Global Two-Group Multiple Endpoint Adjustment Methods Applied to Clinical Trials

Abstract

Ojien, clinical trial results are reported and used for claiming a treatment effect without adjusting for the multiplicity arising from the presence of multiple endpoints. It is well recognized that this practice is likely to inflate the type I error rate unless the endpoints are highly correlated. To control such an inflation of the type 1 error rate, two approaches, both appropriate on their own grounds, are in use. One is the 'global' approach which aims at demonstrating overall treatment effect and considers all given endpoints simultaneously. The other approach, often called the 'endpoint-specific ' approach, assesses significance for each endpoint separately, assuring meaningful control of the overall type 1 error rate. This paper addresses and discusses some of these global multiple endpoint adjustment methods in the context of clinical trial applications where two treatments are being compared (for example, active treatment with a placebo), andgives some simulation results to assess their peqormance regarding the a-level protection and power:

The effects of soy protein containing phytoestrogens on menopausal symptoms in postmenopausal women

“…As a preliminary, we used a global approach and analyzed treatment effects by multivariate analysis of variance (MANOVA), based on mean change in symptom score 21 . Dependent variables were combined in five symptom categories (See Table 3) with the independent categorical variable in each of the five models being treatment status.…”

Comparing Treatment and Control Groups on Multiple Outcomes

“…In previous studies, total sample size conditions ranged from N = 20 to N = 100 (Sankoh et al, 1999;Mudholkar et al, 2001). Lix and Fouladi (2007) demonstrated that Type I error and power rates of the composite step-down test are sensitive to the ratio N=m.…”

“…Power to detect departures from the null hypothesis is influenced by both the magnitude and pattern of correlation among the outcome measures (Sankoh et al, 1999). Three correlation matrices were investigated: ρ j = Q, ρ j = R, and ρ j = T (j = 1, 2).…”

“…A number of procedures to test directional alternative hypotheses have been proposed in previous studies. However, most of these procedures assume covariance homogeneity and/or multivariate normality (Sankoh, Huque, Russell, & D'Agostino, 1999;Tamhane & Logan, 2003). There has been little consideration of procedures for testing a directional alternative hypothesis when covariances are heterogeneous and the data are nonnormal (Mudholkar, Kost, & Subbaiah, 2001;Tamhane & Logan, 2002).…”

Accurate Critical Constants for the One-Sided Approximate Likelihood Ratio Test of a Normal Mean Vector When the Covariance Matrix Is Estimated

“…This results in a highly liberal ALR test if one uses the null distribution derived by Tang et al which assumes a known covariance matrix; see Reitmeir and Wassmer (1996) and Sankoh et al (1999 The outline of the paper is as follows. In Section 2 we set the notation and review the ALR test for the two sample problem in the known common covariance matrix (homoscedastic)…”