Grantee Research Project Results
Final Report: Development of Population-Based Particle Exposure Models for Human Health Risk Assessment
EPA Grant Number: R825267Title: Development of Population-Based Particle Exposure Models for Human Health Risk Assessment
Investigators: Spengler, John D. , Ozkaynak, Haluk
Institution: Harvard University
EPA Project Officer: Chung, Serena
Project Period: December 2, 1996 through December 1, 1999
Project Amount: $500,065
RFA: Air Quality (1996) RFA Text | Recipients Lists
Research Category: Air Quality and Air Toxics , Air
Objective:
The overall objective of this research project was to develop refined models to assess daily population exposure to particulate matter (PM) air pollution across a large urban area suitable for assessing health impacts of PM10 (<10 µm) and PM2.5 (< 2.5 µm) exposure. We developed new statistical methods for estimating population exposure to PM10 and PM2.5 pollution to incorporate an understanding of the interactions of humans with their environment and the limitations of ambient monitoring networks. A probabilistic model, developed by Harvard and the University of British Columbia (UBC), assesses population exposures to inhalable PM (PM10 and PM2.5) and incorporates both the variability and uncertainty in various exposure factors and model inputs. We applied the new and user-friendly PM10 exposure model to air pollution data from Philadelphia, PA, and Vancouver, Canada.
As the project developed, it was clear that microscale variations in air pollution sources (i.e., vehicles and the interaction of people with traffic-related activities) were substantial contributors to uncertainty in model predictions. In the late 1990s, several articles appeared that associated health effects to urban traffic exposure. To address this critical knowledge gap, we developed an additional objective consistent within the primary mission of our research project. We measured exposures to urban PM to determine the influence of traffic, and to develop indicators based on land use and traffic counts.
Current and projected air quality management plans developed ostensibly to protect public health principally are based on ambient monitoring of pollutants. However, the adequacy of the plans for protecting total human exposures relevant to human health has not been assessed. Even with high quality ambient pollution concentration data, a critical component of such an assessment is an understanding of population exposures to air pollutants of concern. This knowledge, combined with information on the health risks of pollutant exposures, will facilitate the development of targeted management strategies to achieve the greatest reduction in health risk.
This cooperative agreement with the U.S. Environmental Protection Agency (EPA)
involved the development of enhanced exposure models for respirable particles
and provided more relevant inputs to such models to assess PM health impacts.
Exposure models provide a cost-effective means to assess the population health
risks associated with air pollution. Census information, transportation information,
and readily available ambient air pollution data can now be combined indirectly
to estimate population exposures, and to assess the relative importance of human
activity and physical factors, such as building penetration, air exchange, and
particle deposition rates, all of which influence exposure. This multiyear study
extended existing information by:
(1) collecting critical exposure information to assess the microscale variation
of PM concentrations at the street level and across urban census tracks classified
by traffic impact; and (2) building stochastic models to predict multiday exposures
to different segments of a population that lives and works in different parts
of a metropolitan area.
This research was conducted by Harvard School of Public Health researchers with exposure, risk, health, and biostatistical experience, and by statisticians from UBC. Under the direction of Professor James Zidek, the UBC Group expanded their work on probabilistic techniques to model time and space varying ambient pollution profile across Vancouver. Using Monte Carlo simulation methods, they modeled concentrations for populations moving through a fine mesh grid of PM pollution. Harvard researchers Halûk Özkaynak (now at EPA), Jianping Xue (now at EPA), and John Spengler worked with the UBC Group to structure time-activity related pollution exposures primarily from indoor sources.
After joining the EPA in June 1998, Halûk Özkaynak became a principal collaborator with Harvard on this project. Working with his postdoctoral partners, Janet Burke and Maria Zufall, he furthered the development of probabilistic-based PM exposure modeling methods with input from Harvard researchers. Their modeling utilized ambient PM data from Philadelphia, PA, and incorporated indoor contributions from a mass balance calculation and physical factors. Meanwhile, the Harvard group, consisting of John Spengler and Jon Levy, collaborated with biostatistics Professor Louise Ryan and her students to measure and model particle concentrations indoors and outdoors in a variety of microenvironments. In addition, the Harvard component developed a geographic information systems (GIS)-approach to predicting community levels of PM related to traffic.
To date, the combined efforts of the Harvard, UBC, and EPA collaborators have resulted in seven publications and one manuscript that currently is under review.
Summary/Accomplishments (Outputs/Outcomes):
Modeling Ambient PM
Dr. Zidek and his coworkers (Drs. Sun and Le) utilized the unique PM monitoring network operated in Vancouver, Canada. Tapered Element Oscillating Microbalance (TEOM) monitors gathered hourly PM10 concentrations data from 10 locations across metropolitan Vancouver. First, they developed a spatial predictive distribution for the ambient space-time response field of daily ambient PM10 across Vancouver, computing hourly 12-hour and 24-hour average concentrations at 299 additional locations.
Observed PM data had a consistent temporal pattern from one monitoring site to the next. They exploited this feature of the field data by adopting a response model with two components; a common deterministic trend across all sites, and a stochastic residual. The "whitening" temporal residues did not lose much of the spatial correlation in the original log-transformed series. This led to an effective spatial predictive distribution for these residuals at unmonitored sites. They obtained the imputed Vancouver daily PM10 field by transforming the predicted residual back to the original data scales. This was the first step towards generating a higher resolution concentration field for calculating human exposures.
The UBC Group, with inputs from Jianping Xue and Halûk Özkaynak, continued to utilize the TOEM PM10 data available for the greater Vancouver area with an extended model to resolve both spatial and temporal interactions to predict the spatial pollution field. They used a hierarchical Bayesian approach. First, they modeled logarithmic field data as a trend model plus Gaussian stochastic residual. That trend model depends on hourly meteorological predictors and was common to all sites. The stochastic component consists of a 24-hour vector response that was modeled as a multivariate autoregressive ([AR], 3 components) temporal process with common spatial parameters. Removing the trend and AR structure leaves "whitened" time series of vector series. With this approach (as opposed to using 24 separate univariate time series models), there is little loss of spatial correlation in these residuals compared with those found in only the detrended residuals (prior to removing the AR component). Moreover, the multivariate approach that was used enabled predictions for any given hour to "borrow strength" through its correlation with adjoining hours. On this basis, we developed a spatial predictive distribution for these residuals at unmonitored sites. By transforming the predicted residuals back to the original data scales, hourly PM10 fields were imputed for Vancouver.
This work showed that a multivariate spatial predictor can be used to overcome
the problem of space-time interaction for the PM10 short-term average pollution.
The 24 consecutive hourly concentrations at each site were treated as a multivariate
response vector for that purpose.
The method does not require that an hourly AR model be specified. Rather, the
daily response vectors can be modeled as a multivariate (AR [3]) process. The
risk of mis-specifying the hour-by-hour stochastic structure is thereby avoided.
As demonstrated with Vancouver's log PM10 field, there was much hour-to-hour
variation. Simultaneously, the multivariate approach assured coherence between
the interpolated fields for success hours. Moreover, the model made the successive
spatial hourly predictions more accurate through the borrowing of strength from
adjacent hours.
The UBC Group extended this method further. The method's capability to borrow strength is enhanced through an extension to the case of multipollutant fields. Strength will come not only from adjacent hours, but from pollutants correlated with that of central interest.
Measuring Ambient and Indoor PM Microenvironmental Concentrations
Although ambient PM has been associated with a range of health outcomes, the health risks for individuals depends partly on their daily activities. Information about particle mass concentrations and size distributions in indoor and outdoor microenvironments can help to identify high-risk individuals and the significant contributors to personal exposure. To address these issues in an urban setting, we measured particle count concentrations in four size ranges and PM10 concentrations outdoors and in seven indoor microenvironments in Boston, MA. Particle counts and PM10 concentrations continuously were measured with two light-scattering devices. Because of the autocorrelation between sequential measurements, we used linear mixed-effects models with an AR-1 autoregressive correlation structure to evaluate whether differences between microenvironments statistically were significant. In general, larger particles were elevated in the vicinity of significant human activity, and smaller particles were elevated in the vicinity of combustion sources, with indoor PM10 concentrations significantly higher than the outdoors and on buses and trolleys. Statistical models demonstrated significant variability among some indoor microenvironments, with greater variability for smaller particles. These findings imply that personal exposures can depend on activity patterns, and that microenvironmental concentration information can improve the accuracy of personal exposure estimates.
To advance human exposure modeling, it is necessary to gather more information on pollution patterns in homogenous microenvironments linked to activities. Little is known about particle size distributions or composition for most microenvironments. Therefore, during the summer of 2000, Harvard researchers measured ultrafine particles, PM2.5, and particle-bound polycyclic aromatic hydrocarbons (PAHs) outdoors and indoor microenvironments in Boston, MA. In indoor microenvironments, averaged across sample days, mean ultrafine particle concentrations ranged from 3,800 to 140,000 particles/cm3, with 7-200 µg/m3 of PM2.5 and 5-12 ng/m3 of particle-bound PAH. PM2.5 indoor-outdoor ratios generally exceeded 1 in settings with high levels of human activity, with lower ratios for ultrafine particles. Cooking activities significantly contributed to elevated levels of all three pollutants. Using linear mixed-effects models with AR-1 autoregressive correlation structures, 10-minute average outdoor concentrations generally were weak predictors of indoor levels, with stronger relationships in an apartment without mechanical ventilation than in air-conditioned nonresidential settings. Although further study is needed to determine whether these patterns could be generalized beyond the monitored sites, these data support previous finding and enhance our knowledge about nonresidential PM and PAH exposure patterns.
GIS Approach to Modeling Urban Traffic-Related PM
To determine whether local transportation sources significantly contribute to exposures, we conducted a community-based pilot investigation to measure concentrations of fine PM (PM2.5) and particle-bound PAHs in the Roxbury section of Boston, MA, in the summer of 1999. Community members carried portable monitors on the streets in a 1-mile radius around a large bus terminal to create a GIS map of concentrations and gathered data on site characteristics that could predict ambient concentrations. Both PM2.5 and PAH concentrations were greater during morning rush hours and on weekdays. In linear mixed-effects regressions controlling for temporal autocorrelation, PAH concentrations significantly were higher with closer proximity to the bus terminal (p<0.05), and both pollutants were elevated, but not statistically significantly so, on bus routes. Regressions on a subset of measurements for which detailed site characteristics were gathered showed higher concentrations of both pollutants on roads reported to have heavy bus traffic. Although a more comprehensive monitoring protocol is needed to develop robust predictive functions for air pollution, this research project demonstrated that pollution patterns in an urban area can be characterized with limited monitoring equipment, and that university-community partnerships can yield relevant PM exposure information.
Vehicle emissions have been associated with adverse health effects in multiple epidemiological studies, but pollution exposures to populations living alongside roadways or elsewhere in an urban area have not been adequately characterized. Characterization of vehicle-related exposures requires detailed information on spatial and temporal trends of various pollutants and the ability to predict exposures in unmonitored settings. To address these issues, we measured continuously particle-bound PAHs, ultrafine particles, and PM2.5 at a number of sites in a neighborhood of Boston, MA, in the summer of 2001.
We intended to capture a range of traffic densities experienced across an urban road network. We used GIS software and traffic count data from the Massachusetts Highway Department to estimate average annual daily traffic according to actual count information or classification. The Massachusetts Bay Transportation Authority data on bus routes augmented traffic-count data, assuming a 1:42 bus/car ratio. We assigned a traffic density score to each 50 m x 50 m cell, based on traffic counts at 50 m, 100 m, 200 m, and 300 m radii. We selected sampling sites to represent different quartiles in traffic density scores.
Harvard researchers took measurements at the side of the road and at varying distances from the road, with simultaneous collection of traffic counts and meteorological conditions. Across all nine sites, mean roadside concentrations were 8 ng/m3 of particle-bound PAHs (range: 4-60), 16,000 ultrafine particle/cm3 (range: 12,000-54,000) and 54 µm/m3 of PM2.5 (range: 12-93) as measured with a DustTrak. Concentrations of all pollutants were lower at greater distances from the road, upwind, and at higher wind speeds, with stronger relationships seen for PAHs and ultrafine particles. In linear mixed-effects regression models accounting for temporal autocorrelation, large diesel vehicle counts significantly were associated with roadside concentrations of PAHs (p = 0.02), with a moderate association with ultrafine particles and little relationship with PM2.5. Although more comprehensive information is needed for epidemiological applications and human exposure modeling, these data provide information about small-scale spatial variations of traffic-related pollutants in an urban center. Despite limitations with implementing the randomized sampling framework based on GIS-assisted traffic scoring, there was a strong relationship between PAH and ultrafine particle count concentration, and the predefined traffic density scores. Although limited, results encourage the use and refinement of GIS method for traffic exposure classification. The collective results from these microscale spatial monitorings of PM concentrations will be important for quantifying spatial variability in population exposures to PM within a census tract, based on information on roadway and traffic density.
Population Exposure Modeling for PM2.5
The EPA developed and applied a population-exposure model for PM (the Stochastic Human Exposure and Dose Simulation [SHEDS-PM] model) in a case study of daily PM2.5 exposures for the population living in Philadelphia, PA. SHEDS-PM is a probabilistic model that estimated the population distribution of total PM exposures by randomly sampling from various input distributions. A mass-balance equation, based on the construct proposed by Dr. Özkaynak for the cooperative agreement, was used to calculate indoor PM concentrations for the residential microenvironment from ambient outdoor PM concentrations and physical factor data (e.g., air exchange, penetration, and deposition), as well as emission strengths for indoor PM sources (e.g, smoking and cooking). PM concentrations in nonresidential microenvironments are calculated using equations developed from regression analysis of exposures compared to the other microenvironments. The distribution of daily exposures to PM of ambient origin was less variable across the population than the distribution of daily total PM2.5 exposures (median = 7 µg/m3; 90th percentile = 18 µg/m3) and similar to the distribution of available indoor and outdoor measurement data for vehicles, offices, schools, stores, and restaurants/bars. Additional model inputs include demographic data for the population being modeled and human activity pattern data from the EPA's Consolidated Human Activity Database (CHAD). Model outputs include distributions of daily total PM exposures in various microenvironments (indoors, in vehicles, and outdoors), and the contribution from PM of ambient origin to daily total PM exposures in these microenvironments. SHEDS-PM has been applied to the population of Philadelphia using spatially and temporally interpolated ambient PM2.5 measurements from 1992-1993, and 1990 U.S. Census data for each census tract in Philadelphia. The resulting distributions showed substantial variability in daily total PM2.5 exposures for the population of Philadelphia (median = 20 µg/m3; 90th percentile = 59 µg/m3). Variability in human activities, and the presence of indoor residential sources in particular, contributed to the observed variability in total PM2.5 exposures. The uncertainty in the estimated population distribution for total PM2.5 exposures was highest at the upper end of the distribution, and revealed the importance of including estimates of input uncertainty in population exposure models.
The distributions of daily microenvironmental PM2.5 exposures (exposures due to time spent in various microenvironments) indicated that indoor residential PM2.5 exposures (median = 13 µg/m3) had the greatest influence on total PM2.5 ambient outdoor PM2.5 concentrations. This result suggests that human activity patterns did not have as strong an influence on ambient PM2.5 exposures as was observed for exposure to other PM2.5 sources. For most of the simulated population, exposure to PM2.5 of ambient origin contributed a significant percent of daily total PM2.5 exposures (median 37.5 percent), especially for the segment of the population without exposure to environmental tobacco smoke in the residence (median = 46.4 percent). Development of the SHEDS-PM model using the Philadelphia PM2.5 case study also provided useful insights into the limitation of currently available ambient and housing data for use in population exposure models.
Data needs for improving inputs to the SHEDS-PM model, reducing uncertainty and further refinement of the model structure partially were addressed by the modeling and measurement activities conducted under this cooperative agreement. Future SHEDS-PM modeling can now incorporate microenvironmental data developed and published with EPA sponsorship. The SHEDS-PM model can consider particle size and composition as well as enhanced GIS and statistical models to characterize space and time patterns to ambient particle concentration fields within a densely populated urban area. Moreover, future versions of the SHEDS-PM model can incorporate microscale spatial variability in ambient concentrations using the information obtained in Boston by the Harvard researchers.
Journal Articles on this Report : 8 Displayed | Download in RIS Format
Other project views: | All 8 publications | 8 publications in selected types | All 8 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Burke JM, Zufall MJ, Ozkaynak H. A population exposure model for particulate matter: case study results for PM2.5 in Philadelphia, PA. Journal of Exposure Analysis and Environmental Epidemiology 2001;11(6):470-489. |
R825267 (Final) |
|
|
Houseman EA, Ryan L, Levy JI, Spengler JD. Autocorrelation in real time continuous monitoring of microenvironments. Journal Of Applied Statistics 2002;29(6):855-872. |
R825267 (Final) |
Exit |
|
Levy JI, Houseman EA, Ryan L, Richardson D, Spengler JD. Particle concentrations in urban microenvironments. Environmental Health Perspectives 2000;108(11):1051-1057. |
R825267 (Final) |
Exit |
|
Levy JI, Houseman EA, Spengler JD, Loh P, Ryan L. Fine particulate matter and polycyclic aromatic hydrocarbon concentrations patterns in Roxbury, Massachusetts: A community-based GIS analysis. Environmental Health Perspectives 2001;109(4):341-347. |
R825267 (Final) |
Exit |
|
Levy JI, Dumyahn T, Spengler JD. Particulate matter and polycyclic aromatic hydrocarbon concentrations in indoor and outdoor microenvironments in Boston, Massachusetts. Journal of Exposure Analysis and Environmental Epidemiology 2002;12(2):104-114. |
R825267 (Final) |
Exit |
|
Levy JI, Bennett DH, Melly SJ, Spengler JD. Influence of traffic patterns on particulate matter and polycyclic aromatic hydrocarbon concentrations in Roxbury, Massachusetts. Journal of Exposure Analysis and Environmental Epidemiology 2003;13(5):364-371. |
R825267 (Final) |
|
|
Sun L, Zidek JV, Le ND, Ozkaynak H. Interpolating Vancouver's daily ambient PM10 field. Environmetrics 2000;11(6):651-663. |
R825267 (Final) |
Exit |
|
Zidek J, Sun L, Le N, Ozkaynak H. Contending with space-time interaction in the spatial prediction of pollution: Vancouver's hourly ambient PM10 field. Environmetrics 2002;13(5-6):595-613. |
R825267 (Final) |
Exit |
Supplemental Keywords:
human exposure, microenvironments, Stochastic Human Exposure and Dose Simulation, SHEDS, polycyclic aromatic hydrocarbons, PAHs, ultrafine particles, PM2.5, PM10, traffic, indoor air pollution, geographic information systems, GIS, monitoring networks, spatial correlation, space-time models, autoregressive processes, particulate matter, PM., RFA, Health, Scientific Discipline, Air, Geographic Area, particulate matter, State, Epidemiology, Risk Assessments, Atmospheric Sciences, EPA Region, urban air, ecological risk assessment, ambient air monitoring, epidemelogy, HVAC systems, particle exposure, Massachusetts (MA), National Ambient Air Quality Standards, population based particle exposure model, particulate exposure, probablistic exposure model, human health riskProgress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.