Efficient Pooling Methods for Skewed Biomarker Data Subject to Regression Analysis

• Activity Number: 35
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Emily Mitchell*+ and Robert H. Lyles and Michelle Danaher and Neil J. Perkins and Enrique F. Schisterman
• Companies: Emory University and Emory University and Eunice Kennedy Shriver National Institute of Child Health and Development and Eunice Kennedy Shriver National Institute of Child Health and Development and Eunice Kennedy Shriver National Institute of Child Health and Development
• Keywords: Biomarkers ; MCEM ; Pooling ; Regression models

Abstract

Pooling biological specimens prior to performing expensive laboratory tests can considerably reduce costs associated with certain epidemiologic studies. Recent research illustrates the potential for minimal information loss when efficient pooling strategies are performed on an outcome in a linear regression setting. Many public health studies, however, involve skewed outcome data, which often require a log-transformation prior to performing linear regression. Due to the non-linear nature of the log function, however, transforming the observed poolwise outcomes and applying standard methods does not necessarily provide consistent estimates of the coefficient parameters. In this study, we use Taylor series arguments along with simulations to explore the conditions under which applying standard linear regression to a simple log transformation of the poolwise outcomes produces defensible coefficient estimates. When this method is expected to fail, we propose the use of a Monte Carlo Expectation Maximization algorithm to identify maximum likelihood estimates of the parameter vector.