Approaches to Estimate Between-and-Within-Subject Correlation Coefficients in Longitudinal Repeated-Measures Studies

• Activity Number: 550
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jennifer Cooper*+ and Jason P Sulkowski and Katherine J. Deans and Peter C. Minneci
• Companies: Center for Surgical Outcomes Research, Nationwide Children's Hospital and Center for Surgical Outcomes Research, Nationwide Children's Hospital and Center for Surgical Outcomes Research, Nationwide Children's Hospital and Center for Surgical Outcomes Research, Nationwide Children's Hospital
• Keywords: mixed model ; longitudinal ; covariance structures ; pediatrics

Abstract

Multivariate repeated-measures data offer a unique opportunity to examine the joint progression of multiple variables over time. Several distinct types of correlations between variables can be derived from such studies. The overall correlation between two variables can be decomposed into within- and between-subject correlations that reflect associations at the individual and collective levels. A number of different models are useful for estimating these correlations, with bivariate linear mixed models being most commonly used. Currently, the SAS MIXED procedure has few options for Kronecker product covariance structures and all best accommodate equally spaced measurements. Univariate mixed models with separate terms for between- and within-subject associations are an alternative that allows for unequally spaced measurements. We explore differences in between- and within-subject correlation coefficients derived from bivariate and univariate mixed models under a variety of covariance structures. We illustrate these differences using data on markers of intravascular hemolysis and nitric oxide consumption in children treated with extracorporeal membrane oxygenation.