Localization versus Holism in the Analysis of Neurological Imaging Data: Power Comparisons of Multivariate and Univariate Methods

• Activity Number: 281
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2007 : 10:30 AM to 12:20 PM
• Sponsor: WNAR
• Author(s): Dawson Hedges*+ and Suzanne Hendrix and Bruce Brown
• Companies: Brigham Young University and Myriad Pharmaceuticals and Brigham Young University
• Keywords: brain image data ; MANOVA ; posttraumatic stress disorder

Abstract

Clinical studies are often burdened by the constraints of small sample size, due to cost considerations, diagnostic issues, and comorbidity. The resultant loss of power can obscure clinically meaningful information. In a brain imaging (MRI) study, univariate analyses showed no significant differences in volumes of 82 brain areas (in 20 main regions) between five healthy controls and six subjects with posttraumatic stress disorder. Two measures of PTSD, the CAPS and a diagnostic interview, were found to correlate highly (r = .952). Surprisingly, a lens model analysis revealed that of the 20 main regions, 11 accounted for over 35% of the diagnostic accuracy, with one accounting for 95%, and several over 50%. Using this information, an appropriate MANOVA analysis was crafted that showed similar high levels of variance accounted for. A strategy is proposed for dealing with small N data.