Proportional hazards models with frailties and random effects.

摘要: We discuss some of the fundamental concepts underlying development frailty and random effects models in survival. One these was idea a model where each subject has his or her own disposition to failure, their so-called frailty, additional any we wish quantify via regression. Although concept individual can be value when thinking about how data arise interpreting parameter estimates context fitted model, argue that is limited practical value. Individual (frailties), whenever detected, made disappear by elementary transformation. In consequence, unless are take form as unassailable, beyond challenge carved stone, if understand term 'frailty' referring effects, then have no Random on other hand, which groups individuals share common effect, used advantage. Even this case however, prepared sacrifice efficiency, avoid complex modelling using considerable power already provided stratified proportional hazards model. Stratified both seen particular cases partially models, view gives further insight. The added structure viewed with distributional constraints, will, for group sizes five more, provide more than modest efficiency gains, even assumptions exactly true. On moderate large numbers very small groups, two three, study twins being well known example, gains far from negligible. For such applications, rather strong. This especially so good robustness properties models. Nonetheless, simpler analysis, based upon remains valid, albeit making less efficient use resources.