Extending Multiple Contrast Tests to the General Linear Model to Evaluate Dose-Response Relationships

• Activity Number: 320
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Michael Paul McDermott*+ and Jason Leonard Morrissette
• Companies: University of Rochester and University of Rochester
• Keywords: Order restriction ; Multiple imputation ; Combining dependent p-values ; Fisher's combination method ; Covariate adjustment ; Clinical trials

Abstract

In randomized controlled trials of different dosages of a drug assigned to parallel groups of subjects, imposing order restrictions on the mean responses is often more appropriate than assuming a particular functional form for the dose-response relationship. In these settings it can also be useful to include covariates in the statistical model. Likelihood ratio tests for equality of ordered means that incorporate covariate adjustment are somewhat complex and rarely applied in practice. We propose a test that is based on multiple contrasts among the adjusted group means. This test is carried out using Fisher's statistic to combine the dependent p-values arising from these contrasts. The null distribution of this statistic can be well approximated by that of a scaled chi-square random variable. The contrasts can be chosen to yield a test with good power for alternatives ranging throughout the restricted parameter space. The test is generally easy to implement for a variety of partial order restrictions and can be used in conjunction with multiple imputation to accommodate missing data. An example from a randomized clinical trial is used to illustrate the proposed test.