![[logo] US EPA](https://webarchive.library.unt.edu/eot2008/20081107052540im_/http://www.epa.gov/epafiles/images/logo_epaseal.gif){kind=logo}
Final Report: Air Pollution and Hospital Admissions in Washington State
EPA Grant Number: R825266Title: Air Pollution and Hospital Admissions in Washington State
Investigators: Moolgavkar, Suresh H.
Institution: Fred Hutchinson Cancer Research Center
EPA Project Officer: Katz, Stacey , Robarge, Gail
Project Period: October 1, 1996 through September 30, 1999
Project Amount: $420,056
RFA: Air Quality (1996)
Research Category: Air Quality and Air Toxics
Description:
Objective:The main objective of this research effort is the development and application of methodology for the analysis of adverse health effects of air pollution when longitudinal data on outcomes are available on an individual basis on a cohort of individuals. Such an analysis is made possible by the existence in the State of Washington of a rather unique resource, the Comprehensive Hospital Abstract Reporting System (CHARS) database. This database will be described below in more detail. The important feature of this database that we wish to exploit is that each individual admitted to a hospital in the state has a unique personal identification code (PIC), which makes possible data linkage and the construction of a longitudinal history of admissions for an individual. Explicitly, we propose the following specific aims:
- Development of methods for the analyses of repeated outcomes on a cohort of individuals when interest is focused on both the occurrence and timing of the outcome and the severity of the outcome. For example, with hospital admissions data, one is interested not only in the occurrence and timing of each admission for an individual, but also in the number of days spent in the hospital. As a stochastic process, the hospital admissions process is a marked point process in which each occurrence of the event of interest (admission to hospital) is associated with a mark (length of stay in hospital), which is described by a distribution. The idea is to develop a marked point process in which both the time of occurrence and the size of the event depend upon the covariates of interest, such as air pollution, weather, and other relevant factors (e.g., cigarette smoking). The appropriate statistical techniques and software also will be developed for fitting such models to data, and the statistical properties of the model will be investigated.
- Analyses of the hospital admissions in the Seattle area. Data for these analyses are available in the CHARS database, which is available from the Washington State Department of Health and covers the period January 1985?June 1994. These data will be analyzed in two ways and the results compared. The first analysis will replicate the ecological analyses done earlier by other investigators. In contrast to some of the earlier work, however, all available pollutants will be considered both singly and simultaneously in the regression models. For the second analysis, a cohort of individuals will be selected from the CHARS database and the admissions histories of the members of the cohort will be analyzed using the marked point process approach to be developed under the first specific aim.
In recent years, many studies have documented associations between levels of air pollution seen in U.S. and Canadian cities and adverse health effects on human populations. Increases in the levels of pollutants are associated with increased emergency room visits for asthma, with increased hospital admissions, particularly for respiratory diseases, and with increased mortality. Many of the studies documenting the adverse health effects of air pollution are of an ecological nature. In these studies, the total number of events (e.g., daily hospital admissions or daily deaths) in a population are regressed on potential explanatory variables, such as levels of air pollution and weather variables.
The purpose of the research proposed in this grant application was two-fold. Firstly, to develop methods for the analyses of longitudinal outcome data in air pollution research. More precisely, when information is available on the history of repeated events (e.g., hospital or emergency room visits) on single individuals of a cohort, methods will be developed for the analyses of this series of events as a function of the covariates of interest, some of which are measured on a population basis, such as air pollution and weather, and some of which are measured on an individual basis, such as cigarette smoking. The statistical properties of these methods, which will be based on viewing the individual histories as the outcomes of a point process, were investigated.
In the second part of this research proposal, the extensive hospital admissions database available from CHARS was analyzed. The CHARS database contains personal identification codes so that longitudinal histories of admissions for individuals in the database can be constructed. A cohort of individuals were chosen from this database, histories of hospitalizations were created for each of them, and the resultant derived data were analyzed using the approach described above. In addition, daily admissions recorded in the CHARS database also were analyzed using the traditional ecological approach, and the results of the two analyses were compared.
A number of papers using regression methods for analyses of time series have recently reported associations between hospital admissions for respiratory diseases and particulate air pollution and ozone. These associations persisted even after adjustment for weather. Most of these papers have restricted attention to admissions for individuals over the age of 65 years because data for these admissions are readily available from the Health Care Financing Administration (HCFA).
In this study, we undertook analyses of air pollution and daily hospital admissions for chronic pulmonary disease in the Seattle, WA, metropolitan area during the 9-year period 1987?1995. Because the State of Washington maintains detailed admissions records for each hospital in the state, we were able to obtain daily counts of admissions for chronic respiratory disease for all hospitals located in King County, WA. In particular, unlike previous studies, we did not have to restrict our analyses to the elderly. Additionally, we had air quality monitoring information on three particulate matter (PM) measures: PM10, PM2.5, and light scattering by nephelometry, which can be converted into a measure of either PM2.5 or of PM1. Finally, we obtained information on daily pollen counts in the Seattle metropolitan area so that we also could investigate the association between daily fluctuations in pollen counts and hospital admissions for chronic respiratory disease.
Both air pollution and tree pollens were independently associated with hospital admissions. In single-pollutant models, among the gases, we found the strongest association between carbon monoxide and hospitalization. The association with sulfur dioxide was weaker, and there was no evidence of an association with ozone. We also found association of hospital admissions with PM10, and a suggestion of an association with an index of light scattering measured by nephelometry. In two-pollutant models, the effect of carbon monoxide remained stable, whereas the effect of particulate matter, measured either as PM10 or by nephelometry, was attenuated and became unstable. We also examined the association between air pollution, pollens, and hospital admissions in three broad age groups, 0-19 years, 20-64 years, and 65 years and older. Although tree pollens were associated with hospital admissions in each of these age groups, the association between air pollution and hospital admissions was seen only in the youngest age group.
Recurrent events (or failures) occur frequently in studies in which the failures are not necessarily fatal. Examples include asthmatic attacks, epileptic seizures, hospital admissions, etc. A point process formulation is commonly used to describe and analyze such data. Regression analysis in this framework, in which the intensity rate of the point process under consideration is modeled as a function of covariates or covariate processes (time dependent covariates), has attracted a lot of attention since the first paper by Prentice et al., 1981, who considered a partial likelihood approach with arbitrary baseline intensity rates. There has been work on robust regression analyses of recurrent event data using the point process formulation, which, instead of modeling the intensity rates, models some other ?marginal' quantities, thus avoiding strong assumptions on the recurrent event process. These approaches focus on subject-specific covariates requiring multiple subjects and fail for environmental covariates as they are the same for all the subjects at any event time.
Recent work on environmental covariates has focused on the use of generalized additive models for investigating associations between indices of air pollution measured at central monitoring locations and daily counts of events such as death or hospital admissions. This approach fails to incorporate between subjects variation in the baseline parameters and subject-specific covariates, if any.
Our work focuses on a model that views the data on each subject as the realization of a point process, the intensity rate that depends on the environmental covariates. This point process formulation allows us to incorporate subject-specific covariates and also the previous history of the process. Specifically, we make a Poisson process assumption and derive relevant likelihood function for estimating the regression parameters under different assumptions on the baseline Poisson intensity.
We illustrate our method by means of an example of air pollution and hospital admissions for chronic respiratory disease.
Journal Articles on this Report: 2 Displayed | Download in RIS Format
| Other project views: | All 2 publications | 2 publications in selected types | All 2 journal articles |
| Type | Citation | ||
|---|---|---|---|
|
|
Dewanji A, Moolgavkar SH. A Poisson process approach for recurrent event data with environmental covariates. Environmetrics 2000;11(6):665-673. |
R825266 (Final) |
|
|
|
Moolgavkar SH, Hazelton W, Leubeck G, Levy D, Sheppard L. Air pollution, pollens, and admissions for chronic respiratory disease in King County, Washington. Inhalation Toxicology 2000;12(Suppl 1 to Issue 1):157-171(15). |
R825266 (Final) R827355 (2001) R827355 (Final) R827355C009 (Final) |
|
carbon monoxide, sulfur dioxide, ozone, PM10, recurrent event data, environmental covariates, conditional likelihood analysis, chronic respiratory disease, case-crossover design. , Air, Geographic Area, Scientific Discipline, Health, RFA, Susceptibility/Sensitive Population/Genetic Susceptibility, Risk Assessments, genetic susceptability, Epidemiology, air toxics, Children's Health, Atmospheric Sciences, EPA Region, particulate matter, State, adolescents, exposure and effects, Acute health effects, ambient air quality, Washington (WA), weather, air quality, cardiopulmonary response, hospital admissions, lung injury, pulmonary toxicity, statistical analysis, toxics, chronic health effects, human health effects, particulates, respiratory, sensitive populations, statistical methods, air pollution, epidemiological studies, Region 10, age dependent response, exposure, inhaled particles, respiration, health risks, human exposure
Progress and Final Reports:
Original Abstract