Analytic and Study Design Considerations When a Continuous Regression Outcome Is Assessed in Pooled Samples

• Activity Number: 185
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Robert H. Lyles*+ and Amita K. Manatunga and Emily Mitchell
• Companies: Emory University and Emory University and Emory University
• Keywords: Grouped data ; Pooled data ; Regression ; Study design

Abstract

The process of pooling biological samples prior to laboratory assessment has recently received renewed attention because of its potential cost efficiency implications for epidemiological studies. In this talk, we demonstrate how ordinary and weighted least squares are directly applicable to the linear regression analysis of continuous outcomes assessed in equal and unequal-sized pools, respectively. We then consider the use of subject-specific covariate information to obtain potentially marked efficiency gains for regression coefficient estimation, relative to random pooling. This informed pooling strategy is contrasted against the use of targeted sampling of subjects for individual laboratory assessment. As time permits, we will also demonstrate the feasibility of maximum likelihood analysis in a mixed linear model setting when repeated or longitudinal outcome data are pooled within subjects. These results make interesting case studies for students of linear models, while also providing insights that carry over to other types of regression modeling when pooling is employed.