Modeling High dimensional multivariate survival data with clustering using frailty and copula models. (R-11950)

In this project we study statistical frailty and copula models for hierarchical clustered multivariate survival times. These are positive random variables that are commonly used to express the time until an event happens. Since these variables are grouped in sub-clusters and clusters, we develop new frailty and copula models to study the influence of different covariates on the survival times. Hereby we focus ourselves also on the association structure between the different survival times that was introduced by this hierarchical group structure. The motivation for these new models came from the data of the 2014 Kenya Demographic Health Survey. In this health survey information was collected on about 14,000 households on mortality, family planning, reproduction and sociodemographic characteristics. Hereby also the survival time of children under 5 year old in a household was recorded. Due to the large amount of variables in this data set we first used random survival forest –methods to identify the important covariates which had an influence on the survival time. Afterwards we used new frailty and copula models to investigate the effect of these variables on the survival time. Hereby we used sample splitting techniques to divide the different households of this data set such that we can estimate the effect of the covariates.

Keywords:Statistics

Disciplines:Statistics, Machine learning and decision making, Biostatistics