General Departmental Seminar Series
Proportional Hazards Frailty Regression Models
Michael Kosorok, Department of Statistics &
Biostatistics and Medical Informatics, University of Wisconsin-Madison
Wednesday, September 13, 2000, 4:00-5:00 pm
1221 Comp Sci & Statistics, 1210 W. Dayton
We consider inference for a rich class of proportional hazards frailty models which are one parameter extensions of Cox regression for right censored data. The estimated survivor functions are more accurate than those from the usual semiparametric model. We propose efficient estimation methods and establish uniform consistency and weak convergence of the estimators. A novel application of Markov Chain Monte Carlo to the profile likelihood permits inference about the finite dimensional parameter separately from the baseline hazard. The bootstrap is shown to be asymptotically valid when making simultaneous inference about all parameters. The techniques are illustrated in an analysis of a non-Hodgkin's lymphoma dataset for which the standard model fits poorly.
Back to General Departmental Seminar Series