint work with Caiya Zhang and Xiaolu Gu.