2004 Progress Report: Relating Cardiovascular Disease Risk to Ambient Air Pollutants Using Geographic Information Systems Technology and Bayesian Neural Networks: The AHSMOG StudyEPA Grant Number: R830547
Title: Relating Cardiovascular Disease Risk to Ambient Air Pollutants Using Geographic Information Systems Technology and Bayesian Neural Networks: The AHSMOG Study
Investigators: Knutsen, Synnove F. , Beeson, Larry , Ghamsary, Mark , Soret, Samuel
Institution: Loma Linda University
EPA Project Officer: Chung, Serena
Project Period: February 1, 2003 through December 31, 2006 (Extended to January 31, 2009)
Project Period Covered by this Report: February 1, 2004 through December 31, 2005
Project Amount: $964,436
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: Health Effects , Particulate Matter , Air
The primary objective of this research project is to determine the association between cardiovascular disease and long term particulate ambient pollutants in 6,338 nonsmoking California Seventh-day Adventists. The outcomes include both fatal and nonfatal coronary heart disease (CHD) as well as other cardiovascular disease, during 22 years of followup. Further, the objectives are to assess the same effects in sensitive subgroups (e.g., prevalent cardiovascular disease [CVD], hypertensives, diabetics, elderly).
Further, this research project aims to assess whether other pollutants (specifically gaseous pollutants) modify the association between particulate pollution and CVD.
The approach for this study is to:
- Utilize data from the existing Adventist Health and Smog (AHSMOG) Study, which have been updated through March 2000 through the current U.S. Environmental Protection Agency (EPA) Science To Achieve Results (STAR) Grant (R827998). These data include monthly indices of air pollutants to ZIP Code centroids, monthly residence and work location histories, outcome assessment (CHD, fatal and non-fatal) and assessment of relevant confounders (smoking, environmental tobacco smoke, diet, exercise, etc.).
- Develop new indices of ambient air pollutants for the individual subjects in the AHSMOG Study using geographic information systems (GIS) technology and stochastic models that include error estimates of the indices.
- Develop nonlinear statistical models using Bayesian neural networks to develop alternative analytical strategies for modeling the relationship between different ambient air pollutants and risk of CHD where several pollutants and latent (unobserved) and missing values can be incorporated.
- Compare new methods developed under (2) and (3) above to the classic or conventional methods previously used in the AHSMOG Study.
Air Pollution Estimates
Monthly air pollution estimates for each subject are available since the start of the AHSMOG Study in 1973, and for some pollutants (e.g., PM2.5) back to 1966. These were developed using a deterministic method with interpolation to the centroid of each ZIP Code. The AHSMOG study had used information from all relevant monitoring stations in California to develop the ambient air pollution estimates. In meetings between the STAR grantees and EPA, however, it was decided to use a common air pollution database from EPA. This will allow a comparison of estimated ambient air pollution values by the four studies. For AHSMOG, this also will allow a comparison with our previous, comprehensive methods for estimation of ambient air pollution. This decision, however, has delayed the progress of the study in obtaining air pollution estimates using geostatistical methods.
To date, all street addresses have been geocoded for each subject on a monthly basis from baseline in 1977 to 2000 or to date of death or loss to followup. In addition, monthly workplace ZIP Codes have been geocoded for each subject in the same time period. Similarly, we have geocoded the EPA monitoring sites in California, Oregon, Nevada, Arizona, and northern Mexico, when available. These layers enabled us to use various techniques in determining how the spatial distribution of the cohort and monitoring network accurately measures exposure to a variety of air pollutants. Because the monitoring network is denser in the more populated areas, we previously have been unable to calculate accurate exposure levels in rural areas. Currently, we are collaborating with the research team at Environmental Systems Research Institute (ESRI) to develop subject-specific ambient air pollution estimates using geostatistical data analysis.
Assessment of Outcome
Incident CHD. From 1977 to 1982, we have information on incident myocardial infarctions (MI). For the period 1983 to 1999, we have self-reported incidence of acute MI with additional information on name and address of the hospital in which these were diagnosed. Validity of this information is being done by obtaining medical records from the individual hospitals.
Based on updated reports from self-report and surrogate information, a total of 512 subjects have reported that they had an acute MI since 1982. Letters have been sent to the hospitals in which the subjects were diagnosed with their self-reported MI. We are having problems getting medical records for everyone because some records have been destroyed, some cannot be found in the hospital identified by the subject for various reasons (e.g., records are located in distant storage, some hospitals require a recent signature of the subject due to their interpretation of the Health Insurance Portability and Accountability Act of 1996 [HIPAA] legislation).
Cardiovascular Disease Mortality. All death certificates have been coded by a certified nosologist and all mortality outcomes have been updated. A total of 2,462 deaths have occurred in the cohort since 1977. Of these, 2,393 are deaths by natural causes (ICD-9: < 800) and 644 are a result of ischemic heart disease (ICD-9: 410-414).
Outcomes in Sensitive Subgroups
The following sensitive subgroups have been identified:
- Older age (> 64 years and > 74 years)
- Prevalent CHD
- Prevalent CHD, stroke, diabetes, or hypertension
- Past smokers
- Prevalent chronic obstructive pulmonary disease (COPD).
Preliminary analyses of sensitive subgroups using previously estimated ambient air pollution do not indicate any increased risk associated with particulate matter air pollution in sensitive subgroups as compared to nonsensitive subgroups, except for persons with COPD where the risk of fatal CHD seems to be elevated, especially in males. Further analyses using ambient air pollution estimates and GIS methodology, however, will be done in accordance with the objectives of this grant.
Analyses and Results
As mentioned, we still do not have individual estimates of ambient air pollution for the subjects using the EPA air quality database. We have, however, analyzed fatal CHD associated with particulate air pollution using our previous air pollution estimates. The results of these show a detrimental effect of particulate pollution, especially PM2.5, in females, but not in males. The association in females was strengthened in two-pollutant models with ozone.
Progress. Obtaining medical records of self-reported MI has turned out to be much more challenging than first expected. This is because of several factors. First, some hospitals do not keep records for more than 10 years. Secondly, because of the HIPAA regulations coming into effect in April 2003, hospitals are very reluctant to give out any records despite the fact that the HIPAA regulations specifically exempt records of persons having given their consent prior to the implementation of the HIPAA rules. In spite of this, many hospitals are requesting that new consent forms using their specific forms be obtained from the subjects. At this point, many of the subjects are dead and thus it will potentially be necessary to obtain consent from close family members.
The progress on developing GIS-based individual ambient air pollution estimates also is taking longer than expected. We are very pleased with the services of ESRI in this effort.
Likewise, the development of statistical models using neural networks and Bayesian neural networks is progressing. The first paper on this work was presented at the Hawaii International Conference on Statistics in Honolulu in June 2004. The plan is to have the statistical models tested using our previous air pollution estimates by the time the GIS-based air pollution estimates are ready.
Two abstracts were presented at the International Society for Environmental Epidemiology Meeting in New York in August 2004 using previous air pollution estimates.
All manuscripts use previously developed ambient air pollution estimates. One manuscript on particulate matter and fatal CHD has been submitted for publication. Several other manuscripts are in their final stages of preparation:
- Particulate matter and risk of non-Hodgkin’s lymphoma.
- Effect of ambient air pollution on mortality in sensitive subgroups.
- Effect of particulate air pollution on mortality.
- Effect of ambient air pollution on risk of incident and fatal lung cancer.
- Effect of particulate air pollution on risk of hospitalizations for respiratory disease.
During Year 3 of the project, we will obtain the remaining outstanding medical records and verify the self-reported non-fatal MIs from these. We also will complete development of the air pollution estimates using GIS as well as traditional methods from the databases obtained from EPA. The development of the statistical models using Bayesian neural networks also will be completed in Year 3 of the project.
We expect to continue analyses of the air pollution-disease relationship in the latter half of Year 3 using the new estimates of ambient air pollution. Publications that we expect to submit in Year 3 of the project will be methods papers on development of the Bayesian neural network and development of the GIS-based ambient air pollution estimates. We further hope to submit papers on the association between total mortality and fatal CHD and ambient air pollution using GIS-developed air pollution estimates.
Because of the delay in getting started, we expect to request a 1-year, no-cost extension. In Year 4 of the project, we expect to complete the analyses and write papers to address the remaining specific objectives.