1. ^aUS Environmental Protection Agency, National Exposure Research Laboratory, Human Exposure and Atmospheric Sciences Division, Research Triangle Park, North Carolina 27711, USA
2. ^bUS Environmental Protection Agency, National Exposure Research Laboratory, Human Exposure and Atmospheric Sciences Division, Las Vegas, Nevada 89119, USA
3. ^cDepartment of Mathematical Sciences, University of Nevada, Las Vegas, Nevada 89154, USA
4. ^dDepartment of Environmental and Occupational Health, Emory University, Atlanta, Georgia 30322, USA

Correspondence: Peter P. Egeghy, Environmental Health Scientist, US Environmental Protection Agency, National Exposure Research Laboratory, Human Exposure and Atmospheric Sciences Division, MD E205-04, Research Triangle Park, NC 27711, USA. Tel.: +1-919-541-4103. Fax: +1-919-541-0905. E-mail: egeghy.peter@epa.gov

Received 2 February 2004; Accepted 27 October 2004; Published online 15 December 2004.

Abstract

The longitudinal NHEXAS-Maryland study measured metals, PAHs, and pesticides in several media to capture temporal variability. Questionnaires were concurrently administered to identify factors that influenced changes in contaminant levels over time. We constructed mixed-effects regression models for lead, phenanthrene, and chlorpyrifos (including metabolites) in indoor air, dust, dermal wipes, and biological fluids. Significant predictors represented time-varying activities as well as unchanging housing and demographic factors. There was little overlap among the models, with predictors generally reflecting the diverse characteristics of the target compounds. We estimated between- and within-person variance components to evaluate the reliability of the measurements. While only one measurement of lead in blood or chlopyrifos in dust was needed for a dependable estimate of an individual's average level, three to eight measurements were needed for most other compound/exposure medium combinations because of considerable temporal variability. Measurements in biological fluids and dust were generally more consistent than those in indoor air. The significant covariates in the full models preferentially reduced the between-person variance component. Since the regression models explained only 1–37% of the within-person variance, the questionnaires in this study provided only modest insight into the factors responsible for the temporal variability in the contaminant levels.