|
Activity Number:
|
420
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #309687 |
|
Title:
|
A Taxonomy of Hazards for the Generalized Gamma Distribution
|
|
Author(s):
|
Christopher Cox*+ and Haitao Chu and Michael Schneider and Alvaro Munoz
|
|
Companies:
|
Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health
|
|
Address:
|
615 N Wolfe St Room E7642, Baltimore, MD, 21205,
|
|
Keywords:
|
survival analysis ; parametric models ; generalized gamma distribution ; non-proportional hazards ; relative times
|
|
Abstract:
|
The generalized gamma (GG) offers flexibility for analysis of censored and truncated survival data. We present a GG hazard taxonomy. The half plane of the scale and shape parameters is divided into four regions within which hazards are bathtub and arc shaped, increasing and decreasing. The GG includes the Weibull, lognormal and gamma, and is closed under power transformations so inverse distributions are included. Characterization of the quantile function allows use of relative quantiles as effect measures. We apply the taxonomy to describe survival after clinical AIDS during different eras of HIV therapy, with hazard functions of very different shapes. Standard errors for relative times and hazards are computed using the delta method and checked with the bootstrap. These methods allow cohort studies to provide assessment of treatments at the population level of public health relevance.
|
