Effect of Misspecified Correlations in Parametric Multiple Testing

• Activity Number: 599
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Changchun Xie*+ and Xuewen Lu and Din Chen and Radhey S Singh
• Companies: University of Cincinnati and University of Calgary and University of Rochester and University of Guelph, Department of Mathematics and Statisitcs
• Keywords: Weighted multiple testing correction ; Correlated endpoints ; Multivariate normal distribution

Abstract

In clinical trials, multiple endpoints are usually correlated. However, many commonly used multiple testing correction methods proposed to control family-wise type I errors disregard the correlation among the endpoints, for example, the Bonferroni correction and Holm procedure. Recently, some parametric multiple testing methods have been proposed to take into account correlations among endpoints. However, the exact correlations among endpoints are usually unknown. If the correlations are misspecified, how robust are these parametric multiple testing methods in controlling family-wise type I errors? In this paper, simulations are conducted to study the effect of misspecified correlations in these parametric multiple testing methods along with an example to address this question.