Comparison of Generalized Linear Models To Evaluate Factors Associated with Metabolic Syndrome

• Activity Number: 405
• Type: Contributed
• Date/Time: Wednesday, August 6, 2008 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Desta B. Fekedulegn*+ and Michael E. Andrew and John M. Violanti and Tara A. Hartley and Luenda E. Charles and Cecil M. Burchfiel
• Companies: National Institute for Occupational Safety and Health and National Institute for Occupational Safety and Health and The State University of New York at Buffalo and National Institute for Occupational Safety and Health and National Institute for Occupational Safety and Health and National Institute for Occupational Safety and Health
• Keywords: Metabolic syndrome ; generalized linear models ; count data

Abstract

Our goal was to evaluate potential use of several models where the outcome represents a count, rather than presence or absence, and estimate measures of association. Metabolic syndrome defined by Grundy (2005) is a binary outcome that refers to presence of three or more of five components, but components can also be summed as a count. Data from 96 randomly selected Buffalo police officers were used to examine associations between depression score (CES-D) and metabolic syndrome. Measures of association for 1 SD change in CES-D were calculated using both the binary outcome and count of syndrome components. Poisson regression used to estimate ratios of means for associations with count as an outcome were statistically significant yet methods using binary outcomes were not. This example shows the potential for Poisson regression to provide an interpretable alternative measure of association.