A Modified Tukey Linear Trend Test in Dose-Response Studies

• Activity Number: 389
• Type: Topic Contributed
• Date/Time: Wednesday, August 10, 2005 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Chao-Yin Chen*+ and Lan Kong and Huei Wang and Thomas Liu and Brian Wiens
• Companies: University of Minnesota and University of Pittsburgh and Amgen Global Biostatistics and Amgen Global Biostatistics and Amgen, Inc.
• Keywords: dose-response ; multiplicity ; multiple comparisons ; Tukey linear trend test

Abstract

Dose-response studies with three active dose groups and placebo are commonly considered during drug development. In such studies, one of the objectives is to detect whether there is a difference between placebo and active doses. The multiplicity issue arises due to multiple comparisons. The Tukey linear trend test can be used to compare placebo with active doses---it is powerful when the dose-response curve is monotone. However, the test may have low power when the assumption of monotone dose-response relationship is violated. In this paper, a modified Tukey linear trend test using fallback procedure and the information on correlation structure between test statistics is proposed. This modified Tukey linear trend test is shown to be more robust to various dose-response curve patterns than the original. It also is shown that the modified Tukey linear trend test is either comparable or superior to other commonly used methods in many realistic situations.