Varying-association copula models for marginal additive hazards distribution

To study correlation of multivariate survival times with semi-parametric additive hazards model, we propose a class of varying-association copula model．A two-stage estimation algorithm is proposed to study the dynamic correlation structure among survival times．First, the method proposed by Lin was used to obtain estimates of marginal survival functions of additive hazards model．Then the estimated survival functions were inserted into a local pseudo-likelihood function based on copula model, to estimate varying association parameter．Consistency and asymptotic normality of the proposed estimators were established, and simulation studies were conducted to empirically examine the finite-sample performances of the proposed methods．The new approach was illustrated with real data from diabetic retinopathy study (DRS)．