![[logo] US EPA](https://webarchive.library.unt.edu/eot2008/20081105140428im_/http://www.epa.gov/epafiles/images/logo_epaseal.gif){kind=logo}
2004 Progress Report: Air Quality and Reported Asthma Incidence in Illinois
EPA Grant Number: R829402C003Subproject: this is subproject number 003 , established and managed by the Center Director under grant R829402
(EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).
Center: Center for Integrating Statistical and Environmental Science
Center Director: Stein, Michael
Title: Air Quality and Reported Asthma Incidence in Illinois
Investigators: Frederick, John , Draghicescu, Dana , Dukic, Vanja , Naureckas, Edward , Rathouz, Paul , Zubrow, Alexis
Current Investigators: Frederick, John , Draghicescu, Dana , Dukic, Vanja , Eshel, Gidon , Im, Haekyung , Naureckas, Edward , Rathouz, Paul
Institution: University of Chicago , Hunter College/CUNY
Current Institution: University of Chicago
EPA Project Officer: Smith, Bernice
Project Period: March 12, 2002 through March 11, 2007
Project Period Covered by this Report: March 12, 2004 through March 11, 2005
RFA: Environmental Statistics Center (2001)
Research Category: Environmental Statistics , Ecological Indicators/Assessment/Restoration
Description:
Objective:This project focuses on statistical investigations to define the relationship between air quality and respiratory health in the Chicago area. The goal of developing improved statistical models for linking acute asthma events to air quality indicators is a driving force behind most of the work, and development of these models will continue for the duration of the project. Efforts during the past year have centered on: (1) the examination and comparison of different measures of asthma incidence for use in population-based studies of the effects of air pollution on respiratory health; (2) the development of statistical models to link acute asthma occurrence in Chicago's Medicaid population to levels of ozone, particulate matter, pollen, and meteorological conditions in both urban-aggregated and spatially-resolved ways; and (3) the development of statistical models and the application of mechanistic models to describe the spatiotemporal structure in ground-level ozone and other measures of air quality for use in health-related work. Below, we summarize results to date in each of these areas. In addition to ongoing research, recent attention has focused on preparing results obtained in the initial 2 years of the project for publication. Several papers will be submitted for publication in refereed journals prior to the end of the current year’s funding.
Problems encountered are mostly centered on the spatial, and in some cases the temporal, coverage of specific datasets related to atmospheric conditions, air quality, and health outcomes. The health outcome data are organized spatially at the level of ZIP code. On any given day, ozone is measured hourly at a minimum of 11 sites in Cook County, but this still does not provide sufficiently dense coverage for all of Cook County’s 56 ZIP codes. Pollen is a significant factor in asthma occurrence, but the available data include only one daily value for the entire Chicago area. The dataset for coarse particulate matter, PM10, has sparse spatial coverage and contains information at irregular time intervals, where the period between data points at a specific site can be up to 6 days. Therefore, we are faced with significant problems of incompatible space-time scales across the multiple data sources to be used in our project. In response to the issues posed by such problems of scale, we are investigating a variety of statistical spatio-temporal modeling techniques and physical models (Community Multiscale Air Quality [CMAQ]) for use in generating spatio-temporal structure in factors that may influence asthma incidence in specific locations. The problems identified are significant, and they have served as a part of the wider impetus for developing important new statistical methodology that will be applicable in our study as well as in other studies with similar study designs.
Progress Summary:Interactions with Personnel from the U.S. Environmental Protection Agency (EPA)
During the past year, investigators from the project interacted with Mike Rizzo from EPA Region 5, located in Chicago. Mr. Rizzo has arranged for the group to get access to updated datasets for ozone and fine particulate matter (PM2.5) for the year 2000 and later. This information is central to studies to be undertaken in the near future. In addition, Paul Rathouz accepted an invitation from Valerie Garcia of EPA’s National Exposure Research Laboratory to attend a meeting organized by the Centers for Disease Control and Prevention with the participation of EPA. Dr. Rathouz presented a summary of efforts in the air quality and asthma project with emphasis on the development of new statistical techniques for linking environmental exposures and health outcomes. Finally, Thomas Brody from EPA Region 5 communicated regularly with Edward Naureckas during the year regarding issues in respiratory health.
Results to Date
A. Validation of Alternate Measures of Asthma Outcomes (Edward Naureckas).This research has been completed by graduate student Xiaoming Bao as her Master’s thesis. By comparing outcomes from the Illinois Medicaid Database with regional EPA data on outdoor air quality, it became clear that prescription fills of short acting bronchodilators provided the event frequency required. The literature contains numerous studies utilizing claims data for asthma hospitalizations and emergency department (ED) visits. While these traditional outcomes have the advantage of providing an easily definable event, they represent only the most severe instances of asthma exacerbation. The occurrence of these events is also influenced by a number of factors that may be independent of asthma severity, such as availability of a primary care physician to provide an urgent outpatient visit. For these reasons, these outcomes may be neither sensitive nor specific as a marker for asthma exacerbation due to adverse air quality or other factors.
Albuterol prescription refills occur at a rate much higher than the markers described above. An early fill by an individual may be indicative of increased asthma activity in this person, even if the exacerbation does not result in a hospitalization or an emergency department visit for asthma. For this reason, prescription fills for short-acting bronchodilators may provide a much more sensitive marker for low-level respiratory effects of degraded outdoor air quality. The goal of this project was to determine the degree to which short-acting bronchodilator prescription fills are correlated with the more traditional outcomes.
Our analysis has found that the greatest association between ß -agonist prescription refills and ED or hospital visits is observed on the day of a short acting prescription fill. There is a smaller but still strong association for each of the next 4 days following the prescription fills. This association is still seen whether patients are actively using controller therapy, such as inhaled corticosteroids, or not. This relationship also was independent of gender and was seen in both the adult and pediatric populations.
The results of this work support the use of short acting bronchodilator prescriptions as a surrogate marker for asthma exacerbation. The patterns seen in our analysis suggest that a distributed lag approach may be useful in investigating the relationship between outdoor air quality and asthma outcomes in our current database as well as additional databases in the future.
B. Short-term Respiratory Effects of Air Pollution in Metropolitan Chicago: Aggregate Analyses Using Bayesian Model Averaging. The initial study in this project used daily ZIP code-aggregate data, obtained by combining albuterol prescriptions of individual Medicaid patients who live in the same ZIP code within the Chicago metropolitan area, for each of the 52 available ZIP codes.
Independent variables were: previous day city-wide averages of temperature, relative humidity, pollen, and PM10(characterized by an average of the 24-hour integrals estimated for each day and every station), as well as the city-wide average of daily max 8-hour averages of ground-level ozone across all monitors. In addition, the model controlled for day of week, year, as well as for unmeasured time-varying confounding variables that depend only on day of the year, with peaks in spring and late summer/early fall. Results from this aggregated model revealed significant links between the number of daily medical claims and temperature (a negative correlation) and pollen levels (a positive correlation). No statistically significant association with air quality as measured by prior day PM10 and ozone appears in the spatially aggregated model.
The following studies involved fitting a spatially-resolved version of the above asthma model, employing the daily max 8-hour averages of ozone for each ZIP code (obtained as predictions from an ozone spatial model at the ZIP code centroids) instead of the city-wide average. Both frequentist and Bayesian estimation procedures (and in both fixed and random ZIP code-effect formulations) have resulted in a somewhat higher ozone effect which is, however, still statistically insignificant, while the other effects have remained the same as in the above aggregate analysis. The results from these analyses are about to be submitted to the Journal of the American Statistical Association (JASA) Applications and Case Studies.
Two issues still remain to be further explored in our future work with these data:
- Study B1: to examine the nature and impact of different procedures used to adjust for unobserved time-varying confounders on estimates of interest.
- Study B2: to use distributed lag models for air pollutants instead of previous-day variables alone.
Vanja Dukic has been working with a Statistics Ph.D. student, Chava Zibman, to address these issues in Years 3 and 4. In addition, we propose to apply the methods developed above to an alternative dataset more recently obtained from a privately owned large pharmacy chain which would allow us to examine some other (more recently available) measures of air quality, such as PM2.5, thought to be more important in asthma studies than traditionally used Total Suspended Particulates (TSP) and PM10. Finally, we propose to extend the methodology above to accommodate individual patient data, as an alternative approach to the case-crossover approach employed in study C.2.
C. Disease Mapping and Distributed Lag Results. Work done by Paul Rathouz this past year has been organized into three interlocking projects. The first of these is the development of a new statistical method for doing disease mapping when data are aggregated into geographical regions such as ZIP codes. This project is ongoing, and is joint with David Clifford, a Ph.D. student in Statistics. Second, Mr. Rathouz worked with Wen Gu, a Master’s student in Statistics, on a new distributed lag model for air pollution and weather effects on individual health outcomes. Third, he created a new model for longitudinal and spatial environmental epidemiology data. This model is “aggregation consistent” in that it can incorporate data with varying degrees of aggregation over time and space and is also interpretable at these various levels of aggregation. He also proposed methods for estimation of this model.
Study C.1: Mapping asthma outcomes aggregated by ZIP code.
Many environmental health data analyses begin in principle with a map of where geographically incident cases occur. However, because merely plotting the locations of cases can yield a very noisy plot, one that that obscures underlying variation in population density and socio-demographic structure, what is generally plotted is an adjusted case incident rate. When exact location data are available, this incidence rate is often modeled (on the log scale) as a Gaussian random field. This model is then used as the basis for prediction at all points in the geographical region of interest, and smooth disease rate contour plots can then be drawn of this fitted surface. For continuous data, this problem is very well-studied in the geostatistical literature, and for count data, such as the number of disease cases, the problem has received some attention in the past decade.
When the data do not contain exact disease locations, but rather are aggregated over some geographical area, these aggregated data pose special challenges to the recovery of a smooth underlying Gaussian random field. Typical geographic areas include administrative or political divisions such as census tracts, ZIP codes, or counties. This project considers this problem of such aggregated disease incidence data and proposes a new computational method for predicting the underlying random field that represents the logarithm of the disease incidence rate at each point in space.
To fix ideas, consider the Medicaid population in fiscal years 1996–1998 with which we have been working. These data are aggregated by ZIP code. Figure 1 displays the number of individuals at risk in each ZIP code. As expected, some ZIP codes are much more heavily populated with Medicaid enrollees than others, so these ZIP codes will contribute more information than the more sparsely populated ones. In this study, we have been examining the rate of ß-agonist prescription fills per subject per day. Figure 2 displays the observed rates in each ZIP code. This display obscures that fact that, for some ZIP codes, the disease incidence rate is much more accurately estimated than for others. In addition, it imposes artificial jumps (discontinuities) on the fitted surface at the ZIP code boundaries and presents constant risk within any given ZIP code. Figure 3 shows a smooth predicted risk surface that is the result of applying our method. This plot smoothes the variation across the areal units (ZIP codes) and automatically borrows information from the more heavily populated ZIP codes to predict the risk surface at points nearby that ZIP code, naturally accounting for differences in sampling variability from ZIP code to ZIP code. From an epidemiological perspective, the plot displays the background risk of disease and is a starting point for further investigation of risk factors and of exposure studies. It also sheds light on what information is being lost by data aggregation, thereby informing study designs. Finally, when used with a regression model for the risk surface, the model can be used to improve statistical efficiency in estimation of the regression model because it allows the user to account for the spatial auto-correlation in the data.
Figure 1.
Figure 2.
Figure 3.
Statistical work on this problem of data aggregation of disease incidence exists but is not vast. Non-model-based smoothing techniques have been proposed and a few model-based methods have been proposed. These methods rely primarily on Markov chain Monte Carlo (MCMC) methods to carry out the four-dimensional integration necessary to generate the variance covariance matrix of the aggregated disease incidence rates across ZIP codes. The approach we develop exploits new results on the computation of these integrals that reduces them to one-dimensional integrals for regular rectangles.
The following is a brief overview of our approach. Let R(x) be the log-risk of an asthma event at point x in space. Then, to an approximation, the number of events in a given ZIP code is a Poisson random variable with mean P exp (R), where P is the total person-time observed in that ZIP code, and R-bar is the average log-risk over that ZIP code (i.e., R is the integral of R(x) over area A representing a given ZIP code). R-bar is easily estimated as log (number events / P). We suppose that R(x) is a realization of a Gaussian random field with covariance between points x and x′ given by K(x-x′). Then, the variance of R-bar is the four-dimensional integral of K(x-x′) over x and x′ each uniformly distributed over A, with a similar result for the covariance between two ZIP codes. We lay a grid of rectangles over the city, allocating a fraction of each rectangle to one or more ZIP codes; see Figure 4 for ZIP code 60608. The variance of and covariances between the values of R-bar for the rectangles are one-dimensional integrals, and the variances of and covariances between the values of R-bar for the ZIP codes are now easily computed via matrix multiplication.
Figure 4.
With this computational technique developed, it is possible to use the following procedure to predict the logarithm of the risk surface (Figure 3) using the observed disease rates (Figure 2) and corresponding population sizes (Figure 1). First, a Gaussian random field model is fitted to the observed disease rates R-bar, yielding a fitted covariance structure. This fitted covariance structure is then used to compute the covariance between each grid cell and each ZIP code. Finally, the observed ZIP code disease rates are used to predict the cell-wise rates just as in an ordinary kriging model in geostatistics.
In the coming year, we will be studying the properties of this method and further developing it using a variety of covariance structures. We will examine sensitivity to various approximations made along the way, we will test it on several data sets, and we will develop anisotropic covariance models to allow for different degrees of auto-correlation in the north-south direction versus in the east-west direction. We anticipate submitting the paper to Environmetrics.
Study C.2: Parametric distributed lag models in air pollution epidemiology.
Our overall project examines the daily asthma health outcomes as a function of daily fluctuations in air pollution, either in a longitudinal or a time series design. In these studies, one persistent challenge has been to specify the time lag structure for the effects of pollutants and weather variables on health outcomes. Not only is it not clear how quickly these factors induce a health effect when the subject is exposed, but there are many other factors which could act to delay the time between a pollution spike and a health outcome that we can in fact observe. These include, but are undoubtedly not limited to: the fact that individual exposures are not measured; behavioral factors that could delay (or accelerate) an effect; and delay between any manifest health effect and the health care event that we observe through claims data. The concern is that these and other factors serve both to delay and to distribute the effects of variations in pollutants over several days. The goal of this study is to develop a model and estimation technique for capturing these distributed lag effects. The problem is complicated by the fact that each pollutant and/or weather variable is highly auto-correlated with itself from day to day, and also that these variables are strongly correlated with one another on the same day and on nearby days. This problem is addressed in preliminary work in Wen Gu’s Master’s paper, written under the direction of Dr. Rathouz, who will further develop the work for publication in the biostatistical literature.
Ms. Gu developed a model for longitudinal asthma-outcome data with a distributed lag structure for five weather and pollution predictors: temperature, relative humidity, log (pollen count), ozone, and PM10. The outcome variable is whether or not each person filled a β-agonist prescription on each day in the summers of 1995–1998. For this analysis, the data are organized for a case-crossover design. Specifically, time-strata were created for each person. A time stratum consisted of a sequence of five Mondays, or of five Tuesdays, etc., within each person. Data were then analyzed via conditional logistic regression, treating all responses in a person-time stratum as a matched set. This effectively adjusts for all factors that vary at the person-time-stratum level. Only strata with at least one event contribute information to the analysis.
In this model, the effect of a given pollutant is modeled as a combination of the values of that pollutant on the current day and each of L days earlier. For ozone, the regression equation would include
Where ßozone is the overall effect of ozone, ozonet is the daily measure of ozone on day t, ozonet-I is the daily measure of ozone I days before t, and wI is a set of weights such that
and 
We modeled the wI weights as a parametric function governed by a parameter α. In our work, the weight function we used was the probability mass function for the binomial distribution (which has support 0,…, L); in further work, we will explore other more flexible functions. An alternative, more non-parametric approach would be to separately estimate each coefficient (say ß*I) of ozonet-I for I = 0,…, L, and then to define
and
.
The advantage to using a parametric model for the weights wI is that it sufficiently constrains the model to be able to estimate distributed lag contributions from several weather variables and pollutant variables in the same model. In our work, we fitted models with separate distributed lag functions for all five variables listed above for L = 14.
Estimation for non-linear models such as these is a challenge. For a single distributed lag function, we discovered that the Box-Tidwell method worked very well, but that it did not work well when multiple distributed lag functions were included in the model. We therefore developed a Gauss-Seidel algorithm to iteratively fit each of the five distributed lag functions, holding the other ones constant, and cycling to convergence. Because the Box-Tidwell method was very effective for a single distributed lag function, and because the Gauss-Seidel method is effective in general regression problems, this combination yielded results. Finally, one nice feature of the Box-Tidwell method is that it also yields standard errors for the parameters ß and α, and these are correct even if there are other non-linear terms in the model. We exploited this fact to obtain standard errors for all of our parameter estimates.
Results for our model fitting are presented in Tables 1–3. The values of α in Table 1 indicate the fitted weight function, where the weights are given by the probability mass function of the binomial distribution. When multiplied by L = 14, these values indicate the day lag I around which the bulk of the weight lies in the distributed lag function. So, relative humidity and temperature have very short lag structures (between 0 and 1 days being the most important), while the important lag for ozone and pollen is about 11 days. Table 2 presents the fitted values of ß for the five variables, presented on the exponentiated scale so that they are interpretable as odds ratios; the second column is for the model containing all five distributed lag functions together. The results are as expected.
Table 1.
α |
sd(α) |
log likelihood |
|
temperature |
0.0658 |
0.0258 |
-153779.19 |
pollen |
0.7460 |
0.1048 |
|
relative humidity |
0 |
I |
|
PM10 |
0.4989 |
0.0769 |
|
ozone |
0.7514 |
0.0542 |
Table 2.
Odds Ratio for distributed lag models |
||
Separately estimated |
Jointly estimated |
|
temperature (5F) |
0.9810 |
0.9803 |
pollen (1 log (grains/m3)) |
1.0183 |
1.0177 |
relative humidity (10%) |
0.9746 |
0.9848 |
PM10 (15 mg/m3) |
1.0253 |
1.0156 |
ozone (20 ppb) |
1.0433 |
1.0356 |
Significance levels for these results are presented in Table 3. They are presented separately because this is a non-standard statistical problem wherein the nominal Z-statistic computed by dividing the estimate of ß by its standard error is not a valid test statistic for the null hypothesis that ß = 0. The problem is that under the null hypothesis there is no information in the data to estimate α. The tests in Table 3 were therefore computed by fixing α = 0.5. While this may not be the true α value, under the null hypothesis, any α is correct. So, the test is valid even while not being the most powerful.
Table 3
P-value |
|
temperature |
0.076 |
pollen |
0.006 |
relative humidity |
0.013 |
PM10 |
0.011 |
ozone |
0.007 |
These results are important because in previous analyses, we did not find an effect of ozone or PM10at lags of 0 or 1 day. While scanning through all possible lags would not be a valid way of testing the effect of these pollutants (due to inflation of Type I error from multiple tests), jointly modeling all of the lags in a distributed lag structure is valid and yields interesting new results.
Study C.3: Longitudinal and Spatial Analysis of Short-Term Respiratory Health Effects of Air Pollution.
The goal of this project is to develop a statistical model and method of estimation for data in air pollution epidemiology studies that have both a longitudinal and a spatial component. For example, in our study, we have data on each individual at each point in time (each day), while individuals are also arranged by ZIP code. We would like to exploit the longitudinal structure of the data and develop a model that allows for individual differences in respiratory health outcomes. We would also like to describe the spatial structure of asthma event incidence. Finally, we would like a model that is “aggregation-consistent.” By this, we mean that the model can incorporate data from a variety of levels of aggregation. For example, outcomes are essentially aggregated over ZIP codes at a given point in time, because we do not have any person data at finer spatial resolutions than ZIP code. Many models currently in air pollution epidemiology are not aggregation-consistent. They yield inferences on the scale of the data which arise in a given study, but these inferences are not comparable to those in other studies with data on a different scale, and are awkward to use when exposure, outcome, and other data are on differing scales.
The advantage of the longitudinal component of the data is that it allows us to increase statistical efficiency and to control confounding variables in the detection of effects of daily fluctuations in air pollution and weather on acute asthma outcomes. These benefits accrue because the analysis essentially hinges on the degree to which health outcomes and air pollution covary within a person over a given time window. Effects estimated in such an analysis are automatically adjusted for all factors that are constant within a person, within a time-window or within a person-time window.
In addition, a spatial modeling component is also important because it permits the display and study of spatial variability in health outcomes that is not explained by person-time-specific, person-specific, or time-specific covariates that are in the model. Such studies are useful for hypothesis generation, examining changes in explained variability as covariates are added to the model, etc.
These desired features are captured in the following model:
Pr(Yijt = 1) = exp(R(xij) + Rij + Rijk + β'zijt),
where Yijt is the binary indicator for person j in ZIP code i at time t for whether or not an asthma event occurs, Rijk is a random effect for person (i,j) for time window k, Rij is a random effect for person (i,j), xij is the location of person (i,j), and R(x) is the underlying average log-risk of an asthma event for a person living at location x, adjusting for zijt, and averaging over between-person and within-person, between-time window variability. Because Rijk, Rji, and R(xij) may represent factors that confound the relationship between zijt and Yijt , we propose a new semi-parametric random effects structure for these factors. As in a more traditional set-up, we assume that the Rijks and the Rijs are normally-distributed and independent of one-another. In addition, the process {R(x)} is a Gaussian random field independent of the Rijks and the Rijs. However, unlike the traditional model, we do not require that these effects be independent of the zijts.
Estimation of this model has required some new methodology. We estimate β via conditional logistic regression. Then, we remove the effects of zijt and use a new restricted maximum likelihood estimation (ReML)-type of estimation to obtain the variances of Rijks and Rijs without modeling their dependence on the zijks. A similar idea will be pursued for estimation of the process {R(x)}. Note that the part of the model represented by β′ zijt may include components of a distributed lag function (Study C.2 above), and that the process {R(x)} may be estimated using the disease mapping methodology outlined in Study C.1 above. This work is ongoing.
D. Spatio-Temporal Structure in Measures of Air Quality: Comparison of Air Pollution Monitoring Data with Physical Models. In addition to space-time modeling of the ozone monitoring data, we are comparing monitoring data to ozone values generated from a physical model (CMAQ). The goals of this work are threefold: (1) evaluate the performance of the model; (2) produce improved maps of ozone by combining the model and monitor data; and (3) provide and evaluate the model generated PM2.5data as an additional variable for the asthma model. CMAQ was run on site to generate a month’s worth of model data for July 1996. In running the model, we produced three nested domains: the eastern United States at 36 km resolution, Illinois at 12 km resolution, and Northeastern Illinois (including Cook county) at 4 km resolution. To date, we have focused on the evaluation of the model output, and have made three preliminary conclusions. First, the highest resolution model output (4 km) provides little additional predictive power over the 12 km resolution for determining ozone at locations where there are no monitors. The highest resolution domain is the most computationally expensive to produce; therefore, if one limits the model output to 36 km and 12 km resolutions, one can save significant resources in the generation of long model runs. Second, the model output does not capture extreme high values in the monitoring data. Possible explanations include problems with the emission model (an input to CMAQ), small-scale effects in the vicinity of the monitors, and difficulties with the meteorological model (another input to CMAQ). This will require further analysis. Third, the Planetary Boundary Layer (PBL) may be particularly important on days in which the model consistently underreports the ozone values. We stratify the model data by testing if the midday PBL is below a certain level. This initial analysis captures some of the low values produced by CMAQ.
Relevance to EPA’s Mission
The components of this project are of direct relevance to important concerns of EPA. These involve gaining an improved understanding of relationships between air quality and respiratory health, where this understanding provides a basis for policy formulation and evaluation of past actions. The results obtained thus far and to be produced in the future are relevant to: (1) identifying new respiratory health outcomes, such as albuterol prescriptions, that are useful in conjunction with air quality information; (2) determining quantitative links between respiratory health and measures of air quality in a major urban area; and (3) developing improved statistical methodologies for relating environmental factors to human health.
Journal Articles on this Report: 1 Displayed | Download in RIS Format
| Other subproject views: | All 18 publications | 3 publications in selected types | All 2 journal articles |
| Other center views: | All 102 publications | 59 publications in selected types | All 37 journal articles |
| Type | Citation | ||
|---|---|---|---|
|
|
Naureckas ET, Dukic V, Bao X, Rathouz P. Short-acting β-agonist prescription fills as a marker for asthma morbidity. Chest 2005;128(2):602-608. |
R829402C003 (2004) R829402C003 (2006) R829402C003 (Final) |
|
, Ecosystem Protection/Environmental Exposure & Risk, HUMAN HEALTH, Economic, Social, & Behavioral Science Research Program, Air, Geographic Area, Scientific Discipline, Health, RFA, PHYSICAL ASPECTS, Ecosystem/Assessment/Indicators, Engineering, Chemistry, & Physics, Health Effects, Risk Assessments, Environmental Statistics, Applied Math & Statistics, Health Risk Assessment, Physical Processes, Ecological Risk Assessment, Ecological Indicators, Environmental Engineering, Ecological Effects - Human Health, EPA Region, particulate matter, Ecological Effects - Environmental Exposure & Risk, Ecosystem Protection, Monitoring/Modeling, Environmental Monitoring, risk assessment, trend monitoring, ozone , chemical transport models, air pollutant-induced pulmonary inflammation, particulate, stochastic models, statistical methodology, air quality, computer models, ecological risk, ecosystem health, environmental indicators, ozone, chemical transport, health risk analysis, environmental health effects, aersol particles, human health risk, monitoring, statistical models, human health effects, particulates, statistical methods, watersheds, Region 5, air pollution, modeling studies, sediment transport, stratospheric ozone, atmospheric particulate matter, emissions monitoring, data models, exposure, water, chemical transport modeling, airborne particulate matter, ecological models, asthma, ecological effects, ecological health, human exposure
Relevant Websites:
http://www.stat.uchicago.edu/~cises/
Progress and Final Reports:
2002 Progress Report
Original Abstract
2006 Progress Report
Final Report
Main Center Abstract and Reports:
R829402 Center for Integrating Statistical and Environmental Science
Subprojects under this Center:
(EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).
R829402C001 Detection of a Recovery in Stratospheric and Total Ozone
R829402C002 Integrating Numerical Models and Monitoring Data
R829402C003 Air Quality and Reported Asthma Incidence in Illinois
R829402C004 Quasi-Experimental Evidence on How Airborne Particulates Affect Human Health
R829402C005 Model Choice Stochasticity, and Ecological Complexity
R829402C006 Statistical Approaches to Detection and Downscaling of Climate Variability and Change