EPIDEMIOLOGIC EVALUATION OF MEASUREMENT DATA IN THE PRESENCE OF DETECTION LIMITS

Abstract

ISEE-83 Introduction: Quantitative measurements of environmental factors greatly improve the quality of epidemiologic studies, but can pose challenges due to the presence of upper or lower detection limits (DL) or interfering compounds, which do not allow for precise measured values. Aim: This paper considers the regression of an environmental measurement (dependent variable) on several covariates (independent variables), and the calculation of odds ratios based on logistic regression with an environmental measurement as a risk variable. Methods: Various strategies are commonly employed to impute values for interval-measured data, including assignment of the DL value or one-half the DL value (DL/2), or zero to non-detected values. Some have proposed the use of “fill-in” values randomly selected from an appropriate distribution. We conduct a simulation study that compares these methods with truncated data methods (e.g., Tobit regression) and multiple imputation methods for analyzing measurement data with DLs. We illustrate the various approaches using measurements of pesticide residues in carpet dust in control subjects from a case-control study of non-Hodgkin lymphoma. Results and Discussion: Based on a simulation study, we found that the approaches that insert the DL or DL/2 can be biased, unless the percentage of measurements below DLs is small (5-10 percent). The fill-in approach generally results in unbiased parameter estimates, but may produce biased variance estimates and thereby distort inference when 30 percent or more of the data are below DLs. Truncated data methods (e.g., Tobit regression) and multiple imputation offer two unbiased approaches for analyzing measurement data with DLs. If interest resides solely on regression parameters, then Tobit regression can be used. If individualized values for measurements below DLs are needed for additional analysis, such as relative risk regression or graphical displays, then multiple imputation produces unbiased estimates and nominal confidence intervals unless the proportion of missing data is extreme.