Abstract:
|
In analysis of event history data, terminating event usually censors the recurrent events. Ignoring the effect of such terminal event leads to bias results. Some parametric approaches, for example, modeling the terminating event time as a covariate for the recurrent event process, have been considered. In this paper, we propose a semi-parametric approach by stratifying subjects into several groups according to their terminating event times. Then the standard Cox proportion hazard model with random effect is applied to access the covariate effects of each stratum. A multinomial model is used for the terminating time distribution to avoid strictly parametric model assumptions. Due to the right censoring of the terminating event and the unobservable frailty variables, the EM-algorithm is adopted to deal with the missing data problem. The proposed method is applied to a motivating study designed to examine the effect of a bisphosphonate on the incidence skeletal-related complication among patients with breast cancer metastatic to bone.
|