PENALIZED LOG-LIKELIHOOD ESTIMATION FOR COX-FRAILTY MODEL WITH NONINFORMATIVE BIVARIATE CURRENT STATUS DATA

Abstract

A Penalized Maximum Likelihood Estimation (PMLE) procedure is proposed for Cox proportional hazards frailty model with noninformative bivariate current status data. An integrated splines (I-splines) was used to approximate the two unknown baseline cumulative hazard functions of the failure times. The one-parameter gamma frailty distribution was used to model the correlation between the two failure times. An easy to implement computational algorithm is proposed to estimate the regression and splines parameters. Bayesian technique as proposed by Wahba (1983) was employed for the variance estimation. The statistical properties of the estimated parameters were studied through extensive simulation and it was observed that the PMLEs were consistent, asymptotically normal and efcient. In addition, the estimators were robust to the choice of knots, censoring rates and type of frailty distribution used. The proposed methodology is further demonstrated through the analysis of the tumorigenicity experiment data by Lindsey and Ryan (1994).