Practical Issues in Calculating Sample Size for Ordinal Data with Repeated Measures

• Activity Number: 617
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Alfred Balch* and Tom Greene and Colby Hansen and C.M.T. Sherwin
• Companies: University of Utah and University of Utah and and
• Keywords: GLMM ; GEE ; Crossover Trials ; Consulting

Abstract

With the wide availability of statistical software for Generalized Linear Mixed Models (GLMM) and Generalized Estimating Equations (GEE) it is useful to be able to plan studies with ordinal outcomes, e.g. Likert scale and repeated measures. Usual approaches for the problem, using normal approximations are less useful when the data are clustered towards one end of the scale, or when the distribution of responses is not known. It is useful to simulate multiple null and alternative hypotheses to assess the impact of design and sample size on power and cost. In this research we will consider the impact of a number of distributions and designs for a 2 or 4 period crossover trial with an ordinal response. Practical aspects, such as eliciting scientific input for null and alternative hypotheses and presenting power vs. design will be emphasized.