New Insights into Modeling Exposure Measurements Below the Limit of Detection

• Activity Number: 179 - Statistical Methods for Measurement Error and Missing Data in Covariates/Exposures
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ana Maria Ortega-Villa* and Danping Liu and Mary H Ward and Albert S Paul
• Companies: National Institutes of Health and National Cancer Institute and National Institutes of Health and National Institutes of Health
• Keywords: limit of detection; substitution methods; multiple imputation

Abstract

In environmental epidemiology it is of interest to assess the health effects of environmental exposures. Some exposure analytes present values that are below the laboratory limit of detection (LOD), which is the analyte's lowest detectable value that can be differentiated from a blank value. There have been many methods proposed for incorporating biomarkers subject to lower detection limits as measure of exposure in risk modeling using logistic regression. These range from substitution methods that impute a fixed value, multiple imputation, to making distributional assumptions on the exposure value and treating the lower limit of detection (LOD) as being left censored. This work provides a fresh look at these approaches, particularly with respect to assumptions that are inherently non-verifiable (assuming we do not see measurements below the LOD). Specifically, we examine the robustness of relative odds ratios to the distribution of the exposure as well as the to the assumption that the relationship between exposure and risk is the same above and below the LOD.