|
Activity Number:
|
121
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 12:00 PM to 1:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #300877 |
|
Title:
|
Exact Analysis of Correlated Binary Response Data--Test of Treatment Effect for Parallel Group Design
|
|
Author(s):
|
Dar Shong Hwang and James S. Lee*+ and Chyi-Hung Hsu
|
|
Companies:
|
BRSI and Sankyo Pharma and Novartis Pharmaceuticals
|
|
Address:
|
399 Thornall St., Edison, NJ, 08837,
|
|
Keywords:
|
correlated binary response ; parallel group design ; multinomial distribution ; exact conditional distribution ; treatment effect
|
|
Abstract:
|
For large clinical trials with repeated categorical measurements, current statistical methods tend to apply large sample approximation. The adequacy of the approximation could be questionable in certain situations. An exact analysis for parallel group design having repeated measurements with binary response is proposed. When the binomial samples are independent, exact and asymptotic tests for treatment effect assuming no treatment by stratification interaction are well known. This paper attempts to test the same hypothesis but with correlated binary response data over time. We express again the constant odds ratio in terms of the parameters of the multinomial distribution. Existence and optimality of conditional distributions for the inference on these parameters are investigated. Subsequently, exact conditional distributions are derived. Statistical inference to evaluate if there is a constant odds ratio across time may then be conducted. When there is no treatment by time interaction, the same conditional distributions are further modified to test the treatment effect. The above methodology may be extended from two time points to three or more time points.
|
