Properties of Multiple Intersection-Union Tests for Multiple Endpoints in Combination Therapy Trials

Abstract

We consider intersection-union tests involving multiple endpoints in a combination drug trial, for which we control the familywise error rate in the strong sense using closed testing methods. Bonferroni-Holm, Simes-Hommel, and Resampling-Based methods all are considered in this context. Familywise error rate control heuristics are developed and evaluated using a simulation study that is specifically tailored to the intersection-union setting. Both Resampling-Based and Simes-Hommel uniformly outperform the Bonferroni-Holm. Using simulations of power, the choice of Simes-Hommel versus Resampling-Based is seen to depend on the particular alternative of interest. Because it is simpler and has generally good power, we recommend the Simes-Hommel intersection-union tests. The techniques are illustrated using real data from a clinical trial to evaluate a combination asthma therapy.

The effect of budesonide and formoterol in one pressurized metered-dose inhaler on patient-reported outcomes in adults with mild-to-moderate persistent asthma

“…All statistical analyses were conducted as two-sided tests. A hierarch ical approach 27 to control for multiplicity of statistical testing was applied to the data from this study, involving the coprimary endpoints (pre-dose FEV 1 and post-dose FEV 1 ) and the percentage of patients who met pre defined criteria for worsening asthma (all presented elsewhere 19 ), as well as the AQLQ(S) overall score and the percentage of patients with symptom-free days (presented herein). All other variables were analyzed without adjustment for multiple comparisons, with p ≤ 0.05 considered statistically significant.…”

Multiple Comparisons, Statistics of

“…Hochberg's step-up testing method always finds the same significances as Holm's method, but sometimes more. The improved power comes at a price though: The method can fail to control the FWER in cases of negatively correlated p-values (Westfall et al, 2001).…”

Generalized confidence intervals compatible with the Min test for simultaneous comparisons of one subpopulation to several other subpopulations

“…Applications of the Min test and methods based on the intersection-union tests for demonstrating drug efficacy have been addressed in several papers (Hung, Chi and Lipicky 1993, Wang and Hung 1997, Hung 2000, Westfall, Ho and Prillaman 2001. Tamhane and Logan 2004, Buchheister and Lehmacher 2006, Soulakova 2009, Soulakova 2010).…”

“…This framework has been presented elsewhere (Westfall, Ho and Prillaman 2001, Soulakova 2009) and is not discussed in this paper.…”