Marginal hazards model for multivariate failure time data with auxiliary covariates

Abstract

A marginal hazards model of multivariate failure times has been developed based on the ‘working independence’ assumption [L.J. Wei, D.Y. Lin, and L. Wessfeld, Regression analysis of multivariate incomplete failure time data by modeling marginal distributions, J. Amer. Statist. Assoc. 84 (1989), pp. 1065–1073.]. In this article, we study the marginal hazards model of multivariate failure times with continuous auxiliary covariates. We consider the case of common baseline hazards for subjects from the same clusters. We extend the kernel smoothing procedure of Zhou and Wang [H. Zhou and C.Y. Wang, Failure time regression with continuous covariates measured with error, J. Roy. Statist. Soc. B 62 (2000), pp. 657–665.] to correlated failure time data. Through semiparametric estimation of the marginal partial likelihood function, we obtain the estimated partial likelihood based estimator of the regression coefficients. We present asymptotic properties of the induced estimator and demonstrate the performance of the proposed estimator through a finite sample simulation study. Finally, a real data application is conducted to elucidate the use of the method.