Comparisons of multiple treatment groups with a negative control or placebo group: Dunnett test vs. closed test procedur

Abstract

Several treatments are usually compared with a control using the Dunnett test. As an alternative, three variants of the closed testing approach are considered, one with ANOVA-F-tests, one with MCT-GrandMean and one with global Dunnett-tests in the partition hypotheses. The problemComparisons of multiple treatment groups with a control (or placebo) group are frequently performed in biomedical experiments: in a plant molecular experiment new mutants were compared with the wild type [24], in a toxicological assay selected clinical chemistry endpoints in three dose groups with disodium adenosine-triphosphate were compared with a water control [13] and in neurological clinical trial three candidate drugs amiloride, fluoxetine, and riluzole were compared with a placebo group for the primary endpoint volumetric MRI percentage brain volume change [6]. Typically, the Dunnett procedure is used [7] in such randomized one-way designs. The main advantage is the availability of simultaneous confidence intervals, alternatively to the multiplicity-adjusted p-values. An alternative is the closed testing procedure (CTP) [15], because it is available for any test in the GLM (particularly its small n i approximations [19]) and can be easily generalized to further joint analysis, e.g. considering multiple endpoints [5]. Three test versions are considered here for the global and partition hypotheses: the common ANOVA F-test, the global multiple contrast test for comparisons against the grand mean [18] and the global Dunnett test. Since the last two are based on multiple contrast tests, subset contrasts can be formulated for all k-samples, allowing the entire degree of freedom to be used in all sub-tests.

Closed testing procedures for treatment-versus-control comparisons and multiple correlated endpoint

“…Coprimary endpoints (case i)) are analysed be means of an intersection-union test (IUT), where the global claim is proved if each marginal test is under H 1 , each at level α. Primary endpoints (case ii)) are analysed almost differently by means of an union-intersection test (UIT), e.g., by a maxT-t test [10]. Here, the claim is proved if at least one, anyone, marginal test is under H 1 , each at an adjusted level α i whereas the entire procedure controls the FWER in a strong sense.…”