< Back to previous page
Zero-inflated semi-parametric Cox┌s regression models for left-censored multivariate survival data. (R-1829)
In some clinical, industrial or environmental studies, researchers observe a positive random response variable. Due to difficulties in the measuring mechanism, they see for some subjects only an upper bound of this response. This type of data is called left-censored data and we observed for each subject the maximum of the response variable and an independent random censoring variable. Furthermore we get an indicator which indicates which variable is the largest. We notice that in some studies the response variable attains a zero-value with a positive discrete probability. In this project we introduce a semi-parametric regression model for multivariate left-censored data in which the response variables have a positive discrete probability on a zero-value. We model the associations between measurements within a subject by families of copula functions. To investigate the influence of covariates on the probability on a zero-value within each component, we use parametric models. For the strict positive part of the response variables, Cox┌s regression models are given to model the influence of the covariates. As results we show the consistency and asymptotic normality of both the finite- as infinite-dimensional parameters in this model. We also introduce bootstrap procedures to approximate the null-distributions of hypothesis tests of these parameters.
Date:1 Oct 2009 → 30 Sep 2011
Keywords:BOOTSTRAP, COX┌S REGRESSION MODELS
Disciplines:Information and computing sciences, Biological sciences, Computer engineering, information technology and mathematical engineering, Basic sciences, Clinical sciences, Health sciences, Translational sciences