Multiplicity Adjustments for Correlated Binary Outcomes

• Activity Number: 212
• Type: Invited
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Andrew C. Leon*+
• Companies: Weill Cornell Medical College
• Keywords: multiplicity adjustment ; type I error ; correlated endpoints

Abstract

Clinical trials of psychopharmacologic and psychotherapeutic interventions tend to include multiple outcomes. For example, ratings of illness severity, response status, remission status, and functional impairment are commonly used in trials of treatments for bipolar disorder. The FDA recognizes that, at times, there is a need for multiple primaries. However, they make clear that in such a case multiplicity adjustments must be proposed in the protocol. That is not necessarily the case for NIMH funded protocols. Clinical investigators tend to resist multiplicity adjustments because of concern for statistical power. However, false positive treatments do not serve the clinical community.

There are many approaches to multiplicity. Most disregard the correlation among outcomes. This can result in a conservative hypothesis testing strategy. The James procedure (1991) is an alternative that accounts for multiplicity among correlated binary endpoints. A simulation study compares this approach with 3 alternatives: Bonferroni, Hochberg and unadjusted. Familywise type I error and statistical power are examined. The approaches are applied to a study of psychiatric interventions.