ABSTRACT
The classical methods assume that up to the covariates included in the model, the population under study is homogeneous. Individuals with the same value of these covariates are expected to have similar event times. However, in some applications it is more reasonable to consider the population as heterogeneous or as a collection of homogeneous clusters of individuals. Individuals may be exposed to different risk levels, and this even after controlling for known risk factors, or being grouped into clusters so that individuals from the same cluster share some common unobserved exposure. The concept of frailty has therefore been developed to handle unobserved heterogeneity in survival data caused by unmeasured covariates. The remaining individuals at risk will tend to form a selected group with lower risk. The distribution of the covariates among the individuals still at risk will therefore evolves over time, with observations at higher risk leaving the risk set earlier.
