2004 Progress Report: Chronic Exposure to Particulate Matter and Cardiopulmonary DiseaseEPA Grant Number: R830545
Title: Chronic Exposure to Particulate Matter and Cardiopulmonary Disease
Investigators: Laden, Francine , Camargo, Carlos , Schwartz, Joel , Speizer, Frank E. , Suh, Helen H.
Current Investigators: Laden, Francine , Camargo, Carlos , Puett, Robin C. , Schwartz, Joel , Speizer, Frank E. , Suh, Helen H. , Yanosky, Jeff D.
Institution: Brigham and Women's Hospital, Inc.
EPA Project Officer: Chung, Serena
Project Period: January 20, 2003 through January 19, 2006 (Extended to January 19, 2008)
Project Period Covered by this Report: January 20, 2004 through January 19, 2005
Project Amount: $933,602
RFA: Epidemiologic Research on Health Effects of Long-Term Exposure to Ambient Particulate Matter and Other Air Pollutants (2002) RFA Text | Recipients Lists
Research Category: Air Quality and Air Toxics , Health Effects , Particulate Matter , Air
The objectives of this research project were to:
- develop a model estimating long-term exposure to air pollution in the continental United States using existing databases, including the U.S. Environmental Protection Agency (EPA) Air Quality System (AQS);
- and evaluate the association of chronic exposure to air pollution with incident coronary and respiratory disease and total mortality in the Nurses’ Health Study (NHS), an ongoing prospective cohort study of 121,700 women residing throughout the United States.
We hypothesize that the incidence of these diseases and total mortality are associated positively with air pollution, and that exposure to air pollution exacerbates existing disease. We also hypothesize that the association with coronary heart disease will be greater among diabetics than nondiabetics and that consumption of antioxidants will modify the association.
In Year 2 of the project, geocodes received from Geographic Data Technology, Inc., have undergone extensive manual cleaning to ensure the highest rate of geocode matching and accuracy. Errors such as spelling mistakes and abbreviations were identified and corrected.
Intensive effort has gone into the modeling process during Year 2 of the project. We began by developing initial models using monthly average PM10 data by monitoring site from the EPA AQS and various other sources, including the IMPROVE (Interagency Monitoring of Protected Visual Environments) network and Harvard research studies. The initial models were fit to explore the data and to generate preliminary modeling results (as presented at the International Society of Environmental Epidemiology 2004 Conference, New York, New York). We used these models, which do not segregate the spatial and non-spatial components, to predict PM10 over the northeast United States (where most of the subjects in the NHS live) at a grid of points.
Based on an evaluation of results from these initial models, a spatial component has been added to our current prediction models. The current, more complex approach combines a monthly pollution surface (the spatial component) and constant (time invariant) effects of predictors derived from a geographic information system (GIS, the non-spatial component). The following variables were generated for each monitoring location using GIS: block group, tract, and county level population density; distance to nearest road by Census Feature Class Code road class; elevation from the U.S. Geological Survey (USGS) National Elevation Dataset; land use/land cover from the USGS National Land Cover Dataset; primary criteria pollutant emissions information from EPA National Emissions Inventory by county; and meteorological variables, including temperature, wind speed, and relative humidity, from the National Climatic Data Center. All data were imported into generalized additive statistical models with smooth terms of space, time (separate surfaces for each month), and the GIS variables. This exposure modeling process has yielded information about how the spatial surface of PM10 changes through time and across seasons and how each of the GIS derived variables affect the predicted concentration surface. The models have been evaluated using cross-validation. A test set of 10 percent of the observations was removed from the dataset, the model generated without them, and then the predicted values for these locations were compared with the measured values of PM10.
PM2.5 monitoring data were not available on a national level until 1999. Therefore, to predict PM2.5 levels for earlier time periods, we will use observations of horizontal visual range made at Weather Bureau/Army/Navy stations (about 430 nationwide) as a predictor for PM2.5 after correcting for the truncated nature of the observations. We have examined the relationships between PM10 and PM2.5 and currently are developing models that will use the above mentioned corrected visibility observations, PM10 predictions, and the PM2.5 monitoring data collected from 1999 to 2002 to estimate monthly average PM2.5 levels back to 1988. This model will first be applied regionally; with the continental United States divided by state into five regions (Northeast, Southeast, South Central, Southwest, and Northwest) covering nearly the entire continental United States.
During Year 3 of the project, we will finalize development of the PM2.5 exposure models. We will use Cox proportional hazards modeling to examine associations of chronic PM10 exposure with mortality and chronic health outcomes, including cardiovascular disease, lung cancer, asthma, and chronic obstructive pulmonary disease. Through this modeling process, we will examine potential confounders and effect modifiers such as smoking, physical activity, body mass index, physician-diagnosed hypertension, and diabetes.