RAND Statistics Seminar Series
Mixed-Effects Cox Models
Presented by Terry Therneau
Thursday, April 07, 2005, 10:30 a.m.
Forum m-1226-1228, Santa Monica
Over the last few years there has been ongoing interest in frailty models. The label comes originally from demography, where the idea is that some subjects have a larger innate risk of death or 'frailty' than others. The models are, in effect, survival models with a random intercept.
Our group has been interested in models with more general random effects, particularly in the area of random institutional effects in clinical trials, both random intercepts and random treatment effects, and in latent genetic traits for related individuals. We have created a major addition coxme to the S/R survival software which estimates a general mixed effects Cox model.
The talk will focus on the practical aspects of the method: for what data sets has coxme worked well or poorly, implementation issues with the algorithm and areas where it might be expected to do well, and future directions. There is evidence that the method may be useful in a much wider range of problems than the medical center studies that motivated its creation.