I frequently serve as a peer reviewer for governmental organizations on issues dealing with risk assessment. For example, I recently served as a peer reviewer for the South Carolina Department of Health of the document Assessment and Recommendations for the South Carolina Air Toxics Standard; I am a member of the External Review Committee, Los Alamos National Laboratory; I chaired the External Review Committee, United States Department of Agriculture's Office of Risk Assessment and Cost-Benefit Analysis; and I served on the Board of Scientific Counselors, Committee to Review EPA's National Health and Environmental Effects Research Laboratory. I currently serve on a National Academy of Sciences/National Research Council Committee and am a peer reviewer for the Academy. A copy of my Resume is attached.
Purpose of This Testimony
Although Sciences is involved in a wide range of risk assessment issues and investigations, a focus of the research and analysis work conducted by me and my colleagues at Sciences is improvement in the sciences that support human health and environmental exposure and risk assessments, such as those conducted today as part of EPA's comparative risk assessments and the hazardous air pollutant (HAP) residual risk program. A major need in comparative risk assessment is development and use of scientifically supported methods and data to identify and appropriately address areas of most important environmental risk. This need is evident most recently in EPA's mandated HAP residual risk program which involves an unprecedented use of risk assessment and is being required while the science of risk assessment is still very much in flux. In that program, over 175 industry categories subject to maximum achievable control technology (MACT) standards currently being developed must have residual risk assessments completed to serve as the basis for risk management decisions. These risk assessments are required eight years after promulgation of each MACT standard and will in total involve emissions of the 188 HAPs from literally thousands of facilities. The risks must also be estimated for all types of climates and terrains, wide ranges of population distributions, direct and indirect (i.e., multipathway) exposures, human and ecological effects, and consideration of maximum individual as well as total population risk. This would be an extremely difficult and challenging task for any risk assessment program.
My purpose today is to provide my thoughts on the direction and progress of the use of risk assessment by EPA and what actions might be taken that would make that use more effective and more soundly based in science and, thus, more responsive to our nation's needs. My analysis is based largely on two important recent documents which describe how EPA is currently intending to apply risk assessment. The first is the Residual Risk Report to Congress (Report No. EPA-453/R-99-001), which was published in March 1999 and provided the Agency's broad scientific guidelines for managing that program. The second is EPA's first draft residual risk assessment document which was released for Science Advisory Board (SAB) review in January of this year. It provides the first detailed application of the guidance from the Report to Congress for performing residual risk assessments, and was applied in a case study to the secondary lead smelting industry.
Let me briefly describe the comparative risk and residual risk assessment programs and their goals. Then I will identify a number of issues that make risk assessment complex and implementation of these programs exceedingly difficult. Finally, I will offer several recommendations for improving the risk assessment process.
Comparative Risk Assessment
Comparative risk assessment is broadly the process whereby human health and environmental risks are identified and evaluated and the risks compared to assist in setting priorities and in making informed regulatory decisions. At the core of the process must be sound risk assessment science and stakeholder participation to provide the necessary framework for sound and socially responsible decision-making. Comparative risk assessments typically look at all types of risks in all environmental media and seek to provide sufficient information to make appropriately informed decisions. These decisions must necessarily rely on the identification and use of accurate risk assessment methods and data.
EPA's Residual Risk Regulatory Program
Risk assessment is currently being used in EPA's residual risk regulatory program. The 1970 Clean Air Act Amendments first required the EPA to identify and then regulate HAPs to levels that provide an "ample margin of safety to protect the human health." The term "ample margin of safety" was defined by EPA in 1989, after the U.S. Court of Appeals ruled that the first step in the regulation of a hazardous or toxic air pollutant was to determine a safe or acceptable level of risk based only on health factors without regard for technical feasibility or cost. However, the regulation of HAP exposures without consideration of social and economic costs or technical feasibility was difficult to implement and only seven HAPs were regulated under the 1970 Amendments.
Consequently, Congress established in section 112 of the 1990 Clean Air Act Amendments (1990 Amendments) a new regulatory process for HAPs. First, a list of HAPs was specifically mandated by Congress and EPA was required to publish, over an eight-year period, MACT standards for the sources of the listed HAPs. Next, eight years after publication of each MACT standard, the EPA was required to promulgate additional standards if needed to ensure protection of public health and the environment. In other words, the risks remaining after imposition of the MACT standards, the so called residual risks, would be determined and additional controls imposed if those risks are judged not to meet the "ample margin of safety" criterion. The EPA began publishing MACT standards in late 1993 and was supposed to be completed with the entire program this year, although that is unlikely to happen.
While relatively straightforward in concept, implementation of the residual risk requirements under section 112(f) of the 1990 Amendments has been difficult and is far from complete. One hurdle was cleared with the definition of ample margin of safety in 1989. EPA's published risk decision-making policy set as a goal: "(1) protecting the greatest number of persons possible to an individual lifetime cancer risk level no higher than approximately one in one million (1 x 10-6), and (2) limiting to no higher than approximately one in ten thousand (1 x 10-4) the estimated risk that a person living near a source would have if exposed to the maximum concentrations for 70 years." EPA further stated that a maximum individual risk (MIR) of one in ten thousand should ordinarily be the upper end of the range of acceptability. As risks increase above this benchmark, EPA stated that they become presumptively less acceptable under section 112, and would be weighed with the other health risk measures and information in making an overall judgment on acceptability.
This risk policy has largely been accepted and it was codified in the 1990 Amendments in section 112(f)(2)(B). However, it has limitations in that it only addresses cancer, inhalation risks, and risks to individuals. We now know that many HAPs are not carcinogens, that humans can be exposed through ingestion and skin contact, and that broader population risks must also be considered in addition to individual risks. EPA has not yet provided complete guidance for how to treat noncarcinogens and population risks in making decisions under the residual risk program.
Recognizing that substantial work remained to be done in planning and implementing the residual risk requirements of the 1990 Amendments, Congress required in section 112(f)(1) that the EPA submit a report to Congress that describes Agency plans for complying with the requirements of the 1990 Amendments dealing with residual risks. As noted above, EPA submitted in March 1999 the final Residual Risk Report to Congress (Report No. EPA-453/R-99-001). The report describes a residual risk assessment strategy design that involves at least two tiers of risk assessment a screening assessment followed by more refined assessment for those HAPs and sources with a potential for excess human health or environmental risks. A specific concern of mine is that the necessary refined levels of assessment, the methods for estimating the refined risks, and the criteria for determining when and how they are to be used, have not been articulated. Some of the specific scientific and technical issues are described below.
Scientific and Technical Issues
There are several scientific and technical issues that will be important to many future residual risk and comparative risk assessments, issues that have not yet been fully addressed. For example, section 112(f)(2) in the 1990 Amendments requires consideration of the environmental effects (also called ecological effects) of HAPs in addition to human health effects. This requirement was new in the 1990 Amendments and encompasses risks to wildlife, aquatic life, or other natural resources. These risks largely result from HAPs, such as PCBs, dioxins, and mercury, that are persistent and can bioaccummulate. The EPA has published broad guidance for conducting ecological risk assessments (Guidelines for Ecological Risk Assessment, EPA/630/R-95/002F, April 1998), but substantial interpretation and judgment are necessary for their application.
As noted above, HAPs in the past were defined and regulated primarily based on adverse effects resulting from inhalation of the pollutant by humans. More recently, the EPA has begun broadening this to consider all potential routes of exposure. For example, HAPs may deposit on, and be adsorbed into soil, plants, and surface waters with which humans can come in contact. Contaminated food crops, animal food products, and fish that are consumed by humans may result. These risks, too, are largely associated with HAPs that can bioaccummulate. Other EPA programs are requiring multipathway risk assessments in some instances but multipathway risk assessment is new to EPA's residual risk program; to date, the principal experience has been with hazardous waste combustion sources and some hazardous waste sites. One problem is that multipathway risk assessments have typically been designed and applied to individual facilities and they require extensive data and analysis. When these applications are focused on facilities, the use of site-specific data in lieu of generic assumptions is found to make an important difference in the risk outcome. Application to broad source categories is a very different matter. These assessments will tend by necessity to rely on conservative (meaning health protective), generic assumptions. However, intensive and focused efforts are needed to identify the generic parameters that are the risk-drivers through a sensitivity analysis to replace the generic assumptions with more accurate scientific data..
EPA is also considering the estimation of broader population risks in addition to individual risks in the residual risk program. For example, EPA's ample margin of safety language requires the protection of the "greatest number of persons" to a risk no greater than one in one million. However, the manner in EPA will address population risks has not yet been defined. One risk characterization process was described in the 1994 National Research Council report (Science and Judgment in Risk Assessment) as including two population risk metrics: (1) distribution of individual risk across the exposed population (e.g., the number of individuals at risk in various risk intervals such as 10-3 to 10-4, 10-4 to 10-5, and 10-5 to 10-6), and (2) estimated population risk, expressed as average annual incidence.
Current Application of Residual Risk Assessment
To date, as noted above, EPA has completed just one draft residual risk assessment, a case study of the secondary lead smelting industry, which was reviewed by EPA's Science Advisory Board (SAB) on March 1 and 2, 2000. I reviewed this draft assessment in detail and presented oral comments to the SAB. In my comments, I concluded that the SAB's comments on the earlier Residual Risk the Report to Congress had not been fully addressed in formulating the case study. Significant gaps in the science, the methods, and the data remain that can only be resolved through more detailed assessment, often including source and site-specific assessments.
EPA appropriately described a tiered process where an initial, conservative screening assessment is done to conserve resources. Depending upon the results, the tier one assessment is to be followed by a more refined assessment; where risk outcomes are low, this screening assessment can indicate no further study is needed. However, where risks are of possible concern, a clear commitment is needed to refine the screening level assessment and to articulate criteria for when and how to provide a more accurate assessment. Currently, EPA has not provided clear guidance on the necessary levels of refinement, the methods and data to be used, or even the criteria for deciding when and how to initiate the refined assessment. These are critical to making responsible and scientifically sound regulatory decisions. While I am sensitive to the Agency's resource limitations and the resultant inability to conduct full site-specific risk assessments for every HAP source in every source category, I believe that more refined source data can often be reasonably obtained and utilized to further refine the assessments. The Agency must be equally sensitive to the profound potential economic impacts of further residual risk regulation of sources that have already expended tremendous resources in meeting MACT standards. I strongly support the need to further regulate any source that is found to clearly and unambiguously exceed acceptable risk levels. However, I do not believe it is in the nation's best economic interest to force needless expenditures when residual risks are not excessive. A refined risk assessment is needed in order to make these determinations. The use of upper bound generic approaches usually provides a poor basis for regulatory actions.
Even implementation of the two-tiered strategy described in the Report to Congress is associated with a number of likely problems. First, given the growing complexity of the science of risk assessment and the wide variability in HAP sources, more than two tiers of risk assessment will usually be needed to ensure relative accuracy as well as cost- and resource-effectiveness. These considerations are necessary because screening risk assessments, with rare exception, estimate risks that are excessive, which can mislead the regulatory process, unnecessarily raise public concern, and possibly miss identification of the most important risks. Recent studies that I and my colleagues at Sciences have conducted concluded that these two tiers of risk assessment can be associated with risk differences of several orders of magnitude.
The Report to Congress also implied that the EPA would conduct all of the necessary residual risk assessments for the more than 175 source categories. However, our knowledge of the EPA HAP regulatory program and staff, based on past working relationships and recent personal communication, indicate that the Agency almost certainly does not have the resources to accomplish this enormous assignment. The more likely outcome is that the EPA will rely on more simplified models and averaged, rather than site-specific, data. This approach will typically define residual risk estimates that are greater than actual risks. These simplified approaches cannot adequately inform regulatory decisions.
With inadequate resources and substantial data gaps, I can see no way for EPA to carry out the residual risk program within the prescribed time without outside partnerships to aid in developing appropriate information, working together, where possible, to ensure that the best data and methods are used in the Agency's analyses, and filling the EPA's resource shortfall with analytical and data gathering support. In addition, with risks estimated using "model plants," and other approaches that rely on averaged data, industrial facilities will often need to ensure that more site-specific data and methods are employed to determine whether the model plant risks are realistic. In our work, we have found that these averaged approaches typically lead to risk overestimates.
For many industrial source categories, the initial conservative screening assessment could find that residual risks are unlikely to exceed levels of concern at any industrial facility and no further risk-related regulation would be forthcoming. However, for many other source categories, much more accurate and rigorous assessments may be needed in order to determine whether further regulation is required. In other words, if a screening approach indicates risks near or above presumptively acceptable risk levels, or ecological, population, or multipathway risks are potentially indicated, much more detailed and accurate assessments will be indicated. A sensitivity analysis should be performed early in the process to determine those site-specific parameters that are the most important to the risk outcome. These are called the risk drivers. Data collection can then be focused more cost-effectively. In some cases, individual facility, site-specific risk assessments using probabilistic exposure and risk assessment techniques will be required to define the most realistic risks for the facility. The criteria and methods for conducting these more refined assessments must be established, including the following:
1. Detailed characterization of the industry sources (point and area) including location and dimensions of all emission sources, emission quantities, building sizes and shapes, and other relevant factors.
2. Detailed characterization of the surrounding terrain, including U.S. Geological Survey topographical and digital elevation maps.
3. Detailed characterization of the population distribution, often within 50 kilometers around the facilities.
4. Hourly, on-site meteorological data or, if not available, long-term data from the nearest National Weather Service station.
5. Specific emission characteristics, including release height, temperature, and velocity, and duration and upset conditions.
6. Agreement on appropriate dispersion models to be used.
7. Agreement on, and often reanalysis of, appropriate health criteria to be used for the emissions (see the discussion of the IRIS database later in this testimony).
8. Agreement on, and often reanalysis of, other health effects besides cancer risks to be considered.
9. Identification and assessment of possible ecological concerns.
10. Identification and assessment of multipathway effects, including defining realistic pathways and receptor considerations (e.g., it is rare that a farmer eats 100% of his daily diet from farm grown poultry, beef, pork, and produce).
Specific Areas of Concern with EPA's Residual Risk Case Study
The process used by EPA in its first residual risk assessment was an appropriately conservative first step that is useful for setting priorities for assessment while conserving Agency resources. However, as presently developed and described by EPA, the process remains a screening tool that can reasonably exclude sources from further assessment, but cannot with accuracy determine whether the residual risks associated with any specific source are above or below accepted levels of risk concern. If used more broadly, the process is almost certain to result in a significant number of false positives namely, sources for which additional regulation appears needed when, in fact, the actual risks are below acceptable levels of concern. This outcome is likely to occur because screening level risks are calculated without properly accounting for the many limitations and uncertainties in the data, models, and methods used by EPA in conducting the assessment, and because of the intentional bias to protect public health where data are uncertain. Inappropriate regulation can only be prevented by the use of much more refined assessment, often including site-specific or category-specific data, thus allowing the decisions needed to support further regulation of those sources that require further control. One illustrative example of the problem caused by reliance on screening tools was an initial screening level analysis that indicated that most of the nation's coke ovens were above EPA's acceptable cancer risk level as provided in the 1989 benzene decision. Site-specific analysis of several facilities using improved data, a better model which we developed, and health criteria that we re-evaluated, actually found residual risks to be on average three orders of magnitude below the screening level risks and actually well below EPA's acceptable cancer risk level.
In cases where substantial uncertainties and significant limitations in data exist, risk management decisions should not be made until these limitations are appropriately addressed. In its initial residual risk assessment of the secondary lead smelting industry, EPA concluded that the risk estimates likely fell "between the estimates made with and without fugitive dust emissions." This risk range spanned more than two orders of magnitude and is so large that the results are impossible to interpret. In this case, at a minimum, better quality, site-specific data of such risk-drivers are needed to refine the risk assessment so that the results are useful.
EPA's use of conservative methods, models, data, and assumptions early in a source category analysis is appropriate to conserve Agency resources and help prioritize actions for later analysis. However, the use of implausible and unrealistic methods, models, data, and assumptions, particularly when better methods and data are easily obtained, is clearly inappropriate and in this case led to a number of potentially erroneous conclusions. For example, EPA's first draft residual risk assessment resulted in risk estimates high enough that (if true) serious adverse human health and ecological effects would likely be easily observable in the nearby areas. However, the lack of apparent evidence of significant human or ecological impacts near the sources of concern gave every indication that EPA's estimates were unrealistic. Inaccurate and incomplete data, coupled with excessively conservative assumptions, lead to excessive risk estimates. Two specific examples are described below:
1. The assumption in the residual risk assessment that a local farmer obtains drinking water from an untreated local surface water source that exceeds maximum contaminant levels (MCL) for antimony is unrealistic. More realistically, water would be obtained from wells or public water systems. This assumption was not conservative, it was implausible.
2. The modeled surface water concentration near one facility resulted in estimates of huge fish tissue concentrations and large potential risks to the recreational fisherman. Moreover, the modeled concentrations were sufficient to cause serious effects for aquatic organisms, such that there would be a question of fish availability for consumption. However, the lack of collaborative evidence (e.g., fish kills in the vicinity of secondary lead smelters) indicated that the estimates substantially over predicted actual concentrations. High estimated risks need to be flagged and confirmatory information needs to be developed.
EPA's first residual risk assessment process also was incomplete because it left unaddressed many potentially important issues that could have significant impacts on the ultimate residual risk estimates. For example:
1. While EPA clearly put a lot of effort into many aspects of the risk assessment process, it did not focus its data collection efforts on the most sensitive risk-driving parameters. EPA further described the assessment as an "iteration," stating that it would be replaced with a more refined, and possibly site-specific, assessment pending SAB's comments and any additional emissions information, as necessary. Although EPA did not provide further details on the next iterations, I believe that the SAB clearly should review the most refined assessment intended to provide the best basis for their scientific evaluation and comment.
2. EPA identified major issues and uncertainties at the end of each section of the assessment but the issues and uncertainties were only dealt with qualitatively and no indication was provided as to how these issues and uncertainties would be addressed in the context of the downstream risk management decisions on this or other industry categories.
3. While EPA acknowledged substantial gaps in data, methods, and procedures, it was not clear what the Agency will do about the missing information. EPA could choose to move ahead with residual risk regulation based on this level of assessment for this source category and plan to keep returning to the source category to further revise the regulations every time new guidance or methods are finalized; however, regulation based on inadequate assessment information will undoubtedly lead to inefficiency and waste.
4. There was no discussion on how the risk assessment results will be applied in a risk management decision.
EPA's first draft residual risk assessment also utilized emissions data gathered from limited, short-term stack tests at limited numbers of facilities, which was then assumed to represent long-term averages for all facilities. Fugitive emissions estimation procedures were also admittedly poorly characterized and uncertain. Use of limited emissions data can dramatically affect the risk results. The uncertainties are compounded by the fact that modeling of fugitive emissions is much more difficult than for stack emissions. For example:
1. In its first residual risk assessment, EPA frequently stated that its estimation of fugitive emissions is uncertain, but continued to use the uncertain data, which can be a risk-driver, to develop risk estimates. The uncertainty in fugitive emission estimates, especially for particulate matter, is common, owing to the technical difficulty in capturing or measuring the emissions. This fact argues strongly that better data be gathered to minimize the uncertainties in the estimation of fugitive emission rates and composition, before these uncertainties are carried through the risk assessment.
2. EPA also used after-MACT emissions which were estimated to support the proposed MACT published in 1995, rather than actual, measured current after-MACT emissions. Because industries are typically complying with the MACT standards by the time (i.e., eight years after the MACT standard is published) the residual risk assessment is conducted, actual after-MACT emissions data should be used. The use of real world data, when available, even in the initial screening level assessment is appropriate.
Recent multipathway risk assessments typically prepared or overseen by EPA are much more site-specific than the evaluation presented in the first draft residual risk assessment. This assessment utilized numerous assumptions and procedures that were not only implausible, but easily corrected. Furthermore, many of the input parameters were questionable. If realistic comparative risk and residual risk regulatory and economic decisions are to be made using risk assessment, it is essential that the models, methods, data, and assumptions be appropriate, validated, and properly used. A major uncertainty arises from the use of models that are both incomplete and not designed to rigorously address the issues involved with residual risk assessment and regulation. In the residual risk assessment, for example:
1. The oral exposure-dose equations do not include a bioavailability factor, thereby assuming that 100 percent of a chemical is absorbed upon exposure. The assumption that the bioavailability of all chemicals is 100 percent is contrary to the scientific literature and has the potential of leading to considerable overestimates of exposure dose (e.g., absorption of a chemical that is adsorbed to particulates or soil may be significantly hindered).
2. The assessment also assumed that particulate matter concentrations are available in the breathing space of a resident near the source. Air concentrations directly produced from the air dispersion modeling are simply multiplied by an inhalation rate to calculate the inhalation dose, and, hence, assume 100 percent retention and absorption of this air concentration. In fact, a smaller percentage of inhaled particles are retained in the lung, and depending on the size of the particulates, some of the inhaled particulates will be deposited in the respiratory tract, where a considerable fraction will ultimately be swallowed and should, therefore, be added to the ingestion pathway where oral bioavailability would govern the absorption of the chemical.
3. In many instances, default parameter assumptions were relied upon without accounting for the characteristics of a site. The reliance on generic default values for key parameters in lieu of site-specific data significantly decreases the likelihood that the modeled exposures will provide a reliable indication of actual exposures. Examples of inappropriate generic assumptions are: the assumption that persons are exposed 24-hours per day to outdoor air at their residence; use of home grown produce and animal products representing 100% of an individual's intake of these products; use of default soil-to-plant uptake factors (these vary considerably depending on soil types and local geochemistry); and, selection of inappropriate exposure pathways.
4. Uptake of metals into fruits and vegetables often drives the home gardener's indirect pathway risks and is one of the pathways for which great uncertainties exist. Plant uptake also has important implications for the meat and milk pathways. The generic guidance used in the first residual risk assessment greatly simplified the methodology for assessing concentrations in fruits and vegetables. This process has historically been subject to many different methodologies and data sources and this pathway has been one of the most difficult areas for risk assessors to model. The empirical values used to predict soil-to-plant transfer of metals are approximate over a wide range of soil conditions. Soil geochemistry (e.g., pH) is an important factor in the bioavailability of metals to plant roots and governs metals uptake into the edible portions. It is recognized that sufficient data often may not exist to characterize uptake using geochemical soil parameters and, therefore, default uptake factors are often used. Nevertheless, where pathways that rely on plant uptake drive the risks, a site-specific assessment must account for the effect that local soil conditions have on plant uptake; the risk outcome most likely will be changed substantively.
5. The draft residual risk assessment concludes that the drinking water pathway accounts for 73% of the total cancer risk and 70% of the total non-cancer risk for a subsistence farmer at the site with the highest ingestion (indirect pathway) risks. Drinking water also accounts for 97% of the farmer's indirect exposure cancer risk and 79% of the indirect exposure non-cancer risk. Notwithstanding the fact that the surface water concentrations are likely to be overestimated due to the assumptions regarding fugitive and stack emissions and shortfalls in the dispersion modeling, the unrealistic assumption was made that nearby surface water is used untreated for drinking water purposes. It would be illegal for any public water supply system to supply drinking water with the estimated levels of contaminants.
6. Several key inputs to the air dispersion model were not included. For example, all of the emission points at a facility were assumed to be co-located at the center of a facility; fugitive emissions were modeled using the same source area at each facility; and building downwash and local terrain features were not accounted for. Inclusion of these inputs is easily accomplished and could have a significant impact on the resulting concentration estimates and risks.
The uncertainty/variability analysis in the first residual risk assessment was limited in its scope and usefulness. The purpose of an uncertainty/variability analysis is to focus on the facilities, pollutants, and exposure pathways with the highest risks identified in the multipathway analysis. The first residual risk assessment does not say how the quantitative uncertainty analysis will be used in the regulatory processes. In addition, the evaluation implied that most of the meaningful uncertainties had been accounted for and that the results supported and validated the point estimates. However, of four possible input parameter categories emissions, transport and fate, exposure, and dose-response only two, namely emissions and exposure, were addressed in the uncertainty analysis. Omission of two parameter categories sidesteps the issue of developing a complete framework for treating the risk quantification in a realistic manner. Furthermore, many exposure variables have not been studied to establish their chemical and physical distributions (e.g., variability in consumption of farm grown produce and animal products).
Even given the limited scope of the uncertainty analysis, the input parameters used in the Monte Carlo analysis addressed only a narrow subset of the factors that influence the deterministic risk outcomes. Emission variables that could not be quantified for probabilistic analysis included: detection limits; test results from one facility used to quantify emissions at another facility; frequency of plant closing; and, fugitive emission estimates. The study ultimately relied on very limited emissions data. These are challenging issues that are resource intensive to address but, nevertheless, can have an enormous impact on the risk outcome.
In the case study residual risk assessment for the secondary lead smelting industry, screening level ecological risk assessment was conducted to estimate potential risks to aquatic and terrestrial communities from HAPs emitted from the facilities that remained after the initial screening. This screening level ecological analysis was intended to identify which HAPs require more analysis and was designed to be conservative, with assumptions generally overestimating actual exposure concentrations, thus overestimating the actual potential for ecological risks. The ecological screen indicated a high potential for ecological impacts but the results are misleading for the following reasons:
1. The screening level ecological assessment does not provide sufficient information to draw appropriate conclusions as to whether an adverse environmental effect, as defined in section 112(a) of the 1990 Clean Air Act Amendments, has occurred. Section 112(a) states that "[T]he term 'adverse environmental effect' means any significant and widespread adverse effect (emphasis added), which may reasonably be anticipated, to wildlife, aquatic life, or other natural resources, including adverse impacts on populations of endangered or threatened species or significant degradation of environmental quality over broad areas." This appears to provide a much broader definition of ecological effects than was used in the case study, requiring significant and new methodological and data needs.
2. The ecological receptors are representative of sensitive species and communities at a generic site; no regard was given to site-specific information.
The Role of EPA's Integrated Risk Information System Database (IRIS)
The accuracy of comparative risks and residual risks relies heavily on toxicity criteria from EPA's IRIS database. Let me present here a brief history of the development of IRIS and why it is uniquely important to the risk assessment process. I will also discuss the limitations and problems of IRIS in meeting the Agency's current risk assessment goals.
IRIS is an electronic data base containing information on human health effects that may result from exposure to various chemicals in the environment. I played a significant role in establishing the IRIS data base at EPA. The initial purpose of IRIS was to compile health information into one central database, and to ensure internal consistency among the various EPA Regions' and Offices' health assessments. It was originally planned for internal use only, and was never intended for direct regulatory use without the careful scrutiny by the Agency. Indeed, the original disclaimer to the preface of each IRIS file clearly indicated that the IRIS summaries are subjected to constant revisions to incorporate new data and new methodology, are subject to review by EPA scientists, and are designed to be used to support risk assessments. There was no mention of any direct regulatory purpose for the IRIS data base.
In recent years, there has been increasing reliance on IRIS for toxicological information and regulatory guidance, even though the latter is inconsistent with its original purpose. In recognition of the need for a more streamlined approach to preparing IRIS assessments and to establishing consensus, the Agency recently initiated a commendable IRIS pilot program. Briefly, the program entails the development of chemical-specific "Toxicological Review" health assessment documents prior to updating or developing an IRIS summary, input from the public, and external peer review process. On April 1996, EPA announced in the Federal Register that 13 substances will be reviewed under the pilot program. To date, ten of the thirteen substances have been updated. Obviously, the progress made by the pilot program in updating the IRIS files has been slow, which could have serious impacts on a program, such as the residual risk assessment program, which have a required completion schedule. The IRIS database currently contains over 500 chemicals, including the 188 HAPs required to be regulated in the residual risk program. Many of the IRIS files are outdated and, while updating is laudable, ten updates in four years is an entirely inadequate response. Reasons for the slow progress include resource limitations. In addition, new advanced methods to perform RfD's/RfC's and cancer assessments are being developed, including new methods for dosimetric adjustments, benchmark dose methodology, categorical regression analyses, biologically-based models with consideration of mechanism of action, and physiologically-based pharmacokinetic (PBPK) models. Thus, updating an IRIS file necessitates not only updating and evaluating the most recent literature but also reassessing the data using these new and complex methods.
As mentioned earlier, EPA's first residual risk assessment relied on toxicity criteria obtained from IRIS. However, the IRIS files were not reviewed to determine whether they were outdated. If a particular chemical is a risk driver in the residual risk assessment, the validity of the toxicity factors used must be investigated. Acceptance of published IRIS criteria without review can lead to considerable uncertainty in the final residual risk results. For example, reevaluation of EPA's cancer unit risk value for coke oven emissions by my company found, using updated epidemiological data and techniques, that the actual cancer unit risk factor is about one-fourth of the IRIS "official number" for coke oven emissions which was prepared in 1984 under my direction. That evaluation was intended only as an initial value pending publication of epidemiology studies that were undertaken at the time. However, our reevaluation was provided to EPA about two years ago, but IRIS still has not yet been updated. Clearly, a 1984 evaluation based on limited and unpublished data is inadequate for use in regulatory decision-making in 2000.
In summary, IRIS was not originally intended to be used for regulatory purposes or for that matter to provide complete toxicological data on a particular chemical. It is most useful as a screen that allows one to quickly access toxicity information that may be of help for risk assessment purposes. It is clear that there is a serious need to update IRIS. The use of outdated IRIS information has serious implications to the use of any risk assessment in decision-making.
Conclusions and Recommendations
Risk assessment provides a thorough evaluation of scientific literature as a basis for regulatory decision-making and provides the impetus for improving the process in order to facilitate better decision-making. Risk assessments support many different kinds of decisions. In comparative risks assessments, the totality of the risks are viewed to improve priority setting, decision-making, and stakeholder involvement. Residual risk assessments support specific decisions mandated under the risk requirements of the Clean Air Act. These decisions demand well thought out, comprehensive, and scientifically supported risk assessment methodology. This need poses an unprecedented challenge for EPA to develop the processes and then conduct comprehensive risk assessments across a range of chemicals, sources, and regulatory programs. This places heavy resource demands on EPA, and the required resources may not be available. Nevertheless, it would not serve our nation well to make regulatory decisions under any program by defaulting to generic risk assessment approaches or using out of date data files in IRIS because of resource constraints. The best science should form the basis for risk management decisions; otherwise our decisions are not well informed and can be flawed. It is also crucial for all stakeholders to have at least some understanding of how EPA intends to use the information developed in the residual risk assessment to make risk management decisions. Only then can the assessment's adequacy be judged.
The first residual risk assessment conducted by EPA is not sufficient to meet the needs of decision-makers under the residual risk program. At best, it presents a process that can only be described as a screening assessment, even when including multipathway and ecological analyses. A screening assessment is a necessary part of the initial risk assessment process for this program, but it falls far short of the refined risk assessment (based on more source-category specific data) that is required to make regulatory decisions.
Earlier this year, the SAB reviewed the EPA draft residual risk assessment method and its first application to the secondary lead smelting industry. Because the method requires substantial improvement, it is essential for the SAB to review the next draft of this approach. More refined risk assessments should not simply be screening assessments with more data; rather, they should rely on new methods and approaches to address important risk factors such as when to refine conservative screening values, how to assess population risks, how to characterize after-MACT standard emissions, how to assess "significant and widespread adverse" ecological risk, when monitoring data are more suitable than modeled data, and what criteria will be used to determine when the risk assessment process will be triggered. These methodological issues need to be reviewed by the SAB because they are not settled matters in the realm of risk assessment and will be of generic importance across most, if not all, of EPA's risk assessment programs.
Any more refined risk assessment methodology should be characterized by the use of category-specific, and selected site-specific, data for the elements that are the risk-drivers as identified in screening assessments. The generic assumptions used in a screening assessments such as those conducted by EPA are designed to be conservative; consequently, they can generate many false positives. Regulatory decisions cannot be based on such assessments; more specific data are needed to determine whether or not actual residual risks of concern exist. While this task varies in degree of difficulty, EPA can focus its further data collection on those elements identified by the screening assessment as the risk-drivers. Much of those data should be readily obtainable.
The residual risk assessment methodology should also explicitly incorporate realistic assumptions and data in both the screening and more refined phases of the assessment. As seen in EPA's case study assessment of the secondary lead smelting industry, implausible or unrealistic assumptions, methods, and data can fatally skew the results of an assessment. Such results have the potential to cause needless data gathering by EPA and industry in order to demonstrate that nonexistent risks are not real; unfounded public concerns may also be avoided if the results of a screening assessment are immediately scrutinized to determine if they are realistic.
Finally, if the IRIS database is to be used for regulatory purposes or for that matter to provide complete toxicological data on any particular chemical, it must be updated across the board and then maintained in an up-to-date manner. The use of outdated IRIS information has serious implications to the use of any risk assessment in decision-making.
CONCLUSIONS
Extraordinary complexity of the risk assessment is called for by the comparative risk and residual risk programs. In 1976 when the first risk assessment process began at EPA, risk assessment and risk management focused largely on single chemicals such as air pollutants and pesticides. The complexity of risk assessment has grown over the years with the most complex risk assessments being conducted at Superfund sites and for combustion sources. These risk assessments addressed multipathway risk assessment issues and ecological risks but have been focused on single facility or single sites. The comparative risk program and the residual risk program as prescribed under the Clean Air Act Amendments of 1990, require unprecedented use of risk assessment across the board for over 175 industry source categories and literally thousands of facilities. In addition, the residual risk program, for example, must address the toxicity of all 188 chemicals on the hazardous air pollutant; the comparative risk program needs to address considerably more chemicals. Adding considerably more to the complexity, the residual risk program calls for the use of multipathway risk assessment and regulation of environmental risks with significant and widespread effects to wildlife, aquatic life and other natural resources, including impacts on endangered or threatened species or significant degradation of environmental quality over broad areas.
The current EPA guidelines for conducting the residual risk program (and there are no such guidelines for the comparative risk program but it probably will follow much the same process), primarily employ available data and generic and upperbound risk assessment approaches. The guidelines mention a tiered approach but do not make a clear commitment to proceeding to a clear approach where risks appear to be high nor are there any guidelines or criteria for when or how to do so.
RECOMMENDATIONS
1. The unprecedented complexity and cost of conducting the risk assessment program called for by both comparative risk and residual risk must be conducted by a carefully orchestrated tiered approach. The first tier should employ the very best available data and commit to an approach that provides the greatest accuracy possible in the risk assessment at this stage so as to avoid unfounded public health concerns over issues that may not be of substance. The risks that are so identified should also be subjected to an analysis to see if they appear to be unreasonably high. Several examples of such unreasonably high risk have been given in this testimony. For those sources that appear to have low risk after the Tier 1 assessment, no further work is needed. For other sources, if the risks appear high, a sensitivity analysis should guide further data collection to focus resources on refining those parameters, including toxicity values, to arrive at a more accurate risk assessment. Refined risk assessments that are focused on defining real risk are necessary to guide the risk management decisions. The resources necessary to conduct these risk assessments both within EPA and on the part of involved parties, need to be recognized. In addition, I see no way that these risk assessments can be refined without some kind of partnership to refine data with involved parties.
EPA's current draft guidelines fall far short of addressing a process that will ensure that any of these steps are followed beyond providing the initial upperbound risk assessment. The shortcomings of effort are laid out in considerable detail in this testimony.
2. Policy related issues need to be clarified. Historically, EPA has limited its risk guidance in the hazardous air pollutant regulatory program to carcinogens, inhalation risks, and risks to individuals, and has not addressed how broader population risk must be considered. Further, language in the 1990 Clean Air Act Amendments for residual risk states that environmental risks must also be considered. These policy issues together with how multipathway risk assessment will be considered under the residual risk program should be clearly articulated at this time. There should be an opportunity for public comment to arrive at final criteria to define how these issues are to be addressed and guidance must be developed to more accurately estimate the effects.
3. Emissions data must be improved. Historically, EPA has used readily available emissions data, for example using estimated rather than measured post-MACT emissions, to estimate current risk. The risk assessment can be no more accurate than are the emissions that are used in the dispersion modeling.
I recommend first that the most accurate available emissions data be used in the Tier 1, screening level risk assessment. Second, for those sources that appear to be associated with risks of concern, I recommend that subsequent refinements in the exposure data be sought before a final risk assessment is completed. Again, I see no way for EPA to arrive at the necessary accuracy in risk assessment without a partnership with the organizations and facilities involved.
4. Use of the IRIS database, as a repository of operational, regulatory toxicity values, must be revised. Historically, the IRIS database was established to provide a repository for all of the information available in the agency with respect to toxicity for particular chemicals. Initially, it was primarily established to ensure consistency across agency programs and to serve as an internal data system to make available work that had been completed to date on individual chemicals. The Risk Assessment Forum, for which I was the first director, set up the IRIS database and commenced the stewardship program to enter new chemicals and to put in information that was existing in the agency at the time the data system was established. In the beginning, it was clear that the database was not necessarily intended for direct use in regulatory decisions without refinement; the preface reflected the fact that the database was not intended for these purposes. Since that time, however, the IRIS database has become not only the most important source of regulatory toxicity values for use across all of EPA's programs, but is widely used across state programs and internationally as well. The files for over 500 chemicals that are contained in this database are in many, many cases vastly out of date both with respect to the current literature and the use of current methods for dose response extrapolation. To illustrate the difficulties, EPA began in 1996 a full-scale review of 13 chemicals in IRIS. At this time, only 10 of those updates have been completed.
The current guidance that has been issued for the residual risk program, and practices for the comparative risk program, use these toxicity values as if they are intended for immediate regulatory use without refinement. This is inappropriate. All risk assessment programs in EPA, in particular in this instance, the comparative risk and residual risk programs, should explicitly recognize that these toxicity values are vastly out of date and must be refined where risk drivers are identified. This recognition should be part of the iterative process for the Tier 2 and subsequent stages of risk assessment refinement for these programs.
I recommend, as everyone else does, that resources be committed to EPA to update all of the over 500 chemicals in IRIS and to keep the database refined and up to date. In making this recommendation, I realize that this is an almost impossible task. Nevertheless, I think all efforts should be made to carry it out to the extent possible. To be totally practical, I further recommend that the preface to the IRIS database be re-stated to recognize reality, that is that the IRIS database can probably never be a current source of all the latest information in the literature with application of the latest risk assessment methodologies necessary to provide the accuracy essential in risk assessment to inform risk management decisions. I further recommend that a partnership be established between EPA and private institutions to refine toxicity values, particular in the Tier 2 part of a risk assessment process where risk drivers have been identified through a sensitivity analysis. By this method, we can focus precious resources on those most important factors that can improve the scientific basis for our decisions.
5. Uncertainty and variability analyses must be applied more explicitly. Historically, EPA has come to recognize the importance of performing variability and uncertainty analysis but to date has been able to do so only for a limited number of factors. the purpose of uncertainty and variability analysis is to focus on the facilities, pollutants, and exposure pathways of greatest concern to the estimated risks identified in the multipathway analysis. Four important parameter categories are emissions, transport and fate, exposure, and dose-response. Further, historically the uncertainty and variability analysis has been stated in some cases only qualitatively and not quantitatively or, in some limited cases, there have been quantitative statements of uncertainty. All too often, this section of the risk assessment is an afterthought that is never mentioned again in the risk management process.
I recommend that there be clear guidelines developed for the use of uncertainty and variability analysis in making risk management decisions. I recognize that this may be a difficult task to undertake but I think it is an important one. For example, the first guideline could be that where uncertainty is so great, a next tier of risk assessment is necessary before any informed risk management decision can be made. Secondly, where a risk management decision is based on a risk assessment that is highly uncertain, the decision should be considered an interim decision until more refinements in the data and methods can be accomplished. There should be a clear commitment to revisit that decision as soon as the improved risk assessment information is made available. I feel certain that guidelines such as I have suggested here could be developed that would ensure that appropriate use of uncertainty and variability analysis is made during the entire risk assessment and risk management process.
Thank you for the time to address this Committee, I would be happy to entertain any questions you may have.
1. Risk assessments today must involve a range of policy-related issues such as consideration of population, ecological, and multipathway effects. Detailed criteria are required to determine how and when these issues are to be addressed and guidance must be developed to more accurately estimate the effects.
2. When EPA's risk assessment program began over 20 years ago, it was intentionally biased to be health protective because of the many gaps in the knowledge and data at that time. Today, multiple tiers of refined assessment are appropriately being planned to be resource effective and to provide the necessary technical accuracy. However, specific criteria must be developed to determine when more refined assessments are needed and at what levels of detail. At the same time, specific methods for the application of refined risk assessment techniques are essential.
3. Accurate emission data are critical to the estimation of accurate risks. Methods must be sought that will improve the quality of the emission data used.
4. As currently structured, IRIS is primarily useful for screening purposes. It must be updated and then maintained if it is to be used in risk management decisions.
5. Variability and uncertainty analyses must be conducted to understand the quality and accuracy of the risk assessment and its impact on risk management decisions.