Survival models and health sequences

摘要

Medical investigations focusing on patient survival often generate not only afailure time for each patient but also a sequence of measurements on patienthealth at annual or semi-annual check-ups while the patient remains alive. Sucha sequence of random length accompanied by a survival time is called a survivalprocess. Ordinarily robust health is associated with longer survival, so thetwo parts of a survival process cannot be assumed independent. This paper isconcerned with a general technique---time reversal---for constructingstatistical models for survival processes. A revival model is a regressionmodel in the sense that it incorporates covariate and treatment effects intoboth the distribution of survival times and the joint distribution of healthoutcomes. It also allows individual health outcomes to be used clinically forpredicting the subsequent survival time.