Multiple Testing Procedures Controlling the FWER in a Two-Stage Group Sequential Design

• Activity Number: 459
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Li He*+ and Sanat K. Sarkar
• Companies: Merck and Temple University
• Keywords: Multiple Testing ; Familywise Error Rate ; Two-stage Group Sequential Design

Abstract

We consider testing multiple hypotheses by controlling the familywise error rate (FWER) in a two-stage group sequential design. Given prespecified boundaries for the first stage p-values, we determine boundaries for the second-stage p-values by controlling the FWER at some predetermined level either through a combination function combining the two p-values from both stages or by directly utilizing the dependence among them. We derive such two-stage procedures as a generalized version of the Bonferroni procedure or of the Holm and Hochberg procedure and prove their FWER control in the strong sense. We performed simulation studies to show that our procedures can provide significant power improvement over the corresponding one-stage procedures based only on the first stage data compared to the one-stage procedures based on all the data. We also illustrate our procedures by applying them in a real data set.