Protecting Human Health From Air Pollution: Shifting From a Single-pollutant to a Multipollutant Approach

Dominici, Francescaa; Peng, Roger D.b; Barr, Christopher D.a; Bell, Michelle L.c

doi: 10.1097/EDE.0b013e3181cc86e8
Air Pollution: Opinion

To date, the assessment of public health consequences of air pollution has largely focused on a single-pollutant approach aimed at estimating the increased risk of adverse health outcomes associated with the exposure to a single air pollutant, adjusted for the exposure to other air pollutants. However, air masses always contain many pollutants in differing amounts, depending on the types of emission sources and atmospheric conditions. Because humans are simultaneously exposed to a complex mixture of air pollutants, many organizations have encouraged moving towards “a multipollutant approach to air quality.” Although there is general agreement that multipollutant approaches are desirable, the challenges of implementing them are vast.

From the aDepartment of Biostatistics, Harvard University School of Public Health, Boston, MA; bDepartment of Biostatistics, Johns Hopkins University Bloomberg School of Public Health, Baltimore, MD; and cYale University, School of Forestry and Environmental Studies, New Haven, CT.

Editors' note: A commentary on this article appears on page 195.

Correspondence: Francesca Dominici, Department of Biostatistics, Harvard University School of Public Health, 655 Huntington Ave, Boston, MA 02115. E-mail:

Article Outline

In this commentary, we discuss a multipollutant approach for controlling ambient air pollution that describes multipollutant concepts for different aspects of air quality management and science: (1) scientific estimation of the health risk of multiple pollutants, (2) setting of regulatory standards for multiple pollutants, and (3) simultaneously implementing compliance with regulatory standards for multiple pollutants.

Air pollution policies worldwide are typically geared toward control of a single pollutant. In the last few decades, most epidemiologic studies of air pollution and health have focused on estimating the adverse effects associated with ambient exposure to a single pollutant, adjusted for exposure to other pollutants and potential confounders. However, the human body is actually exposed to multiple air pollutants at once, in a complex mixture.

The rationale for the historical focus on single-pollutant research compared with multipollutant research is clear: single-pollutant research is easier to conduct, the results are more clearly interpretable, and most air quality policies to protect human health are based on single-pollutant strategies. In fact, under the Clean Air Act, the US Environmental Protection Agency (EPA) has the responsibility for setting a separate air quality standard for each criteria pollutant. For example, the scientific literature provides evidence on how carbon monoxide (CO) affects various human health endpoints, the EPA sets a numerical standard for CO intended to protect human health with an adequate margin of safety, and other agencies such as a state governments may develop and implement emissions-control strategies to achieve those standards. This approach does not account for the health responses associated with the simultaneous exposure of multiple pollutants.

The scientific community and the US EPA are moving toward a multipollutant approach to quantify the health consequences of air pollution mixtures as a whole, while recognizing that such a paradigm shift will be challenging.1–5 In 2004, 2 reports of the US National Research Council recommended the development of a multipollutant approach to air quality control. These reports questioned whether the current focus on single-pollutant science, and on separate National Ambient Air Quality Standards for each of the 6 criteria pollutants, truly addresses the health burden experienced by the population.3,4 The move toward a “one-atmosphere” approach has been in progress for some time, as evidenced by the EPA's development of the Community Multiscale Air Quality regional air-quality modeling system6 that simultaneously estimates levels of multiple air pollutants, including ozone and particulate matter (PM).

There are many elements that support a multipollutant approach. For example, this approach could: (1) characterize more fully the complexity of the exposure and their health impacts (eg, humans breathe ozone, PM, and other pollutants simultaneously); (2) identify the most harmful pollution emission sources (eg, industrial smokestack or automobiles), thus allowing more targeted regulation; and (3) aid effective management of air quality (eg, reduced hydrocarbon emissions from mobile sources, which would affect levels of PM, ozone and air toxics). However, a multipollutant approach to air quality presents many challenges.

In this commentary, we discuss the following multipollutant concepts: (1) the estimation of how exposure to multiple pollutants simultaneously affects the risk of adverse health response, (2) the design of policies aimed at controlling air quality of multiple pollutants simultaneously, and (3) the design of compliance strategies for multiple pollutants.

Back to Top | Article Outline


Multipollutant Approach to Estimation of Health Risk

Many studies from a variety of disciplines (including epidemiology, human exposure assessment, and toxicology) have been conducted to characterize the human health response to air pollution. However, most of these studies report the health effect of one pollutant adjusted for the exposure to other pollutants and confounders (eg,7,8). In this paper, we define a “multipollutant approach to estimation of health risk” as an investigation that focuses on estimating the total health effect associated with the exposure to multiple pollutants. This total health effect requires acknowledgment that the health burden from simultaneous exposure to multiple pollutants may differ from the sum of individual effects estimated from single pollutant models. As an example, synergistic effects have been reported for the simultaneous exposures of cigarette smoking and asbestos.9,10

A variety of approaches can be applied to estimate total health effect from multiple pollutants. A first approach is the use of statistical regression models having as predictors a main effect for each pollutant (adjusted for exposure to the other pollutants) and an interaction term for each pair of pollutants (“the statistical interaction approach”). From the fitted values of the regression model, we can estimate total effect and the associated statistical uncertainty. Similar statistical models can be defined to account for higher-order interactions, and therefore to capture the health burden associated with the simultaneous exposure to more than 2 pollutants. A note of caution: the results of any regression model become highly unstable when incorporating 2 or more pollutants that are highly correlated (eg, PM10 and nitrogen dioxide [NO2] both indicators of urban pollution). In this case, the regression model cannot reliably estimate the main effects of these 2 pollutants nor their interaction. Ideally, knowledge about the biologic mechanisms of how air pollution adversely affects health could be incorporated in the development of the statistical model, as several pollutants might lead to the same biologic mechanism of injury.

Statistical regression models with interaction terms for estimating total effect of multiple exposures have started to appear in the literature. In animal studies of environmental exposures, approaches to interaction terms include the use of multiple combinations of exposures. For example, one study used 4 groups of exposures to estimate the total effect on pulmonary function associated with simultaneous exposure to ozone and nanoparticles.11

Current statistical methods are inadequate for addressing the health risks of multiple pollutants and estimating high order interactions. Dimension reduction of some form is needed to reduce the data to a set of key predictors and to remove predictors that do not have explanatory power. Regression shrinkage and penalization methods such as the lasso12 and its numerous variants attempt to identify a small subset of individual predictors that are highly associated with the response. Penalization methods in general have the advantage that the estimated regression coefficients are directly interpretable and the response variable is used in the estimation procedure, unlike with standard dimension reduction methods such as principal components or factor analysis. Bayesian analogues, such as stochastic search variable selection,13 have similar estimation and model selection properties. The estimation of interactions between pollutants is of particular scientific interest14; however, incorporating interaction effects into any regression model results in an explosion of parameters that need to be estimated. Current methods such as the lasso, SSVS (statistical search variable selection), and related methods handle large numbers of parameters by removing spurious predictors in the model, or by grouping highly correlated predictors.15,16 A challenge requiring methodologic development is that current methods treat all predictors symmetrically and do not take advantage of the hierarchical nature of the air pollution mixture as a whole. Air pollution can be divided into groups of constituents (eg, based on sources or chemical properties) that can be further divided into their constituents. Current methods for high-dimensional regression analysis approach the problem in an “unsupervised” manner, grouping correlated predictors without incorporating prior scientific information.

Statistical methods for clustering components of an air pollution mixture based on their biologic mechanism, emission sources, or ability of predict the outcome can be borrowed from other disciplines. New statistical methods have been developed in genomics to handle large amounts of data. Logic regression has been proposed to explore interactions in high-dimensional genomic data.17 There have also been relevant developments in survival regression analyses with a large number of correlated covariates18 and methods for supervised clustering of genes (see for example19,20).

A second approach for estimating the total health effect of multiple exposures is to use the ambient levels of one pollutant to represent the combined exposure to several pollutants or to an emission source (“the indicator approach”). For instance, selenium PM2.5 has been used as an indicator of the overall coal combustion mixture.21 Other examples include sulfate as a marker of regional pollution22 and PM2.5 filter absorbance as a marker for diesel particles.23

A third approach is to define the exposure to one or more emission sources through source identification methods, including factor analysis and source apportionment techniques. For example, one can assign PM chemical constituent data to emission sources by using statistical methods for data reduction, such as principal component analysis, factor analysis, and hierarchical clustering.24,25 Specifically a Bayesian structural equation model was developed that jointly specifies an exposure model, (using a factor analysis to assign PM chemical constituents to emission sources, eg, source profiles) and a health-effect model (estimating the association between health outcomes and the source profiles).26 These statistical methods for data reduction summarize the levels of many pollutants to define emission profiles (eg, crustal, oil combustion, traffic, vegetative burning) (see for example27–32).

Regardless of the approach, there are additional challenges in estimating the health effects of multiple exposures.

Back to Top | Article Outline

Inadequate Data

Different air pollutants are measured with varying frequency and with different measurements by season. For example, ozone is typically measured daily and PM is measured every 6 days. Ozone is often measured during the warm season only, whereas PM is measured throughout the year. Therefore, a study of these pollutants in combination would have a much smaller dataset available than a study of either pollutant alone. Another example of inadequate data is mercury exposure from industrial sources (primarily through coal and oil combustion). Such exposure is not routinely measured in its airborne form, although efforts are underway to establish comprehensive mercury monitoring networks.33

Better methods for collecting and processing exposure data include land-use regression modeling, satellite imagery, air quality modeling, kriging and other spatial interpolation methods, and human exposure modeling.7,34–38 These methods can generate estimates of pollution levels in locations and time periods where monitoring data are not available; however, as these are estimated values, they introduce their own uncertainties and limitations. Methods such as kriging and inverse-distance weighting are constrained by the data and location of the existing monitoring network. Air-quality modeling produces more accurate estimates for some pollutants than for others. Researchers often face a tradeoff between the certainty of data and the statistical power of a larger dataset.

Back to Top | Article Outline

Data Analysis

Studies of air pollution and health address a growing number of pollutants, particularly in the chemical components of PM. Many pollutants have similar emission sources or formation pathways, or are precursors to other pollutants. Estimation of the total effect of multiple exposures is challenged by these interconnections among individual pollutants. For example, nickel and vanadium PM levels can be highly correlated, as both often result from oil combustion. The effect estimate for an association between a health outcome and nickel could represent the effect of vanadium, or the effect of other products of oil combustion. In such cases, researchers may apply the indicator approach described above using the levels of a single pollutant, such as nickel, to represent the overall source mixture. However, as the pollutants have multiple sources, this technique does not provide a unique identifier for a source. Another complexity of data analysis is that some pollutants are in the formation pathway of others. For instance, volatile organic compounds are precursors to tropospheric ozone, and nitrogen oxides (NOx) are precursors to ozone as well as secondary particles.

Statistical approaches to a large set of correlated exposure variables include dimension reduction techniques aimed at transforming several correlated exposure variables into a smaller set of almost independent key predictors. Dimension reduction techniques have been used to conduct multipollutant studies of air pollution and health. These include the source apportionment methods described (eg, factor analysis to identify mobile combustion and soil particles32). Dimension reduction methods are attractive from a regulatory standpoint because they help to identify specific targets of regulatory intervention. However, these approaches also have limitations. The focus on just a few sources might omit other important (but perhaps more-difficult-to-measure or less-understood) sources. In addition, some source-related approaches require information that may be poor quality or not available at all. For example, comparison of factor analysis results generated by different research teams on the same dataset found similar conclusions but with specific differences such as how to disentangle diesel from gasoline vehicle pollution.29 It is also difficult to generalize results from source-based analyses to other areas because of the location-specific nature of most source signals. Further, source-based approaches have the general limitation that they substitute one complex mixture (the air) with another complex mixture (the source). The scientific questions involving sources and pollutants can also be somewhat circular. When a given pollutant is associated with adverse health outcomes, the next scientific question may be whether that pollutant is acting as a marker for a particular source; however, if a given source is identified, the next scientific question may be what pollutant or set of pollutants in that source make it harmful.

Back to Top | Article Outline

Exposure Assessment

Strategies to improve exposure measurement may work better for some pollutants than for others. While statistical models can account for measurement uncertainty to some degree, the uncertainty may not be fully known or quantified, and will vary by pollutant. Different monitoring networks provide varying quality of estimates based on the number of monitors available, the position of monitors (eg, distributed more evenly across a county or clustered), the nature of the monitor location (eg, near major roadway versus background monitor), and potentially different frequency of measurement.

Broad classes of measurement error include: (1) instrument measurement error, (2) detection limits, (3) exposure measurement error relating to the discrepancies between ambient monitor values and personal exposure, and (4) spatial misalignment when the exposure area (eg, county) and the exposure estimates (eg, a small number of monitors within a county) are not spatially matched. Instrument measurement error can produce monitor values that are either higher or lower than the true value. For example, one study found sulfate PM2.5 levels to be under estimated by 30% to 40%.39 The quality and consistency of measurement methodologies can differ by pollutant providing, for example, less certain and comparable estimates for organic carbon matter PM than for many other pollutants. Levels of some pollutants may be below current detection limits, and such data are often addressed by substituting a zero or other fixed value, thereby obscuring variation in pollutant concentrations at low levels. The relation between ambient monitor values and personal exposure varies by indoor-outdoor activity patterns, the person, and the pollutant.40 Personal monitors have been used to estimate air pollution in some studies,41 and yet the use of personal monitors to estimate exposure can be cost-prohibitive and impractical for large or long-term studies.

The spatial misalignment between the exposed populations and the monitor sites can obscure within-community variability in pollution concentrations and can create a type of measurement error that has been largely ignored. Without proper adjustment, the health-effect risk estimates of the different pollutants can be biased. Recent work has described various approaches for addressing this problem and has introduced statistical approaches for estimating health risks with multivariate adjustment for the spatial variation of the pollutants.42,43 The degree of spatial misalignment will differ by pollutant, as some pollutants are more homogenous across large areas than others, and will also differ by spatial area. Even if a standard protocol of a given spatial unit (eg, county-level resolution) is applied, spatial misalignment may vary by area. For example, some counties in the United States are considerably larger than others, creating a higher probability of exposure error depending on the monitor coverage. For example, San Bernardino County, California is over 1500 times larger in land area than Kalawao County, Hawaii.44

While issues of exposure error are relevant to health studies of a single pollutant, such errors take on a larger role in multipollutant studies. Each type of measurement error can affect the exposure estimates for various pollutants differently. A multipollutant study often has better exposure estimates for some pollutants than for others, and the true nature of the exposure error for each pollutant may be poorly understood. This can complicate the estimation of the health burden of an air pollution mixture.

Back to Top | Article Outline

The Special Case of Particulate Matter

Multipollutant research has already been widely conducted with respect to particulate matter. Most studies of PM and health have based exposure on total mass of a particular size distribution (eg, PM10, PM2.5). Studies on PM size distribution (eg,45,46) have been a step towards identifying the part of the PM mixture that is more harmful to human health. As total suspended particles (TSP), PM10, and PM2.5 represent somewhat different sources, with crustal materials falling generally in the larger range and combustion particles falling in the smaller range (<2.5 μm). However, even within each size range, particulate matter is inherently a multipollutant, with a varying mixture of chemical components. The PM chemical composition varies greatly across the United States and by season, and chemical composition may affect toxicity.46–49

While research on PM total mass does study a complex mixture, current research has made substantial progress in studying health effects of PM in a multipollutant context by examining which PM sources or chemical components are more strongly associated with human health. High-priority questions on PM and human health have shifted from the health effects of total mass to the sources and attributes of PM (eg, size fraction, chemical components, etc.) primarily responsible for various health outcomes. Multipollutant epidemiologic studies of PM chemical components can be conducted on local and national scales now that daily concentration data are available through EPA monitoring of pollutants at a large number of US locations.47,50–52 This represents an opportunity for multicity research of PM chemical components that was not previously feasible.

Back to Top | Article Outline

Multipollutant Approach to Regulation

Air quality standards are based mostly on a summary of the scientific evidence on the health impact of each pollutant separately, with input from a scientific advisory committee and public comment. For example, the National Ambient Air Quality Standards are established for the 6 criteria pollutants separately, and PM is considered as a single pollutant based on total mass and size distribution without consideration of its chemical form. While regulators are fully aware that the ambient levels of the criteria pollutants are related (eg, NO2 is a precursor to ozone, while both are criteria pollutants) and that air pollution is a complex mixture, each standard is designed to reduce harm to human health for that individual pollutant, regardless of the levels of other pollutants.

The ability of decision-makers to establish multipollutant policies is currently limited by the availability of multipollutant science on how air pollution mixtures affect health. However, as the scientific community embraces a multipollutant framework and provides epidemiologic evidence on air pollution mixtures, multipollutant air quality standards will become possible. Some examples are described below to illustrate how a multipollutant framework could be applied to air quality regulation.

Back to Top | Article Outline

Setting Standards for Combinations of Air Pollution Levels

If the total health effects of multiple exposures (and their statistical uncertainty) could be reliably estimated, then air standards could be based on levels of multiple pollutants. For example, if strong evidence were found that the risk of the adverse effects of PM2.5 is higher on days with ozone levels higher than a certain level, then, in principle, it would be possible to define a standard for PM2.5 that would take into account the levels of ozone.

The US EPA Air Quality Index provides a single score for air quality by converting air pollution levels for 5 pollutants to an overall quality scale of 0–500 (0 representing the best air quality and 500 the worst).53 The EPA has assigned a general health condition to the index level. For example, an Air Quality Index of 0–50 represents “healthy” air quality. Air Quality Index levels of 101–150 represent air quality levels that may be “hazardous” for sensitive individuals. An individual air quality index is calculated for every pollutant, and an overall Air Quality Index is based on the highest index for any single pollutant. While this index is not an air quality regulation, it provides an example of how multiple pollutants could be considered together to set a health-based standard.

Back to Top | Article Outline

Setting Standards for Emission Sources

If a specific source is identified as particularly harmful to human health, the development of regulations could target the emissions from that source (eg, by capping total emissions from a specific industry). This strategy would cover all primary and secondary pollutants from that source. For example, if oil combustion-related particles were identified to be more toxic than crustal-related particles, then a regulation could target the levels of the PM2.5 chemical components corresponding to an identified source rather than target the PM2.5 total mass. Examples include Mexico's Hoy no Circula restrictions on driving based on license plate numbers and London's fee program for driving during peak traffic times in designated high traffic areas.54

Back to Top | Article Outline

Setting Standards for Health Risk

Standards could be set more generally to achieve some health benchmark. For example, regulations could require that the excess number of deaths attributable to air pollution exposure not exceed a given percentage of the baseline mortality for a given area. Such a regulatory standard could specify the various air pollutants or sources eligible for adjustment to obtain the desired reduction health risk. For instance, one community might achieve lower mortality through reducing particulate matter levels, while another community could lower ozone levels to achieve the same health benefit. This type of multipollutant regulation would require an understanding of the health risks of the various different pollutants in key indicators of health. The structure of such a regulation would have to be based on an understanding of the interaction between pollutants. For example, if there is no interaction between 2 pollutants, decision-makers could reduce either of them to obtain the required reduction in health risk. If there were interaction, this interaction would have to be understood and quantified so decision-makers could incorporate this interaction in estimates of the health benefits achieved by various combinations of the pollutant levels.

Back to Top | Article Outline

Multipollutant Approach to Achieving Compliance

We have described how multipollutant science could in principle estimate the total health effect associated with simultaneous exposure to several pollutants, and how regulators could develop air quality standards that are based on the levels of multiple pollutants. A third aspect of a multipollutant approach to air quality management is the development of strategies for achieving compliance with single-pollutant air quality standards while at the same time accounting for the relationships among the various pollutants. For example, a strategy to lower levels of one pollutant, say PM, may also affect the levels of other pollutants, say ozone. Rather than construct a separate compliance strategy for each air pollutant standard, decision-makers could evaluate various compliance strategies and determine their collective impact.

Recently, there have been efforts to develop compliance strategies for multiple pollutants. As an example, the Detroit Multipollutant Pilot Project aims to develop a framework to control air quality by a collective assessment of control strategies to address criteria pollutants, hazardous air pollutants, visibility, and ecosystems, and with consideration of other issues such as energy use and climate.55 The project involves detailed analysis of emissions, air quality modeling, etc, and is specifically focused on multipollutant approaches to manage air quality including regulatory compliance. The project evaluates how various potential control strategies, could affect the levels of a variety of air pollutants, rather than developing a separate strategy for each pollutant. While such an approach to compliance would be a significant improvement over considering compliance for each pollutant alone, it still relies on existing single-pollutant standards, which in turn rely on risk estimation for single pollutants.

Multipollutant approaches to air-quality management improve upon current approaches by understanding how emissions affect concentrations of numerous pollutants (including chemical and physical transformation). A variety of technologies and tools are available to assess the effects of various control strategies on multiple pollutants simultaneously, including assessment of compliance with regulatory standards. These include air quality modeling, mathematical optimization, and other approaches to develop the most efficient compliance policies. For instance, a state could evaluate several compliance policies with the objectives of meeting specific air quality targets (ie, regulatory standards for each pollutant), minimizing cost, and providing relatively even distribution of health benefits, or whatever goals the state designates.

Back to Top | Article Outline


Greater public health protection from air pollution can be achieved by shifting from a single-pollutant approach to a multipollutant approach. For this transition to succeed, fundamental changes in the way science approaches air pollution studies and new methodologic developments are needed. Researchers are moving toward this new approach, as evidenced by the case of particulate matter, which was first treated as a single pollutant through studies of total mass, then as a set of different size distributions, then a set of single pollutants through studies of separate chemical components, and most recently to studies of the complex mixture of particles from specific sources. This trajectory is also evidenced by the US EPA's recent request for proposals on Clean Air Centers, which highlights the importance of multipollutant research.56

Several organizations and policy-makers have encouraged a move from single-pollutant approaches towards a “one atmosphere” system to control air quality.3–5,57,58 The EPA's Scientific Advisory Board Particulate Matter Research Centers Advisory Panel recommended that future attention should be directed towards integrated assessments of multiple air pollutants.59 These reports provide compelling reasons for a multipollutant approach. However, the actual implementation of their recommendations is daunting. This area of research remains incompletely developed, to the point that scientists and regulators may use the term “multipollutant approach” in different ways—one referring to the estimation of the health burden from simultaneous exposure to a complex air mixture, and the other to compliance with multiple air pollutant regulations.

Each aspect of a multipollutant approach (scientific assessment of health risk, the setting of regulations, and compliance with regulations) has to recognize the involvement of multiple pollutants. Most important, the development of a multipollutant approach requires scientific knowledge of how pollutant mixtures affect health. There are several hypothesized mechanisms, including mechanisms that are likely shared by multiple pollutants. Future efforts to address health effects of mixtures should rely on our understanding of these biologic mechanisms. For example, scientific knowledge explaining the potential synergism among pollutant exposures (ie, increased total pollutant solubility/bioavailability, oxidative potential, etc), should inform the characterizations of air pollution mixtures—similar to what is done in genomics, where genes tend to be clustered based on their biologic functions.

As scientists, policy-makers, and other decision-makers move toward a multipollutant approach to air quality control, it is critical that the various constituencies understand the differences and overlap in multipollutant concepts, and the need for future research in all of these areas.

From an atmospheric-science point of view, the relationships among pollutants are well understood. However, the state of the science on how simultaneous exposure to multiple pollutants affects human health in real-world settings is in its early stages. One of the important components of a multipollutant approach to risk estimation is the development of statistical models that can estimate the total health effect associated with the simultaneous exposure to multiple pollutants including their potential interactions. This is a promising area of future development of statistical methods. However, multipollutant statistical models face greater risks for misspecification and spurious conclusions than single pollutant models. To extend beyond exploratory analysis, information from other disciplines and past epidemiologic investigations is necessary to develop an initial model formulation by (1) determining which pollutants to include, (2) selecting the relevant health outcomes to study, and (3) building an appropriate functional form for the relationships among pollutants. Findings contributing to understanding biologic mechanisms, such as from toxicology or human exposure studies, can help reduce dimensionality and inform the structure of statistical models. In turn, epidemiologic studies can test specific hypotheses in real-world settings and human populations that link back to biologic findings.

The study of multiple pollutants will require collaboration across various disciplines. This presents its own challenges, such as defining common language. In studies of air pollution and human health, the synergistic effect of more than one environmental agent is often defined as an interaction. However, the word “interaction” has different meanings across scientific disciplines. From a statistical standpoint, “interaction” can refer to the risk associated with changes in the product of 2 variables (in this case, the concentrations of 2 pollutant levels). If the statistical interaction between 2 pollutants is not null, then the total effect of multiple pollutants cannot be estimated by taking the sum of effects based on single-pollutant research. From a biologic standpoint, “interaction” can refer to multiple pathways of various exposures or factors in physiological changes (eg, gene-environment interaction refers to how environmental exposures affect individuals differently based on their genetics). These definitions are separate from “interaction” in atmospheric chemistry, such as how nitrogen oxides and volatile organic compounds interact together to form tropospheric ozone.

In a multiple-pollutant approach to air quality, connections between epidemiology and toxicology could become increasingly important. Toxicology can study multiple pollutants in a controlled setting, whereas epidemiology can examine effects of ambient air pollution levels in large real-world human populations. Animal-exposure studies have investigated exposure to multiple pollutants, including investigations of the pollutants for confounding and interaction.14,60,61 However, a multipollutant approach to toxicology is also challenged by the increasing cost and time required to expose a large number of animals to several pollutants simultaneously.

Multipollutant concepts could be further expanded to consider other pathways of environmental exposure, such as ingestion, dermal exposure, or mother-to-fetus pathways. Some air pollutants have well-established multiple exposure pathways (eg, lead). Understanding the health effects of the total environment (ie, air, water, agriculture) can perhaps be seen as a long-term objective. As we move toward a “one atmosphere” approach and as scientific understanding of the health impacts of air pollution grows, decision-makers may eventually be able to regulate air pollution sources and the overall air pollution mixture more effectively, perhaps even to achieve a “one environment” approach.

Back to Top | Article Outline


1. Board of Scientific Counselors (BOSC). Mid-Cycle Review of the Office of Research and Development's Air Research Program at the U.S. Environmental Protection Agency. U.S. Environmental Protection Agency; 2008.
2. Health Effects Institute (HEI). HEI Strategic Plan for Understanding Health Effects of Air Pollution 2005–2010. Boston, MA: Health Effects Institute; 2005.
3. National Research Council (NRC) Board on Environmental Studies and Toxicology. Research Priorities for Airborne Particulate Matter: IV. Continuing Research Progress. Washington, DC: National Academies Press; 2004.
4. NRC Committee on Air Quality Management in the United States. Air Quality Management in the United States. Washington, DC: National Academies Press; 2004.
5. U.S. EPA. The Multi-Pollutant Report: Technical Concepts and Examples. Washington, DC: U.S. EPA; 2008.
6. U.S. EPA. Science Algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System. Washington, DC: EPA Office of Research and Development; 1999.
7. Bell ML, Kim JY, Dominici F. Potential confounding of particulate matter on the short-term association between ozone and mortality in multisite time-series studies. Environ Health Perspect. 2006;115:1591–1595.
8. Rojas-Martinez R, Perez-Padilla R, Olaiz-Fernandez G, et al. Lung function growth in children with long-term exposure to air pollutants in Mexico City. Am J Respir Crit Care Med. 2007;176:377–384.
9. Erren TC, Jacobsen M, Piekarski C. Synergy between asbestos and smoking on lung cancer risks. Epidemiology. 1999;10:405–411.
10. Henderson DW, Rödelsperger K, Woitowitz HJ, Leigh J. After Helsinki: a multidisciplinary review of the relationship between asbestos exposure and lung cancer, with emphasis on studies published during 1997–2004. Pathology. 2004;36:517–550.
11. Han SG, Andrew R, Gairola CG, Bhalla DK. Acute pulmonary effects of combined exposure to carbon nanotubes and ozone in mice. Inhal Toxicol. 2008;20:391–398.
12. Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc Series B. 1996;58:267–288.
13. George EI, McCulloch RE. Variable selection via Gibbs sampling. J Am Stat Assoc. 1993;88:881–889.
14. Mauderly JL, Samet JM. Is there evidence for synergy among air pollutants in causing health effects? Environ Health Perspect. 2008;117:1–6.
15. Maclehose RF, Dunson DB, Herring AH, Hoppin JA. Bayesian methods for highly correlated exposure data. Epidemiology. 2007;18:199–207.
16. Yuan M, Lin Y. Model selection and estimation in regression with grouped variables. J R Stat Soc Series B. 2006;68:49–67.
17. Ruczinski I, Kooperberg C, LeBlanc ML. Exploring interactions in high-dimensional genomic data: An overview of logic regression with applications. J Multivar Anal. 2004;90:178–195.
18. Witten DM, Tibshirani R. Survival analysis with high-dimensional covariates. Stat Methods Med Res. 2009 Aug 4. [Epub ahead of print].
19. Bair E, Tibshirani R. Semi-supervised methods to predict patient survival from gene expression data. PLoS Biol. 2003;2:E108.
20. Leek JT, Storey JD. Capturing heterogeneity in gene expression studies by surrogate variable analysis. PLoS Genet. 2007;3:1724–1735.
21. Laden F, Neas LM, Dockery DW, Schwartz J. Association of fine particulate matter from different sources with daily mortality in six U.S. cities. Environ Health Perspect. 2000;108:941–947.
22. Sarnat SE, Suh HH, Coull BA, Schwartz J, Stone PH, Gold DR. Ambient particulate air pollution and cardiac arrhythmia in a panel of older adults in Steubenville, Ohio. Occup Environ Med. 2006;63:700–706.
23. Janssen NA, Lanki T, Hoek G, et al. Associations between ambient, personal, and indoor exposure to fine particulate matter constituents in Dutch and Finnish panels of cardiovascular patients. Occup Environ Med. 2005;62:868–877.
24. Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning: Data Mining, Inference and Prediction. Stanford, CA: Springer-Verlag; 2001.
25. Harrell FE Jr. Regression Modelling Strategies. New York: Springer-Verlag; 2001.
26. Nikolov M, Coull B, Catalano P, Goldeski J. An informative bayesian structural equation model to assess source-specific health effects of air pollution. Biostatistics. 2007;8:609–624.
27. Sarnat JA, Marmur A, Klein M, et al. Fine particle sources and cardiorespiratory morbidity: An application of chemical mass balance and factor analytical source-apportionment methods. Environ Health Perspect. 2008;116:459–466.
28. Vedal S, Hannigan MP, Dutton SJ, et al. The Denver Aerosol Sources and Health (DASH) study: Overview and early findings. Atmos Environ. 2009;43:1666–1673.
29. Hopke PK, Ito K, Mar T, et al. PM source apportionment and health effects: 1. Intercomparison of source apportionment results. J Expo Sci Environ Epidemiol. 2006;16:275–286.
30. Ostro B, Roth L, Malig B, Marty M. The effects of fine particle components on respiratory hospital admissions in children. Environ Health Perspect. 2009;117:475–480.
31. Park SK, O'Neill MS, Stunder BJ, et al. Source location of air pollution and cardiac autonomic function: trajectory cluster analysis for exposure assessment. J Expo Sci Environ Epidemiol. 2007;17:488–497.
32. Cakmak S, Dales RE, Vida CB. Components of particulate air pollution and mortality in Chile. Int J Occup Environ Health. 2009;15:152–158.
33. Lubick N. Funding struggle for mercury monitoring. Nature. 2009;459:620–621.
34. Brauer M, Hoek G, van Vliet P, et al. Estimating long-term average particulate air pollution concentrations: application of traffic indicators and geographic information systems. Epidemiology. 2003;14:228–239.
35. Isakov V, Touma JS, Burke J, et al. Combining regional- and local-scale air quality models with exposure models for use in environmental health studies. J Air Waste Manag Assoc. 2009;59:461–472.
36. Kim SY, Sheppard L, Kim H. Health effects of long-term air pollution: influence of exposure prediction methods. Epidemiology. 2009;20:442–450.
37. Leem JH, Kaplan BM, Shim YK, et al. Exposures to air pollutants during pregnancy and preterm delivery. Environ Health Perspect. 2006;114:905–910.
38. Bell ML. The use of ambient air quality modeling to estimate individual and population exposure for human health research: a case study of ozone in the Northern Georgia region of the United States. Environ Int. 2006;32:586–593.
39. Yao X, Shairsingh K, Lam PH, Evans GJ. Underestimation of sulfate concentration in PM2.5 using a semi-continuous particle instrument based on ion chromatography. J Environ Monit. 2009;11:1292–1297.
40. Brown KW, Sarnat JA, Suh HH, Coull BA, Koutrakis P. Factors influencing relationships between personal and ambient concentrations of gaseous and particulate pollutants. Sci Total Environ. 2009;407:3754–3765.
41. Magari SR, Schwartz J, Williams PL, Hauser R, Smith TJ, Christiani DC. The association of particulate air metal concentrations with heart rate variability. Environ Health Perspect. 2002;110:875–880.
42. Gryparis A, Paciorek CJ, Zeka A, Schwartz J, Coull B. Measurement error caused by spatial misalignment in environmental epidemiology. Biostatistics. 2009;10:258–274.
43. Peng RD, Bell ML. Spatial misalignment in time series studies of air pollution and health data. Biostatistics. In press.
44. U.S. Census. Census 2000 Summary File 1. Suitland, MD: U.S. Census Bureau; 2000.
45. Dockery DW, Pope CA III, Xu X, et al. An association between air pollution and mortality in six U.S. cities. N Engl J Med. 1993;329:1753–1759.
46. Peng RD, Chang HH, Bell ML, et al. Coarse particulate matter air pollution and hospital admissions for cardiovascular and respiratory diseases among Medicare patients. J Am Med Assoc. 2008;299:2172–2179.
47. Bell ML, Dominici F, Ebisu K, Zeger SL, Samet JM. Spatial and temporal variation in PM2.5 chemical composition in the United States for health effects studies. Environ Health Perspect. 2007;115:989–995.
48. Dominici F, Peng RD, Bell ML, et al. Fine particulate air pollution and hospital admission for cardiovascular and respiratory diseases. J Am Med Assoc. 1006;295:1127–1134.
49. Peng RD, Bell ML, Geyh AS, McDermott A, Zeger SL, Samet JM. Emergency admissions for cardiovascular and respiratory diseases and the chemical composition of fine particle air pollution. Environ Health Perspect. 2009;117:957–963.
50. Ostro B, Feng WY, Broadwin R, Green S, Lipsett M. The effects of components of fine particulate air pollution on mortality in California: results from CALFINE. Environ Health Perspect. 2007;115:13–19.
51. Franklin M, Schwartz J. The impact of secondary particles on the association between ambient ozone and mortality. Environ Health Perspect. 2008;116:453–458.
52. Bell ML, Ebisu K, Peng RD, Samet JM, Dominici F. Hospital admissions and chemical composition of fine particle air pollution. Am J Respir Crit Care Med. 2009;179:1115–1120.
53. U.S. EPA. Air Quality Index: A Guide to Air Quality and Your Health. Washington, DC: U.S. EPA; 2003.
54. Tonne C, Beevers S, Armstrong B, Kelly F, Wilkinson P. Air pollution and mortality benefits of the London congestion charge: Spatial and socioeconomic inequalities. Occup Environ Med. 2008;65:620–627.
55. U.S. EPA. Detroit Multi-pollutant Pilot Project. Presented at: USEPA Technology Transfer Network. Philadelphia; 2008.
56. U.S. EPA. U.S. EPA Office of Research and Development, National Center for Environmental Research, Science to Achieve Results (STAR) Program, Clean Air Research Centers. 2009. Funding Opportunity Number EPA-RC2009-STAR-C1.
57. NARSTO. NARSTO Assessment: Technical Challenges of Accountability and Multipollutant AQ Management. Research Triangle Park, NC: NARSTO; 2007.
58. Weiss L, Manion M, Kleiman G, James C. Building momentum for integrated multipollutant planning: Northeast states' perspective. Environ Manage. May 2007;25–29.
59. U.S. EPA Science Advisory Board (SAB) Particulate Matter (PM) Research Centers Program Advisory Panel. Particulate Matter Research Centers Program Advisory Report: An SAB Advisory Report. Washington, DC: U.S. EPA; 2009.
60. Reed MD, Barrett EG, Campen MJ, et al. Health effects of subchronic inhalation exposure to gasoline engine exhaust. Inhal Toxicol. 2008;20:1125–1143.
61. Hamade AK, Rabold R, Tankersley CG. Adverse cardiovascular effects with acute particulate matter and ozone exposures: interstrain variation in mice. Environ Health Perspect. 2008;116:1033–1039.

Cited By:

This article has been cited 1 time(s).

First Steps Toward Multipollutant Science for Air Quality Decisions
Greenbaum, D; Shaikh, R
Epidemiology, 21(2): 195-197.
PDF (118) | CrossRef
Back to Top | Article Outline
© 2010 Lippincott Williams & Wilkins, Inc.