Comparison of Power from Area Under the Curve and Mixed Effects Models Methodologies for Profile Analysis

• Activity Number: 452
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Robbie Beyl*+ and Jeff Burton and William Johnson
• Companies: Pennington Biomedical Research Center and Pennington Biomedical Research Center and Pennington Biomedical Research Center
• Keywords: AUC ; Mixed model ; Profile analysis ; OGTT ; Power

Abstract

Assume that study subjects are randomly assigned to one of two treatments and assessed for a specific response at time 0 (baseline) and each of three post-treatment times. Further, assume the objective is to compare the treatments with respect to their response profiles across time. This setting is representative of an oral glucose tolerance test used in diabetes research. While the analytical method of choice is to employ mixed effects models for repeated measures, many biomedical researchers prefer comparing the treatments in terms of area under the curve (AUC). Thus, the goal is to determine if one of the analytical methods is more powerful than the other. Data are generated under the null hypothesis and for various configurations of means under the alternative hypothesis. These cases include changes to none, one, two, and all three of the post treatment means. The power curves estimated from these data are adjusted such that estimated power for both tests is equal to the nominal significance level under the null hypothesis. Once adjusted, we compare power between the two testing methodologies across the specified mean configurations.