SEMINARS
|
|
Spring 2006
STATISTICS
COLLOQUIUM
Friday, May 5, 2006
3:30-4:00—Refreshments
4:00-5:00—Talk
Yost Hall, Room 101
Jason Fine
University of Wisconsin--Madison
Department of Statistics
Department of Biostatistis & Medical Informatics
Functional Regression Modelling of Survival Proceses
We consider functional regression modelling of point processes which are continuously observed until censored.
Such processes play a central role in applications of multistate and multivariate survival analysis. Varying coefficient mean and association models are formulated for covariate effects on the processes. The continous observation scheme is exploited: the coefficients are estimated nonparametrically by extending GEE to the functional set-up. Optimal weighting for functional GEE is discussed; the results from the finite dimensional case to do not extend automatically. The estimators converge at the parametric rate and without smoothing, unlike with varying coefficient intensity models. Uniform consistency and weak convergence is established, which provides the theoretical basis for new tests for covariate effects, parametric sub-modeling of the effects, and goodness-of-fit testing. Several real examples, including prevalence of CGVHD in a leukemia clinical trial and familial aggregation of alcoholism in a genetic epidemiologic study, are used to illustrate the practical utility of the proposed methods.
|
