George Mason University
Statistics Colloquium Series
Seminar Announcement


Vaccines and Survival

Ken Berk
FDA and Illinois State University


and

Peter Lachenbruch
FDA


ABSTRACT

How do you assess the potency of 30 lots of vaccine, as compared to a standard? Each of the lots is used to vaccinate animals with three different dilutions of the vaccine. The animals are later "challenged" with the disease, and they are observed for 10 days. Survival (especially, median survival time) is compared among the lots and dilutions. We discuss the fitting of various models for the censored data, including the Cox proportional hazards model and several parametric models. The generalized gamma family (exponential, Weibull, lognormal, gamma) has the nice property that the ratio of median survival times is independent of dilution and of the quantile (any other percentile gives the same result as the median). The generalized gamma is of particular interest and a limiting case of it, the Pareto distribution, has the interesting property that there is a guarantee time - for values below it there are no deaths. Fitting the Pareto model is an intriguing problem, and the Pareto has the nice property of the generalized gamma family. The ratio of pth lifetime percentiles does not depend on dilution, and it is the same for each percentile.


Friday, April 13, 2001
Johnson Center, Assembly Room D
Seminar at 10:45 a.m.
Refreshments at 10:30 a.m.
For the 2001 Spring Seminar Schedule, go to
www.science.gmu.edu/statseminars