You are here:
MIXED MODELS ANALYSIS OR URBANIZATION LEVEL ON CHLORPYRIFOS EXPOSURE
Citation:
Egeghy, P P. MIXED MODELS ANALYSIS OR URBANIZATION LEVEL ON CHLORPYRIFOS EXPOSURE. Presented at International Society of Exposure Analysis 2002 Conference, Vancouver, Canada, August 11-15, 2002.
Impact/Purpose:
The NHEXAS analysis projects being conducted by NERL address the following scientific issues/questions:
- Do exposure distributions vary by demographic group?
- What is the impact of censoring (BDL) on multimedia distributions and associations?
- Which questionnaire/diary items are most useful in explaining variability or identifying "high exposure" groups?
- How do exposure classifications and measurements compare in their ability to explain variability in exposures and biomarkers?
- Can changes in exposures over time be explained by dietary and questionnaire/activity information?
- How well do NHEXAS findings compare with existing models/assessments (e.g., NATA)?
Description:
The National Human Exposure Assessment Survey (NHEXAS) pilot studies were conducted from 1995 through 1997 to examine human population exposure to a wide range of environmental contaminants. In one of the studies, NHEXAS-Maryland, a longitudinal design was used to repeatedly measure aggregate residential chlorpyrifos exposure in a stratified random sample of 80 individuals. Chlorpyrifos is a semi-volatile insecticide which has been found to persist for weeks following application. Until its recent removal from the retail market, chlorpyrifos was commonly used for indoor and outdoor treatments by both residents and commercial applicators and also as a termiticide during residential construction. The effect of urbanization level (i.e., characterization of participant's community as urban, suburban, or rural) on chlorpyrifos exposure was investigated using mixed-effects regression models to accommodate longitudinal data and to estimate variance components. Three surrogates of exposure, namely, indoor air concentrations (n = 97), surface dust loading (n = 123), and creatinine-adjusted urinary metabolite (3,5,6-trichloro-2-pyridinol) concentrations (n = 341), were considered separately. The effect of the specified covariance structure was explicitly evaluated to determine if the pooling of variance components, as is often performed by default, is appropriate. Likelihood ratio tests comparing possible covariance structures suggested that the pooling of within- and between-person variance components among urbanization levels may not be appropriate for indoor air concentrations (p < 0.001). The variance estimates indicated that indoor air concentrations were far more variable among households in rural communities than in urban or suburban communities. Furthermore, since maximum likelihood estimates of the means depend upon the variance components, the choice of covariance structure exerted a large effect upon the estimates of indoor chlorpyrifos concentrations. Mean indoor air concentrations for urban, suburban, and rural households in this study were estimated to be 7.3, 20.2, and 23.0 ug/m3, respectively, when allowing for distinct variance components, but were estimated to be 11.9, 16.6, and 11.7 ug/m3, respectively, when pooling the components. Similar analyses, on the other hand, indicated that variance components may be pooled among urbanization levels when evaluating surface dust and urinary metabolite concentrations. These preliminary results, albeit based on relatively small sample sizes, indicate that the choice of covariance structure can have a large effect on the results of an analysis and must be given appropriate consideration.
This work has been funded wholly or in part by the United States Environmental Protection Agency. It has been subjected to Agency review and approved for publication.