Abstracts: ISEE 20th Annual Conference, Pasadena, California, October 12–16, 2008: Contributed Abstracts
Exposure assessment for large-scale, panel-based epidemiology studies can be challenging even in a data-rich study. Typically, data are highly diverse both spatially and temporally. In order to take advantage of these data to predict exposures, models accounting for changes in both scales must be incorporated. A ten year epidemiologic study of the progression of asthma in young children in relation to air pollution has enrolled 302 children who have quarterly health and exposure evaluations which include semiannual spirometry in the field office; each child also participates in three two-week panels each year, during which daily symptoms, medication use, and exposure-related activities are recorded and pulmonary function tested twice daily. The study is evaluating acute effects associated with daily exposure to air pollutants (e.g., PM, elemental carbon, ozone) and bioaerosols (fungal spores, pollen grains, endotoxin).
The importance of both spatial and temporal aspects of a rich data set is illustrated for representative pollutants from the class of polycyclic aromatic hydrocarbons (PAHs), a class of compounds often used as a surrogate for diesel exhaust. PAH data were collected daily by the PAS2000 monitor for a year at three fixed locations, an EPA Supersite and two trailer sites at elementary schools. The PAS2000 provides realtime measurements of total particle-bound PAH. During the same year, intensive air pollution sampling took place at 84 homes including five to ten 24-hr PAH filter measurements. However, since the measurements were not made at all 84 homes on the same day, the between-home and within-in home variability must be disentangled. Therefore, the data were used to formulate a land use regression model that utilizes the mixed modeling approach to account for both longitudinal and cross-sectional variability in the samples. The PAH filter concentrations at participant homes were the dependent variable; the independent variable included the particle-bound PAH concentrations at the fixed sites, meteorological data (wind direction, wind speed, relative humidity (%), temperature and precipitation), source data (traffic and land use), and other temporal and spatial variables (agricultural burning, season, etc) were. Highly predictive variables in the models were PAS2000 measurements, wind direction, wind speed, season and road type closest to home. LUR allows for extraction of spatial relationships in a more deterministic fashion than purely stochastic methods such as kriging or co-kriging.
One of the important PAHs for this study is benzo[a]pyrene. The overall mean of benzo[a]pyrene is 0.41 μg/m3, with a standard deviation of 0.88 μg/m3. The highest seasonal levels are in the winter (November through February) with an average of 0.79 μg/m3, and a range of 0.00 to 8.13 μg/m3. The average daily range for benzo[a]pyrene is 0.60 μg/m3, reflecting the degree of spatial variability. The daily range is higher, on average, in the winter than in the remainder of the year.
Important variables for predicting benzo[a]pyrene are the PAS2000 measurements, season of sampling, average daily temperature, direction of sampling site from the EPA Supersite, and amount of agricultural burning on the day of sampling.