|
Abstract:
|
In cohort studies, restricted mean survival time (RMST) has attracted much attention as a summary measure for its intuitive interpretation and clinical meaning. While the mean of failure times may not be estimable due to right censoring, RMST always serves as a useful alternative tool to summarize the underlying survival time distribution over a specific time duration of interest. When the data are subject to length-biased sampling, the existing regression methods on RMST are not available. In this paper, we consider a regression model that directly relates the covariates and the RMST for length-biased right-censored data. Unbiased estimation methods are developed to obtain consistent estimators for covariate effects by properly adjusting for informative censoring induced from the unique data structure. Stochastic process and martingale theories are used to establish the asymptotic properties of the proposed estimators. We investigate the finite sample performance through simulations and illustrate the method by applying it to a prevalent cohort study of Canadian dementia patients.
|
