![[logo] US EPA](https://webarchive.library.unt.edu/eot2008/20081107175850im_/http://www.epa.gov/epafiles/images/logo_epaseal.gif){kind=logo}
2005 Progress Report: Spatial Exposure Models for Assessing the Relation Between Air Pollution and Childhood Asthma at the Intra-urban Scale
EPA Grant Number: R831845Title: Spatial Exposure Models for Assessing the Relation Between Air Pollution and Childhood Asthma at the Intra-urban Scale
Investigators: Jerrett, Michael
Institution: University of Southern California
EPA Project Officer: Saint, Chris
Project Period: November 1, 2004 through October 31, 2007
Project Period Covered by this Report: November 1, 2004 through October 31, 2005
Project Amount: $449,966
RFA: Environmental Statistics Research: Novel Analyses of Human Exposure Related Data (2004)
Research Category: Environmental Statistics , Health Effects
Description:
Objective:The objectives of this research project: (1) to derive proximity-based, geostatistical, land use regression, and dispersion models of intra-urban exposure as well as an individual exposure model (IEM) for O3, NO2, NO, and fine particles (PM2.5) for the 12 Children’s Health Study (CHS) communities; (2) to assess, with empirical and simulation models, which of the ambient exposure models assigned to subjects in the CHS cohort results in the lowest measurement error when compared to household measurements; (3) to use the modeled ambient and personal exposures to assess air pollution-health associations in CHS data sets for determining whether more refined exposure models produce larger health effects; and (4) to apply resulting exposure and health effects estimates to derive the burden of incident asthma attributable to air pollution in the 12 communities.
Progress Summary:This report documents accomplishments for the first year of the U.S. Environmental Protection Agency (EPA) Science To Achieve Results (STAR) grant entitled, “Spatial exposure models for assessing the relation between air pollution and childhood asthma at the intra-urban scale”. We are reporting on work completed to June 2006 because administrative delays in the initial contract award and Institutional Review Board Reviews postponed the start of the project until June of 2005. Since that time, we have made excellent progress toward the study objectives, and we have published results in high impact journals. We have also contributed to the training of a Ph.D. student in Biostatistics (Roger Chang), a Ph.D. student in Epidemiology (Ketan Shankardass), and a Postdoctoral Fellow (John Molitor).
Although the scope of the project remains unchanged, we obtained significant new funding from EPA and the National Institute of Environmental Health Sciences (NIEHS) to renew our Children’s Environmental Health Center and to continue our project operating grant on childhood asthma incidence. With these two related projects, we have gained access to over 1,000 field measurements of O3, NO, and NO2. These field measurements were suggested as a possibility for supporting our exposure assessment program, but the results of the competition were unknown at the time of the STAR grant application. Because these measurements have the potential to improve the robustness of our exposure measurement error modeling, we decided to expand the scope of the study to include the communities where these measurements were taken and to include the relevant health data associated with these measurements. Specific accomplishments are detailed below.
Bayesian Exposure Measurement Error Analysis
We have progressed well in implementing the main Bayesian exposure measurement model. Our first paper details this model and focuses specifically on the issue of missing exposure information. This paper was published in the number-one ranked American Journal of Epidemiology. The paper laid groundwork for our ongoing assessment of the performance of different spatial exposure models. The key finding was that incorporation of exposure information from multiple sources, with explicit allowance for measurement error, could alter the results of health effects assessment such that results would become significant with the full integration of information in comparison to more naïve models.
Although numerous epidemiologic studies now use models of intra-urban exposure, there has been little systematic evaluation of the performance of difference models. In our second paper, now in review for publication in Environmental Health Perspectives, we develop a modeling framework for assessing exposure model performance and the role of spatial autocorrelation in health effects estimation. Data were obtained from an exposure measurement substudy of subjects from the Southern California CHS. We examined how the addition of spatial correlations to a previously described unified exposure and health outcome modeling framework affects estimates of exposure-response relationships using the substudy data. The methods proposed build upon previous work by Molitor, et al. (2006), which developed measurement-error techniques to estimate long-term NO2 exposure and its effect on lung function in children. This paper further develops these methods by introducing between- and within-community spatial autocorrelation error terms. The analytical methods developed are set in a Bayesian framework where multistage models are fitted jointly, properly incorporating parameter estimation uncertainty at all levels of the modeling process. Findings suggest that the inclusion of residual spatial error terms improves the prediction of health effects. The results also demonstrate how residual spatial error may be used as a diagnostic for comparing exposure model performance. In general, the California Line Source (CALINE) model outperformed simple distance metrics of exposure in terms of maximum explanation of residual spatial variance.
Incident Asthma Analysis
Background. Studies examining associations between traffic-related air pollution and asthma have produced inconsistent results, possibly due to exposure measurement error and an emphasis on prevalent rather than incident disease. The purpose of this study is to measure the association between estimates of traffic pollution exposure and incidence of asthma in the children.
Methods. A sample of 229 children (30 incident cases) was selected from participants in the Southern California CHS, a prospective cohort investigation of air pollution and respiratory health. Individual covariates and new onset asthma incidence were reported annually through questionnaire surveys during 8 years of followup (beginning when the children were 10 years old). Children had nitrogen dioxide (NO2) monitors placed near their home for 2 weeks in the summer and 2 weeks in the winter. A line source CALINE4 dispersion model also was used to predict nitrogen oxide (NOx) concentrations. Cox proportional hazards models with allowance for community random effects were used to assess the risk of asthma onset in relation to pollution exposures.
Results. There was a positive association between incident asthma and markers of traffic pollution. Modeled NOx levels were significantly associated with incident asthma, with a hazard ratio (HR) across the interquartile range (IQR) of 1.65 (95% CI: 1.17-2.32). Household NO2 had a HR of 3.25 (95% CI: 1.35-7.85) across the IQR.
Conclusions. We found significant associations between markers of traffic pollution exposure and asthma incidence. These findings suggest that traffic-related air pollutants contribute to the development of asthma in children.
Because few previous studies had demonstrated a significant association between incident asthma and pollution, it was an important first step for us to establish this relationship. In the next year we will apply the Bayesian framework to incident asthma in the larger cohort, where we now have available over 1,000 monitoring locations. An abstract summarizing these findings was published. A paper reporting these and other results is in preparation for submission to a high-ranking journal.
Individual Exposure Model Applications
The IEM was developed to estimate exposure of CHS participants to vehicle-related air pollutants (Wu, Lurmann, et al., 2005). The IEM tracked children’s time-activity patterns and integrated their exposures from the five microenvironments in which they spent most of their time—residential indoor, residential outdoor, school indoor, school outdoor, and in-vehicle. All other locations were grouped into a single category for which residential indoor concentrations were assigned.
In this study, the IEM model was applied to update and extend the database of estimated personal exposure of approximately 3,600 CHS participants. Specifically, it was applied using local-scale traffic inputs derived from the latest CALINE4 simulations (as described above), and it was applied for more years of the study to facilitate longitudinal analyses of personal exposures and health outcomes. New IEM simulations were made for CHS Cohorts A, B, C, and D for 1994, 1997, and 2000. The results showed the expected differences over time and were otherwise consistent with previous results.
Comparisons of Traffic Activity Databases
Spatial Accuracy of Roadways. The accuracy of traffic activity and roadway data is important in estimating air pollutant concentrations caused by motor vehicle emissions. In a previous study, we reported significant spatial discrepancies between the California Department of Transportation (Caltrans) roadway network and the TeleAtlas MultiNet™ (TAMN) roadway network up to hundreds of meters (Wu, Funk, et al., 2005). Although the TAMN data showed better spatial accuracy in roadway geometry than the Caltrans data, they did not contain traffic activity information. Therefore, a geographic information system (GIS) algorithm was developed and applied to transfer Caltrans traffic data to the TAMN network, yielding a Caltrans_Tele traffic database, which contains Caltrans traffic counts for the TAMN roadway network.
Recently, the TeleAtlas Company released a new roadway database, Dynamap®/Traffic Counts, which was originally developed by Geographic Data Technology (GDT). The GDT traffic data (version 8.2) were obtained in July 2005. The GDT data had similar geographical information as TAMN but were loaded with annual average daily traffic volume data as point coverage on the streets where the counts were taken. As part of this study, the traffic count data in the GDT database were compared with the Caltrans data.
Methods. The GDT traffic count data were compiled for different years (1996 to 2003), while Caltrans counts represent a single year, 2000. The GDT data were projected to 2000 using an annual growth rate of 3 percent in order to compare the data. The GDT traffic counts were transferred from their measurement locations to as many street segments as possible, based on street names and relative locations. Then, a point coverage map of traffic volumes at the center of each GDT street segment with a valid traffic count was created. These points were snapped to the nearest Caltrans roadways within a 200 m buffer. Traffic counts at the snapped GDT points were compared to the corresponding nonhighway Caltrans roads with the same name. The comparison of freeways and highways was conducted without name-matching because Caltrans freeway and highway names were coded (e.g., 069.300/074.973) and were difficult to match with actual names.
Forty-five percent of all the GDT road links (including the smallest roads) were assigned valid traffic volume data after the traffic counts were transferred from point locations to roadway segments. Of all the GDT links with valid traffic counts, 94 percent of the link center-points were snapped to the nearest Caltrans roads within a 200 m buffer. Of all the snapped GDT points, 54 percent had the same names as the Caltrans surface roads. These GDT and Caltrans matching pairs had a mean traffic count of 13,100 and 18,600, respectively, with a correlation coefficient of 0.64. The significantly lower GDT count was mainly due to the fact that GDT may use two or more lines to represent a freeway or a surface street, while Caltrans generally uses a single line to represent a two-direction, multiple-lane roadway. After the GDT data were adjusted for the multiple-lane effect, the average GDT traffic count increased to 16,700, and the correlation increased to 0.69.
Six percent of the GDT points were snapped to the nearest Caltrans freeways. The matches were based entirely on geographic coordinates because of the difficulty in street-name matching. Our earlier evaluation of Caltrans roadway accuracy suggested that freeways were located more accurately than arterials and collector roads. Prior to adjusting for the multiple-lane effect, the snapped GDT and corresponding Caltrans freeways had a mean traffic count of 95,000 and 190,000, respectively, and a correlation coefficient of 0.89. After adjusting for the multiple-lane effect, the average GDT traffic counts increased to 196,000, and the correlation coefficient became 0.81. Although the correlation is lower than what we expected for freeways, the agreement of average traffic volumes within 3 percent for freeways is excellent.
The results showed that, for the matching roadways, GDT traffic counts were comparable to the Caltrans data for freeways (within 3% on average) but were 14 percent lower for surface streets. The correlation coefficient was higher for freeways than for surface streets, which is consistent with our previous experience that traffic data for freeways were more accurate than for surface streets. There are significant differences in and problems with the data for surface streets. We found that 40 percent of the GDT center points were either snapped to the wrong Caltrans surface roads or not snapped to any road in the Caltrans database. A large number of GDT local roads with traffic counts (small roads—average daily traffic count ≈3,000) are not incorporated in the Caltrans database. In contrast, about 30 percent of the Caltrans surface streets were not covered by GDT traffic count data. Caltrans surface streets missing in the GDT database had average volumes of 9,000 vehicles per day.
Sensitivity tests were conducted with the CALINE4 air quality dispersion model using the Caltrans_Tele and GDT data for 324 residential locations in Long Beach, California. Heavy-duty (HDV) and light-duty vehicles need to be modelled separately with the CALINE4 model because these vehicles have significantly different emission rates. The GDT database did not contain HDV information; we assigned the HDV fractions from the Caltrans data to the matching GDT roads. HDV fractions of 0.05 and 0.02 were assigned to the unmatched GDT freeways and surface streets, respectively. The results showed that CALINE4 estimated concentrations using the Caltrans_Tele, and the GDT data had a correlation coefficient ranging from 0.85 to 0.95, depending on the type of pollutant (CO, NO2, NOx, PM10, and PM2.5) and roadway (freeways and surface streets). The estimated concentrations using GDT data were 15 percent and 40 percent lower on freeways and surface streets, respectively, than those using the Caltrans_Tele data. The lower concentration estimates derived from the GDT data were due to a number of factors: (1) the GDT database did not contain major roadways that were included in the Caltrans database; (2) our extrapolation of the GDT point data to roadway segments left gaps in the coverage compared to the Caltrans data which have continuous coverage on all its segments; and (3) our assumptions of 5 percent and 2 percent HDV fractions for the unmatched GDT freeway and surface roads, respectively, may underestimate HDV volumes on these road types.
Future Activities:We now have assembled all the data necessary to implement land use regression models for the multiple communities in our study. Over the next year, we will calibrate these models and incorporate the larger data set of field measurements into our Bayesian exposure measurement error framework. We also will submit the incident asthma analysis for review and begin our health burden assessment for selected communities. Finally, we will continue our assessment of associations between the various metrics of traffic pollution and the measured pollutants.
Journal Articles on this Report: 6 Displayed | Download in RIS Format
| Other project views: | All 7 publications | 6 publications in selected types | All 6 journal articles |
| Type | Citation | ||
|---|---|---|---|
|
|
Gilliland F, Avol E, Kinney P, Jerrett M, Dvonch T, Lurmann F, Buckley T, Breysse P, Keeler G, de Villiers T, McConnell R. Air pollution exposure assessment for epidemiologic studies of pregnant women and children: lessons learned from the Centers for Children’s Environmental Health and Disease Prevention Research. Environmental Health Perspectives 2005;113(10):1447-1454. |
R831845 (2005) R826708 (2000) R826708 (2001) R826708 (2002) R826708 (Final) R826710 (Final) R827027 (2002) R831861 (2004) R831861 (2005) R832141 (2006) |
|
|
|
Jerrett M, Arain A, Kanaroglou P, Beckerman B, Potoglou D, Sahsuvaroglu T, Morrison J, Giovis C. A review and evaluation of intraurban air pollution exposure models. Journal of Exposure Analysis and Environmental Epidemiology 2005;15:185-204. |
R831845 (2005) R831861 (2005) |
not available |
|
|
Molitor J, Molitor N-T, Jerrett M, McConnell R, Gauderman WJ, Berhane K, Thomas D. Bayesian modeling of air pollution health effects with missing exposure data. American Journal of Epidemiology 2006;164(1):69-76. |
R831845 (2005) R831861 (2005) |
not available |
|
|
Schlesinger RB, Kuenzli N, Hidy GM, Gotschi T, Jerrett M. The health relevance of ambient particulate matter characteristics: coherence of toxicological and epidemiological inferences. Inhalation Toxicology 2006;18(2):95-125. |
R831845 (2005) R831861 (2005) |
not available |
|
|
Wu J, Lurmann F, Winer A, Lu R, Turco R, Funk TH. Development of an individual exposure model for application to the Southern California Children’s Health Study. Atmospheric Environment 2005;39(2):259-273. |
R831845 (2005) R827352 (Final) R827352C015 (Final) R828172 (Final) R831861 (2004) R831861 (2005) |
|
|
|
Wu J, Funk T, Lurmann F, Winer A. Improving spatial accuracy of roadway networks and geocoded addresses. Transactions in GIS 2005;9(4):585-601. |
R831845 (2005) R827352 (Final) R827352C015 (Final) R831861 (2005) |
|
geographic information science, spatial analysis, lung function,
,
HUMAN HEALTH, ENVIRONMENTAL MANAGEMENT, Economic, Social, & Behavioral Science Research Program, INTERNATIONAL COOPERATION, Scientific Discipline, Health, RFA, PHYSICAL ASPECTS, Molecular Biology/Genetics, Risk Assessment, Risk Assessments, Environmental Statistics, Health Risk Assessment, Physical Processes, Children's Health, Biochemistry, Environmental Policy, Exposure, exposure assessment, exposure model, health effects, developmental effects, developmental toxicity, immunotoxicology, children's environmental health, computer models, health risk analysis, spatial correlation, dispersion models, human health risk, statistical models, model assessment, dietary exposure, air pollution, children, environmental risks, vulnerability, children's vulnerablity, asthma, Human Health Risk Assessment
Relevant Websites:
http://hydra.usc.edu/scehsc/default.asp
Progress and Final Reports:
Original Abstract