Change-Point Proportional Hazards Model for Clustered Event Data

• Activity Number: 454
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Yu Deng*+ and Jianwen Cai and Donglin Zeng
• Companies: University of North Carolina at Chapel Hill and University of North Carolina at Chapel Hill and University of North Carolina at Chapel Hill
• Keywords: Clustered event ; Proportional hazards model ; Change-point ; m out of n Bootstrap ; Empirical process

Abstract

In analysis of clustered time-to-event data, the effect of some covariates may vary depending on whether another variable exceeds an unknown threshold. For example, in a study of diabetes where information on family members is collected, it is found that the effect size of a marker depends on whether leukocyte telomere length is larger or smaller than the first telomere quartile. In this work, we propose a change-point proportional hazards model for clustered event data. The model incorporates the unknown threshold of the threshold variable as a change-point in the regression. Marginal partial likelihood functions are maximized for estimating both regression coefficients and the change point in the model. Furthermore, we propose a test using score statistics to check the existence of a change point. M out of n bootstrap method is used to make inference for the estimator of the change point. We establish the asymptotic distribution of the proposed estimators. The small-sample performance of the proposed method is demonstrated via simulation studies. Finally, a diabetes data set is analyzed to illustrate the method.