A Risk-based Assessment And Management Framework For Multipollutant Air Quality

Abstract

The National Research Council recommended both a risk- and performance-based multipollutant approach to air quality management. Specifically, management decisions should be based on minimizing the exposure to, and risk of adverse effects from, multiple sources of air pollution and that the success of these decisions should be measured by how well they achieved this objective. We briefly describe risk analysis and its application within the current approach to air quality management. Recommendations are made as to how current practice could evolve to support a fully risk- and performance-based multipollutant air quality management system. The ability to implement a risk assessment framework in a credible and policy-relevant manner depends on the availability of component models and data which are scientifically sound and developed with an understanding of their application in integrated assessments. The same can be said about accountability assessments used to evaluate the outcomes of decisions made using such frameworks. The existing risk analysis framework, although typically applied to individual pollutants, is conceptually well suited for analyzing multipollutant management actions. Many elements of this framework, such as emissions and air quality modeling, already exist with multipollutant characteristics. However, the framework needs to be supported with information on exposure and concentration response relationships that result from multipollutant health studies. Because the causal chain that links management actions to emission reductions, air quality improvements, exposure reductions and health outcomes is parallel between prospective risk analyses and retrospective accountability assessments, both types of assessment should be placed within a single framework with common metrics and indicators where possible. Improvements in risk reductions can be obtained by adopting a multipollutant risk analysis framework within the current air quality management system, e.g. focused on standards for individual pollutants and with separate goals for air toxics and ambient pollutants. However, additional improvements may be possible if goals and actions are defined in terms of risk metrics that are comparable across criteria pollutants and air toxics (hazardous air pollutants), and that encompass both human health and ecological risks.

INTRODUCTION

The National Research Council recommends adoption of a multipollutant approach to air quality management that is both risk- and performance-based (NRC, 2004). Management decisions should be based on minimizing the exposure to, and risk of adverse effects from, multiple sources of air pollution. The success of these decisions would be measured by how well they achieved this objective. Here, we provide a brief introduction to risk analysis and how it is applied within the current approach to air quality management. Recommendations are made as to how current practice could evolve to support a fully risk- and performance-based air quality management system.

Canada, the United States, and Mexico generally follow a single pollutant approach to air quality management, e.g. standards are set for individual criteria air pollutants such a NOx, ozone, and particulate matter (PM), and separate regulations are developed to address air toxics. There are numerous examples in which formal risk assessment methods have been applied to criteria pollutants (CPs) and hazardous air pollutants (HAPs) (U.S. EPA, 2005, 2006, 2007). Conceptually, transition from current practice to an integrated multipollutant air quality management strategy would entail:

• Hazard identification with respect to public health and ecosystem protection.

• Assessment of human and ecosystem exposure and response to multiple pollutants.

• Risk characterization of human individual, human population, and ecosystem risks in order to support risk management decision making.

• Reduction of emissions of multiple pollutants and resulting ambient concentrations to minimize exposure risk.

• Accounting for the influence of emissions from local, regional, national, and international sources on exposure and risk.

• Development of data and tools to support these activities.

• A system for measuring performance of the management system that will assess the effectiveness and efficiency of management actions and dynamically adjust them to more optimally achieve stated goals.

Committing to such a transition implies that we can determine which pollutants or mixtures sensitive sub-populations are being exposed to, and that we can determine the relative risks associated with these exposures. Further, such a transition requires that objective procedures be designed to assess these risks, identify appropriate actions for reducing them, and measure the effectiveness of these risk-reduction actions.

RISK ASSESSMENT AND AIR QUALITY

The concept of risk analysis has been applied in a variety of programs for protection of human health and welfare. The formalism of risk analysis has been discussed in a number of monographs and reports, such as NRC (1983).

Risk analysis involves development of decision alternatives for reducing or managing risk, valuation of the alternatives, decision making to choose an alternative, implementation of the decision, evaluation of the effectiveness of the decision (also labeled “accountability”) and iteration on the decision over time. The last two steps, accountability and iteration, are an inherent part of the proposed multipollutant framework put forward in Figure 1.

Figure 1.

Major Components of a Risk-Based and Accountability-Based Approach to Multipollutant Air Quality Management.

The starting point for the logic of Figure 1 is human activity, the source of air pollution. Human activities include many functions such as the extraction of raw materials, conversion of these materials to goods and services, and use of these goods and services (e.g., heating, cooling, transportation) by consumers. Human activity is influenced by factors such as age, gender, socio-economic status, and disease status. It can also be influenced at high temporal and spatial resolution by weather (i.e., rain, snow, and storms can lead to short-term changes in transportation and other energy use).

Human activities emit gases and particles to the atmosphere where they may become pollutants or alter the environment in some way (as through anthropogenic climate change). Once in the atmosphere, these “primary” emissions combine with emissions from natural sources and become subject to physical transport (horizontal or vertical advection), physicochemical transformations and removal. Chemical transformation can result in the production of new gas or particulate pollutants (called secondary pollutants), and can lead to changes in the chemical composition of atmospheric particles.

When primary and secondary pollutants come in contact with humans, ecosystems, or other receptors, an exposure occurs. Dose differs from exposure. Dose is what is delivered to an affected organ or ecosystem component. Exposure is sometimes referred to as potential dose because it is an upper bound on the actual (but often unknown and harder to quantify) internal dose. Once a dose occurs, the receptor organ or system reacts in ways that if the dose is high enough or prolonged enough could, in the case of humans, range from symptoms such as acute respiratory distress to chronic illness or cancer. The portion of risk analysis that is concerned with quantifying the frequency and severity of exposure and dose outcomes is risk characterization.

When conducting a risk analysis, it quickly becomes apparent that there are substantial differences in the quality and quantity of information available for individual components of the analysis. Initially it may only be possible to quantify the first few components of the source-to-outcome continuum. For example, in the absence of exposure-response data for a particular pollutant, or regarding interactions among multiple pollutants, either simplifying assumptions are made or the exposure-response portion of the analysis is omitted. If the latter, then decision making may be based on a surrogate measure of risk, such as community level exposure or ambient concentration, rather than on a direct estimate of risk. Such surrogates can be useful as interim information until a more complete characterization of risk is possible, but there is also the possibility that a surrogate could be misleading with regard to the actual effect of management options on risk.

Risk characterization can include (but not be limited to) one or more of the following: a) risk to human individuals (e.g., lifetime individual risk of cancer); b) risk to human populations (e.g., expected number of premature deaths due to excess cancers attributable to air pollution exposure or dose); and c) ecological impacts – usually focused on populations rather than individuals (e.g., survival of a plant, insect, or animal species). Risk characterization also provides information on the timing of the risk and information on uncertainty and variability generated through the risk analysis. It also provides a context for characterizing and communicating the magnitude and breadth of the risk, for example, by relating it to other more familiar human activities (e.g., a one in a million risk of death is about half the risk of dying in a train accident). However, these types of risk comparisons must be made with care. Risks differ not only in their magnitude, but also in other attributes, such as whether the risk is voluntary or involuntary, how dreaded the risk is (e.g. cancer versus sudden death from a car accident), and the temporal latency of the risk. Hence, risk communication must be done with care so not to inappropriately co-mingle these attributes that affect risk perception.

RISK MANAGEMENT

The discussion above characterizes risk assessment, which is a science-based process. Risk management involves using the results of a risk assessment in the decision- making process. One potential aspect of the risk management decision process is to place a value on the adverse effects associated with the risk or, conversely, on the damages that could be avoided by reducing the risk. Therefore, once risks have been identified and, in some cases, quantified, a valuation process can proceed. This part of the process depends on societal values. Thus, while it can be framed as part of a formal decision methodology, risk management is inherently subjective and based on policy considerations.

The valuation process deals, either explicitly or implicitly, with the relative value placed on avoiding adverse effects identified during risk characterization (e.g., should there be more focus on preventing excess cancers in humans or on avoiding damage to ecosystems, if there are such trade-offs), and also the amount of resources that should be devoted to risk management. However, these issues are separate, at least conceptually, from the scientific assessment aspects of dealing with the emissions-to-risk characterization aspects of a risk assessment. For several perspectives on the valuation of risk reductions, see Bateman et al. (2002); Hoffstetter and Hammitt (2002); and Miller et al. (2007).

Risk management options are policy choices on how to modify emissions or human activity to change downstream exposure, dose, and risks. Such options could focus on technology-based solutions (e.g., end-of-pipe emission control technologies, substitution of feedstocks, etc.) or on changing demand for goods and services in the economy. The latter can include production processes (e.g. energy efficiency improvements) and consumer demands (e.g. purchase of hybrid vehicles) to reduce pollution emissions.

Iteration is an important follow-on to the initial risk management and decision making process. Periodic evaluation of the effectiveness of management options once selected and implemented is an inherent part of the combined risk assessment and risk analysis process.

An important feature of Figure 1, which is relevant to multipollutant approaches, is that most of the current focus of multipollutant management and analysis has been on emissions and air quality components. To date, little has been done on characterizing multipollutant exposures, doses, and outcomes. As such, multipollutant management has been limited to prescribing control options that reduce multiple pollutants, without regard to the potential interactions between or relative magnitudes of exposures and risks associated with different combinations of pollutants. In part this reflects the institutional and legal frameworks within which management decisions are made. In the United States, for example, the structure of the Clean Air Act leads to the setting of separate standards for individual pollutants (e.g., NAAQS in the U.S.) or for individual source categories (e.g., New Source Performance Standards, National Emission Standards for Hazardous Air Pollutants, mobile source emissions in the U.S.), which restricts how multi-source and multipollutant risks might be incorporated into management strategies. Similar policy constraints exist in Canada and Mexico.

ACCOUNTABILITY

The risk-based framework is a structure that can be used both for integrated assessment of multipollutant air quality management (MPAQM) actions and for accountability. The components of this framework that apply to accountability assessments include measurement of the effect of implemented risk management strategies on changes in emissions, pollutant transport and fate (air quality), exposure, dose, and effects (risk characterization). These components can be simplified in some cases. For example, exposure and effects could be combined into an exposure-response component that also includes risk characterization. For each component, there are a variety of data and modeling tools that can support integrated assessments at correspondingly varying degrees of confidence. For example, for determining emissions, there are direct measurements of some pollutants from some sources, and surrogate or proxy measurements of other source/pollutant combinations. Measurements should be relevant to the averaging times of interest for the integrated accountability assessment. For other cases in which emission data may be unavailable, emissions are estimated using mass balance or other methods (NARSTO, 2005). Current methods used for developing emission inventories would be a starting point for conducting a multipollutant integrated assessment.

While integrated assessments support the design and execution of air quality management actions, accountability is a process for verifying whether predicted reductions in emissions, air quality, exposure, dose, and/or risk were actually achieved. If not, the accountability process entails determining why not and what can be done to correct the air quality management actions or to improve the prediction/estimation process. To determine whether ambient exposure actually decreases as a result of an emission control strategy or regulation, one could begin with continuous emission monitoring (CEMs) or conduct emission source tests before and after the change to verify the actual change in emissions. In order to do this in a manner that is useful with respect to human or environmental health objectives, one would verify corresponding changes in ambient conditions at temporal and spatial scales relevant to the risk-based adverse effect endpoints.

A key consideration in conducting a prospective analysis to predict and manage risk reduction, as well as to conduct a retrospective accountability assessment, is to use a parallel structure, as illustrated in Figure 2. The prospective analysis, whether it is referred to as risk analysis, integrated assessment, or by some other name, is intended to predict the results of management actions, usually incorporating reductions in emissions, ambient concentrations, exposure, and risk. Such an assessment is done to support a decision-making process. The accountability assessment is done after actions have been taken to determine whether they actually accomplished the predicted reductions. The outcome of an accountability assessment can feed into a new or updated risk or integrated assessment or directly support decision-making on whether to modify the risk management strategy. In order for both the predictive and retrospective approaches to be most meaningful, they should be based on common metrics and take into account similar framings of the problem. For example, the spatial and temporal resolution of modeling and monitoring needs for emissions, air quality, exposure, and risk should be commensurate and based on the health and environmental outcomes of concern. If the predictive and retrospective approaches do not match with respect to spatial and temporal resolution and key metrics, then the ability to make causal inferences regarding accountability is compromised.

Figure 2.

Parallelism Between Prospective Risk Analysis for Air Quality Management Planning and Retrospective Risk Analysis for Accountability

RISK ASSESSMENT IN CURRENT REGULATORY PRACTICE

Risk assessments are conducted for a number of purposes within Canada, Mexico, and the United States. In the three nations, risk analyses have been used to help inform the setting of ambient air quality standards for public health and ecosystem protection, to evaluate risks from exposure to air toxics, and as inputs to cost-benefit analyses.

While risk assessments for human health effects are relatively common, assessments of ecological risks from air pollution are less so, and are generally more limited in scope and execution. This relative lack of ecological risk assessments reflects the additional analytical elements that are required to conduct them, the much greater heterogeneity in the susceptibility of receptors, and the diversity of receptors. SAB (2007) noted that there is value in advancing the methods for ecological risk assessments by developing tools to “aid the proper consideration of temporal and spatial scale, biological complexity, and the environmental influences that amplify or detract from the level of risk associated with a single or multiple stressors in play.” The SAB also indicated a need for greater specificity in problem formulation in ecological risk assessments – in other words, a need to better define the objectives of the risk management problem.

In recent years, assessments of ecosystem risk changes have been conducted by EPA for the Clean Air Interstate Rule (CAIR), focusing on the change in risk of acidification related effects in freshwater rivers and lakes. In addition, limited assessments of the impacts of O₃ on commercial forest productivity and agricultural productivity have been conducted; however, these analyses have only focused on quantifying the direct commercial impacts of changes in forest and crop yields, and have not looked at overall sets of ecosystem services.

Of course, one of the challenges for multipollutant risk assessments including ecosystem risks is how human health and ecosystem risks can be compared and/or combined. Similar to combining disparate risk estimates within an ecosystem risk analysis, economic and noneconomic valuation techniques can be used to translate human health and ecosystem risks into common metrics for combining and comparing different bundles of risks. However, the level of uncertainty in these types of valuation estimates may limit their usefulness to decision makers in some circumstances.

COMBINING AND COMPARING RISKS ACROSS POLLUTANTS

While it is relatively straightforward to compare human health impacts across CP goals or standards, it is not an easy task to compare health risks between CPs and HAPs, or to compare overall human health and ecosystem risks. Furthermore, there is a need to combine the assessment of exposure and risk associated with multiple pollutants, which requires consistent metrics. While it is possible to compare the incidence of health effects between existing analyses of CPs and HAPs, the interpretation is not straightforward because the objectives of these analyses differed with respect to risk endpoints and policy goals. For many HAPs, the relative risks may seem small in magnitude compared to those addressed by national-scale regulations on NAAQS pollutants such as PM, or precursors such as VOCs. However, the purpose of the HAPs regulations is not only to reduce population level impacts, but to reduce maximum individual risks as well, and these reductions are not easily compared to NAAQS programs which are intended to reduce broad public health impacts. In a multipollutant framework, a consistent set of goals and metrics are needed.

ASSESSMENT OF MULTIPOLLUTANT AIR QUALITY MANAGEMENT

To date, most assessments of risks and benefits have focused on either a single pollutant, as in the case of NAAQS risk analyses, or a set of pollutants controlled by a regulation focused on one or several pollutants, as in the case of the benefits analyses of CAIR and the Non-road Diesel Rule. In general, these analyses have not focused on identifying the set of pollutants or their levels that would optimally reduce overall risk from exposure to atmospheric pollution.

To develop a risk-based air quality management (AQM) process, a clear understanding of the definition of risk needs to be established. For example, risk can be defined in terms of individual risk (e.g., individual risk of cancer) or population risk (e.g., overall population risk of death from pollution-related causes). In addition, a framework for combining and comparing risks across different pollutants is necessary. Estimates of multipollutant emissions, air quality, exposure, and controls are far more readily available than for health effects associated with pollutant mixtures. However, this is not to say that there are no scientific and/or data demands to better characterizing all aspects of the source-to-outcome continuum for multiple pollutants. Because different sets of dose-response or concentration-response functions are available for different types of pollutants, a significant challenge exists for the technical community. For example, many HAPs have dose-response functions based on laboratory animal studies, while for CPs such as NO_x and SO₂, large sets of concentration-response functions based on multi-city human population studies are available.

An important policy question is whether or not human health risks and risks to ecosystem functions should be combined in a single AQM optimization framework. Because of the differences in the spatial and temporal scales of each type of risk, it may be impractical to develop a fully integrated framework. However, conducting separate optimization exercises may lead to contradictory or suboptimal AQM plans.

Of the various steps in the risk assessment or accountability process, the quantification of multipollutant emissions is typically the step that can be performed with a high degree of confidence. Even so, the development of consistent multipollutant emission inventories remains a significant challenge (see NARSTO 2005). With regard to estimating ambient concentrations, air quality models are often used for multiple pollutants, including precursors to ozone and PM formation, acid deposition and some HAPs. These models undergo constant improvement, with the physics and chemistry of secondary pollutant formation as an example of an area of ongoing research. Exposure assessment can also be conducted for vectors of multiple pollutants. For all of these steps, emissions and ambient concentration data are required and may not be consistently available for all pollutants of interest. Thus, there are data gaps that need to be filled for exposure modeling applications; if not, “plausible” assumptions are often made to fill these gaps. However, the greatest challenges in multipollutant risk assessment are the current highly limited portfolios of risk characterization through concentration-response or dose-response information that includes exposure.

Dose (or concentration)-response assessment and risk characterization is typically considered for specific, individual pollutants. In many cases, there are multiple health effects endpoints for a given pollutant, as well as disparate, parallel health effects associated with each of many pollutants. There are challenges to the assessment of interactions among multiple pollutants (e.g., synergisms, or antagonisms) with respect to dose and response, and thus there are many uncertainties and research gaps that need to be filled in order to refine these aspects of the risk assessment. As noted earlier, even if a “perfect” dose and risk characterization can be performed, there are challenges regarding how decisions should be made by taking into account the multipollutant nature of sources and chemistry, given the range of exposed populations and potential health effects from exposure to multiple pollutants at varying ambient concentrations.

These complexities of decision-making are even greater due to the complex interactions between pollutants in the atmosphere. For example, under certain types of conditions, reductions in emissions of some pollutants such as NO_x can lead to increases in ambient concentrations of O₃, PM or HAPs in some locations while reducing ambient concentrations in other locations. Air quality management decisions for health and welfare then must weigh both the positive and negative impacts across different populations. Depending on the policy objectives, decisions could be made to allow for an increase in health or ecological risks in some populations to reduce risk in others.

In spite of these challenges, there are distinct advantages to conducting risk assessments in a multipollutant framework. One such advantage is the ability to use a single model run to characterize risks for multiple pollutants of interest. Given the costly nature of detailed photochemical modeling and exposure modeling, this advantage can reduce the overall costs of risk analyses, and allow for more sensitivity calculations, or an evaluation of a larger number of policy scenarios.

Another advantage of a multipollutant approach is the avoidance of an “ordering effect”, where the incremental risk changes resulting from single pollutant policies are dependent on the order or sequence in which they are assessed. For example, in EPA’s recent benefit-cost analyses of the PM and O₃ NAAQS, attainment of the PM NAAQS was analyzed first, and resulted in substantial estimated net benefits. In that analysis, significant levels of NO_x emission reductions were applied in portions of the western United States to reduce nitrate concentrations that contributed to nonattainment of the annual and daily PM standards. In a subsequent analysis of the proposed O₃ NAAQS, the proposed NO_x emission reductions were included in the baseline for the analysis, and more extensive VOC and NO_x reductions were then applied to attain the O₃ standard incremental to attainment of the PM standards. The results of the O₃ analysis showed a range of net impacts including both net costs and net benefits, which were substantially below the net benefits for the PM standard (U.S. EPA, 2006, 2008). The results would likely have been different had the O₃ standard been analyzed first. More importantly, had the two standards been analyzed together, incremental to a baseline with attainment of the previous set of PM and O₃ standards, the results could have created an “optimized” management strategy.

One could envision a benefits analysis where a suite of all six of the current criteria pollutants (O₃, PM, lead, SO₂, NO_x, and CO) are analyzed as a group, taking into account common emission source and meteorological conditions. Given the current U.S. 5-year regulatory review cycle, only a single benefit-cost analysis would be conducted at the end of the 5-year cycle, with consideration of a new suite of proposed standards compared against a baseline with implementation of the old suite of standards. The current structure of EPA’s NAAQS reviews would still require individual risk assessments to inform the standard review process for individual NAAQS, but the assessment of the overall projected health and welfare impacts of implementing a revised suite of NAAQS would be reframed to be fully multipollutant in nature. This evolution would correspond to the regulatory transition levels (1–4) discussed in Chapter 2.

DISCUSSION, SUMMARY, AND RECOMMENDATIONS

Risk assessment can be a useful multipollutant framework for integrating and evaluating diverse sets of data and models to produce metrics to aid in decision making. The ability to implement a risk assessment framework in a credible and policy-relevant manner depends on the availability of component models and data, which are scientifically sound and developed with an understanding of their application in integrated assessments. The same can be said about accountability assessments used to evaluate the outcomes of decisions made using such frameworks:

• Current approaches and experience with single pollutant risk assessments show the utility of this information in decision-making. Formal risk analyses are used in support of the setting of national standards in North America, recognizing uncertainties and limitations in the available analytical data and methods. Results from these analyses point to the targets, goals and metrics that can similarly be used in evaluating the outcomes under an accountability framework.

• Key strengths of single pollutant risk assessments include well-defined health endpoints, use of weight-of-evidence approaches to determine the likelihood of causal relationships, use of well characterized monitoring and modeling data for air quality, and careful reporting of quantified and unquantified uncertainties.

• Key limitations of air pollution risk assessments include incomplete coverage of health effects, inability to completely control for co-pollutant confounding and effect modification, and lack of individual exposure based health functions.

Some of the important uncertainties and gaps in knowledge regarding risks associated with multipollutant exposures, which have yet to be resolved:

• Uncertainty in primary emissions, particularly for many hazardous air pollutants

• Lack of systematic data on emissions and air quality for multiple pollutants in an integrated framework and database, beyond the criteria pollutants across North America.

• Scientific knowledge regarding the role of secondary processes for the formation of many air pollutants, such as fine particulate matter associated with secondary organic aerosols

• Dearth of personal exposure monitoring data for many pollutants and averaging times of interest, as well as lack of microenvironmental data and penetration rates of outdoor pollutants to indoor microenvironments

• Extrapolation from ambient concentrations to personal exposures of ambient origin, particularly in the context of interpreting results from epidemiologic studies

• Role of potential confounding by co-pollutants in evaluating the epidemiological evidence for individual pollutants

• Role of effect modifiers in epidemiological studies; e.g., demographic and lifestyle attributes, socioeconomic status, genetic susceptibility factors, occupational exposure, and medical care

• Concentration-response functions, including identification of potential population level thresholds and temporal stability of concentration-response functions

• Effect of individual components of PM or mixtures of PM relative to their total mass concentration, combined with the presence of gaseous pollutants

• Transferability or applicability of findings from an epidemiological study conducted in one location to other locations

• Lack of multipollutant exposure and epidemiological studies, little knowledge of additive, synergistic or antagonistic interactions between pollutants, and lack of knowledge about societal preferences for different bundles of risks across pollutants.

• Time scales for recovery of ecological systems from apparent changes from atmospheric deposition.

While progress has been made in developing the models and tools necessary to conduct multipollutant risk analyses (e.g., development and application of the CMAQ and BenMAP models, and establishment of monitoring networks for a range of pollutant species), challenges remain in informing and conducting these analyses. Some of the challenges are in the policy hierarchy, including the development of clear, specific, prioritized sets of objectives for multipollutant management, especially given the potential for tradeoffs (e.g., between an increase in risk for some populations and a decrease in risk for others that may occur when attempting to optimize emission reductions which potentially could increase or decrease ambient exposure by location). Others are technical in nature, and include the need to develop clear definitions of risk metrics and methods for combining and comparing risks across CP and HAPs pollutants. In addition, the challenges of integrating or prioritizing human health and ecosystem risk analyses remain quite formidable. Finally, some challenges are posed by the need to fill the most important gaps in the scientific knowledge that informs risk analyses. While the epidemiological literature is relatively large and robust in the United States and Canada, there are some important gaps in the available data on concentration-response functions for Mexico, especially away from Mexico City. Knowledge of vulnerable aquatic and terrestrial ecosystem response to reductions in pollutant deposition also remain limited in North America. To provide complete assessments of the economic (or other) benefits associated with risk reduction, a more complete and current set of estimates of the public’s willingness to pay for reductions in risk for a variety of health and ecological outcomes need to be generated for all three countries.

In all cases, improvements in the communication of the results of risk assessments are needed, especially for communicating uncertainty. Improved understanding of how decision-makers use quantitative information from risk analyses is critical for improvements on how quantitative results are communicated. The forthcoming report from the recently established U.S. National Academy of Sciences/Institute of Medicine, Committee on Decision Making and Communication Under Uncertainty, may provide useful insights in this area.

Key findings and recommendations on risk analysis include the following:

• The existing risk analysis framework is conceptually well suited for analyzing multipollutant management actions. Many elements of this framework, such as emissions and air quality modeling, already exist with multipollutant characteristics. However, the framework needs to be supported with information on exposure and concentration-response relationships that result from multipollutant health studies;

• Accountability and risk analysis are two realizations of the same risk framework. Most risk analyses to date have been prospective in nature-- estimating the health or ecological risk reductions expected to occur as a result of a management action. Accountability involves a process of retrospective risk analysis -- estimating the actual risk reductions that have occurred from implementation of a management action. Because the causal chain that links management actions to emission reductions, air quality improvements, exposure reductions and health outcomes is parallel between prospective and retrospective risk analyses, both types of assessment can and should be placed in a single risk management framework with common metrics and indicators where possible.

• Goals and actions need to be defined in terms of risk metrics that are comparable across criteria pollutants and air toxics (hazardous air pollutants), and can encompass both human health and ecological risks. These common risk metrics are essential to implement a risk-based multipollutant air quality management paradigm. While these metrics are challenging to develop, they are necessary in order to achieve multipollutant air quality management goals that meet public health and environmental needs in an effective and efficient manner, as conceived by the NRC (2004) committee.