Lan Kong

Doctoral student

Department of Biostatistics

University of North Carolina at Chapel Hill

 

                        Analysis of survival data from case-cohort studies

                            with semiparametric transformation models

 

Prentice (1986) proposed a case-cohort design to reduce the cost and effort of

large cohort studies of rare diseases.  Under this design, the covariate information 

is ascertained for only a random sample called subcohort from the entire study

population and additional cases outside the subcohort. The Cox model has been 

the most popular model for survival data analysis. There are several statistical

methods available for fitting the Cox model to the case-cohort data. However,

the proportional hazard assumption in the Cox model may not always be true in

some applications or one may be interested in modeling the association from 

different aspects.  In this research, we consider a so-called semiparametric

transformation model for the case-cohort data. The fruitful class of semiparametric

transformation models includes the Cox model and a variety of non-proportional

hazard models. The estimating methods for regression parameters and survival

function under such models are proposed for case-cohort studies. We theoretically 

derive the asymptotic properties of the proposed estimators.  The finite sample

properties of the proposed estimators, as well as the efficiency relative to the full

cohort estimators are assessed via simulation studies. A data set from the Atherosclerosis Risk in Communities (ARIC) study is used to illustrate the proposed

methods.