ABSTRACT
Individuals in a survival study with the same values of measured explanatory variables may nonetheless exhibit variation between their observed survival times. Some may be more frail than others, and will have a greater hazard of death at any given time than others who are less frail. This frailty effect can be incorporated in a survival model by adding a random component to the linear part of the model. Situations where the survival times of individuals within a group are not independent can be modelled by assuming that they share a common frailty. This enables repeated event times within an individual, such as the times to a migraine attack, to be analysed. After a review of the impact of frailty on the survivor and hazard functions, models that include individual or shared frailty effects are summarised, and it is shown how they can be fitted using likelihood methods. Procedures for comparing models with frailty and for assessing the extent of any frailty effect are also presented.
