Some Bayesian Semiparametric Approaches to Modeling Survival/Reliability with a Noisy Longitudinal Marker

Tim Hanson
Division of Biostatistics
University of Minnesota

Wednesday, October 11th 2006
3:30pm
Moos 2-690
Minneapolis Campus

Abstract:
What do hemodialysis patients, Mediterranean fruit flies, and automated test chips have in common? Give up? Their lifetimes can be modeled as a function of a noisy longitudinal marker collected over time! In this talk I will present two detailed analyses of time-to-event data involving noisy longitudinal makers: the time to death of Medflies modeled as a function of their daily egg-laying trajectories, and the time to failure of an electrical component due to atmospheric corrosion (essentially rust) as a function of electrical resistance across the chip. Novel findings include: the overarching assumptions on the survival mechanism play a very important role in predicting survival (so assuming proportional hazards may or may not be dumb depending on the data), the model for the longitudinal trajectory can highly affect the model's predictive ability (so modeling data that exhibit big jumps with a smooth curve might get you into trouble), and while "nonparametric" implies "flexible", it doesn't necessarily imply "good prediction".

A social tea will be held at 3:00 P.M. in A434 Mayo. All are Welcome.
For more details contact 612-624-4655 or see http://www.biostat.umn.edu/seminar_academic.html