Spatial Analysis of Cardiovascular Data: A Simulation Study

• Activity Number: 348
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Samantha Seals*+ and Charles Katholi and Inmaculada Aban
• Companies: The University of Alabama at Birmingham and The University of Alabama at Birmingham and The University of Alabama at Birmingham
• Keywords: spatial analysis ; cardiovascular MRI ; Monte Carlo simulations

Abstract

The goal of this simulation study is to investigate the choice of working correlation structures in the analysis of spatially correlated data using mixed models. This study is motivated by the question of what impact does myocardial infarction (MI) have on the rotation of the left ventricle (LV) of the heart which is responsible for ejecting the blood out of the LV. Rotation values may be obtained from 16 segments of the LV via cardiovascular MRI (cMRI), resulting in spatially correlated data. To achieve the goal of this study, we performed simulation studies based on the assumption that rotation follows a multivariate normal distribution with spatial correlation. We simulated under four correlation structures: exponential, Gaussian, Matern, and spherical. We examine simulated type I error and power of tests that compares the mean rotation for two different populations (for instance, MI patients versus control patients) and the effect on the parameter estimate and its standard error. We also examine and compare the performance of goodness-of-fit indices (AIC and BIC) in the choice of the working correlation.