Saturday, February 25
PS3 Poster Session 3 and Continental Breakfast Sat, Feb 25, 8:00 AM - 9:15 AM
Conference Center AB

A Guide to Modeling Strategies for Immunological Count Data (303456)

Pete Anderson, University of Colorado Skaggs School of Pharmacy and Pharmaceutical Sciences 
Elizabeth Connick, University of Arizona 
Gary Grunwald, University of Colorado at Denver 
Sam Mawhinney, University of Colorado at Denver 
Amie Meditz, Boulder Community Health, Beacon Center for Infectious Diseases 
*Claire Palmer, University of Colorado at Denver School of Medicine 

Keywords: count data, generalized linear mixed models, repeated measures, HIV

Immune cell count is an informative and frequently used outcome measure in the field of immunology. It is typically obtained through multiple sampling methods, whereby tissue samples are taken from several areas and subsequently divided into smaller sections before they are assayed. For analysis, data can be aggregated to a single measure at the subject level or kept in clustered form. Additionally, total tissue area collected is rarely uniform and must be properly accounted for. We used example data from an HIV drug study and simulations to compare the following modeling strategies: ordinary least squares regression, generalized linear models, linear mixed models, generalized linear mixed models and generalized estimating equations. Simulation results were evaluated on the basis of bias and confidence interval coverage. We ultimately provide a modeling guide for different scenarios encountered when analyzing this type of data, with special attention paid to the issues of low counts, non-uniform areas and small sample sizes.