Setting Priorities for Drinking Water Contaminants (1999)

A number of schemes that prioritize chemicals according to their importance as environmental contaminants have been developed by government agencies and private industries. This chapter reviews several of these existing chemical prioritization schemes. The objective is to understand the extent to which existing ranking schemes provide relevant guidance for developing a prioritization scheme for drinking water contaminants.

The committee selected a total of 10 schemes (several at the recommendation of EPA) for evaluation; these are listed in Table 2-1. The first three (the Cadmus approach, the American Water Works Association screening process, and the Regulation Development Process) are prioritization schemes specifically intended for drinking water contaminants. The next three (Waste Minimization Prioritization Tool, section 4[e] of the Toxics Substances Control Act [TSCA], and California EPA Proposition 65), are general prioritization tools for environmental contaminants. The remaining four (the Hazard Ranking System; Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA) priority listing; Hazard Analysis of Releases Inventory, and the Pesticide Leaching Potential Index), are prioritization tools for specific environmental sites or media. These are included because hazardous wastes sites, contaminated sediments, and pesticide-contaminated soils all have the potential to contaminate waters that may ultimately serve as drinking water sources. The 10 schemes are intended to be representative of contaminant prioritization schemes for a variety of functions and do not include all existing contaminant ranking schemes. The prioritization schemes were evaluated using the common framework shown in Figure 2-1. "Selection of contaminant pool" refers to the universe of chemicals

Figure 2-1

Framework for evaluating existing chemical ranking schemes.

considered in each prioritization scheme. "Determination of exposure" refers to the factors considered in evaluating the extent to which ecosystem and human receptors become exposed to the contaminant. "Determination of toxicity" refers to the factors considered in evaluating the potential human health effects of the contaminant. "Prioritization scheme" refers to the means by which exposure and toxicity are combined to provide a metric for ranking or prioritizing the contaminant.

The Cadmus approach is a health-risk-based methodology for ranking a candidate list of drinking water contaminants. The major components of the approach are laid out in a report entitled "Development of a Priority Pollutants

List for Drinking Water" (Cadmus Group, 1992). Using this approach, a risk index is derived to identify and prioritize chemicals that pose a potential health threat in drinking water. The risk index is based on the following chemical parameters: quantity produced, quantity released to water, persistence in water, frequency of detection in water, and toxicity to human health. The ranking scheme therefore incorporates both toxicity and exposure criteria.

Cadmus compiled a list of candidate chemicals from a variety of sources, including the Integrated Risk Information System (IRIS) and a set of data from the Agency for Toxic Substances and Disease Registry (ATSDR) called HAZDAT. Other chemicals were obtained from the EPA's pesticides and ground water database and EPA's storage and retrieval system (STORET). Of approximately 600 chemicals, 380 were found to have defined toxicity criteria, and 155 of these were found in water. These 155 compounds formed the candidate list that was prioritized using the Cadmus approach.

In the Cadmus approach, assessment of exposure is based on three components: annual production quantity, exposure quantity, and occurrence in water. Annual production quantity (PQ) was chosen as a possible measure of exposure because the more chemical that is produced, the greater the likelihood the chemical will be released into the environment and, ultimately, the greater the likelihood people will be exposed to it. PQ values ranges from 1 to 10, based on the division of all available (log-transformed) production quantity data into ten intervals. For example, a score of 4 indicates that the chemical had an annual production volume of 90,992 to 517,606 pounds. At the time of the development of the Cadmus risk index approach, the annual production data for all evaluated chemicals ranged from approximately 0 to 17.6 billion pounds per year.

The exposure quantity (EQ) score is a function of the quantity of a chemical released to water and the persistence of that chemical in water. The quantity released to ambient water is the sum of the quantity released to surface water as determined from the Toxics Release Inventory (TRI) database, the quantity released to surface water as determined from the Permit Compliance System chemical release database, and the quantity released to ground water as determined from the TRI database. Persistence is defined as a function of the half-life of the chemical in water and its tendency to partition to nonaqueous media as determined by the octanol-water partition coefficient. The persistence factor is assigned a value between 0.001 and 1.0. Chemicals that are most persistent in water receive a higher score than those that are degraded readily or are adsorbed onto sediments and removed from the environment. The assignment of the

persistence factor value, although not clearly explained in the Cadmus report, appears in some cases to be subjective. One important simplification in assessing persistence is that chemicals introduced into the ground water are assumed to persist indefinitely. This assumption ignores the potentially important degradation or removal mechanisms of hydrolysis, adsorption, and biological degradation that can occur in an aquifer.

The occurrence in water (OW) score is a function of the frequency of detection and the maximum concentration found in ambient waters. These two noninteractive parameters are combined in an additive fashion to determine the OW score. This approach implies that a chemical may be deemed important if it is frequently detected even if at low concentrations or if it occurs at high concentrations even when found infrequently.

In the Cadmus approach, human health risk (HR) is defined as an average of the toxicity scores for both carcinogenic and noncarcinogenic risks. The carcinogenicity score is a weighted average of the concentration corresponding to the 10^-5 lifetime cancer risk level; the average based upon the designations given by EPA as to whether the chemical has been found to cause cancer in laboratory animals or humans or has yet to be demonstrated as a carcinogen. The noncarcinogenicity score is also a weighted average and is based on a concentration known as the drinking water equivalent level (or DWEL) and a severity coefficient that is a measure of the type of effect produced by a particular chemical. Assignment of the weights used in the averaging relationships that ultimately determine the HR score was inadequately explained in the report and is another example of the subjective nature of the determination of the various components of the risk index.

The overall risk index (RI) for each chemical is computed using the following equation:

where W₁ through W₄ are the weights assigned to each parameter. Within the brackets is a summation of weighted exposure information as measured by production quantity (PQ), exposure quantity (EQ), and occurrence in water (OW). This sum is then multiplied by the human health risk (HR) and by another weight factor.

An evaluation of the ranked list of chemical names at the back of the Cadmus report and the component scores attached to them illustrates an interesting point

with respect to the treatment of insufficient data. If a critical element of data (e.g., defined toxicity criteria or positively detected in water) was missing for a chemical, the chemical was not included on the candidate list for subsequent risk prioritization. For example, the solvent trichloroethylene was not included on the candidate list, and yet this is the most prevalent ground water contaminant in the category of organic compounds known generally as volatile organic chemicals. It is difficult to imagine how a risk index could successfully identify chemicals that may cause problems in the future if requisite data are not available to calculate the risk index score in the first place.

The AWWA has developed a process that evaluates a chemical's toxicity and occurrence in the environment, along with technical and economic feasibility of removing the chemical from drinking water. This process is described in a report entitled A Screening Process for Identifying Contaminants for Potential Drinking Water Priority Listing and Regulation (RCG et al., 1993).

The AWWA screening process was not applied to any specific list of chemicals. Instead, the AWWA presented the methodology as part of a process that could be applied to a single contaminant or group of contaminants with similar characteristics.

The screening process includes an evaluation of potential exposure to a chemical in the environment. A variety of data sources, including the National Organics Monitoring System, National Pesticide Survey, STORET, the Federal Reporting Data System, and U.S. Geological Survey data bases are examined for the existence of occurrence data on the chemicals of interest. Data quality is considered in the ranking or inclusion of chemicals. For example, the data of the survey and the age of the information is taken into consideration, along with published quality control and quality assurance data. Also considered are "hot spots" that may give the chemical of interest more attention than would be expected if it were simply a target of monitoring on a routine basis.

Data that serve as input to the toxicity screening step come from government data bases, including IRIS, Health Effects Assessment Summary Tables, Agency for Toxic Substances and Disease Registry, Hazardous Substance Data Bank,

Registry of Toxic Effects of Chemical Substances, Chemical Carcinogenesis Research Information System, and Developmental and Reproductive Toxicology Database. Each of the different data bases has different formats and methods of reporting toxicity. The practitioner must choose which of the various reporting formats should be used to categorize the chemicals.

Data quality for carcinogens is based on weight-of-evidence classification. For noncarcinogens, data quality is based on the IRIS level of confidence, which is reported as either high, medium, or low. Based on the data and an evaluation of its quality, the screening approach suggests that a compound be included on the drinking water priority list if it has at least one of the following characteristics:

1. existence—i.e., toxicity data exist in IRIS or other credible sources;
2. quality—i.e., the IRIS conclusions have high or medium confidence results and/or a carcinogen classification of A, B1, or B2; and
3. applicability—i.e., oral ingestion effects data, or high or medium confidence data on absorption after ingestion exist.

These scores are qualitative and represent an approach to assessing health effects on the basis of existence and quality of data rather than magnitude of toxic effect.

As described above, the AWWA screening approach first screens chemicals based on toxicity criteria. Once a chemical has been judged to have significant health effects, the screening process evaluates the potential for exposure. The AWWA screening approach also includes considerations of technical and economic feasibility. Technical feasibility is defined as the ability to control a contaminant using existing treatment technology; economic feasibility is the determination of whether reasonable treatment costs would result in an attempt to control a particular chemical.

The AWWA screening process is quite flexible and does not depend on any uniformly quantitative parameter calculation. Thus, it can include compounds on a drinking water priority list even if all the needed data to calculate a particular factor are not available. Although the toxicity to humans and prevalence of a compound in the environment should be paramount when prioritizing chemicals for regulatory action, technical and economic feasibility cannot be discounted when considering implementation and remediation strategies for those chemicals of high concern. Chemicals scoring high for toxicity and prevalence but low for technical and economic feasibility may be good candidates for new technology research.

The Regulation Development Process (RDP) is a proposed process developed by the AWWA, National Association of Water Companies, Association of Metropolitan Water Agencies, and Association of State Drinking Water Administrators for selection of drinking water contaminants for regulation and for the analysis used to make regulatory decisions (Cook, 1998).

The RDP has not been applied to any specific list of contaminant candidates. RDP recommends that chemicals be considered for analysis only if they have been found in ambient waters. While the RDP does not spell out what criteria should be used to select chemicals for occurrence monitoring, the process does emphasize that EPA should give significant weight to occurrence data. Thus, according to the RDP, the agency should establish a robust unregulated contaminant occurrence monitoring program that can be used to select contaminants for the Drinking Water Contaminant Candidate List (CCL). Contaminants that occur frequently with wide geographic distribution and at levels that cause health concerns, would be included on the CCL.

The RDP assumes that the National Contaminant Occurrence Database, which is currently under development by EPA, will be the primary source of occurrence monitoring survey data on unregulated contaminants. The RDP proposes that exposure data be compiled as a plot of frequency of occurrence versus concentration. When combined with population data, this plot could be translated into a plot of national (or regional) exposure potential, providing an estimate of populations exposed to various concentrations.

According to the proposed scheme, toxicity data are needed to determine if there is a genuine public health threat associated with a contaminant. Health effect studies should answer the question: "Are there significant adverse health end points associated with exposure to a contaminant at concentrations seen or likely to be seen in finished drinking water?" If such data are not available, the contaminant should not be listed on the CCL.

Maximum contaminant level goals (MCLGs) may be useful screening tools, however the RDP suggests that an uncertainty range be provided with the MCLG that "describes the full range of higher MCLGs that are possible without risking disease, albeit with lower margins of safety." For nonthreshold effects (such as

cancer), the RDP indicates that dose-response curves should be plotted. If extrapolation models must be used to determine risks at low doses, all uncertainties should be fully identified and carried through the analysis.

Under the RDP, the information from exposure analysis and health effects data would be combined to produce a frequency distribution of risk. That is, the concentration metric from the exposure analysis would be converted to a risk metric using the dose-response relationship. This conversion would translate the frequency distribution from the exposure analysis into a plot of population exposed versus risk level. Integration of the curve would yield the number of people exposed above the risk level, which could be used in evaluating the severity and magnitude of the hazard. The RDP also recommends using this plot to determine the potential for health risk reduction that would be accomplished by various maximum contaminant levels (MCLs) achievable with current treatment technologies. This estimate would serve to indicate whether setting a national MCL can produce ''a meaningful reduction in the public health risk." Under the RDP, the benefits and costs of regulating a contaminant would have to be carefully considered in the risk management phase. An advantage of this approach, unlike the AWWA screening process, is that it does not eliminate contaminants from consideration based on technological or economic feasibility of treating the contaminant; instead it leaves such decisions to risk managers. A drawback to this approach, however, is that the absence of health effects data on a contaminant may significantly curtail the development of a frequency distribution of risk and would preclude inclusion of the chemical on the CCL even though there are occurrence monitoring data.

The WMPT was developed in response to the Waste Minimization National Plan by EPA's Office of Solid Waste and Office of Pollution Prevention and Toxics. It prioritizes source reduction and recycling activities based on risk (EPA, 1997). The WMPT provides a screening-level assessment of potential chronic risks to human health and the environment through prioritization of chemicals based on their persistence, bioaccumulation potential, toxicity, and quantity in the environment.

Because the purpose of the WMPT is to assist in making decisions related to generation of environmental contaminants, the number of chemicals (4,700) covered by the WMPT is very large. The list consists primarily of chemicals on the

Toxics Substances Control Act Inventory and, in particular, those that are actually in commerce. Data exist for 880 chemicals.

The WMPT assesses exposure based on a chemical's inherent potential to result in significant environmental exposure and thus does not include site-specific information. Exposure is based on three factors: (1) bioaccumulation potential, which is based on the octanol/water partition coefficient, bioaccumulation factor, and bioconcentration factor; (2) persistence, which is based on biodegradation rates and hydrolysis rates; and (3) potentially releasable mass of contaminant, which is based on amounts of the chemical in production waste streams. An overall measure of exposure is computed through a multiplicative relationship of factors of each of these three contributors to exposure.

The WMPT considers both human and ecological toxicity. Human toxicity is assessed using indicators for both cancer and noncancer effects. Indicators for cancer effects include cancer slope factors or potency factors. Indicators of a chemical's potential to cause chronic noncancer effects, such as hepatic toxicity, include EPA reference doses and reference concentrations.

The WMPT assigns a score to each of the three exposure factors (bioaccumulation potential, persistence, and mass) and the toxicity factor. It uses three approaches to generate these scores from quantitative data elements.

• The "binning" or "fence line" scoring approach involves comparing the quantitative value for a given chemical data element against predefined "high" and "low" threshold values, termed ''fence lines." This approach is used for determining bioaccumulation potential and persistence and for some of the toxicity assessments. One advantage of the binning approach is that it accounts for chemical data that often are not very precise, and grouping data into similar "bins" avoids the false sense that such data are highly precise.
• The "continuous-scale" scoring method involves mathematically transforming the actual chemical value for a given data element into a factor score. This approach is used for determining the mass factor.
• The "decision rule" scoring method calculates factor scores based on a single or a combination of multiple data elements, following a specified set of rules. This approach is used for scoring human toxicity, depending on the data available for evaluating cancer effects.

Once factor scores are computed, they are then combined in a multiplicative relationship to obtain an overall chemical score. In recognition that the factor scores vary by orders of magnitude, the algorithm is presented as an additive relationship in which the logarithms of the factor scores are added together to generate the overall chemical score. That is,

where T denotes the logarithmic toxicity factor, M the logarithmic mass factor, P the logarithmic persistence factor, and B the logarithmic bioaccumulation potential factor. It is important to emphasize that the additive relationship results simply from a mathematical transformation of a multiplicative relationship of the factor values.

The Interagency Testing Committee (ITC) was established in 1976 under Section 4(e) of TSCA to screen and recommend chemicals and chemical groups for consideration by the EPA administrator for priority testing and potential rule making. It consists of representatives from 15 member and liaison agencies. The EPA administrator is required to take action on the ITC recommendations within 12 months by requiring the manufacturers of these chemicals to conduct testing or by informing the public why the testing should not be implemented. By congressional mandate, the committee must revise the Priority Testing List at least every six months.

The ITC has used three chemical selection processes to screen and identify chemicals for priority testing consideration. From 1977 to 1980, the ITC's process consisted of examining large lists of chemicals and designating chemical categories that satisfy generic definitions. From 1980 to 1989, the committee used sequential exposure and biological scoring processes followed by in-depth review. Since 1989, the committee has used computerized processes to identify chemical groups that are associated with adverse health or ecological effects or that are likely to involve occupational or environmental exposure. These computerized processes were developed to evaluate several thousand chemicals and to incorporate several feedback loops to ensure that chemicals are reconsidered as new estimates or new data become available (Walker and Brink, 1989; Walker, 1991; Walker, 1995).

Between 1977 and 1983 the ITC conducted scoring exercises focused on high annual production chemicals (e.g., > 10⁶ kg/year), analogues of known car-

cinogens, and chemicals of concern to federal or state agencies and others. In 1986, the ITC organized expert panels to identify substructures indicative of chemicals with potential to cause adverse human health and ecological effects.

Prior to 1986, a panel of experts assigned consensus scores to 11 exposure factors, including annual production, fraction released in the plant, number of workers potentially exposed, fraction released to the environment, number of people exposed in the general population, frequency of general population exposure, intensity of general population exposure, persistence, penetrability, influence on the environment, and bioaccumulation potential. Scores ranged from 0 to 3. Consensus scores for the 11 exposure factors were incorporated into algorithms developed to estimate occupational, general population, and environmental exposures.

Currently, ITC's computerized system assesses the potential for environmental exposure by matching Chemical Abstract Service (CAS) numbers of a large quantity of discrete organic chemicals in a computerized database with CAS-numbered chemicals in publicly available sources of exposure data (e.g., STORET database and the Chemical Evaluation Search and Retrieval System).

Prior to 1986, biological scoring was conducted to provide scores for mutagenicity, carcinogenicity, and developmental toxicity or reproductive effects, because TSCA section 4(e) directs the ITC to give priority attention to chemicals with the potential to cause "gene mutations, cancer and birth defects." Biological scoring was also conducted to provide scores for acute toxicity, subchronic toxicity, bioconcentration, and ecotoxicity, because the ITC considered these to be critical factors for estimating the biological effects that could be caused by chemicals. Each chemical was scored for each biological effect by a number of experts, who reviewed available data or predicted toxicity based on structure-activity relationships. Assigned scores ranged from-3 (strong suspicion of an adverse effect, but no data) to +3 (data supporting an adverse effect at the dose level). In averaging specific effect scores assigned to a chemical, no mixing of positive and negative scores was permitted. Discrepancies were discussed among the experts and resolved.

Under the current ITC approach, substructures likely to be associated with chemicals causing adverse health or ecological effects are used to probe a large universe of chemicals on the TSCA inventory. The process for selecting chemical substructures involves soliciting the opinions of internationally recognized experts with substructure questionnaires developed by ecological and health effects panels. The ITC has conducted reliability assessments of substructures used

to identify potentially hazardous groups by retrieving and analyzing toxicity data on discrete chemicals in the chemical groups and supplementing these data with data extracted from studies submitted under TSCA.

From 1980 to 1986, the ITC used sequential exposure and biological scoring processes to prioritize chemicals for more in-depth review. By 1984, the ITC was aware that the methods used during these years had identified hundreds of easy-to-recognize chemicals in need of testing, and that it was becoming more difficult to identify priority chemicals. ITC then developed a new system, the Substructure-based Computerized Chemical Selection Expert System, which used cost-effective, computerized processes that simultaneously integrate effects and exposure information. Chemicals from sources of environmental monitoring data and substructures with potential to cause adverse health or ecological effects are assigned codes, and these values are used to select chemicals with environmental exposure and adverse effect potential. (Chemicals with no or low effects potential or low or no exposure potential are removed from the list and recycled.) This list of chemicals is then screened for low consumption chemicals. A minimum production-importation ceiling of 22,000 kg/year is used to select chemicals. An algorithm is then applied for scoring chemicals for potential adverse effects, potential exposure, and production volume. Information profiles are developed on those chemicals with high scores and are reviewed at the ITC Chemical Selection Workshop, where chemicals are selected for in-depth review based on consensus decisions. The ITC approach of combining automated sorting of chemicals with expert input is advantageous in that it allows a large number of chemicals to be considered while still allowing for expert judgment.

In 1986, the State of California passed the Safe Drinking Water and Toxic Enforcement Act, better known as Proposition 65 (OEHHA, 1997). This act requires the governor of California to publish lists of chemicals known to cause cancer and reproductive toxicity. The California Environmental Protection Agency's (Cal/EPA) Office of Environmental Health Hazard Assessment (OEHHA) is the lead agency responsible for implementation of Proposition 65. Two committees of the Science Advisory Board, known as the Carcinogen Identification Committee and the Developmental and Reproductive Toxicant Identification Committee, serve as the state's qualified experts for rendering an opinion as to whether a chemical is known by the state to cause cancer or developmental and reproductive toxic effects.

Under Proposition 65, OEHHA has established a tracking database to categorize chemicals, their relevant data, and evaluation status. Chemicals are included in the tracking database as a result of suggestions by state agencies and other sources, or based on literature searches. The sources of literature include scientific journals, the Chemical Carcinogenicity Research Information System, the carcinogenic potency database compiled by Dr. L. Gold and published by Environmental Health Perspectives, the National Cancer Institute's survey of compounds tested for carcinogenic activity, and pesticide registrant data submitted to Cal/EPA. The basis for identifying a chemical as a potential candidate may be, for example, positive cancer or reproductive toxicity bioassays or evidence of very high production or use volume accompanied by evidence of relevant toxicity. To date, more than 580 potential candidate carcinogens and more than 320 potential candidate developmental or reproductive toxicants have been entered.

OEHHA does not use exposure information in the assignment of hazard priorities, but it accounts for this information in the selection of chemicals from the "Candidate List" for committee consideration, as discussed below. The tracking database records chemical use and occurrence information, such as whether the chemical is used in California industries, is a byproduct of industries operating in California, is a pesticide used on food crops grown or imported into California, or is a component of consumer products or drugs sold in California. Information on current restrictions on exposure to the chemical is also noted when readily available. In the absence of information specific to California, evidence of exposure, production, or use in the United States will be assumed to reflect the experience in California. A qualitative evaluation of the level of exposure concern is expressed as high, medium, low, no identified concern, or inadequate data.

The Proposition 65 approach assigns a draft hazard priority based on the potential for developmental or reproductive toxicity or carcinogenicity. A qualitative appraisal of this level of concern is based on an evaluation of available information, including epidemiological and animal toxicity studies and other relevant data. The list of chemicals and their draft prioritization status is then released for public and scientific comment, after which final hazard priorities are set.

Under Proposition 65, chemicals are selected from the tracking database for assignment of a hazard priority status based on toxicological concerns. Chemicals in the tracking database that have not yet been assigned a final priority status are included in "category I." Chemicals with a final priority status of high level of carcinogenic, reproductive, or developmental hazard concern are put on the "Candidate List." Chemicals are then chosen from the Candidate List for the preparation of a hazard identification document. The selection of chemicals from the Candidate List is based on the level of exposure concern. Thus, chemicals get on the Candidate List based solely on toxicological considerations and then are addressed in order of priority based on exposure potential.

Category II includes chemicals in the tracking database that have been assigned a final priority status other than high. This includes chemicals with a hazard priority status of medium, low, no identified concern, or inadequate data. Action is not anticipated for category II chemicals until all high priority chemicals have been identified from the tracking database, assigned to the Candidate List, and brought before the appropriate committees for evaluation. The Proposition 65 approach is thus a largely qualitative approach, involving several iterations of expert judgment.

The Hazard Ranking System (HRS) is the principal mechanism that EPA uses to place hazardous waste sites on the National Priorities List of sites to be cleaned up under CERCLA. It is used to screen releases of uncontrolled hazardous substances for their potential to cause human health or environmental damage. Although it ranks sites, not chemicals, it is relevant to this evaluation because the threat posed by a contaminated site is determined to a large extent by the contaminants.

Contaminants considered in the HRS are determined by the hazardous substances, as defined in CERCLA regulations, found at the site.

For a given site, the HRS evaluates exposure in ground water (gw), surface water (sw), soil (s), and air (a). For each of these pathways, the system computes a "likelihood of release" factor (LR) based on the likelihood that the waste has been or will be released to each of the four pathways for transport through the environment. It computes a "waste characteristics" factor (WC) based on the

toxicity, mobility, persistence, and/or bioaccumulation potential of the hazardous material. Also considered in evaluating the WC factor is the estimated quantity of the contaminant based on an assessment of the mass of waste present at the site.

The HRS also computes a target factor (T) for each pathway, considering four possible types of receptors: human individuals, human populations, natural resources, and sensitive environments. This factor is intended to describe the magnitude of the hazard with respect to the number of targets at risk. The factor is quantified by counting the number of targets involved and assessing the severity of the contamination. Severity of contamination is determined by comparing media-specific concentrations with benchmarks such as MCLGs.

As stated above, toxicity is a determinant used in evaluating the WC factor for a pathway. The toxicity factor for a particular hazardous substance is determined primarily from toxicological responses, represented by slope factors for cancer and reference dose values for noncancer effects. If neither of these is available, the toxicity factor can be determined from acute toxicity parameters, such as the LD₅₀ (the dose of a chemical calculated to cause death in 50 percent of the test population). If no toxicity information is available for any of the hazardous substances found at the site, the toxicity factor is set at a minimum value.

The hazard score for each of the four pathways is computed from a multiplicative relationship of three factors: LR, WC, and T. For example, for the ground water pathway, the hazard score is computed as:

where SF is a scaling factor appropriate for the ground water pathway. After the hazard scores for each of the four pathways are computed, the overall site score is computed:

This relationship is an additive averaging relationship, which implies that it is not necessary for all four exposure pathways to be important for the site to rank high.

On the other hand, the overall hazard score for the site is likely to be high if all four pathways have potential to transport the contaminants.

In addition to the ranking of hazardous waste sites, CERCLA also requires a ranking of the hazardous substances themselves. Section 104(I)(2) of CERCLA, as amended (42 U.S.C. 9604[I][2]) requires that the ATSDR, together with EPA, prepare a prioritized list of hazardous substances found at sites on the National Priority List (NPL). Prioritization must be based on a determination of significance of the threat to human health. To facilitate this task, ATSDR developed the HAZDAT database. This database contains information on frequency of occurrence of substances at NPL sites, potential for human exposure at these sites, and potential health effects. The prioritization scheme for the 1995 list is described in an ATSDR document published in April 1996 (ATSDR, 1996). Each substance on the CERCLA priority list is a candidate for a toxicological profile to be prepared by ATSDR and the subsequent identification of priority data needs.

All contaminants present at NPL sites are considered for the CERCLA priority list of hazardous substances. Currently, the HAZDAT database lists more than 2,800 substances occurring at NPL sites. Only substances found at three or more NPL sites were considered for the priority list, which consists of more than 750 substances. Petroleum-related substances are excluded from the prioritization process because they are regulated by legislation other than CERCLA.

Under this prioritization scheme, the potential for human exposure is based on two factors: relative source contribution (SC) and exposure status of populations. SC is computed as:

where RQ is the reportable quantity, which is an inverse measure of toxicity (discussed below). The theoretical daily dose is the sum of the daily doses of the contaminant from exposure to contaminated air, soil, and water. Each of these is estimated as the product of the concentration of the contaminant in that medium and the average exposure rate, using standard EPA guidelines. The concentration in each medium is representative of the maximum concentration found at a par-

ticular site and is computed as the geometric mean of maximum concentrations at observed NPL sites. The rationale for using the mean of the maximum observed concentrations, as opposed to the mean of an average of observed concentrations, is not explained in the documentation of this process.

The exposure status of populations is a categorical variable indicating whether the population has been exposed to the contaminant, has been exposed to a medium containing the contaminant, may potentially be exposed to the contaminant, or may potentially be exposed to a medium containing the contaminant. A point value is assigned to the exposure status, depending on severity, and this value is added to a logarithmic transformation of SC to give an overall score for the potential for human exposure. This approach for evaluating exposure potential is somewhat unconventional inasmuch as the score depends on a measure of toxicity (RQ).

The CERCLA hazardous substance prioritization scheme considers toxicity by using the RQ approach, which was developed by EPA for guidance regarding environmental releases of hazardous substances. Any person in charge of a vessel or facility from which a hazardous substance has been released in a quantity that equals or exceeds its RQ must immediately notify the appropriate authorities. RQs have been established for listed hazardous substances based on a wide variety of toxicity information, including acute toxicity, chronic toxicity, carcinogenicity, aquatic toxicity, ignitability, and reactivity. The inclusion of ignitability and reactivity is consistent with EPA's definition of hazard characteristics, even though these are generally not considered toxicity characteristics. For a substance for which an RQ value has not been established, ATSDR estimates a value for this substance and refers to it as a toxicity/environmental score (TES). An RQ (or TES) for a particular hazardous substance can be adjusted for potential hydrolysis, photolysis, or biodegradation in the environment. Thus, this parameter has information not only regarding human health effects but also regarding potential for human exposure, and yet in this prioritization scheme uses the parameter solely as a metric of toxicity.

In addition to measures of the potential for human exposure and toxicity, the CERCLA hazardous substance prioritization scheme uses frequency of occurrence (NPL Frequency) as a third criterion for ranking. This parameter is a scaled measure of the number of NPL sites at which a substance has been observed. The overall hazard potential of each candidate substance is computed according to the following algorithm:

where each quantity has been scaled to a value with a maximum of 600 points so that the total score has a possible maximum value of 1,800 points. ATSDR documentation (ATSDR, 1996) provides some rationale for the sum of the three quantities being roughly logarithmic, which would justify the additive algorithm. However, little explanation is provided for the decision to transform the toxicity metric to a suitable weighting for this criterion. Specifically, the toxicity points value is equal to 2/3 raised to the exponent of the cumulative ordinal rank, multiplied by 600, which apparently results in the desired weighting of toxicity relative to the other two components of the algorithm.

The Hazard Analysis of Releases (HAZREL) score was developed by EPA's Office of Science and Technology as a screening-level hazard analysis to indicate sediment contamination potential and to predict where sediment problems have occurred (EPA, 1996). The objectives of the sediment inventory and analysis include generation of a relative ranking of chemicals in industrial categories using 1993 Toxics Release Inventory (TRI) and Permit Compliance System (PCS) chemical release data and prioritization of watersheds for collecting additional information to establish a baseline for future inventories. The HAZREL socre is an index of the magnitude of potential sediment contamination based on specific releases, physical and chemical properties, and potential environmental risk.

EPA selected chemicals for HAZREL ranking from the TRI and 1993 PCS chemical release data. The available list of chemicals included 25,500 individual TRI and PCS records of point source releases of 111 chemicals. About 1,020 watersheds and 31 individual industrial categories were represented by these two sources of data.

HAZREL assesses exposure through quantification of a "fate" score, which is calculated as the product of an air/water partitioning subfactor, a sediment adsorption subfactor, and a biodegradation subfactor.

HAZREL determines toxicity with respect to effects of chemicals on aquatic life. A ''tox" score is calculated by taking the inverse of the sediment chemistry screening value, which is based on a combination of equilibrium partitioning and biological effects related to the protection of aquatic life. The tox score is also based on a theoretical evaluation of bioaccumulation.

The HAZREL score is the product of the sediment hazard score (SHS) and the annual chemical load (ACL) in pounds per year:

The sediment hazard score is a product of fate and tox scores. Total HAZREL scores at the watershed level ranged from 0 to 312. Approximately 1,000 watersheds were evaluated, and 17 belonged in priority group 1.

The HAZREL scores are useful because they are quantitative values that can be calculated and ranked. Estimates of the fate score includes subfactors associated with physical, chemical, and biological fate, although there is little information to actually determine the subfactors necessary for an overall evaluation. However, the HAZREL score relies primarily on a determination of aquatic toxicity, which is not applicable for setting drinking water standards, and certainly the sediment aspects of the HAZREL score are not directly applicable to drinking water.

The Groundwater Technology Section of the Environmental Fate and Groundwater Branch of EPA developed a numerical scale called the Pesticide Leaching Index, or Groundwater Leaching Index, to determine the annual risk or hazard from pesticide use with respect to ground water contamination (Wolf, 1996).

The index has been applied only to pesticides used on apples and potatoes. There is no reason, of course, why it could not be expanded to cover other pesticides, but as yet this has not been done.

The Pesticide Leaching Index uses a pesticide mobility index with environmental fate data and other information to compare the relative mobility of pesticides in a particular soil. Calculations for the pesticide leaching potential shown in the document describing the approach are based on Paxton sandy loam soil. The depth to ground water, a critical aspect of the pesticide leaching potential calculation, was set at 0.2 meters, which is very shallow and probably not representative of the depth of ground water in most areas of the United States.

No toxicity information is incorporated in the calculation of the leaching potential. Therefore, the index is only an estimate of potential exposure.

The Pesticide Leaching Index is a function of an attenuation factor. The index is given a score of 1, 2, or 3, depending on the level of the calculated attenuation factor. The attenuation factor is a function of soil parameters, the Henry's constant, and the pesticide half life. The index uses well developed fate and transport equations based on an understanding of how these chemicals move in soil. In general, the parameters needed for calculating the factors that make up the index are available from pesticide manufacturers who must provide these data to EPA and state agencies before a pesticide is registered and approved for use. Unfortunately, the index is narrowly focused and has only been applied to pesticides.

The common theme throughout all the reviewed schemes is the prioritization of contaminants on the basis of risk to human health and/or the environment, which depends on both exposure and toxicity. The only exception to this is the Pesticide Leaching Index, which does not consider toxicity. Table 2-2 summarizes the chemical schemes with respect to the exposure and toxicity considerations incorporated into the prioritization process. Each scheme is unique in its use of data and ranking criteria, and all rely to some extent on subjective (albeit expert) judgment. This derives from the unique purpose of each prioritization scheme and the lack of a universally suitable risk ranking tool.

The exposure potential for a contaminant is determined by the likelihood of its release to the environment, the quantity released, the persistence in the environment, the proximity of the source to receptors, and the mechanisms governing its transport through the environment to a receptor. For drinking water contami-

nants, the exposure pathway includes transport through the water distribution system. The characterization of exposure is accomplished either in an observational or a predictive fashion. Prioritization schemes that characterize exposure in an observational fashion include the CERCLA prioritization scheme, the AWWA screening process, and the proposed Regulation Development Process. These methods use monitoring data for the concentrations of contaminants in the environment to indicate exposure potential. Ideally, prioritization decisions for research and regulation would be based on environmental occurrence data, and for drinking water contaminants this may include observations of contaminants in the drinking water distribution system. Such data are often not available or are incomplete. In the absence of sufficient occurrence data, exposure may be predicted using information about the quantity of a contaminant that is produced and the frequency or rate of release to the environment, combined with estimates of persistence and mobility in the environment. The schemes that characterize exposure predictively include the WMPT, HRS, HAZREL and Pesticide Leaching Index. The ITC approach, the CADMUS approach, and California Proposition 65 employ a combined approach inasmuch as data for production and release are used along with data for observed concentrations in the environment.

The advantage of the predictive approach to evaluating exposure potential is the dependence on chemical properties that indicate fate and transport tendencies and the avoidance of site-specific environmental information. The predictive approach also relies on production and release data, and such data are easier to inventory and measure than observed concentrations in the environment, which must originate from comprehensive monitoring programs. The disadvantage, of course, is the inherent difficulty in predictive fate and transport modeling and the subsequent large uncertainty in predicted behavior. For example, none of the hazard ranking schemes evaluated in this chapter accounts for stable degradation products that may result from a variety of environmental transformations, which in some cases can be more toxic to humans than the parent compounds. Furthermore, if production rate data are used alone, they may be a poor surrogate indicator of levels of contaminant released to the environment.

The toxicological impacts of an environmental contaminant may be defined in relation to a number of receptors, broadly categorized as either human or ecological. For drinking water contaminants, prioritization according to human health impacts is relevant, whereas ecosystem impacts are not. For the prioritization schemes that account for human health impacts, toxicity is quantified using measures that indicate both cancer and noncancer effects. Typically, these are cancer slope factors and reference doses, both associated with ingestion exposure, that are taken from sources such as EPA's IRIS database. More qualitative indicators include the EPA weight-of-evidence classification scheme for carcinogenicity. The CERCLA hazardous substance prioritization scheme is unique in that it includes information about a chemical's ignitability and reactivity as part of the characterization of the toxicity score.

While all the prioritization schemes in some manner consider both exposure and toxicity, they differ in the way this information is combined. Some schemes, such as the CERCLA hazardous substance prioritization scheme, use an additive approach in which some measure of a contaminant's exposure potential is added to a measure of the contaminant's toxicity. This is equivalent to an "either/or" conceptualization in which a contaminant may rank high if it has either high exposure potential or high toxicity. The majority of schemes use a multiplicative approach, in which a measure of a contaminant's exposure potential is multiplied by a measure of the contaminant's toxicity. This is equivalent to an "and" conceptualization in which a contaminant can rank high only if it has an appreciable exposure potential and an appreciable toxicity.

If a contaminant has a high potential for exposure and is highly toxic, it will rank high under both approaches. The difference between the approaches is most apparent when considering contaminants that rank high only in one category. For example, if a contaminant is known to be very toxic, but has no known potential for exposure, then according to the multiplicative approach its rank is zero or very small. With the additive approach, such a contaminant may rank quite high. A particular problem with the additive approach is the need to consider scaling of the quantities. That is, if a toxicity metric is being added to an exposure metric, then one must decide how to scale the metrics so that they are weighted appropriately. Such decisions are often arbitrary, as in the toxicity scoring in the CERCLA hazardous substance prioritization scheme. Whether an additive or a multiplicative approach is appropriate is entirely dependent on the objectives of the prioritization scheme. It is a matter of judgment whether a contaminant should be considered important if it is only toxic, or only abundant in the environment.

Clearly, one of the greatest difficulties in constructing a prioritization scheme is determining the best way to handle uncertain and missing data. Few of the schemes are designed to address systematically the issue of statistical precision of data and how to propagate this information through the prioritization scheme. Schemes designed to serve as quantitative risk-ranking schemes rely on complete, high quality data for both exposure potential and toxicity. Of the ten schemes examined, those that fall into this category include the WMPT, the HRS, the CERCLA priority listing, the Cadmus approach, HAZREL, the Pesticide Leaching Potential, and Section 4(e) of TSCA. These schemes are extremely useful for processing large quantities of data for well-characterized contaminants, and they do not rely extensively on subjective expert judgment, which may be biased or inadequate. However, in most cases, if a contaminant is missing critical data it either is not ranked or it drops to the lowest priority. This limits the usefulness of a scheme for prioritizing emerging contaminants for which quantitative metrics of exposure and toxicity may not exist. The contaminants that may cause the biggest risk to humans may be those about which we have the least information. The AWWA screening process, the proposed Regulation Development Process, and California Proposition 65 directly consider data quality and

completeness. Necessarily, these schemes rely heavily on expert judgment to make decisions in the face of uncertainty about exposure potential and health effects. These schemes do not necessarily rank contaminants, but may simply categorize them in groups with high and low priority for further action. As is clearly explained in the proposed Regulation Development Process, further action may be a decision to regulate with specification of MCLGs and MCLs, or may be a decision to prioritize research to fill data gaps.

For the present purpose of delineating an appropriate prioritization procedure for selection of contaminants currently on the CCL, none of the quantitative ranking schemes is directly applicable. This derives from the fact that contaminants may be placed on the CCL on the basis of sparse data. To make a decision in this case requires significant involvement of expert judgment. Furthermore, a variety of non-mutually-exclusive actions, including prioritization for research, promulgation of a health advisory, promulgation of a drinking water standard, or removing the contaminant from further consideration, may be recommended. A simple ranking scheme is not likely to sufficiently capture the complexity of this decision-making process.

ATSDR (Agency for Toxic Substances and Disease Registry). 1996. 1995 CERCLA Priority List of Hazardous Substances That Will Be the Subject of Toxicological Profiles & Support Documents. Agency for Toxic Substances and Disease Registry, U.S. Department of Health and Human Services.

Cadmus Group. 1992. Development of a Priority Pollutants List for Drinking Water. Prepared for EPA, Office of Ground Water and Drinking Water. Washington, D.C.

CFR (Code of Federal Regulations). 1997. The Hazard Ranking System (Appendix A to Part 300). 40 CFR Ch. 1, pp.108-210.

Cook, P. 1998. A proposed regulation development process for the drinking water program: Recommendations to EPA from AWWA, NAWC, AMWA, and ASDWA. National Association of Water Companies (unpublished report).

EPA (U.S. Environmental Protection Agency). 1996. The National Sediment Contaminant Point Source Inventory: Analysis of Facility Release Data. First Draft. EPA/823/D/96/001. Washington, D.C.: EPA, Office of Science and Technology.

EPA. 1997. Waste Minimization Prioritization Tool Beta Test Version 1.0: User's Guide and System Documentation. Draft. Report No. EPA 530/R/97/019. Washington, D.C.: EPA, Office of Solid Waste and Emergency Response and Office of Pollution Prevention and Toxics.

OEHHA (Office of Environmental Health Hazard Assessment). 1997. Procedure for Prioritizing Candidate Chemicals for Consideration Under Proposition 65 by the 'State's Qualified Experts'. Sacramento, Calif.: Office of Environmental Health Hazard Assessment, California Environmental Protection Agency.

RCG et al. 1993. A Screening Process for Identifying Contaminants for Potential DWPL Listing and Regulation. Prepared for the American Water Works Association. Denver, Col.

Walker, J. D. 1991. Chemical selection by the Interagency Testing Committee: Use of computerized substructure searching to identify chemical groups for health effects, chemical fate and ecological effects testing. The Science of the Total Environment 109/110:691-700.

Walker, J. D. 1995. Estimation methods used by the TSCA interagency testing committee to prioritize chemicals for testing: exposure and biological effects scoring and structure activity relationships. Toxicology Modeling 1(2):123-141.

Walker, J. D., and R. H. Brink. 1989. New cost-effective, computerized approaches to selecting chemicals for priority testing consideration. Pp. 507-536 in Aquatic Toxicology and Environmental Fate: Eleventh Volume, G. W. Suter II and M. A. Lewis, eds. Philadelphia, Pa.: American Society for Testing Materials.

Wolf, J. 1996. Pesticide Leaching Potential to Aid in Contaminant Selection. Memorandum to E. Washington. EPA, Office of Pesticide Programs. Washington, D.C.