vvEPA United States Environmental Protection Agency Office of tVlarine and Estuarine Protection (WH-556F) Washington, DC 20460 ^ 503/8 ~B9-Oci^-. Office of Water Regulations r lObQ and Standards (WH-552) ■i'^'^^1 Washington, DC 20460 Water September 1989 EPA-503/ 8-89-002 Assessing Human Health Risks from Chemically Contaminated Fish and Shellfish: A Guidance Manual Printed on Recycled Paper ACKNOWLEDGMENTS This document was prepared under EPA Contract No. 68-03-3319. Dr. Kim Devonald of the Office of Marine and Estuarine Protection was the Project Monitor for EPA, and also drafted the Background section of the Introduction. John Maxted contributed the sections on Relationship of Fisheries Risk Assessment to Water Quality Standards and Relationship of Fisheries Risk Assessment to Monitoring under the Clean Water Act. The EPA Office of Pesticide Programs prepared Appendix F. Portions of this report were based on the Guidance Manual for Health Risk Assessment of Chemically Contaminated Seafood, a document prepared for the Puget Sound Estuary Program in EPA Region X. Information on quality assurance/quality control (QA/QC) plans and documentation was taken from the Puget Sound Protocols, also developed for the Puget Sound Estuary Program. Dr. John Armstrong of EPA Region X Office of Puget Sound was instrumental in developing the original concept of a guidance manual for assessing human health risk from contaminated fish and shellfish. Comments on the EPA Region X risk document were provided by participants at a 1987 workshop on health risk assessment related to consumption of fish and shellfish. The workshop was sponsored by EPA Region I and the New England Interstate Water Pollution Con- trol Commission. Valuable comments on earlier drafts of this report were received from the following reviewers: John R. Bagby Donald Barnes Bruce R. Barrett Robert Cantilli Richard L. Caspe Dave DeVault Kim Devonald Carol Finch Jeffeiy Foran Kevin G. Garrahan Rebecca Hamner James E. Harrison John L. Hesse Robert A. Kreiger A.R. Malcolm Alvin R. Morris Edward V. Ohanian Kenneth Orloff Gerald Pollock Vacys J. Saulys Malcolm Shute Paul E. Stacey Brian Toal Paul White Missouri Department of Health EPA Office of Toxic Substances EPA Water Management Division, Region IV EPA Office of Drinking Water EPA Water Management Division, Region II EPA Great Lakes National Program Office, Region V EPA Office of Marine and Eistuarine Protection EPA Great Lakes National Program Office, Region V National Wildlife Federation EPA Exposure Assessment Group EPA Deputy Assistant Administrator for Water EPA Marine and Estuarine Branch, Region IV Michigan Department of Public Health Minnesota Department of Health EPA Physiological Effects Branch EPA Water Management Division, Region III EPA Health Effects Branch, Office of Drinking Water EPA Water Management Division, Region IV California Department of Health Services EPA Great Lakes National Program Office, Region V Connecticut Department of Health Services Connecticut Department of Environmental Protection Connecticut Department of Health Services EPA Exposure Assessment Group DOCUMENT LIBRARY Woods Hole Oceanographic Institution CONTENTS Page CHAPTER 1: Introduction 1 Objectives 1 Organization 2 Background 2 1986 EPA Risk Assessment Guidelines 3 Regulatory Roles and Coordination of Federal, State, and Local Agencies 3 Applicability of this Guidance Manual 4 Relationship of this Manual to Other EPA Documents 5 Relationship of Fisheries Risk Assessment to Water Quality Standards 5 Relationship of Fisheries Risk Assessment to Monitoring Under the Clean Water Act 6 Relationship of EPA Risk Assessment Methods to FDA Risk Assessment Methods 7 CHAPTER 2: Overview of Risk Assessment and Risk Management 11 Major Steps in Risk Assessment 11 Need for Risk Assessment Appproach 12 Uses of Risk Assessment 13 CHAPTERS: Hazard IdentiHcation 15 Contaminants of Concern 15 Toxicity Profiles 19 Sources of Information 22 CHAPTER 4: Dose-Response Assessment 23 Exposure and Dose 23 Dose-Response Relationships 23 Carcinogenic Potency Factors 25 Reference Doses 26 Sources of Information 27 Carcinogenic Potency Factors 27 Reference Doses 27 CHAPTERS: Exposure Assessment 29 Measurement of Contaminant Concentrations in Tissues 29 Study Objectives and General Sampling Desifft 32 Selection of Target Species and Size Classes 36 Sampling Station Locations 41 Time of Sampling 43 Kinds of Samples 44 Sample Replication 47 Selection of Analytical Detection Limits and Protocols 47 QA/QC Program 48 Documentation and QA Review of Chemical Data 50 Statistical Treatment of Data 51 Analysis of Sources, Transport, and Fate of Contaminants 52 Analysis of Exposed Populations 53 Comprehensive Catch/Consumption Analysis 54 Assumed Consumption Rate 57 Exposure Dose Determination 59 Single-Species Diets 59 Mixed-Species Diets 60 Sources of Information 61 CHAPTER 6: Risk Characterization 63 Carcinogenic Risk 63 Noncarcinogenic Effects 64 Chemical Mixtures 66 CHAPTER 7: Presentation and Interpretation of Results 67 Presentation Format 67 Summary Tables 67 Summary Graphics 69 Risk Comparisons 70 Summary of Assumptions 70 Uncertainty Analysis 72 Sources of Uncertainly 72 Approaches to Uncertainty Analysis 74 Supplementary Information 75 CHAPTERS: References 77 APPENDICES A. EPA/FDA Summary Policy Statement on Chemical Residues in Fish and Shellfish B. Integrated Risk Information System (IRIS) C. Sources of Information for Toxicity Profiles D. Evaluation of the Effects of Composite Sampling on Statis- tical Power of a Sampling Design E . Evaluation of the Effects of Sample Replicaiton on Statistical Power of a Sampling Design F. Estimation of Fish/Shellfish Consumption from a National Database G. EPA Office of Research and Development, Environmental Research Laboratories H. Compilations of Legal Limits for Chemical Contaminants in Fish and Fishery Products LIST OF FIGURES LIST OF TABLES FIGURES Number Title Following Page 1 Overview of risk assessment and risk 11 management Hypothetical example of dose-response 23 curves for a carcinogen and a noncar- cinogen Interaction between environmental fac- 35 tors and exposed population factors Summary of recommended marine and es- 40 tuarine indicator species 5 General sampling station layouts for prob- 41 ability sampling in two dimensions 6 Conceptual structure of quantitative 63 health risk assessment model Example graphic format for display of 69 quantitative risk assessment results for hypothetical study area and reference area Plausible-upper-Hmit estimate of lifetime 70 excess cancer risk vs. concentration of a chemical contaminant in fish or shellfish at selected ingestion rates TABLES Number Title Page 1 Organic priority pollutants and 301(h) 16 pesticides ranked according to octanol- water partition coefficients (Kow) 2 Inorganic priority pollutants ranked 19 according to bioconcentration factor 3 Toxicity profile for mercury and PCBs 20 4 Criteria for selecting target species 37 5 Approximate range of cost per sample for 49 analyses of EPA priority pollutants m tissues as a function of detection limits and precision 6 Example tabular format for display of quan- 68 titative risk assessment for consumption of fish and shellfish 7 Summary of assumptions and numerical 71 estimates used in risk assessment approach Introduction Contamination of aquatic resources by toxic chemicals is a well recog- nized problem in many parts of the U.S. High concentrations of potentially toxic chemicals have been found in sediments and in aquatic organisms from Puget Sound, the Southern California Bight, northeast Atlantic coastal waters, the Hudson River, the Great Lakes, and elsewhere (Malins et al. 1984; Brown et al. 1985b; De Vault et al. 1986; Capuzzo et al. 1987). Heavy consumption of contaminated fisheries products by humans may pose a substantial health risk. This concern has prompted recent studies of catch and consumption patterns for recreational fisheries and associated health risks (e.g.. Puffer et al. 1982; Humphrey 1983, 1987, 1988; Sonzogni and Swain 1984; Swain 1988). To protect the health of consumers offish and shellfish, information is needed on relative health risks associated with various edible aquatic species, geographic locations, and consumption rates. In the past, diverse models have been used to estimate human health risks from exposure to toxic substances in food [e.g., Cordle et al. 1978; U.S. Office of Technology Assessment 1979; U.S. Environmental Protec- tion Agency (EPA) 1980b; Food Safety Council 1980, 1982; Connor 1984a; Tollefson and Cordle 1986]. In the present report, a stand- ardized procedure is recommended for assessing human health risks from consumption of chemically contaminated fish and shellfish. The purpose of this manual is to provide guidance for health risk assessment related to chemically contaminated fisheries, based on EPA approaches (e.g., U.S. EPA 1980b, 1986a-e, 1987a). The objec- tives of the guidance manual are to: • Describe the steps of a health risk assessment procedure for consumption of contaminated fish and shellfish • Define the conceptual basis for standard toxicological vari- ables [e.g., Carcinogenic Potency Factors or Reference Doses (RfD) for chemicals] and criteria [e.g., U.S. Food and Drug Objectives Organization Background Administration (FDA) action levels] related to risk assess- ment, and information sources for updating these Vcilues • Provide guidance on presentation of risk assessment results • Summarize assumptions and uncertainties of the recom- mended procedure for risk assessment. The guidance provided in this manual is directed primarily at risk assessment related to recreational fisheries. Although assessment of human health risks from commercial fisheries products is not ad- dressed specifically in the examples provided herein, the concepts discussed throughout the manual are relevant to risk analysis for commercial fisheries. This manual provides guidance only, and does not constitute a regulatory requirement of any kind. The technical content is consistent with approved EPA procedures for risk assessment, as published in the Federal Register (U.S. EPA 1986a-e). The guidance manual is intended to describe what EPA believes to be the most scientifically defensible methods for assessing environmental health risks. These are the methods EPA will use in conducting health risk assessments re- quired in its statutorily mandated programs. The relationship between these procedures and risk assessment approaches used by FDA is described briefly in the background section below. Background information on available health risk assessment guidance and use of this manual is provided in the remainder of this introduction. An overview of risk assessment is provided in the following section, including a discussion of the distinction between risk assessment and risk management, and a review of their possible uses. The major steps of the risk assessment process recommended herein are described in subsequent sections. Guidance is provided on mathematical models used to estimate chemical exposure and risk. Sources of information on toxic chemicals and model variables are noted. Finally, suggestions for presentation of risk assessment results are provided and uncertain- ties are summarized. Risk analysis encompasses both risk assessment and risk management. Risk assessment is a scientifically based procedure to estimate the probability of adverse health effects from a specific exposure to a toxic agent. Risk assessment differs from risk management, although both are elements of regulatory decision-making (National Research Coun- cil 1983). Risk assessment provides the scientific basis for public policy and action. In risk management, risks are interpreted in light of legis- lative, socioeconomic, technical, and political factors, and appropriate controls are determined. Risk management often involves evaluating risks relative to potential benefits associated with an activity and defining an acceptable risk level (i.e., the maximum risk considered tolerable). For example, a ri.sk manager might weigh the risks as- sociated with chemical contamination of fish and shellfish against the health benefits (e.g., decreased risk of heart disease) associated with consumption of fish and shellfish in place of red meat. 1986 EPA Risk Assessment Guidelines In September 1986, EPA published final guidelines for assessing health risks related to environmental pollutants. The guidelines are in five parts (U.S. EPA 1986a-e): • Carcinogen Risk Assessment • Exposure Assessment • Mutagenicity Risk Assessment • Health Assessment of Suspect Developmental Toxicants • Health Risk Assessment of Chemical Mixtures. These guidelines pertain to health risk assessment for all environmen- tal exposures [e.g., air exposure; ingestion of water or environmentally contaminated foods; and other direct human contact with con- taminated soils, water, sediments, or other materials (Federal Register 51 No. 185, p. 34049)]. The guidelines were developed through a 2-year process that included contributions and review by the larger scientific community; full Agency consideration of public comments in response to proposed guidelines on November 23, 1984; and review and approval by the EPA Science Advisory Board (Federal Register 51 No. 185, p. 33992). While U.S. EPA's risk assessment guidelines (1986a-e) apply to all exposure routes, they do not contain detailed information on applica- tion of the basic principles for each exposure route. This guidance manual provides such step-by-step assistance for assessing health risks from exposure through consumption of chemically contaminated aquatic organisms. The guidance is applicable to freshwater, brackish water, and saltwater fish and shellfish. The risk assessment methods recommended in this manual are consistent with the principles set forth in U.S. EPA (1986a-e). As described in a recent policy statement by EPA's Risk Assessment Council and FDA (see Appendix A), FDA, EPA, and the states have somewhat differing roles in assessing and managing risks from fish consumption. FDA has the lead responsibility for risk management of foods in interstate commerce or other products of national importance including fish and shellfish. For some chemicals in foods (specifically pesticides), EPA assists FDA in performing the technical evaluations that support risk management decisions. The federal government is not directly responsible for managing risks to individuals who consume unusually large amounts of foods not in interstate commerce or foods harvested from locally contaminated areas (e.g., some recreational fisheries). Environmental agencies and health departments at the state and local levels have responsibility for protecting consumers of local fisheries products. These agencies are responsible for issuing public health advisories and regulations related to local fisheries. Regulatory Roles and Coordination of Federal, State, and Local Agencies Applicability of this Guidance Manual Section 408 of the Federal Food, Drug and Cosmetic Act authorizes EPA to establish tolerances (maximum permissible concentrations) or action levels for pesticides in raw agricultural commodities, including fish and shellfish. FDA is responsible for setting action levels and tolerances for concentrations of other chemicals in fish, shellfish, or other foods. FDA also has responsibiUty for enforcing the guidelines developed by both EPA and FDA, which may involve removal of adulterated foods (i.e., foods contaminated in excess of an action level or tolerance) from interstate commerce. An action level is the mini- mum concentration of chemical in food that may be cause for FDA to take enforcement action. An action level is promulgated when a tolerance or exemption authorizing the presence of a substance in food has not been established or has been revoked. Action levels are estab- lished and revised according to criteria in the Code of Federal Regula- tions (21 CFR 109 and 509). An action level is revoked when a formal tolerance for the same substance is established. In developing action levels and tolerances, FDA and EPA take into account both the magnitude of the health risks to consumers and the economic impacts of banning food from a particular source. FDA and EPA set hmits on chemical contaminants in fisheries products to achieve an optimal balance of health protection and minimization of economic impacts on food-producing and harvesting industries (e.g., commercial fisheries and fish marketers). All action levels and tolerances to date have been developed to provide national protection rather than on a regional or local basis. These national standards protect the average consumer of a food product, assuming the consumer eats foods from a typical "national market basket" (U.S. FDA 1984). Action levels and tolerances are not in- tended to protect certain local subpopulations, such as individuals whose consumption of fish and shellfish from a given water body may exceed the national average (Appendix A). EPA and FDA recognized the need to coordinate their activities and guidance in assessing health risks from contaminated fish and shellfish. The Standing Committee on Fish Contamination has been formed to resolve potential differences in risk assessment calculations for specific chemicals or specific exposure situations (Appendix A). The EPA/FDA policy statement in Appendix A provides further discussion of the evolving coordination between EPA, FDA, other federal agen- cies, and the states. The EPA/FDA policy statement also describes procedures whereby states can obtain further information or assistance pertaining to risk management in specific local situations. EPA's nonregulatory technical guidance, including this manual and the 1986 final guidelines for risk assessment (U.S. EPA 1986a-e), is avail- able to state and local governments responsible for fisheries manage- ment, environmental protection, and public health. This manual is intended for use as a handbook by those state and local agencies that are responsible for assessing potential risks from local fish or shellfish consumption. For example, it maybe useful in assessing risks to highly exposed regional populations (e.g., certain fishermen or families who may eat unusually large amounts of fish). This manual does not provide guidance on policy issues that are beyond the scope of the technical risk assessment process (e.g., selection of acceptable risk levels, and methods for performing local cost-benefit anailyses). For specific technical assistance in applying the risk assessment methods described in this manual, users may contact EPA national offices (see the last page of Appendix A) for updated information on regional EPA facilities that can provide on-site assistance in applying risk assessment techniques. This manual is not intended as an exhaustive guide to all aspects of sampling, statistical design, laboratory analysis, exposure assessment, and lexicological risk analysis. Citations are provided to references that provide details on these topics. In addition, several other EPA documents that provide relevant information are listed below: • U.S. EPA (1987a) Integrated Risk Information System (IRIS) Manual - A regularly updated electronic database on the toxicity and carcinogenicity of individual chemicals (see Ap- pendix B herein) • General guidelines on exposure and risk assessment (U.S. EPA 1986a-e). • Guidance documents on risk assessment approaches for specific chemicals [e.g., dioxins and dibenzofurans (Bellin and Barnes 1986)]. • Superfund Risk Assessment Information Directory (U.S. EPA 1986g). • Risk Assessment, Management, Communication: A Guide to Selected Sources (U.S. EPA 1987b) - A general bibliography which is updated periodically. It should also be noted that the National Oceanic and Atmospheric Administration (NOAA), FDA, and EPA have recently completed a joint study of PCB contamination in Atlantic coast bluefish and poten- tial human health effects (NOAA, FDA, and EPA 1986, 1987). The design of that study, the statistical analysis of the data, and the estimates of dietary intake of PCBs by bluefish anglers and their families provide examples of some of the concepts illustrated in this guidance manual. Environmental quality guidelines may be developed from risk assess- ment models to complement available water quality standards. For example, this manual contains recommended procedures for develop- ing guidelines on concentrations of contaminants in edible tissues of fish and shellfish based on risk assessment. Comparisons of data on tissue concentrations of contaminants with such guidelines may be used by state agencies in regulating the harvest, transportation, and sale offish and shellfish used for human consumption, and in develop- ing health risk advisories. In contrast, state water quality standards are designed to regulate discharges of contaminants to surface waters. Relationship of this Manual to Other EPA Documents Relationship of Fisheries Risk Assessment to Water Quality Standards Relationship of Fisheries Risk Assessment to Monitoring Under the Clean Water Act State water quality standards include two primary elements: desig- nated uses and criteria. Recreationcil fishing and shellfishing are ex- amples of designated uses that may be applied to a water body. Criteria are concentration levels of contaminants in surface water that provide protection from the effects of toxic chemicals, with an ample margin of safety. There are two basic kinds of criteria: those that protect aquatic life and those that protect human health. The criteria incorporated into state water quality standards are en- forceable requirements used by the states to regulate dischargers. In support of the state programs, and to meet the requirements of Section 304(a) of the Clean Water Act, EPA periodically issues national water quality criteria recommendations for use by the states in setting their enforceable standards. In developing national criteria recommenda- tions to protect public health, EPA considers human exposure to chemical contaminants in fish and shellfish as well as drinking water. The Criteria and Standards Division of EPA's Office of Water Regula- tions and Standards is responsible for developing national criteria recommendations under Section 304(a) of the Clean Water Act. The current criteria are summarized in Quality Criteria for Water - 1986 (U.S. EPA 1986h). The technical procedures for deriving human health criteria for water are described in Water Quality Criteria Docu- ments, Availability (U.S. EPA 1980b). The development of water quality criteria to limit human exposure to contaminants in fish and shellfish requires the translation of the con- taminant level not to be exceeded in the animal tissues to a level in the water in which the animal resides. This is accomplished through the use of the bioconcentration factors (BCF). A BCF is a measure of the potential of a chemical to accumulate in biological tissues. A BCF value is defined as the ratio of the concentration of a chemical in tissues of a given aquatic species to the concentration in water. Each chemical BCF may be estimated either directly from the results of bioassay testing or from an octanol-water partitioning coefficient for the chemi- cal, if test data are not available. The calculation of a water quality criterion to protect human health from exposure to contaminants in fish and shellfish is accomplished through the use of the BCF and toxicological and epidemiological data (e.g., data on the amount, or dose, of the contaminant that results in a defined human health risk). The coefficients used in this manual to define the critical dose or the toxic potency for each chemical (see Dose-Response Assessment) are the same as those used to develop water quality criteria. IRIS (U.S. EPA 1987a) is the central location for human health-related data and information used by all EPA programs. States routinely conduct chemical analyses of fish and shellfish tissue as part of their environmental monitoring programs. The results of fish contamination monitoring are documented in state reports and in the National Water Quality Inventory Report to Congress (as required by Section 305(b) of the Clean Water Act). The information presented in this guidance manual can be used to support these activities through the identification of guidelines on levels of contaminants in tissues that correspond to a defined risk to human health (e.g., tolerable risk levels). In addition to these ongoing monitoring activities, the 1987 amend- ments to the Clean Water Act, in particular the new Section 304(1), require states to develop Usts of impaired waters, identify point source discharges of toxic substances and the amounts of pollutants present, and develop individual control strategies (permits) for each point source discharger. The information in this guidance manual may be useful in evaluating data on concentrations of chemical contaminants in fish and shellfish tissue and associated human health risks to identify waters impaired by toxic contamination. Because of differences in legislative and regulatory responsibilities among EPA, FDA, and state and local governments, these entities have developed differing procedures for risk assessment and risk manage- ment. As an EPA guidance manual, this document presumes the use of standard EPA risk assessment procedures. However, certain pro- cedures recommended in this manual can be modified to make the risk assessment compatible with alternative approaches used by FDA and some states. This section explains how conversion factors can be used to make risk assessment procedures recommended herein compatible with certain assumptions used in FDA risk assessments. A major difference between EPA and FDA risk assessment ap- proaches concerns the methods for extrapolating the toxic potency of chemicals in small experimental animals (e.g., rats and mice) to es- timate potential effects in humans. U.S. EPA (1986a) pointed out several species-specific factors that may influence the response to a carcinogen, including life span, body size, genetic variability, concur- rent diseases, and the rates and products of metabolism and excretion. To account for at least some of the differences between experimental animals and humans, the estimate of exposure in laboratory animals is multiplied by a scaling factor to obtain an estimate of equivalent dosage in humans. EPA uses the ratio of animal-to-human surface area, whereas FDA uses the corresponding ratio of body weights as a scaling factor. Thus, EPA uses mg of carcinogen per m body surface area per day as a standardized scale for expressing dosages, whereas FDA uses mg carcinogen per kg body weight per day. This difference in inter- species extrapolation factors results in approximately a five- to ten-fold difference in estimates of carcinogenic potency (and risk) derived by the two agencies. In recognition of the difficulties that differences in interspecies extrapolation procedures between EPA and FDA may pose for state agencies and others who rely on federal guidance on risk assessment, EPA's Risk Assessment Council and FDA reviewed the pros and cons of their respective methods for dosage scaling. They concluded that Relationship of EPA Risk Assessment Methods to FDA Risk Assessment Methods the most appropriate method for interspecies dosage extrapolation may vary depending on exposure conditions and chemicals involved. For example, one procedure may be more realistic for lipophilic chemicals, whereas the other would be more appropriate for hydrophilic chemicals. Differences in target organs (i.e., primary site of toxicity) also affect the preferred extrapolation procedure. Because the EPA extrapolation procedure results in a higher estimate of risk than the FDA procedure (by approximately an order of mag- nitude), the former is considered more protective. For most EPA assessments, the surface-area based extrapolation is appropriate. The technical basis for EPA's approach relies primarily on a demonstrated relationship between pharmacological effects (e.g., balance of rates of metabolism and excretion) and body surface area (Pinkel 1958; Freireich et al. 1966; Dedrick 1973). If state or local policymakers decide that the body-weight based extrapolation is more appropriate for local risk management needs, then procedures recommended in this manual can be modified by converting EPA's dose-response data using a ratio of human body weight to surface area. This would allow the risk assessor to use carcinogenic potency factors in EPA's com- puterized database, IRIS (U.S. EPA 1987a). IRIS is a database main- tained by EPA to provide regularly updated toxicological data for use in risk assessment. The use of IRIS would greatly increase the ability of a state to perform risk assessments for chemicals of local concern while increasing consistency among jurisdictions sharing responsibility for common waters. Although the conversion of EPA estimates of toxic potency to es- timates based on equivalent dosage scales related to body weight is not technically complex, the modified procedure should preferably be carried out only by experienced toxicologists. The conversion factor will vary depending on whether the dose-response data were derived from rats or from mice. Thus the original data set must be reviewed to determine an appropriate conversion factor. In general, an EPA es- timate of carcinogenic potency would be multiplied by a factor equal to the ratio of surface area per unit body weight (m /kg) of the laboratory animal to that of humans. For example, if the EPA car- cinogenic potency factor is C and the surface area per unit body weight is X for the laboratory animal and Y for humans, the corresponding potency factor based on dosage scaled to body weight is C multiplied by X divided by Y. Because specific data on surface area are often unavailable, body weight to the two-thirds power is typically used as an estimate of surface area. Note that some EPA carcinogenic potency factors are derived from epidemiological studies and therefore do not require conversion. Other steps in the process to estimate carcinogenic potencies may vary somewhat among regulatory agencies. For example, different agencies may choose different data sets to derive a carcinogenic potency factor for the same chemical. The mathematical expression used to model the dose-response relationship may also differ among agencies. Hogan and Hoel (1982) and Cothern et al. (1986) discuss various models for extrapolating data from high doses used in laboratory experiments to the low doses of concern in carcinogenic risk assessment. At low doses corresponding to risks of 10" to 10 or less, different models may produce results that vary by as much as several orders of magnitude. Nevertheless, the linearized multistage procedure used by EPA (U.S. EPA 1986a; also see below, Dose-Response Assessment) yields results that correspond approximately (within a factor of two) to those produced by the linear model used by FDA. The interagency Subcom- mittee on Fish Residue Issues of the EPA Risk Assessment Council, which included representatives from FDA, concluded that the dif- ferences in procedures for modeling dose-response relationships be- tween EPA and FDA were small relative to the uncertainties in other steps of a risk assessment. Therefore, the EPA/FDA policy statement (Appendix A herein) does not discuss procedures to reconcile these differences. A final distinction between EPA's risk assessment procedures and other potential approaches is that EPA does not yet provide a stand- ardized approach for assessing carcinogenic effects on children and fetuses. Information on perinatal carcinogenicity is presently being developed by EPA and others. Overview of Risk Assessment and Risk Management The objective of risk assessment is to estimate the probabihty of adverse health effects from exposure to a toxic agent. The elements of the risk assessment process and their relationship to risk management are shown in Figure 1. U.S. Office of Technology Assessment (1987) provides a review of general policies and technical approaches of federal agencies in assessing risks to human health associated with exposure to chemicals . Background information on food safety evalua- tion by Federal and state agencies is provided by U.S. Office of Technology Assessment (1979) and Food Safety Council (1980, 1982). Examples of approaches used by FDA to assess human health risks from toxic chemical exposures are described in Cordle et al. ( 1978) and Flamm and Winbush (1984). The following sections provide an overview of the steps in risk assess- ment, the need for a risk assessment approach to evaluate human health risks from chemically contaminated fisheries, and potential appHcations of the results of fisheries risk assessment. The general format for risk assessment and all definitions of terms used in this report are consistent with those provided by National Research Coun- cil (1983) and U.S. EPA (1986a-e, 1987a). A complete risk assessment includes the following steps: • Hazard identirication: Quahtative evaluation of the potential for a substance to cause adverse health effects (e.g., birth defects, cancer) in animals or in humans • Dose-response assessment: Quantitative estimation of the relationship between the dose of a substance and the prob- ability of an adverse health effect Major Steps in Risk Assessment 11 u N C E R T A I N T Y A N A L Y S I S T HAZARD ASSESSMENT I DOSE-RESPONSE ASSESSMENT I EXPOSURE ASSESSMENT I RISK CHARACTERIZATION MONITORING l_ ANALYSIS OF CONTROL OPTIONS I MANAGEMENT DECISION I ECONOMICS POLITICS LAW SOCIAL IMPLEMENTATION OF CONTROLS L I A S S E S S M E N T M A N A G E M E N T Figure 1 . Overview of risk assessment and risk management Need for Risk Assessment Approach • Exposure assessment: Characterization of the populations ex- posed to the toxic chemicals of concern; the environmental transport and fate pathways; and the magnitude, frequency, and duration of exposure • Risk characterization: Integration of qualitative and quantita- tive information from the first three steps, leading to an es- timate of risk for the health effect of concern. Because uncertainties are pervasive in risk assessment, uncertainty analysis is a key element of each stage of the assessment process. Assumptions and uncertainties are summarized in the risk charac- terization step. The risk characterization includes a balanced discus- sion of the strengths and weaknesses of the data presented. Direct measurement of human health risks is possible in certain limited circumstances. Such circumstances generally involve a single high exposure or repeated moderate exposures of humans to a specific chemical with a clear adverse effect. For example, direct measurement of the incidence of chloracne (a skin disorder) might be possible in a population of workers exposed to a PCB spill. In contrast, it is virtually impossible to directly measure the incidence of cancer associated with consumption of chemically contaminated fish or shellfish. The long latency period for cancer, the potential for contamination of fisheries by multiple chemicals, and confounding exposures through other routes would complicate the interpretation of such data. Mathematical models are therefore used by EPA, FDA, the Agency for Toxic Sub- stances and Disease Registry, states, and other regulatory agencies to estimate human health risks from exposure information. Risk assess- ment procedures discussed in this manual focus on estimating potential health risks from long-term exposure to relatively low levels of contamination. This prospective approach is also useful for developing regulations to limit exposure to toxic chemicals and reduce associated risks. Scientific knowledge of the effects of toxic chemicals on humans is still rudimentary. Much of the present information is extrapolated from results of laboratory tests on animals (e.g., rats and mice). For example, animal test data may be used to estimate levels of chemical exposure that are unlikely to cause toxic effects in human populations. Toxicologists are faced with many uncertainties when estimating the potential for human health risks associated with intake of toxic chemi- cals. Despite these uncertainties, regulatory decisions must be made. Many assumptions and subjective judgments may enter into an evalua- tion of human health risk. The risk assessment approach provides a framework for consistent, systematic estimation of health risks, with clear statements of assumptions and uncertainties. The risk assessment framework offers an alternative to some common approaches to evaluation of data on chemical residues in fish and shellfish. As noted by Kneip (1983) and Peddicord (1984), many investigators have evaluated chemical residue data in light of human health concerns simply by comparing tissue concentrations of selected 12 chemicals to action levels or tolerances established by U.S. FDA (1982, 1984). This approach is limited for the following reasons: • FDA action levels or tolerances are available for only a few chemicals (mercury, 12 organic pesticides or related degrada- tion products, and PCBs). • FDA has not estabhshed regulatory limits for some of the most potent suspected human carcinogens (e.g., 2,3,7,8- tetrachlorodibenzo-p-dioxin) or for some of the common con- taminants in surface waters (e.g., polynuclear aromatic hydrocarbons and most heavy metals). • Action levels and tolerances were intended to be used only for regulation of interstate commerce of food products. • When setting regulatory limits, FDA and EPA consider economic impacts of food regulation as well as the potential human health risks on a national basis (U.S. FDA 1984). When using action levels or tolerances to interpret bioaccumulation data, investigators implicitly adopt economic policies of the federal agencies responsible for setting the limits. Thus, risk management issues at a national level are not clearly separated from site-specific risk assessments. • Action levels and tolerances were developed from a national perspective. They were not intended to protect localized sub- populations of recreational anglers that may consume con- taminated fish or shellfish at a rate substantially above the national per capita average. Use of regulatory limits on toxic chemicals in food products established by other countries (Nauen 1983) would suffer from many of the limita- tions listed above for FDA values. Moreover, a concise review of the basis for each of these limits is not available. Uses of Risk Assessment Risk assessment may be applied to data on chemical residues in fish and shellfish for the following purposes: • Identify and rank toxic chemical problems in specific locations • Develop environmental criteria or guidelines at the national, state, regional, or local level • Develop public information and advisories. The first two applications fall within the general category of regulatory decision-making. In this context, one goal of EPA is to define, identify, and set priorities for reducing unacceptable risks. Risk assessment and management provide a framework for balanced analysis of environ- mental problems (e.g., Tetra Tech 1986a) and consistent policies for reducing health risks (e.g., through reduction of toxic pollutant dis- charges and cleanup of polluted areas). Risk assessment can be used to identify and rank environmental problems in several ways. First, contaminated sites can be ranked according to the relative risks associated with consuming fish and 13 shellfish harvested nearby (e.g., Versar 1985). Site rankings may be used to establish priorities for investigation of contaminant sources and for cleanup. Maps of chemical residue data or risk estimates provide a geographic overview of the condition of resources linked to human exposure. Second, priority chemicals can be identified according to associated health risks or indices of relative hazard (e.g., Ames et al. 1987) . Finally, various fishery species and size (or weight) classes within species can be ranked according to relative risks. Risk assessment is an important analytical tool for developing environ- mental criteria and guidelines. For example, water quality criteria derived by U.S. EPA (1980b, 1986h) are based in part on human health risk assessment. FDA uses quantitative risk assessment to estimate potential human health risks, which are considered together with economic factors in developing action levels for chemical con- taminants in fishery products (U.S. FDA 1984). Risk assessment models can be used to develop guidelines on maximum advisable contaminant concentrations in recreationally harvested species. Such guidelines can contribute to development of target cleanup criteria established to develop remedial actions for contaminated sites. The results of risk assessments may be used to inform the public about the relative health risks of various fishery species and geographic locations. Providing the recreational public with such information allows for individual choice in determining harvest area, target species, consumption rates, and other factors based on relative risk. Further- more, risk assessment may contribute to management decisions by federal, state, and local agencies, which may include: • Investigating sources of pollution • Reducing exposure potential by implementing pollution con- trols • Restricting fishery harvests by geographic area or by species 9- Issuing public advisories or controls to limit: - Geographic area of harvesting - Harvest season - Harvest methods - Species harvested - Catch number - Size range harvested - Consumption rate. Further information on the relationship between risk assessment and risk management maybe found in Lowrance (1976), U.S. EPA (1984b), Lave and Menkes (1985), Ames et al. (1987), Lave (1987), Russell and Gruber (1987), and Travis et al. (1987). 14 Hazard Identification The first step in the risk assessment process is to define toxicological hazards posed by the chemical contaminants in samples of fish and shellfish. These hazards are summarized in a toxicity profile for each contaminant of concern. The EPA chemical database, IRIS, can be easily accessed to obtain summaries of key toxicological data to include in toxicity profiles. The results of the hazard assessment infiuence the nature and extent of subsequent steps in risk analysis. For example, the endpoint of concern in dose-response assessment may be selected based on the most severe adverse effect identified in the hazard assessment. In the absence of quantitative data for other steps in the risk assessment process, the results of the hazard assessment constitute the final product for a qualitative evaluation of risk. The contaminants of concern to be included in a particular risk assess- ment should be selected based on the following criteria: • High persistence in the aquatic environment • High bioaccumulation potential • High toxicity to humans (or suspected high toxicity to humans based on mammalian bioassays) • Known sources of contaminant in areas of interest • High concentrations in previous samples of fish or shellfish from areas of interest. General information on persistence, bioaccumulation potential, and toxicity may be obtained from references such as Lyman et al.(1982) and Callahan et al. (1979). Other key sources that are periodically updated are the Registry of Toxic Effects of Chemical Substances (e.g., Tatken and Lewis 1983) and the Annual Report on Carcinogens (e.g.. National Toxicology Program 1982, 1985). Specific information that is Contaminants of Concern 15 directly useful in risk assessment can be obtained for many chemicals from IRIS (see below, Sources of Information and Appendix B). TABLE 1. Organic Priority Pollutants and 301(h) Pesticides Ranked According to Octanoi-Water Partition Coefllcients (Kqw) (updated from Callahan et al. 1979) Priority Pollutant Substance log(Kow) Reference 69 di-n-octyl phthalate 8.06 83 indeno(l,2,3-cd)pyrene 7.66 89 aldrin 7.40 79 benzo(ghi)perylene 7.05 111 PCB-1260 6.91 ..q mirex 6.89 75 benzo(k)nuoranene 6.85 74 benzo(b)nuoranene 6.60 82 dibezo(a,h)anthracene 6.50 107 PCS- 1254 6.48 73 benzo(a)pyrene 6.42 91 chlordane 6.42 92 4,4'-DDT 6.36 90 dieldrin 6.20 129 TCDD (dioxin) 6.10 94 4,4'-DDD 6.02 106 PCB-1242 6.00 72 benzo(a)anthracene 5.91 112 PCB-1016 5.88 76 chrysene 5.79 93 4,4'-DDE 5.69 99 endrin aldehyde 5.60 53 hexachlorocyclopentadiene 5.51 9 hexachlorobenzene 5.47 100 heptachlor 5.44 101 heptachlor expoxide 5.40 39 fluoranthene 5.22 84 pyrene 5.18 41 4-bromophenyl phenyl ether 5.08 64 pentachlorophenol 5.00 40 4-chlorophenyl phenyl ether 4.92 20 2-chloronaphthalene 4.72 81 phenanlhrene 4.57 98 endrin 4.56 78 anthracene 4.54 109 PCB-1232 4.48 80 fluorene 4.38 ..1 methoxychlor 4.30 52 hexachlorobutadiene 4.28 66 bis(2-ethylhexyl)phlhalate 4.20 68 di-n-butyl phthalate 4.13 77 acenaphthylene 4.07 67 butyl benzyl phthalate 4.05 108 PCB-1221 4.00 8 1,2,4-trichlorobenzene 3.98 12 hexachloroethane 3.93 1 acenaphthene 3.92 102 alpha-HCH 3.85 m o i d b k d i i n o i i a d k d d g d h d h d b f d m k b b P 16 Table 1 (Cont.) Priority Pollutant Substance log(Kow) Reference 103 beta-HCH 3.85 P 104 delta-hexachlorocyclohexane 3.85 h __r parathion 3.81 e 7 chlorobenzene 3.79 d 105 gamma-HCH 3.72 h 21 2,4,6-trichlorophenol 3.69 c 95 alpha-endosulfan 3.60 96 beta-endosulfan 3.60 97 endosulfan sulfate 3.60 49 fluorotrichloromethane(renioved) 3.53 c 26 1,3-dichlorobenzene 3.48 k 25 1,2-dichlorobenzene 3.38 k 27 1,4-dichlorobenzene 3.38 k 55 naphthalene 3.36 h 113 toxaphene 3.30 38 ethylbenzene 3.15 62 N-nitrosodiphenylamine 3.13 b 22 para-chloro-meta cresol 3.10 a 31 2,4-dichlorophenol 3.08 a 28 3,3'-dichlorobenzidine 3.02 37 l,2-diphenylhydra7ine 2.94 g 58 4-nitrophenol 2.91 d r malathion 2.89 e 60 4,6-dinitro-o-cresol 2.85 6 tetrachloromethane 2.64 d 42 bis(2-chloroisopropyl)ether 2.58 g 85 tetrachloroethene 2.53 b 11 1,1,1-trichloroethane 2.47 b 34 2,4-dimethylphenol 2.42 b 87 trichloroethene 2.42 b 15 1,1,2,2-tetrachloroethane 2.39 b 47 bromoform 2.30 32 1,2-dichloropropane 2.28 86 toluene 2.21 b _r guthion 2.18 14 1,1,2-trichloroethane 2.18 24 2-chlorophenoI 2.16 b 50 dichlorodifluoromethane (remove d) 2.16 c 4 benzene 2.11 d 51 chlorodibromomethane 2.08 35 2,4-dinitrotoluene 2.00 36 2,6-dinitrotoluene 2.00 33 1,3-dichloropropene 1.98 30 l,2-/ran5-dichloroethene 1.97 c r demeton 1.93 23 chloroform 1.90 b 48 dichlorobromomethane 1.88 56 nitrobenzene 1.83 b 5 benzidine 1.81 g 13 1,1-dichloroethane 1.78 57 2-nitrophenol 1.77 54 isophorone 1.67 b 71 dimethyl phthalate 1.61 b 17 Table 1 (Cont.) Priority Pollutant Substance log(Kow) Reference 16 chloroethane 1.54 59 2,4-dinitrophenol 1.53 29 1,1-dichloroethene 1.48 65 phenol 1.46 a 10 1,2-dichloroethane 1.45 b 70 diethyl phthalate 1.40 b 63 N-nitrosodipropylamine 1.31 44 dichloromethane 1.30 19 2-chloroethylvinylether 1.28 g 43 bis(2-chloroethoxy)methane 1.26 g 3 acrylonitrile 1.20 b 18 bis(2-chloroethyl)ether 1.12 b 46 bromomethane 1.00 2 acrolein 0.90 b 45 chloromcthane 0.90 88 vinyl chloride 0.60 61 N-nitrosodimethylamine -0.58 g a Veith et al. (1979a). b Veith et al. (1980). c Gossett et al. (1983). d Veith et al. (1979b). e Kenaga and Goring (1980). f L^o, A., 20 November 1984, personal communication. g U.S. EPA (1980b). h i J Kanckhoff (1981). Rapaport and Eisenreich (1984). Miller etal. (1985). k 1 Means et al. (1980). Miller et al. (1984). ni McDuffie (1981). n Chiou et al. (1981). o Briggs (1981). P Solubilities of the various isomers of HCH indicate that they A'ill have similar log(Kow) values. q Chlorinated pesticides that are not on ih ; prionty pollutant list but are included in Section 301(h) (Clean Water Act) monitoring programs. r Organophosphorus pesticides that are not on the priority poll utant list but are included in Section 301(h) (Clean Water Act) monitoring programs. Recommendations regarding specific contaminants of concern are beyond the scope of this guidance manual. A general list of con- taminants with available EPA toxicological data listed in IRIS is provided in Appendix B. The procedures for quantitative risk assess- ment outlined in this manual are designed for use only with chemicals having toxicological indices [Reference Doses (RID) or Carcinogenic Potency Factors). In addition to the availability of toxicological in- dices, the relative bioaccumulation potential of various chemicals is a key consideration in selecting contaminants of concern. EPA priority- pollutant organic chemicals and selected pesticides are listed in Table 1 in descending order of bioaccumulation potential, according to their octanol-water partition coefficients (Tetra Tech 1985a). Note that 18 organic compounds with a log octanol-water partition coefficient greater than or equal to 2.3 were recommended by Tetra Tech (1985a) for inclusion in EPA Section 301(h) (Clean Water Act) monitoring programs. EPA priority-pollutant metals are listed in Table 2 in descending order of bioaccumulation potential, according to their BCF (Tetra Tech 1985a). TABLE 2. Inorganic Priority Pollutants Ranked According to Bioconcentratlon Factor (BCF) Priority Pollutant No. Substance log BCF^ 4.602 4.602 3.447 3.073 2.762 2.544 2.513 2.253 2.190 2.104 2.000 1.699 1.176 ND ND ND ND ND BCF = Bioconcentraction Factor. The value shown is the geometric mean BCF among studies summarized by Tetra Tech (1985a). U.S. EPA (1986h) provides additional information on BCF values for selected chemicals. ND = No data. Screening of potential contaminants of concern should be done on a case-by-case basis during preparation of risk assessments. When data on concentrations of contaminants in edible tissues of fishery or- ganisms are available, preUminary calculations of potential risks may be made to rank chemicals by relative priority for detailed evaluation. If contaminant concentration data are available for soils, air, and water (at a hazardous waste site, for example), U.S. EPA (19860 methods for selecting indicator chemicals for public health evaluations at Su- perfund sites may be used to gain perspective on contaminants of concern. For potential carcinogens, the qualitative weight of evidence for carcinogenicity should be considered. Those chemicals with suffi- cient evidence of carcinogenicity in humans should generally be con- sidered as contaminants of concern. Toxicity profiles are summaries of the following information for the selected chemicals of concern: • Physical-chemical properties (e.g., vapor pressure, octanol- water partition coefficients) 123 methylmercury 123 phenylmercury 123 mercuric acetate 120 copper 128 zmc 115 arsenic 118 cadmium 122 lead 119 chromium VI 119 chromium III 123 mercury 124 nickel 127 thallium 114 antimony 117 beryllium 121 cyanide 125 selenium 126 silver Toxicity Profiles 19 • Metabolic and pharmacokinetic properties (e.g., metabolic degradation products, depuration kinetics) • Toxicological effects (e.g., target organs, cytotoxicity, car- cinogenicity, mutagenicity) according to specific uptake route of concern (i.e., ingestion). A toxicity profile may consist of an IRIS chemical file. An example file taken from IRIS is provided in Appendix B. The key elements of a hazard assessment should be summarized in a concise tabular format. The examples shown in Table 3 and in the first two sections of the IRIS file (Chronic Systemic Toxicity; Risk Estimates for Carcinogens) in Appendix B illustrate the kinds of information used to evaluate toxicological hazards. Neither toxicity profile in Table 3 is intended to be comprehensive. TABLE 3. Toxicity Profile for Mercury and PCBs* Property Mercury *" PCBs' CAS Number 7439-97-6 1336-36-3 Physical-Chemical Molecular weight 200.6 - 318.7 154.2 - 498.7 Vapor pressure (mm Hg) 0.012 - 0.028 2.8x10-9-7.6x10-5 Solubility (mg/L) 0.056 - 400,000 -5.9 LogKowd N/A" 4.0-6.9 Lx)g Bioconcentration Factor 2.0-4.6 1.9-5.2 Carcinogenic Status Noncarcinogen Probable human car- cinogen Group B2 - Sufficient animal evidence - Inadequate human evidence Acute Toxicity Human LD50 (mg/kg body wt) 29« Mammal LD50 (mg/kg body wt ) 1 .0 - 40.9 Chronic Toxicological Effects Humans Mammals Critical endpoint for risk as- sessment Motor and sensory im- pairment leading to paralysis, loss of vision and hearing, and death. Kidney dysfunction. Reproductive impair- ment and teratogenic ef- fects. 1,010-16,000 Skin lesions, liver dysfunctions, and sensory neuropathy. Possible reproduc- tive and develop- mental impairment. Ilcpaloxicity, fetotoxicity, skin lesions, and hepato- cellular carcinoma. Reproductive and developmental im- pairment. Central nervous system Hepatocellular car- effects (e.g., ataxia and cinoma . parathesia) . 20 Table 3 (Cont.) This is an example toxicity profile and is not intended to be comprehensive. Mercury may occur in its elemental form, as inorganic salts, or as organic complexes. Consequently, the chemical and toxicological properties vary tremendously depending on the degree of complexation or metal speciation. Q Physical-chemical properties and toxicity vaiy according to the degree of chlorine substitution, the number of adjacent unsubstituted carbons and steric configuration. Bioconcentration Factors are the ratio of a chemical concentration in tissues of marine or estuarine organisms and the concentration in water to which the organism is exposed (Tetra Tech 1985a). N/A = not applicable. ^ U.S. EPA (1980a,b, 1986f; lARC 1978). ° For mercury (II) choride via oral route of exposure (Tatken and Lewis 1983). Relevance to consumption of mercuiy (primarily methylated) in Tish is questionable. ^ Clarkson et al. (1973). Information in a toxicity profile is used to support the weight of evidence classification for the likelihood of a chemical causing a given health effect. The endpoints considered should include noncar- cinogenic as well as carcinogenic effects. EPA has developed a weight- of-evidence classification scheme which indicates the state of knowledge on the carcinogenicity of chemicals (U.S. EPA 1986a, 1987a). It includes the following categories: • Group A - Human Carcinogen: This group is used only when there is sufficient evidence from epidemiologic studies to sup- port a causal association between exposure to the agent and cancer. • Group B - Probable Human Carcinogen: This group includes agents for which the weight of evidence of human carcinogeni- city based on epidemiologic studies is "limited." It also includes agents for which the weight of evidence of carcinogenicity based on animjil studies is "sufficient." The group is divided into two subgroups. Usually, Group Bl is reserved for agents for which there is limited evidence of carcinogenicity from epidemiologic studies. It is reasonable, for practical purposes, to regard an agent for which there is "sufficient" evidence of carcinogenicity in animals as presenting a carcinogenic risk to humans. Therefore, agents for which there is "sufficient" evidence from animal studies and for which there is "inade- quate" evidence or "no data" from epidemiologic studies would usually be categorized under Group B2. • Group C - Possible Human Carcinogen: This group is used for agents with limited evidence of carcinogenicity in animals in the absence of data on humans. It includes a wide variety of evidence: e.g., (a) a malignant tumor response in a single, well-conducted experiment that does not meet conditions for sufficient evidence; (b) tumor responses of marginal statistical significance in studies having inadequate design or reporting; (c) benign but not malignant tumors with an agent showing no response in a variety of short-term tests for mutagenicity; and (d) response of marginal statistical significance in a tissue known to have a high or variable background rate. 21 Sources of Information • Group D - Not Classifiable as to Human Carcinogenicity: This group is generally used for agents with inadequate human and animal evidence of carcinogenicity or for which no data are available. • Group E - Evidence of Noncarcinogenicity for Humans: This group is used for agents that show no evidence for carcinogeni- city in at least two adequate animal tests in different species or in both adequate epidemiologic and animal studies. The clas- sification of an agent in Group E is based on the available evidence and should not be interpreted as a definitive con- clusion that the agent is not a carcinogen under any circum- stances. The above descriptions for the categories were taken from U.S. EPA (1986a). At present, a weight-of-evidence classification for car- cinogenicity is available in IRIS for each chemical assigned a Car- cinogenic Potency Factor. In many cases, EPA regions and others may rely on toxicity profiles generated previously. IRIS is a key source of chemical toxicity data, including information from critical studies and weight-of-evidence classifications for carcinogens. The first step in a hazard assessment should be to consult IRIS chemical files for potential contaminants of concern. IRIS chemical files are available for approximately 260 chemicals (as of August 1988). Further information on IRIS is provided in Appendix B. The primary sources of toxicity profiles are the EPA Office of Waste Programs Enforcement and Office of Health and Environmental As- sessment (e.g., Appendix C, Table C-1). EPA toxicity profiles are available for approximately 195 chemicals. Additional sources are shown in Appendix C, Table C-2. Under the Superfund Amendments and Reauthorization Act of 1986, EPA and the Agency for Toxic Substances and Disease Registry are preparing toxicity profiles for 100 hazardous substances considered as high priority contaminants at Superfund sites. Supplementary information on the toxicity of contaminants of concern may be obtained from bibliographic or chemical/toxicological databases. DIALOG, a comprehensive bibliographic database system (Dialog Information Services, Inc., 3460 Hillview Avenue, Palo Alto, CA 94304), offers access to databases such as Pollution Abstracts, National Technical Information Service, and ENVIROLINE. Chemi- cal and toxicological information can be obtained from the databases listed in Appendix C, Table C-3. In particular, MEDLARS and its associated databases (e.g., Toxline, RTECS, and AQUIRE) provide extensive toxicological information. 22 Dose-Response Assessment After the potential hazard associated with each contaminant of con- cern is characterized, the relationship between dose of a substance and its biological effect is determined. Dose-response data are used to determine the toxicological potency of a substance, a quantitative measure of its potential to cause a specified biological effect. The concepts of exposure, dose, dose-response relationship, and toxicological potency are discussed in the following sections. The concepts of exposure and dose, as defined below, are central to risk assessment: • Exposure: Contact by an organism with a chemical or physical agent • Dose: The amount of chemical uptake by an organism over a specified time as a consequence of exposure. The "ingested dose," or amount of chemical ingested, is distinct from the "absorbed dose." For the oral route of exposure, the absorbed dose is the amount of chemical assimilated by absorption across the lining of the gastrointestinal system. Exposure level or exposure concentra- tion is used to denote the concentration (mg/kg wet weight) of con- taminant in edible tissue of fish or shellfish. As shown below, the absorbed dose is estimated from food consumption rate, the exposure concentration, and an absorption coefficient (see Exposure Assess- ment). The form of the dose-response relationship for carcinogens is assumed to be fundamentally different from that for noncarcinogens (U.S. Office of Science and Technology Policy 1985). Examples of general dose-response relationships are shown in Figure 2. The lack of a demonstrated threshold in dose-response relationships for car- Exposure and Dose Dose-Response Relationships 23 CO cr o s 3 u z HI o UJ LOW-CXDSE REGION OF CONCERN \^ DOSE OF CARCINOGEN OBSERVED DATA POINTS • CHEMICAL A ▲ CHEMICAL B ■ CHEMICAL C MODELS Low dose extrapolation Models fit within observed data range i Ol UJ u. u- olti Z X UJ O o UJ cc Rfd Frequercy - RfD- UF- NOAEL- Dose- Propcnion of animals tested Reference Dose Uncertainty Factor No Observed Adverse Effects Level Ingested Dose DOSE OF NONCARCINOGEN Figure 2 Hypothetical example of dose-response curves for a carcinogen and a noncarcinogen. cinogens (U.S. EPA 1980b, 1986a; U.S. Office of Science and Tech- nology Policy 1985) implies a finite risk of cancer even at very low doses of the carcinogen. For noncarcinogenic effects, there is usually a threshold dose (i.e., a dose below which no adverse biological effects are observed in the animal bioassay). The threshold dose is termed the No-Observed-Adverse-Effect-Level (NOAEL), as shown in Figure 2. Note that a nonzero mean response may be a NOAEL if that mean response is not significantly different from zero as determined by a statistical test. The Lowest-Observed-Adverse-Effect-Level (LOAEL) is the lowest concentration that results in a statistically significant health effect in the test population. A measure of toxicological potency is derived from an empirical dose-response relationship for the chemical of interest. Toxicological potency indices for two broad categories of toxicants are defined as follows: • Carcinogens are individually characterized by a Carcinogenic Potency Factor, a measure of the cancer-causing potential of a substance estimated as the upper 95 percent confidence limit of the slope of a straight line calculated by the linearized multistage procedure or another appropriate model • Noncarcinogens are individually characterized by an RfD, an estimate of the daily exposure to the human population (includ- ing sensitive subpopulations) that is unlikely to produce an appreciable risk of adverse health effects during a lifetime. Carcinogenic Potency Factors are also referred to as Slope Factors. The RfD is conceptually similar to an Acceptable Daily Intake (U.S. EPA 1987a). The data set used to define toxicological indices is determined by the quality of available data, its relevance to modes of human exposure, the similarity of the species to humans, and other technical factors. Adequate data from clinical or epidemiological studies of humans are preferred over animal data. If adequate human data are not available, a data set for the animal species most similar to humans or for the most sensitive species is used in the dose-response assessment. Data are evaluated by EPA to ensure high quality (e.g., U.S. EPA 1980b, 1985a). The main source of dose-response data for deriving Carcinogenic Potency Factors and RfDs is Ufetime cancer bioassays performed on rats or mice. Because most of these experiments are designed to be cost-effective, doses in bioassays may be orders of magnitude above those encountered in the human environment. High doses are used in laboratory bioassays for several reasons: 1) to reduce the time re- quired to produce a response and thus overcome problems related to latency period (i.e., length of time between exposure and appearance of health effects), 2) to obtain sufficient statistical power to detect health effects, and 3) to decrease the absolute number of animals required and thereby reduce costs. Doses in animal bioassays for oral uptake of contaminants are usually the administered (ingested) dose, not the absorbed dose (i.e., uptake across the lining of the gastro- intestinal system). 24 Carcinogenic Potency Factors and RfD values derived by EPA are listed in the IRIS database. At present, values for these toxicological indices are being standardized for agency-widp use. A brief overview of methods by which these indices are derived is presented below. The Carcinogen Assessment Group of EPA currently uses the linear- ized multistage procedure to derive Carcinogenic Potency Factors (U.S. EPA 1980b, 1985a, 1986a, 1987a). The multistage model assumes that carcinogenesis results from a series of interactions between the carcinogenic chemical and DNA, with the rate of interactions linearly related to dose. For example, a chemical may induce a mutation in the DNA of a cell to initiate carcinogenesis. However, a series of further interactions between DNA and the same chemical (or another one) may be necessary to promote carcinogenesis and induce a tumor. The multistage model is the model most frequently used by EPA when there is no convincing biological evidence to support application of an alternative model. Other models include the logit, probit, single-hit, and WeibuU models (Food Safety Council 1980, 1982; Hogan and Hoel 1982; Cothern et al. 1986). At high doses (corresponding to lifetime risks greater than about 10"^), all currently used models yield generally similar risk estimates. Below risks on the order of 10'^ the models diverge increasingly as dose declines. In the low-dose range, the linearized multistage model generally predicts risks similar to the single-hit (i.e., linear) model. For many data sets, both of these models yield higher estimates of low-dose risk than do other models (U.S. EPA 1980b; Hogan and Hoel 1982; U.S. Office of Science and Technology Policy 1985). The mathematical form of the multistage model for a specified car- cinogen is: where: R(d) R(d) = 1 - exp [-(qid + q2d2 + . . . + qtd'')] (1) = Excess lifetime risk of cancer (over background at dose d) (dimensionless) qi values = Coefficients [kg day mg'^ (i.e., the inverse of dose units)] d = Dose (mg kg' day" ) Jc = Degree of the polynomial used in the multistage model. U.S. EPA (1987a) described the linearized multistage procedure as follows: • The multistage model is fitted to the data on tumor incidence vs. dose • The maximum linear term consistent with the dose-response data is calculated, which essentially defines the linear portion of the dose-response function at low doses Carcinogenic Potency Factors 25 Reference Doses • The coefficient of the maximum linear term, designated as qi*, is set equal to the slope of the dose-response function at low doses • The resulting estimate of qi* is used as an upper-bound es- timate of the Carcinogenic Potency Factor (termed Slope Fac- tor in U.S. EPA 1987a). qi* is usually calculated as the upper 95 percent confidence limit of the estimate of the coefficient qi in Equation 1. The model commonly used to estimate plausible-upper-limit risk for low levels of exposure over a lifetime is therefore: R (d) = qi* d (2) where: R (d) = Upper-bound estimate of excess lifetime risk of cancer (dimensionless) qi* = Upper-bound estimate of carcinogenic potency (kg day mg-^) d = Dose (mg kg' day" ). Equation 2 represents a linear approximation of the multistage model. Because the slope of the dose-response function at high doses could be different from that at low doses, the use of qi* as an upper-bound estimate of potency is not valid at high levels of exposure. Thus, qi* should not be used as the upper-bound estimate of potency at ex- posures corresponding to excess lifetime risks greater than ap- proximately 10'" per individual (i.e., one excess tumor per 100 exposed individuals). If a potency factor is derived from nonhuman data, as is usually the case, it must be extrapolated to humans. Before being applied to humans. Carcinogenic Potency Factors derived from animal data are corrected using surface-area differences between bioassay animals and humans (U.S. EPA 1980b, 1986a). The rationale for using surface-area extrapolations is detailed in Pinkel (1958), Freireich et al. (1966), Dedrick (1973), and Mantel and Schneiderman (1975). The relation- ship between surface-area extrapolation and body-weight extrapola- tion approaches is discussed in the Introduction above (see Background, Relationship of EPA Risk Assessment Methods to FDA Risk Assessment Methods). Current methods for predicting human health effects from exposure to noncarcinogenic chemicals rely primarily on the concept of an RfD (U.S. EPA 1987a). The RfD is derived from an observed threshold dose (e.g., NOAEL or LOAEL if the NOAEL is indeterminate) in a chronic animal bioassay by applying an uncertainty factor, which usual- ly ranges from 1 to 1,000 (Dourson and Stara 1983). The relationship between the NOAEL, the RfD, and the uncertainty factor are il- lustrated in Figure 2 above. The uncertainty factor accounts for dif- ferences in threshold doses among species, among intraspecies groups differing in sensitivity, and among toxicity experiments of different 26 duration. Dourson and Stara (1983) and U.S. EPA (1987a) discuss the methods for deriving RfD values and the criteria for selecting uncer- tainty factors. In brief, an uncertainty factor of 1000 is based on combining a factor of 10 to account for animal-tb-human extrapolation, a factor of 10 to protect sensitive individuals, and a factor of 10 to account for use of a LOAEL in place of a NOAEL. In many cases, EPA regions and other agencies will be able to rely on dose-response assessments generated previously. Current values for Carcinogenic Potency Factors and RfDs are given in IRIS (U.S. EPA 1987a; e.g., see Appendix B). Before using these values, investigators should consult the IRIS database and current EPA health assessment documents for information on their derivation and associated uncert- ainties. Contacts for information on specific chemicals are listed in IRIS Chemical Files. The Carcinogenic Potency Factors calculated by the EPA Carcinogen Assessment Group are pubUshed in IRIS and in each health assess- ment document produced by the Office of Health and Environmental Assessment (e.g., U.S. EPA 1985a). The EPA Carcinogen Assessment Group develops these carcinogenic potency values and updates them periodically. Before being entered into IRIS, Carcinogenic Potency Factors and supporting documentation are reviewed by the Car- cinogen Risk Assessment Verification Endeavor (CRAVE) work group. The list of Carcinogenic Potency Factors published in each health assessment document is intended only to provide comparative information for various chemicals. IRIS should be used as the primary source of Carcinogenic Potency Factors for risk assessment. IRIS is the primary source of RfD values. An example of an IRIS data sheet for the pesticide hndane is shown in Appendix B. The data sheet provides information on the RfD, the endpoints (biological effects) of concern, experimental data sets, doses, uncertainty factors, additional modifying factors, confidence in the RfD, reference documentation, and dates of agency RfD reviews. Individual program offices within EPA may need to be consulted for information on chemicals not yet incorporated into IRIS. For example, the Office of Drinking Water is a source of RfDs for selected chemi- cals. In May 1987, the Office of Drinking Water released draft Health Advisories containing RfDs and guidelines for short-term effects for 16 pesticides: alachlor, chiordane, l,2-dibromo-3-chloropropane (DBCP), 2,4-dichlorophenoxyacetic acid (2,4-D), 1,2-dichloro- propane, endrin, ethylene dibromide (EDB), heptachlor and hep- tachlor epoxide, lindane, methoxychlor, oxymyl, pentachlorophenol, toxaphene, and 2,4,5-trichlorophenoxypropionic acid (2,4,5-TP). Of- fice of Drinking Water Health Advisories will eventually be incor- porated into IRIS. Sources of Information Carcinogenic Potency Factors Reference Doses 27 Exposure Assessment Exposure assessment is the process of characterizing the human populations exposed to the chemicals of concern, the environmental transport and fate pathways of those chemicals, and the frequency, magnitude, and duration of the exposure dose (U.S. EPA 1986b). For exposure assessment of contaminated fish and shellfish, the following factors should be considered: • Concentrations of contaminants in aquatic biota of concern • Potential environmental transfer of contaminants from sources through aquatic species to humans • Fisheries harvest activities, diet, and other characteristics of exposed human populations • Numerical variables (e.g., food consumption rate, contaminant absorption efficiency) used in models to estimate exposure • Purpose of the exposure assessment (e.g., assessment of poten- tial closure of sport or commercial fishery; documentation of health risk from local contaminant sources such as hazardous waste site or wastewater discharges; development of sportfish consumption advisories). Information on contaminant concentrations and the exposed popula- tion is combined to construct an exposure profile, which includes estimates of average rates of contaminant intake by exposed in- dividuals. Key stages of an exposure assessment for contaminated fish and shellfish are discussed in the following sections. Measurement of Contaminant Concentrations in Tissues Guidance on development of study designs to measure concentrations of toxic substances in edible tissues of fish and shellfish is provided in 29 this section. The guidance provided below focuses primarily on field surveys or monitoring programs involving the collection of samples directly from aquatic environments, or from harvesters when the specific geographic origin of samples is known. Such guidance is directly relevant to analysis of recreational fisheries. The present document does not specifically address approaches to marketplace sampling of commercial fisheries products, although some of the con- cepts discussed below apply to marketplace surveys. Sampling designs for collection of fisheries products from the marketplace are available in FDA Compliance Program Guidance Manuals (e.g., U.S. FDA 1986). Sampling of commercial fisheries directly at the source is preferred over marketplace sampling because the former generally allows documentation of the samphng location. If the exposure assessment is designed to include contaminant intake from consumption of commercial fish and shellfish, samples may be obtained in two ways. First, samples of target species can be obtained directly from commercial fishermen. In this case, a strict quality assurance/quality control (QA/QC) program should be implemented to ensure proper handling, storage, and documentation of samples. Documentation should include sampling location, species name, size (length, carapace width, or shell height/width), weight, sex, reproduc- tive condition, time and date of sampling, and preservation technique. In most cases, a technician or observer should be on board the fishing vessel to maintain proper sample handling and documentation. Alter- natively, samples may be collected by monitoring program personnel using vessels other than commercial fishing boats. In this case, samples should be collected in a way that simulates commercial fishing prac- tices as closely as possible (e.g., same species, size classes, season, fishing area, sampling method, and water depth). Regardless of the general approach to sampling, the organisms collected should be placed directly in temporary storage on board the sampling vessel. Upon return to shore, resection of samples should be accomplished as quickly as possible using an adequate clean-room. If an extended sampling cruise necessitates resectioning on board, an adequate clean- space should be set aside to ensure that samples are not contaminated. Analysis of chemical residues in tissue to support an exposure assess- ment is one kind of bioaccumulation study. Bioaccumulation is defined here as the uptake and retention of a contaminant (e.g., a potentially toxic substance) by an organism. The term bioconcentra- tion refers to any case of bioaccumulation wherein the concentration of contaminant in tissue exceeds its concentration in the surrounding medium (i.e., water or sediment). The phrase "bioaccumulation sur- vey" will be used below to refer to measurement of chemical residues in tissue samples from fish and shellfish collected in the field. The elements of a study design for analysis of chemical residues in tissue include: • Objectives • Target species and size (age) class • Sampling station locations • Target contaminants 30 • Sampling times • Kind of sample (e.g., composite vs. grab, cooked vs. raw; fillet vs. whole organism) • Sample replication strategy • Analytical protocols, including detection limits • Statistical treatment of data. Because the complexity and specific features of a sampling design will depend on the objectives of the exposure assessment, no single design Ccm be recommended here. Nevertheless, some basic steps in the study design process can be summarized as follows: • Define concise objectives of the study and'any hypotheses to be tested. • Define spatial and temporal characteristics of fisheries relative to harvesting activities (e.g., seasonality, catch or consumption rates, species composition, size ranges, demersal vs. pelagic species). • Define harvesting activities and methods of preparing food for consumption that potentially affect exposure to contaminants. • Define kinds of samples to be collected (species, type of tissue, mode of preparation) and variables to be measured, based on a preliminary exposure analysis. • Evaluate alternative statistical models for testing hypotheses about spatial and temporal changes in measured variables. Select an appropriate model. • When possible, use stratified random sampling for each fish and shellfish species, where the different strata represent dif- ferent habitat types or kinds of harvest areas that may infiuence the degree of tissue contamination. • When practical, specify equal numbers of randomly allocated samples for each stratum/treatment combination (e.g., habitat type in combination with species or season). • Include samples from a relatively uncontaminated reference or control area to help define local contamination problems. • Perform preliminary sampling or analyze available data to evaluate the adequacy of alternative sampling strategies (e.g., composite samples vs. tissue from individual organisms) and statistical power as a function of the number of replicate samples. • Develop a QA/QC program that covers: sample collection and handling; chain of custody; data quality specifications; analyti- cal methods and detection limits; data coding; data QA/QC steps to assess precision, accuracy, and completeness; database management specifications; data reporting require- ments; and performance audits. • Define data analysis steps, including statistical tests, data plots, summary tables, and uncertainty analysis. 31 Study Objectives and General Sampling Design Note that the second and third steps above depend on information developed as part of the characterization of the exposed population (see Exposed Population Analysis below). Also, practical limitations of field sampling may dictate compromises in the sampling design. For example, use of equal sample sizes is generally recommended because statistical analysis of data sets with unequal sample sizes may be difficult or unnecessarily complex. However, collection of equal num- bers of replicate samples for each treatment (or stratum) may be impractical if both dominant and rare species are to be sampled at a series of harvest locations with a broad range of harvest yields. Depending on the specific objectives and corresponding study design, a series of statistical analyses rather than a single test may be ap- propriate. Detailed guidance on sampling strategies is provided by Phillips (1980), Green (1979), Tetra Tech (1985b,c; 1986b), Phillips and Segar (1986), and Gilbert (1987). Much of the guidance provided in the following sections incorporates the suggestions of these authors. The statement of objectives is a critical step in the study design process, since specification of other design elements depends on the survey objectives. The study objectives must in turn relate to the objectives of the exposure assessment in which the data will be used. The relation- ships between study objectives and general features of a sampling design are addressed in the next section. Specific objectives of a chemical residue study should be defined to ensure collection of appropriate data for the exposure assessment. Different objectives may require radically different sampling designs. Although the primary objective of a field study may be to estimate the mean concentrations of specified chemical contaminants in edible tissues of harvested species, it may be necessary to specify additional objectives to meet the needs of exposure assessment or risk manage- ment. For instance, statistical discrimination among mean con- taminant concentrations in samples from different harvest areas, seasons, or species maybe desired. Such information might be needed to manage relative risks among harvest areas and to impose fisheries closures on a site-specific basis. Example Objectives-Some examples of objectives for exposure assess- ments paired with appropriate bioaccumulation survey objectives are given below. These objectives are provided to illustrate the ways in which the elements of a bioaccumulation study design depend on the exposure assessment objectives. They are not intended to be recom- mended objectives for an actual exposure assessment. In these ex- amples, the bioaccumulation study design involves specifically the measurement of chemical residues in edible tissues of fishery species. Information on the exposed population, including an analysis of their dietary habits (e.g., fisheries species consumed, food preparation method, and consumption rate), is discus.sed later (see Exposed Population Analysis). Such information may influence the objectives of the exposure assessment and the bioaccumulation survey. 32 Example 1: • Exposure Assessment: Estimate the worst-case exposure for a wide range of contaminants over a predefmed geographical area. • Bioaccumulation Design: Estimate mean concentrations of contaminants in edible tissues of a selected narrow size range of individuals of the most contaminated species during the season of peak contaminant concentrations. Example 1 represents a screening survey to evaluate the need for further work. Edible portions of a limited number (e.g., 3-5) of in- dividual organisms or composite samples would be analyzed for a large number of compounds and the risk assessment conducted assuming moderate or high (but plausible) consumption rates. The species and size range selected would be the ones most likely to accumulate high concentrations of contaminants. Typically, the target species for a screening survey would be the largest individuals of a bottom dwelling species associated with soft sediments. Example 2: • Exposure Assessment: Estimate the long-term average ex- posure to each of the contaminants A, B, and C through consumption of aquatic species L, M, N, and O combined from harvest area Z for the average person in the exposed human population. • Bioaccumulation Design: Estimate the mean concentrations of contaminants A, B, and C in edible tissues of aquatic species L, M, N, and O combined from harvest area Z over an annual period. Example 2 illustrates a simple case involving the consumption of multiple species from a single harvest location. Individual or com- posite samples of each species would be analyzed separately during different seasons or during a single season expected to represent the annual average. If samples are analyzed separately during different seasons (e.g., see discussion of Example 4 below), the mean annual exposure for all species could still be calculated from the seasonal data. In general, highly composited samples are not recommended because information on different factors (e.g., species, seasons) that affect contaminant concentrations is lost. Example 3: • Exposure Assessment: Estimate a plausible-upper-limit of exposure to each of the contaminants A, B, and C through consumption of aquatic species L, M, N, and O combined from harvest area Z for a seasonal harvester in the exposed popula- tion. • Bioaccumulation Design: Estimate the upper bound of the 95 percent confidence interval of the mean concentration for each of the contaminants A, B, and C in edible tissues of aquatic 33 species L, M, N, and O combined from harvest area Z during the season of highest contamination. The general sampling design for the objectives of Example 3 would require rephcate composite samples to estimate upper bounds of 95 percent confidence intervals for the mean concentrations of con- taminants across species. To meet these objectives, samples could be composited across species, although this is generally not recom- mended. Multispecies composites would not provide data for assess- ing exposures corresponding to different dietary habits. To obtain an upper-limit estimate of exposure, it might be sufficient to analyze samples from only one season if available information on seasonal variation was sufficient to select one season as the expected worst case. Example 4: • Exposure Assessment: Estimate the probability distribution of exposure to each of the contaminants A, B, and C through consumption of each of aquatic species L, M, N, and O from harvest area Z for various segments of an exposed population (e.g., ethnic groups) over an annual period. • Bioaccumulation Design: Estimate the probability distribu- tion of concentrations of contaminants A, B, and C in edible tissues of each of aquatic species L, M, N, and O from harvest area Z over an annua! period. To accomplish the objectives of Example 4, extensive seasonal data on the dietary composition of several subgroups of the exposed popula- tion must be available. Separate replicate composite samples of each harvested species could be analyzed for each season. During each season, the species analyzed would correspond to those represented to a significant extent in the diet. Probability (frequency) distributions and means of contaminant concentrations would be derived for each species during each season. By combining data from different species, the probability distribution of exposure and the mean exposure weighted by species representation in the diet could be calculated for each population segment. Note that data to support the analyses required by Example 4 are seldom available before a specially designed study is conducted. Example 5: • Exposure Assessment: Estimate an average and a plausible- upper-limit of exposure to each of the contaminants A, B, and C through consumption of each of aquatic species L, M, N, and O from each of the harvest areas X, Y, and Z over an annual period. • Bioaccumulation Design: Estimate the mean concentration and the upper bound of the 95 percent confidence interval of the mean concentration for each of the contaminants A, B, and C in edible tissues of each of species L, M, N, and O from each of the harvest areas X, Y, and Z during each of the harvest seasons. 34 The sampling strategy appropriate for Example 5 is complicated by the occurrence of discrete harvest areas. Replicate composite samples of a given species would generally be required for each season and area in which the species is harvested. Because the characteristics of the exposed population may differ among harvest areas, it may be ap- propriate to divide the exposed population into segments correspond- ing to geographic areas, ethnic groups, age classes or other factors. The seasonal and total annual exposure for each segment of the exposed population would be calculated for each species as in Example 4 above. Influence of Environmental and Population Factors-The four ex- amples just given illustrate the variety of general study designs that may be needed to meet diverse objectives. The specific design of a chemical residue study will depend on the interplay between dietary patterns of the exposed population and environmental factors that influence con- centrations of contaminants in tissues of aquatic organisms. Some of the important environmental factors are: • Conventional water quality (i.e., hardness, salinity, temper- ature, suspended solids) • Habitat location, depth, proximity to contaminant sources • Contaminant concentrations in water • Contaminant concentrations in sediments • Species available for harvest, as influenced by habitat, migratory cycles, and fisheries management practices • Organism activity pattern, food habits, and habitat • Seasonal biological cycles (e.g., stage of sexual cycle) in rela- tion to the frequency and seasonality of contaminant inputs (e.g., industrial discharges, waste dumps, dredging) • Organism size (or weight), age, and sex • Lipid content of tissue analyzed (where lipophilic organic contaminants are of concern). Examples of the interaction between these factors and parameters of the exposed population are given in Figure 3. Seasonal variation in environmental factors or activities of the exposed population may correlate with contaminant concentrations in con- sumed fish and shellfish. Therefore, at least general knowledge of seasonal changes in contaminant concentrations and human consump- tion patterns may be needed to design an appropriate sampling ap- proach for estimating long-term exposure. Two extreme examples of contamination and diet patterns are provided below: Homogeneous Diet and Contamination: • Each of the species is present in the harvest area all year • There is no seasonal variation in contaminant concentrations • Contaminant concentrations do not vary among species 35 EXPOSED-POPULATION FACTORS ^ 4^ ,,T 4^ ^ # i? O ■y / .^^ ^ c? ENVIRONMENTAL FACTORS ^ CONVENTIONAL WATER QUALITY ^ PROXIMITY TO CONTAMINANT SOURCES CONTAMINATION OF WATER/SEDIMENTS SPECIES AVAILABLE FOR HARVEST ORGANISM ACTIVITY MODE ^ SEASONAL BIOLOGICAL CYCLES « ORGANISM SIZE ORGANISM AGE ORGANISM SEX LIPID CONTENT OF TISSUE / /////#/,/ ^ ^ ^ ^ C9 /<^ ^ cT 4-4 A. %, \ ^^ v?. z g o o UJ -I d ? z < z cc o < z 3 o o ^ I > 6 « o oc cc u. a. i3 00 o> u .2 o c 0} (D CC 0) 'o CD Q. W 1- O o _C (U c 03 w 0) ■D c (0 (U c CO E •D CD ■D e CD E E o o CD O 03 E E CO 1- Z3 example, when contaminant concentrations are positively correlated with fish (or shellfish) size, frequent consumption of the smaller in- dividuals may be acceptable even though consumption of larger in- dividuals should be severely limited. Two general approaches to field sampling are possible. First, the investigator can obtain samples directly from harvesters. This ap- proach has the advantage that the sampled population is the popula- tion of direct interest for the exposure and risk assessments. However, one drawback of this approach is the potential for contamination or degradation of samples due to handhng of the samples by the har- vesters. Moreover, the precise sampling locations may be unknown if samples are collected at dockside from recreational or commercial fishing boats. The second approach is to obtain samples independent of the normal harvesting efforts, allowing standard sample handling practices to be implemented. Independent sampling also facilitates the collection of adequate samples for stratification by organism size, habitat, or some other variable. The remainder of this section address- es a sampling effort that is independent of normal harvesting activities. Sampling stations should generally be located in known harvest areas. However, additional stations in relatively uncontaminated reference or control areas should also be sampled. By comparing results among harvest areas and between each harvest area and the reference station, one can establish not only the degree of spatial heterogeneity but also the magnitude of elevation above reference of contaminant concentra- tions (and corresponding health risks) at each harvest area. Because sampling depth or vertical position on the shore may influence con- taminant concentration in aquatic organisms, reference station char- acteristics should be closely matched to those for the harvest areas. Sampling stations may be located within a study area according to one of several probability sampling designs (Figure 5). Gilbert (1987) provides a concise summary of conditions under which each sampling design is preferred. Simple random sampling implies that each individual organism of a species within a specified area has an equal chance of being selected for measurement and that selection of one individual does not in- Huence selection of others. A simple random sampling strategy is appropriate if there are no major trends or patterns of contamination in the study area. Note that sampling of fish or shellfish with sampling gear (e.g., hook and line, nets) will often be nonrandom with respect to species and size classes because of the selective nature of the gear. Stratified random sampling involves random sampling within nonover- lapping strata of a population (e.g., subareas where recreational fishing effort is concentrated or where contamination is greatest). This sam- pling approach is appropriate when localized geographic areas within a harvest region are heterogeneous relative to the kind or degree of contamination. Two-stage sampling involves random or systematic subsampling of primary units selected by a random sampling technique. For example, fish could initially be collected randomly from a given stream reach. In Sampling Station Locations 41 SIMPLE RANDOM SAMPLING STRATIFIED RANDOM SAMPLING • STRATA PRIMARY- UNITS TWO-STAGE SAMPLING • • ^ • • • • CLUSTER SAMPLING • - •/ (^ • CLUSTERS SYSTEMATIC GRID SAMPLING • • • • • • • • • • • • RANDOM SAMPLING WITHIN BLOCKS • •_ Reference: Gilbert (1987) Figure 5 General sampling station layouts for probability sampling in two dimensions. the second stage of sampling, subsamples of fillet from each fish could be selected randomly for chemical analyses. Multistage sampling is an extension of two-stage sampling. Cluster sampling involves choosing groups of individual organisms at random, then measuring contaminant concentrations in all individuals within each cluster. Cluster sampling is sometimes used to estimate means if clusters of sampling units (e.g., individual organisms in a clump) can be selected randomly more easily than can individual units. Systematic sampling consists of sampling at locations and/or times according to a pattern. For example, samples may be collected at equidistant points on a spatial grid or at equsilly spaced time intervals. Systematic sampling is generally preferred for mapping patterns of contamination. As such, it is more appropriate for soil or sediment sampling than for bioaccumulation studies. The random-sampling- within-blocks strategy shown in Figure 5 combines systematic and random sampling. Such procedures produce more uniform coverage than does simple random sampling. Gilbert (1987) describes systematic sampling approaches for locating "hot spots" or highly contaminated local areas. He addresses the fol- lowing questions: • "What grid spacing is needed to hit a hot spot with specified confidence?" > • "For a given grid spacing, what is the probability of hitting a hot spot of specified size?" • "What is the probability that a hot spot exists when no hot spots were found by sampling on a grid?" If grid sampling is to be applied to a bioaccumulation study, the target species must exhibit limited mobility. Grid sampling can also be ap- plied to collection of aquatic sediment samples. Gilbert (1987) provides guidance on spacing of grid samples. Grid sampling is especially appropriate for identifying environmental contamination associated with discrete sources of pollution such as industrial discharges, storm drains, and combined sewer overflows. The use of caged mussels is a promising approach for identifying sources through chemical residue analysis. As part of the Long Island Sound Estuary Program, EPA Region I is using caged mussels to monitor chemical contaminants entering the Sound from tributaries. The California mussel watch program (e.g., Ladd et al. 1984), the U.S mussel watch (Goldberg et al. 1978, 1983; Farrington et al. 1983), and the NOAA status and trends program (Boehm 1984) illustrate the use of both resident and caged transplant mussels to monitor bioaccumula- tion of toxic chemicals over space and time. Toxic chemical residues in mussels are excellent indicators of point source discharges as well as pollution gradients (Phillips 1976; Popham et al. 1980; Phelps et al. 1981). U.S. EPA (1982) described recommended protocols for caged mussel studies. A combination of two-stage and stratified-random (or stratified-grid) sampling is recommended here for most studies of fisheries contamina- 42 tion to support exposure assessment. The two stages correspond to an individual organism and edible tissue. Samples of individual organisms may or may not be composited depending on specific study objectives (see below, Kinds of Samples, Composite Sampling). Sampling strata may include harvest areas, species, and size classes. Other sampling strategies may be either too simple or inappropriate to meet the typical objectives of exposure assessment studies. The timing of bioaccumulation surveys should be based on the tem- poral distribution of harvest seasons and inherent biological cycles of target species. The timing of harvest periods depends on the availability of fishery resources, which may be influenced by the migratory patterns and feeding cycles of target species. Biological cycles that influence an organism's susceptibility to bioaccumulation should also be considered when choosing a sampling period. The most important of these is the reproductive cycle, which is discussed further below. In crustaceans (e.g., crab and shrimp), the molting cycle also determines the potential for bioaccumulation of toxic chemicals. The rate of uptake of con- taminants by crustaceans is highest just after molting, before hardening of the integument limits its permeability. The reproductive cycles of aquatic organisms may exert a major in- fluence on tissue concentrations of many contaminants, especially organic compounds (Phillips 1980). If a worst-case analysis is desired, the target species should be sampled at a time during the harvest period when tissue contaminant concentrations are expected to be at their highest levels. In some species, contaminant content of edible tissues may reach a seasonal maximum at or just before the peak of reproduc- tive ripeness, before gametes or offspring are released. This may be especially characteristic of species that are consumed whole (e.g., clams and oysters). In other species (e.g., salmonids), lipid and as- sociated contaminants may be mobilized and transferred from muscle tissue to eggs before they are released. In such species, the peak of contamination may occur in edible tissue (muscle) well before spawn- ing. Because the time of sampling should be tailored to the reproduc- tive characteristics of the target species, sampling periods may vary among species. However, once a sampling period is chosen, it should remain constant over time if an ongoing monitoring program is planned. An alternative approach is to sample throughout the harvest season for each target species. In this way, representative values can be obtained for estimating means within sampling periods and for detecting seasonal or long-term trends. In most cases, exposure assessments will be performed over relatively short periods of time (e.g., a year), and multiyear sampling may not be required. Within a harvest season, however, sufficient samples should be collected to estimate the mean concentrations of contaminants during the harvest period. To estimate temporal variation or to obtain worst-case estimates, replicate samples will be needed at several times within the harvest season. The frequency of sampling should be related to the expected rate of change in tissue concentrations of contaminants. For an extensive review of temporal changes in bioaccumulation and body burdens of contaminants in aquatic organisms, the reader should consult Phillips (1980). Time of Sampling 43 Kinds of Samples The kind of tissue sampled and the sampling unit (i.e., bdividual organisms vs. composites of several organisms) greatly influence the sensitivity, precision, and representativeness of an exposure assess- ment. The issues of composite sampling and sample preparation tech- niques are addressed in the following sections. Composite Sampling-An alternative to the analysis of tissue from individual organisms is the analysis of composite samples. Composite tissue sampling consists of mixing tissue samples, each called a sub- sample, from two or more individual organisms typically of a single species collected at a particular site and time period. The mixture is then analyzed as a single sample. The analysis of a composite sample therefore provides an estimate of an average tissue concentration for the individual organisms that make up the composite sample. Com- posite sampling is a cost-effective strategy when the individual chemi- cal analyses are expensive but the cost of collecting individual samples is relatively small. The collection of composite samples is required in cases where the tissue mass of an individual organism is insufficient for the analytical protocol. Bioaccumulation surveys designed to support exposure assessments may use a composite sampling strategy. Current risk assessment models used by EPA are based on estimates of long-term average exposure. Estimates of the mean concentrations of contaminants in edible tissue samples from harvested organisms are used as estimates of the exposure concentrations for human consumers of fish and shellfish. Composite sampling of the tissue from selected organisms is a method for preparing a sample that will represent an average con- centration. The collection of replicate composite tissue samples at specified sampling locations will result in a more efficient estimate of the mean (i.e., the variance of the mean obtained with replicate com- posite samples is smaller than that obtained with the collection of replicate samples of individual organisms). One major disadvantage of composite sampling is the inability to directly estimate the range and the variance of the underlying popula- tion of individual samples. Such information is extremely useful in bioaccumulation monitoring programs as an early warning signal of increasing levels of contamination. For example, only a few individuals within a sample may contain high contaminant concentrations. Mixing these individuals with less contaminated organisms in a composite sample at a given station may dilute the contaminants and mask a potential problem. In exposure assessment, the patchy distribution of highly contaminated fish or shellfish may indicate the spatial distribu- tion of sources of contaminants. The benefits of compositing individual samples from a single station within a given sampling period often outweigh the disadvantages just discussed. In such cases, Rohde (1976) and Tetra Tech (1986b) provide a method for calculating the variance of the underlying population (X) of individual samples when the variance of the composite samples (Z) is known: Var X = n (Var Z) (3) 44 where: VarX VarZ n = variance of the mean of individual samples from all composites = variance of the mean of composite samples = number of subs2imples constituting each composite sample. This equation assumes that replicate observations from individual and composite samples are normally distributed. Also, the composites must each consist of subsamples of equal mass (i.e., the same mass of tissue is taken from each organism). For unequal proportions of com- posite subsamples (i.e., tissue mass), the variance of the series of composite samples increases and, in extreme cases, exceeds the variance of grab samples. Thus, it is recommended here that the same mass of tissue be taken from each organism contributing to a composite sample of a single species (Tetra Tech 1986b). For the analyses presented below, it was assumed that the composite samples consist of subsamples of equal proportions. Two special cases of composite sampling are "space-bulking" and "time-bulking" (PhiUips and Segar 1986). Space-bulking involves sam- pling of individual organisms from several locations and combining tissue samples into one or more composite samples for analysis. Time- bulking involves taking multiple samples over time from a single loca- tion and compositing these samples. Time-bulking over a harvest season is especially appropriate where short-term variations in con- taminant concentrations in tissue samples are significant and budget constraints preclude repeated analyses over time. The adoption of space-bulking or time-bulking strategies ultimately relates to the objectives of the exposure assessment. Because exposure concentrations are typically averaged over time in risk assessment models, time-bulking may be more justified than space-bulking. In any case, one should use these strategies with extreme caution since sig- nificant information on spatial and temporal heterogeneity may be lost. Selection of space-bulking or time-bulking techniques should be sup- ported by analyses of available data or data from preliminary sampling. Tiered analyses of samples can also be used to evaluate the ap- propriateness of compositing strategies. For example, individual samples may be stored separately over the entire harvest season. At the end of sample collection, preliminary analyses of individual tissue samples from a selected series of sites and times could be performed to evaluate temporal and spatial heterogeneity and to devise an ap- propriate compositing strategy. Tetra Tech (1986b) evaluated the effects of composite sampling on the statistical power of a sampling design (see Appendix D). Their results demonstrate that the confidence in the estimate of the mean concentra- tion of contaminant in tissue increases as the number of individual samples in the composite increases. The statistical power (i.e., the probability of detecting a specified minimum difference among treat- ments) increases dramatically with the number of individual samples in each replicate composite sample. However, the increase in power associated with adding more individual samples to each composite 45 eventually becomes negligible (e.g., at greater than 10 individuals per composite at typical levels of data variability). For moderate levels of variability in chemical residue data, 6 to 10 individual samples within each of 5 replicate composite samples should be adequate to detect a treatment difference equal to 100 percent of the overall mean among treatments. Rohde (1976), Schaeffer et al. (1980), Brumelle et al. (1984), and Gilbert (1987) also discuss statistical aspects of composite sampling. Sample Preparation— Tissue samples should be removed from target organisms and prepared for analysis according to a well-defmed protocol. Tissue preparation methods can greatly affect the results of bioaccumulation analyses (Smith et al. 1973; Skea et al. 1981; Puffer and Gossett 1983; Landolt et al. 1987). In specifying a tissue prepara- tion protocol, the following issues should be addressed: • Type of tissue (e.g., muscle fillet, whole body, internal organs) • Location of tissue in organisms' body • Removal of any or all of shells, scales, skin, and subcutaneous fat • Raw vs. cooked samples and cooking method • Homogenization method • Minimum sample mass for each kind of analysis. The kind and location of tissue analyzed may influence the realism of the exposure assessment. For example, most humans consume only fillets of fish, not internal organs or whole fish. Because internal organs are often more contaminated by toxic chemicals than are fillets, ex- posure estimates based on chemical analyses of organs or whole fish could be unrealistically high. Removal of skin and subcutaneous fat from samples before chemical analysis generally reduces the mean concentrations of chlorinated organic compounds. In species with a subcutaneous fat layer, this practice may also reduce the variability of replicate data, allowing more sensitive discrimination among statistical treatments (e.g., species or sampling locations). Within the fillet tissue, contaminant concentrations may vary depending on the original loca- tion of the sample on the fish's body. The effect of cooking on the ultimate health risk from a mixture of chemicals (including any transformation or degradation products produced by heating) is unknown. Some studies have shown decreases in concentrations of lipid-soluble organic compounds such as DDT and PCBs following pan-frying, broiling, or baking offish fillets (Smith et al. 1973; Skea et al. 1981; Puffer and Gossett 1983). For example, cooking of fillets before chemical analysis may result in a 2 to 65 percent decrease in the concentration of PCBs relative to the uncooked sample, but the results vary greatly with species and cooking method. However, cooking may also activate or transform chemicals to create carcinogens [e.g., formation of benzo(a)pyrene during charbroiling]. Regardless of method of tissue preparation, an adequate mass of each sample and adequate homogenization of samples before they are analyzed are necessary to obtain representative results (e.g., see Tetra Tech 1986e). 46 Because information on the effects of tissue preparation methods on the results of chemical residue analyses is limited, it is recommended that a pilot survey be performed to establis|i consistent, reliable methods. Relevant protocols for sample storage and preparation are available in a bioaccumulation monitoring guidance document issued by the EPA Section 301(h) (Clean Water Act) program (Tetra Tech 1986e) and in the EPA Interim Methods for the Sampling and Analysis of Priority Pollutants in Sediments and Fish Tissue (U.S. EPA 1981). Because many decisions about sample preparation depend on the specific objectives of the study, no single protocol for sample prepara- tion covers all of the possible approaches. For example, samples are usually blotted dry before being weighed to obtain an estimate of wet weight. However, when bivalve molluscs are being prepared for analysis, it may be desirable to retain excess water for later analysis. In general, field studies to support exposure assessment should focus on the kind of tissue that is most commonly consumed (e.g., fillet). Analysis of raw edible tissues is recommended to provide data on the concentrations of contaminants initially present in tissues that are normally consumed. Eventually, it may be possible to mathematically account for cooking effects in the exposure assessment. At present, however, data on cooking effects are highly variable. Replicated measurements of contaminant concentrations in tissue samples are needed to perform uncertainty analysis (e.g., charac- terizing the precision of the estimates of mean contaminant concentra- tions). Replicated data are also needed for many statistical tests of spatial and temporal trends. Sample replication is recommended here for all bioaccumulation measurements to be used in exposure assess- ments. Guidance on selection of a sample replication scheme is provided in Appendix E. In most cases, at least five replicate samples of individual fish (or shellfish) are required to provide minimal statis- tical power (e.g., ability to discriminate a treatment difference equal to 200 percent of the overall mean among treatments). Increases in sample replication beyond about 10 individual replicates clearly do not provide sufficient benefits in statistical power to justify added costs of sampling and analysis (Appendix E). Greater power can be achieved in a cost-effective manner by composite sampling if information on contamination of individual organisms is not needed (Appendix D). Criteria for selection of method detection limits for analytical protocols may be based on risk assessment models explained below (see Risk Characterization). For example, the analytical chemistry methods may be chosen to enable detection of a chemical concentra- tion associated with a specified minimum risk level defined as accept- able by risk managers. Other factors may dictate choice of a lower detection hmit. For example, routine analytical methods may attain much lower Hmits than required by the specified minimum detectable risk level. Also, lower detection limits may be desired if an objective of the study is to develop baselme bioaccumulation data as well as health risk data. In some cases (e.g., 2,3,7,8-tetrachlorodibenzo-/7-dioxin, ben- zidine, dieldrin, N-nitrosodimethylamine), the minimum detection Sample Replication Selection of Analytical Detection Limits and Protocols 47 QA/QC Program limit that can be achieved with current technologies corresponds to a plausible-upper-limit risk that is substantially above risk levels of potential concern (e.g., greater than 10'^ to lO"*). Tetra Tech (1985c) provides further guidance on detection limits for bioaccumulation surveys. Approved routine EPA methods for sampling and full-scjm analysis of chemical contaminants in tissues are not available. U.S. EPA (1981) published interim methods for sampling and analysis of priority pol- lutants in tissues. EPA-approved protocols for chemical analysis of water samples were adapted for application to tissue samples as part of the Section 301(h) (Clean Water Act) marine discharge waiver program of the Office of Marine and Estuarine Protection [see Tetra Tech 1986e for 301(h) sampling and analysis protocols]. Specifically, 301(h) analytical methods for extractable organic compounds were adapted from Method 1625 Revision B (U.S. EPA 1984a) and addi- tional guidance from the EPA Contract Laboratory Program for Or- ganic Analysis (U.S. EPA 1984c). When applicable, the 301(h) protocols incorporate established EPA advisory limits for precision, accuracy, and method performance (U.S. EPA 1984c). The EPA Office of Acid Deposition, Environmental Monitoring, and Quality Assurance is developing further guidance on sampling and analysis methods to support exposure assessments. Other available methods for analysis of chemical contaminants in tissue samples include those used by U.S. FDA (1978), NOAA (MacLeod et al. 1984), and Ozretich and Schroeder (1985). These analytical protocols are designed to apply to specific subsets of the EPA priority pollutants. U.S. FDA (1978) methods, as described in the Pesticide Analysis Manual, include variations in procedures for tissues differing in lipid content. The choice of an analytical protocol may be influenced by available financial resources. Chemical analysis of samples is often the most costly portion of a sampling and analysis program. Higher analytical costs may be required to achieve greater sensitivity (i.e., lower detec- tion limits). Examples of analytical costs are shown in Table 5. At a given level of sensitivity, a wide range of precision is encountered among diverse organic compounds. For example, the low end of the range of variation shown for extractable compounds in Table 5 can usually be achieved for hydrocarbon analyses, whereas substantially more variability is common for analyses of phthalates and some organic acid compounds. A wide range of analytical costs is also encountered at a given level of sensitivity (Table 5). Differences in analytical tech- niques, laboratory experience with these techniques, and pricing policies of laboratories account largely for the wide variation in cost. An adequate QA/QC program is essential for any sampling and analysis effort to support exposure assessment. U.S. EPA (1984c, 1985c) provides guidance on QA/QC for chemical analysis. Tetra Tech (19860 describes QA/QC procedures for field and laboratory methods used by the EPA Section 301(h) (Clean Water Act) program. Horwitz et al. (1980) provide guidance on QA/QC in the analysis of foods for trace contaminants. Brown et al. (1985a) describe QA guidelines 48 followed by NOAA for chemical analysis of aquatic environmental samples. TABLE 5. Approximate Range of Cost per Sample for Analyses of EPA Priority Pollutants in Tissues as a Function of Detection Limits and Precision^ EPA Priority Approximate Pollutant Detection Typical Approximate Group Limit Precision Cost Range'' Extractable acid/base/neutrals/ PCBs/pesticides < 1-20 ppb <5%->100% $900- > $2,000 Volatiles < 5-20 ppb < 10%- > 100% $250 - $350 Metals 100 ppb <10%->30% $250- $300 NOTE: Range of per sample cost is based on multiple quotes compiled for specific applications and 5 samples. The actual costs may vary from the range shown. This information is provided solely for perspective on relative differences in cost and should not be interpreted as a recommendation of appropriate costs for any given cir- cumstance. Each cost range is mainly the result of laboratory differences in technique and pricing, NOT the range in precision or detection limits shown. A QA/QC plan should be developed as part of the study design for sampling and analysis of chemical residues. The QA/QC plan should include the following information: • Project objectives • Project organization and personnel • QA objectives for precision, accuracy, and completeness for each kind of measurement • Summary of sampling procedures, including sample con- tainers, preparation, and preservation • Forms for documenting sample custody, station locations, sample chziracteristics, sample analysis request, and sample tracking during laboratory analysis • Detailed description of analytical methods • Calibration procedures for chemical measurements • Internal QC checks for analytical laboratories • Performance and system audits for sampling and analysis operations • Preventive maintenance for equipment • Procedures for data management, data QA review, and data reporting for each kind of measurement • Corrective actions 49 Documentation and QA Review of Chemical Data • Procedures for QAyQC reporting and responsible federal and state QA officers • Mechanisms for approval of alterations to the monitoring pro- gram, for suspending sample analyses, and for stopping sample analyses within a tiered design. Relevant portions of the QA plain should be incorporated in the statement of work for each contract laboratory involved in sample analyses. Adequate documentation of the results of chemical analyses are needed to ensure proper interpretation of the data. If a contract laboratory is performing the sample analyses, such documentation should be specified in the original statement of work. The documenta- tion listed below is recommended for chemical residue data: • A cover letter discussing analytical problems (if any) and referencing or describing the procedure used • Reconstructed ion chromatograms for each sample analyzed by gas chromalography/mass spectrometry (GC/MS) • Mass spectra of detected target compounds for each sample analyzed by GC/MS • Chromatograms for each sample analyzed by gas chromatog- raphy/electron capture detection (GC/ECD) and/or gas chromatography/flame ionization detection (GC/FID) • Raw data quantification reports for each sample • A calibration data summary reporting calibration range used [and decafluorotriphenylphosphine (DFTPP) and bromofluorobenzene (BFB) spectra and quantification report for GC/MS analyses] • Final dilution volumes, sample size, wet-to-dry ratios, and instrument detection limit • Analyte concentrations with reporting units identified (to two significant figures unless otherwise justified) • Quantification of all analytes in method blanks (ng/sample) • Method blanks associated with each sample • Tentatively identified compounds (if requested) and methods of quantification (include spectra) • Recovery assessments and a replicate sample summary (laboratories should report all surrogate spike recovery data for each sample; a statement of the range of recoveries should be included in reports using these data) • Data qualification codes and their definitions. The data reporting forms for the EPA Contract Laboratory Program illustrate an appropriate format for documentation of chemical data. 50 All contamineint concentration data to be used in a risk assessment should undergo a thorough QA review by a qualified chemist inde- pendent of the laboratory that analyzed the samples. In some cases, the analytical laboratory may provide a QA review that is simply checked by an independent chemist. The purpose of the QA review is to evaluate the data relative to data quahty objectives (e.g., precision and accuracy) and performance limits estabhshed in the QA plan. In many cases, qualifiers are necessary for selected data values. These qualifiers should be added to the database. A summary of data limitations should always be included in the risk characterization (see below, Risk Char- acterization). The EPA Office of Acid Deposition, Environmental Monitoring, and Quality Assurance is developing guidelines for quality assurance of chemical data to support exposure assessments. Statistical analyses of data will depend on specific study objectives. For each species, statistical summaries of tissue concentration data should include sample size, estimates of arithmetic mean concentration, range, and a measure of variance (standard error or 95 percent con- fidence limits). Geometric mean concentrations are appropriate measures of central tendency when only estimates of tissue burden of contaminants or exposure dose are desired. However, arithmetic means are needed to compare exposure estimates with RfDs and to calculate health risk for chronic effects because long-term consump- tion is an averaging process. Mean tissue concentrations and variances may be calculated for mixed-species diets if data are available on the proportion of each species m the diet. The one-way ANOVA model discussed earlier or multifactor ANO VA models are appropriate for testing for differences in mean contaminant concentrations among species, among sampling stations, or among time periods (Schmitt 1981; also see Tetra Tech 1986b,d). For small sample sizes and data that do not satisfy the assumptions of ANOVA, nonparametric tests such as the Wilcoxon rank sum test for two treat- ments or the Kruskal-Wallis test for multiple comparisons are recom- mended. These tests have the added advantage of being relatively insensitive to a few missing data points or undetected observations (Gilbert 1987). Long-term data sets may be tested for trends by time series analysis (for reviews, see Montgomery and Reckhow 1984 and Gilbert 1987). Examples of trend analysis for chemical contaminants in fish are provided by Brown et al. (1985b) for PCBs in striped bass of the Hudson River and by DeVault et al. (1986) for PCBs and DDT in lake trout from the upper Great Lakes. Data on concentrations of contaminants of concern in tissue samples will often contain observations below detection limits. Means and variances for tissue concentrations should be calculated twice: once using detection limits for undetected observations and once using 0 for undetected observations. Although alternative approaches are pos- sible (e.g., using one-half the detection Hmit), the approach recom- mended here yields more accurate, complete results by quantifying the range of the estimated values. According to the EPA Exposure Assess- ment Group, calculations of plausible-upper-limit risk estimates based on detection limits should generally be avoided. However, risk es- timates based on detection limits may occasionally be useful to indicate that particular chemicals, species, or geographic locations are not Statistical Treatment of Data 51 Analysis of Sources, Transport, and Fate of Contaminants problems, even assuming undetected contaminants are present at concentrations just below their respective detection limits. The choice of contaminant concentration values to use in subsequent calculations to estimate exposure (and ultimately risk) is partly a risk management decision. Exposure estimates are commonly based on arithmetic mean concentrations of contaminants in edible tissue offish or shellfish. Use of the upper 90 or 95 percent confidence limit in place of the mean would provide a conservatively high estimate of exposure. Calculation of conservative estimates for exposure is an appropriate step in uncertainty analysis. However, U.S. EPA (1986b) guidelines on exposure assessment discourage the use of worst-case assessments. Use of upper confidence limits for chemical concentrations in com- bination with a plausible-upper-limit estimate for the Carcinogenic Potency Factor may lead to an unrealistic (i.e., highly unlikely) estimate of upper-bound risk, especially if a conservatively high estimate of fish consumption is also adopted. In most cases, the best estimate of exposure based on mean contaminant concentrations should be used to develop risk estimates. If upper confidence limits for chemical concentrations are used to develop risk estimates, the effects of compounding conservative assumptions should be evaluated. Exposure pathways and routes are potential mechanisms for transfer of contaminants from a source to a target human population or sub- population. The sources, transport, and fate of chemicals in the en- vironment are analyzed to evaluate exposure pathways and routes. To compensate for a limited database, this analysis often includes mathe- matical modeling of contaminant transport and fate. The modeling of exposure pathways focuses on transfer of contaminants from source to target fishery species, since the transfer step from fishery to humans can be based on knowledge of fishery harvest activities (see below, Exposed Population Analysis). When extensive data on contamination of a fishery is available and source-tracing is not an objective, modeling of chemical transport and fate may be unnecessary. Although the specific uses of modeling in exposure assessment are diverse, several broad objectives may be outlined as follows: • Estimate the spatial and temporal distribution of concentra- tions of chemical contaminants in edible tissues of fish and shellfish • Identify potential sources of contaminants • Evaluate alternative source controls or remedial actions. Estimation of contaminant concentrations in fish and shellfish by mathematical modeling is especially useful when available data on tissue contaminants are limited. If the distribution of contaminants in sediments or water can be estimated from available data or model predictions, estimates of chemical residues in fishery species can be based on relationships of tissue contamination to environmental con- 52 tamination (e.g., laboratory-derived BCFs). Spatial characterization is important for identifying areas of high contamination resulting from heterogeneous transport and deposition of contaminants. Temporal characterization is important for defining time-dependent changes b contaminant concentrations that may mitigate future exposure and risk. Predictions of spatial trends in chemical residues may also aid m identifying and controlling sources of pollutants. For example, when data on sources, sediments, and tissues are available, modeling of chemical transport and transformation processes may help to link the patterns of chemical contaminants observed in the environment with specific individual sources. Information on differential degradation of contaminants and compositional relationships for complex mixtures can be used to support the model analysis (e.g.,'calibration and valida- tion). Finally, modeling of contaminant releases in combination with chemical residues in fisheries may aid in evaluating alternative source controls or remedial actions for waste sites. The results of modeling can indicate the level of source control or remedial action needed to achieve a desired level of environmental quality. In the exposure assessment guidelines, U.S. EPA (1986b) describes general approaches for characterizing sources, exposure pathways, and environmental fate of chemicals. Analysis of chemical transport and fate is a major endeavor, which cannot be addressed in detail here. For additional information, the interested reader should consult Cal- lahan et al. (1979), Burns et al. (1981), Jensen et al. (1982), Mills et al. (1983), Games (1983), Connor (1984b), Thomann and Connolly (1984), Onishi (1985a,b), U.S. EPA (1986b), Pastorok (1986), and references therein. The second stage of the exposure assessment, analysis of exposed populations, includes the following steps: • Identify potentially exposed human populations and map loca- tions of fisheries harvest areas • Characterize potentially exposed populations - Subpopulations by age, sex, and ethnic composition - Population abundance by subpopulation • Analyze population activities - Harvest trip frequency - Seasonal and diel patterns of harvest trips - Time per harvest trip - General activity (e.g., clamming, crabbing, fishing) • Analyze catch/consumption patterns by total exposed popula- tion and subpopulation - -Proportion of successful trips - Catch by numbers and weight according to species - Time since last meal of locally harvested organisms - Number of consumers sharing catch - Parts of organisms eaten - Method of food preparation (e.g., raw, broiled, baked) Analysis of Exposed Populations 53 Comprehensive Catch/Consumption Analysis • Estimate arithmetic average consumption rate by species and by total catch for the total exposed population and for sub- populations. For seasonal fisheries, consumption rates may be estimated on an annual and a seasonal basis. Only selected steps may be performed in a given exposure assessment, depending on data availability, study objectives, and funding Umita- tions. Note that many of the steps to characterize harvest activities and consumption rates apply only to analyses of recreational fisheries. When estimating consumption of fish and shellfish of commercial origin, harvest activities are irrelevant. Also, the specific geographic origin of commercial fisheries products is often unknown. Two approaches to estimating consumption rates are outlined below. In the first approach, a comprehensive analysis of a recreational fishery is performed based on extensive catch/consumption data for the ex- posed population. In the second approach, estimates of consumption rates are based on available values for the U.S. population (or sub- populations) or other assumed values. Most of the available estimates were derived from recall or diary studies (Lindsay 1986) and include commercial fisheries products. It is recommended here that local or regional assessments of fishery consumption be performed whenever possible to avoid possible errors inherent in extrapolating standard values for the U.S. population to distinct subpopulations. Moreover, extrapolation of standard consumption estimates that include com- mercial fisheries products to recreational fisheries should generally be avoided. In developing a profile of the exposed population, there is no single "correct" estimate of consumption rate. Because consumption rates are highly variable, use of a range of values or a probability distribution for consumption rate estimates is recommended. This approach may also be followed when estimating consumption rates for subpopulations of interest. An alternative to the typical practice of basing risk estimates on selected consumption rates involves presenting risk estimates graphi- cally for a wide range of consumption rates that essentially includes all possible realistic values (see below. Presentation and Interpretation of Results). For example, plots of estimated risk vs. consumption rate may be useful for public presentations on recreational fishery resour- ces. In this case, the risk associated with any particular subgroup within the exposed population may be evaluated by selecting a consumption value for the subgroup and reading the corresponding risk from the graphic plot. Use of this approach avoids having to collect extensive data on the exposed population. A similar approach involves selecting an "acceptable" (tolerable) risk level and providing advice on levels of consumption, such that the "acceptable" risk is not exceeded. The advantage of both of these approaches is that consumption rates need not be determined or assumed. Appropriate field survey forms, data analyses, and format for presen- tation of results for a comprehensive catch/consumption analysis of 54 fisheries are provided by Landolt et al. (1985), McCalium (1985), and National Marine Fisheries Service (1986). Details of methods will not be presented here, except to emphasize some important considerations for calculating consumption rates. Examples of analyses of catch/con- sumption data can be found in Puffer et al. (1982) for coastal waters of southern California, in Landolt et al. (1985, 1987) for Puget Sound, in Belton et al. (1986) for New York Bay and Newark Bay, and in National Marine Fisheries Service (1986) and companion documents for other areas of the U.S. Lindsay (1986) reviewed alternatives to field survey methods, including use of food diaries and dietary recall. Gartrell et al. (1986a,b) described methods used by FDA in their total diet studies to estimate rates of consumption of various foods. Note that the results of the FDA total diet studies are of limited use in the present context because fish are grouped with meat and poultry. Estimates of seafood consumption used by FDA to calculate average intake of methylmercury for exposed portions of the U.S. population were based on a diary survey sponsored by the Tuna Research Foundation (Tollefson and Cordle 1986). Sup- plementary information on analysis of fisheries consumption data can be found in SRI (1980). The average rate of consumption of fish or shellfish is the key exposure variable for use in subsequent steps of risk assessment. Consumption rates should be expressed in terms of g/day and meals/year [meals/year may be calculated from g/day by assuming an average meal of fish or shellfish equals about 150 g (0.331b) if the average meal size is un- known]. Average consumption rate for each harvest species is calcu- lated from field data according to the following steps: • For each successful angler trip, calculate the weight of harvest by species based on number and total weight harvested per household • Calculate mean harvest weight consumed per person per time by: - Dividing the total harvest weight for each species by the number of consumers in household and by the days elapsed since last meal from the same area - Multiplying the value obtained in the preceding com- putation by a factor to account for the proportion of cleaned weight to total weight [according to Landolt et al. (1985), this factor equals about 0.5 for squid and crabs, 0.3 for fish, and 1.0 for shucked clams; these estimates should be verified or replaced by local data] • Calculate mean consumption rate per person by geographic harvest area, by subpopulation, and by total exposed popula- tion. Note that the above method (cf. Landolt et al. 1985, 1987) may provide a biased estimate of average consumption rate due to its dependence on a short-term observation (i.e., time since last meal). Averaging of data over a longer time period might be preferable, but such data may be more susceptible to biases from inaccurate recall of consumers (interviewees). Harvest weights should generally be determined direct- ly rather than from length measurements. However, for shellfish and 55 crabs, it may be necessary to establish tissue weights from weight- length regression analysis. The model for calculating mean daily consumption rate (lijk) for fishery species i, human subpopulation j, and area k is therefore: VjlklPi ^'j" " /Vij; ? ^'''" " Mjk ? i/jki Tjk, (4) where: lijkl Nijk Wijkl Pi Hjki Tjki = Mean daily consumption rate of species i for subpopula- tion j, area k, and household 1 (kg/day) = Number of households (successful harvest trips) for species i, subpopulation j, and area k = Weight of species i harvested by household 1 of subpopu- lation j in area k (kg) = Proportion of cleaned edible weight of species i to total harvested weight = Number of people in household 1 of subpopulation j in area k = Time elapsed since last meal by household 1 of subpopu- lation j in area k (days). When consumption rates (lijki) are log-normally distributed, a geometric mean consumption rate may be calculated by log-transform- ing the data before applying Equation 4 to calculate a mean consump- tion rate. Consumption rate data may be summarized further by calculating means across species, subpopulations, and areas. However, it should be recognized that means of Ijjk across species do not represent actual diet patterns for consumers of mixed-species diets. To calculate mean consumption rates for mixed-species diets, all Ijjki should be summed across species within a household before determining mean consump- tion rates across households (Ijk): '- ? e = 2 1 /jjkl (5) N, ijk where: Ijki = Mean daily consumption rate of all fishery species for household 1, subpopulation], and area k (kg/day) Njk = Number of households in subpopulation j and area k and other terms are defined above. Landolt et al.(1985) summarized the assumptions involved in calculat- ing mean consumption rates (lijkl) by household as follows: • Consumption - Pi values are assumed as noted above - Catch was distributed evenly among consumers in household - People in household actually ate the entire cleaned catch 56 - Personal harvest consumption was distributed evenly over the time interval since the last successful trip • Fishing interval - Fishing frequency (days) is related to seasonal fisheries; that is, interviewees did not report average time interval for entire year but only for recent past. Therefore, cal- culated consumption rates cannot be directly extrapo- lated to a yearly basis. Fishing interval was set to 1 day if unreported (Landolt et al. 1985). Despite the hmitation noted in the last item above, calculated con- sumption rates can be extrapolated to an annual average rate by multiplying the liju by 365 days and by a species-specific factor equal to the fraction of the year a fishery is available. Determination of this species-specific factor is somewhat subjective because of large seasonal fluctuations of the harvest (e.g.. Appendix E of Landoh et al. 1985). These factors should be determined on a case-specific basis. In many cases, comprehensive data on fisheries catch and consumption patterns are not available. For some risk assessment problems (e.g., ranking of potential problem chemicals in aquatic organisms or development of consumption advisories) extensive catch/consumption data are not needed. Moreover, catch/consumption patterns undoub- tedly vary over time. Extensive long-term monitoring of catch/con- sumption for all areas of interest within a large water body may not be warranted. Despite its obvious limitations, extrapolating consumption data from one area (or time) to another may be a suitable approach when: • Site-specific data are unavailable • Differences among areas (or times) are expected to be small • Precise estimation of average fish or shellfish consumption is unnecessary to meet the study objectives. In the past, many risk analysts have simply assumed standard values for food consumption rates based on previous analyses of dietary patterns of the U.S. population (U.S. EPA 1980b; SRI 1980). Average values for fish and shellfish consumption for the U.S. population generally range from 6.5 to 20.4 g/day (Nash 1971; National Marine Fisheries Service 1976, 1984; SRI 1980; U.S. Department of Agriculture (USDA) 1984; also see Appendix F). Most estimates include fish and shellfish (molluscs, crustaceans) in marine, estuarine, and fresh waters, but saltwater species form the bulk of consumed items. Most estimates also include commercially harvested fisheries products. Also, estimates of average U.S. consumption do not account for subpopulations in areas such as the Great Lakes that consume large quantities (20 g/day) of locally caught sport fish. An estimate of 6.5 g/day for consumption of commercially and recrea- tionally harvested fish and shellfish from estuarine and fresh waters was used by U.S. EPA (1980b) to develop water quality criteria based on human health guidelines. The value of 6.5 g/day is an average per-capita consumption rate for the U.S. population, including non- Assumed Consumption Rate 57 consumers, based on data in SRI (1980). Consumption rates for por- tions of the U.S. population (e.g., by region, age, race, and sex) show that average consumption of fisheries organisms may vary from about 6 to 100 g/day (e.g., Suta 1978; SRI 1980; Puffer et al. 1982). Finch (1973) determined that approximately 0.1 percent (i.e., the 99.9th percentile) of the U.S. population consumes 165g/day of commercially harvested fish and shellfish. Pao et al. (1982) provided estimates of 48 g/day for the average and 128 g/day for the 95th percentile consumption rates by U.S. consumers of fish and shellfish. Rupp (1980) presented estimates of average daily consumption of freshwater fish, saltwater fish, and all shellfish according to age group within the U.S. population. SRI (1980) presents average and 95th percentile rates of consumption of all fish and shellfish according to age group, race, region and other demographic variables. Estimates of food consumption rates for specific subpopulations in the U.S. are also available from a database maintained by the EPA Office of Pesticide Programs (see Appendix F). Limitations of fisheries consumption data are discussed by SRI (1980) and Landolt et al. (1985). The present status of data on fish consumption in the U.S. is also reviewed by Wagstaff et al. (1986). One or more of the following values of average consumption rate may be assumed when site-specific data are unavailable: • 6.5 g/day to represent an estimate of average consumption of fish and shellfish from estuarine and fresh waters by the U.S.population (U.S. EPA 1980b) • 20 g/day to represent an estimate of the average consumption of fish and shellfish from marine, estuarine, and fresh waters by the U.S. population (USDA 1984) • 165 g/day to represent average consumption of fish and shellfish from marine, estuarine, and fresh waters by the 99.9th percentile of the U.S.population (Finch 1973) • 180 g/day to represent a "reasonable worst case" based on the assumption that some individuals would consume fish at a rate equal to the combined consumption of red meat, poultry, fish, and shellfish in the U.S. (EPA Risk Assessment Council as- sumption based on data from the USDA Nationwide Food Consumption Survey of 1977-1978; see Appendix F). Extrapolation of these values to local populations and recreational fisheries should generally be avoided. Limited estimates of average consumption rates for recreational fisheries are given in SRI (1980). Whenever possible, data on local consumption patterns should be collected or obtained from a current database. Alternatively, risk estimates may be expressed on a unit consumption basis (i.e., per unit weight of fish/shellfish consumed). This latter approach is used by some states in issuing sportfishing advisories. If average consumption values listed above are assumed for local risk assessment, it is recom- mended that a range of values be used. The references cited earlier should be consulted for consumption rate data for fish and shellfi.sh separately, or for individual species (also see references cited in Ap- pendix F). 58 In the next step of the exposure analysis, information on estimated contaminant concentrations and rate of consumption of fish and shellfish are combined to estimate chemical intake by exposed humans. Analyses of single-species diets and mixed-species diets are discussed separately in the following sections. The general model to calculate chemical intake for a single-species diet is: £ijkm — _ Cikm^iik^m (6) W where: Eijkm Cilun lijk w Effective ingested dose of chemical ni from fishery species i for human subpopulation j in area k (mg kg' day" averaged over a 70-year lifetime) Concentration of chemical m in edible portion of species i in area k (mg/kg) Mean daily consumption rate of species i by subpop- ulation j in area k (kg/day averaged over 70-year lifetime) Relative absorption coefficient, or the ratio of hu- man absorption efficiency to test-animal absorption efficiency for chemical m (dimensionless) Average human weight (kg). Values of subscripted terms above may be estimated means or uncer- tainty interval bounds (e.g., 95 percent confidence intervals) depend- ing on the exposure scenario being modeled (e.g., worst case vs. average case vs. lower-limit case). Note that Eijkm is analogous to the dose "d" in Equations 1 and 2. The term "effective" ingested dose (Eijkm) is introduced to emphasize that estimates of chemical intake (i.e., ingested dose) may be modified by the term Xm to account for differential absorption of contaminants by humans and bioassay animals. Absorption coefficients (Xm) are assumed equal to 1.0 unless data for absorbed dose in animal bioassays used to determine toxicological indices (carcinogenic potency or RfD) are available and the human absorption coefficient differs from that of the animal used in the bioassay. Assuming that Xm is equal to 1.0 is equivalent to assuming that the human absorption efficiency is equal to that of the animal used in the bioassay. In the absence of data to the contrary, this is ap- propriate. Toxicological indices are determined from bioassays that usually measure administered (ingested) dose. Therefore, the es- timated chemical intake by humans, Eijkm, is usually the ingested dose, not the absorbed dose. If the toxicological index used to estimate risk is based on the absorbed dose, then an estimate of human absorption efficiency for the chemical of concern may take the place of the term Xm in Equation 6 above. In most cases, however, information or Exposure Dose Determination Single-Species Diets 59 Mixed-Species Diets assumptions about absorption efficiencies has been incorporated into EPA's estimates of RfDs and Carcinogenic Potency Factors. There- fore, Xm is usually dropped from Equation 6 and Eijkm becomes simply the ingested dose. W is usually assumed to be 70 kg for the "reference man" (U.S. EPA 1980b). Assuming other average values to account for growth from a child's body weight to adult weight over a lifetime would not change the results of carcinogen risk assessment substantially. Concerns about exposures over a time period of less than about 15 years may require modehng of early childhood exposures. Standard values for age- specific body weight and other factors used in exposure assessment are provided by Anderson et al. (1985). Estimation of chemical exposure due to a mixed-species diet is com- plicated by variation in the dietary habits of individuals. The various diets of individual humans may differ from one another in the kinds and relative proportions of fishery species consumed. The sum of species-specific exposures (Eijkm) is not equivalent to total exposure for a mixed-species diet. In a diverse fishery, each individual consumer is likely to consume only a subset of the total available species. Thus, the sum of species-specific exposures might overestimate the average consumption rate for mixed-species diets. To estimate average chemical exposure resulting from a mixed-species diet, an exposure dose should first be estimated for each individual in a subpopulation as follows: £hjkm = 2 w (7) where: Ehjkm = Effective exposure dose of chemical m from a mixed- species diet eaten by individual human h in subpopulation j in area k (mg kg" day" averaged over a 70-year lifetime) Ihijk = Average consumption rate of species i by individual h in subpopulation j in area k (kg/day averaged over a 70-year lifetime) and other terms are defined as above. The average exposure dose for mixed species diets is: ijkm = E £hjkn h (8) //jk where: Ejkm = Average effective exposure dose of chemical m from 1 j_..-l\ mixed-species diet for subpopulation j in area k (mg kg' day' ) Hjk = Number of persons in subpopulation] in area k. Uncertainty estimates can be obtained by calculating 95 percent con- fidence limits for Ejkm- 60 Sources of Information References to protocols for sampling and analysis of toxic chemical residues in fish and shellfish are provided above (see Measurement of Contaminants). For the updated status of protocols and new develop- ments, contact a representative of the EPA Office of Water (Appendix A) or one of the EPA Office of Research and Development Laboratories (Appendix G). Information on sampling and analysis of commercial fisheries products collected from the marketplace is avail- able in FDA Compliance Program Guidance Manuals (available from FDA, Freedom of Information (HFI-35), 5600 Fishers Lane, Rock- ville, MD 20857). Compilations of data on concentrations of chemical contaminants in fish and shellfish are available in the EPA Ocean Data Evaluation System (ODES), reports of the NOAA Status and Trends Program (e.g., Matta et al. 1986), Tetra Tech (1985b), and Capuzzo et al. (1987). For current local information, contact a member of the EPA Regional Network for Risk Assessment/Risk Management Issues (Appendix H). Many state health and environmental agencies maintain regional databases on chemical residues in fish and shellfish. For example, the New York State Department of Environmental Conservation and the New Jersey Department of Environmental Protection publish periodic reports on contaminants levels in fish (e.g., Armstrong and Sloan 1980; Belton et al. 1986; Sloan and Horn 1986). The Wisconsin Department of Natural Resources (Bureau of Water Quality) maintains com- puterized records of long-term data on PCS concentrations in fish of the Great Lakes. Summaries of data on contaminant concentrations in a variety of foods are available in Grasso and O'Hare (1976), Lo and Sandi (1978), Stich (1982), U.S. FDA (1982), and Vaessen et al. (1984). FDA is developing a data system called FOODCONTAM for pesticide and industrial contaminant residues in foods. References containing estimates of the rates of consumption of fish and shellfish by the U.S. population were presented above (see As- sumed Consumption Rate). The EPA Office of Pesticide Programs maintains the Tolerance Assessment System (Saunders and Petersen 1987). The Tolerance Assessment System uses a USDA database (based on a 1977-1978 survey) to generate estimates of consumption of various foods stratified by specific subpopulations (e.g., infants, children, and adults in the northeastern U.S.). The Office of Pesticide Programs is also developing information on the effects of food prepara- tion methods on chemical residues in food. 61 Risk Characterization In the risk characterization stage, results of the hazard, exposure and the dose-response assessments are combined to estimate the prob- ability and extent of adverse effects associated with consumption of contaminated fish or shellfish. An overview of the risk characterization process is shown in Figure 6. In human health risk assessment, car- cinogens and noncarcinogens are treated separately. Indices of risk for these different categories of toxicants are based on different dose- response models (see above, Dose-Response Assessment). The procedures for generating quantitative estimates of risk are em- phasized in the following sections. However, it is critical that numerical estimates of risk not be presented in isolation from the assumptions and uncertainties inherent in the process of risk assessment. The risk characterization should include a discussion of assumptions and un- certainties and their potential impact on numerical estimates of risk; i.e., the degree to which the numerical estimates are likely to refiect the actual magnitude of risk to humans. For example, if upper con- fidence limits for mean chemical concentrations are used to develop risk estimates, the effects of compounding assumptions of upper- bound estimates of carcinogenic potency and conservatively high es- timates of consumption rate should be evaluated. A risk characterization should mclude a summary of the preceding steps of the risk assessment: hazard assessment, dose-response assessment, and exposure assessment. The weight-of-evidence classification and other supporting information should be summarized concisely. Ap- proaches to presentation of summary information to be included in risk characterization are presented in the next chapter (see below. Presen- tation and Interpretation of Results). Numerical estimates of carcinogenic risk can be presented in one or more of the following ways (U.S. EPA 1986a): • Unit risk: The excess lifetime risk corresponding to a con- tinuous constant lifetime exposiue to a unit carcinogen con- Carcinogenic Risk 63 Hazard Identification & Dose-Response Assessment < Is Substance Potentially Hazardous'' Physical-Chemical Characterization And Bioaccumulation Potential Environmental Partitioning, Degradation. Transport Mechanisms, and Potential Exposure Media Metabolism and Pharmacokinetic Properties Toxic Effects in Humans and Laboratory Animals Quantitative Relationships Weight of Evidence \ ►T^NoJ ^ Stop .f^ ct <— — o a. _ VI a — © Q = u a <£ E - 5 « 2 e g feu b L> _2 "3 .o n H V 'o. E CS X U « d 1 ^ > od2 "St D J S •o S' >. "Si 5 £ 12 M u n £ 0 " <0 ^ I .9 SH ^^^ <<99=f999 ^'Z.-Z. ZZBJUJCUUJtUU ZZZZZZOv<>OvO;C?;C> ci fs ci ri CN M fs (N rj N M n CQ CQ CQ CO CQ CQ < ^ o *p ^ ^ v^ 999999<<<<<< 3t3t3t;3;aa;<<<<<< f^ fn C*i f*j f*i <*i -'^ -^ "-^ ^-^ -«-v j-^ 999999999999 UUJUUUtUtLltLltLlUJUlU oooooooooooo oooooooooooo o o o o o o o o d d o o r~- r- p- r^ (^ 1^ o o o o o o d: d> d> d> o o 1^ 1^ r^ r^ r^ r^ iSsaB 6 S.E 999999999999 'Osq*r)Tj;popr4'-««-«<><) Conta Rate (mg/k o o o o o o '^^'^ o o o'^'n'^ o d> d> i! oO'-oQt-^tnSp^»oQr^ oooooooooooo On w^ U u C -g S 2 y d8 It should be emphasized that some variables are capable of being measured relatively precisely (e.g., contaminant concentrations in fish tissue), whereas others may only be estimated on an order-of-mag- nitude basis (e^., Carcinogenic Potency Factor). The precision and accuracy of the final risk estimates are directly related to the precision and accuracy of the variables incorporated into the equations used to calculate exposure and risk. Quantitative uncertainty analyses such as sensitivity analysis are easily performed with a spreadsheet by calculating exposure estimates for low, mid, and high values of key variables within their respective plausible ranges. Specification of probability distributions for key variables is an alternative method of uncertainty analysis requiring graphical models (see below, Uncertainty Analysis). In the example shown in Table 6, the average, minimum, and maximum concentrations of each contaminant [PCBs and mercury (Hg)] are used to estimate potential health risk, thereby accounting for uncertainty in chemical analyses. Also, risks are estimated for two consumption rate estimates (6.5g/day and 20 g/day). Note that spreadsheet summaries of quantita- tive information should be supported by a text discussion of qualitative uncertainties such as the weight of evidence for the health effect of concern. Presentation of risk assessment results in graphic form may include: • Plots of estimated risk vs. consumption rate • Plots of estimated risk vs. contaminant concentration in edible tissue of fish or shellfish • Summary maps of risk estimates for different geographic loca- tions or individual sampling stations • Histograms of estimated risk by fishery species, human sub- population, or geographic location. Because estimated risk for a given area and fishery species varies with consumption rate and because consumption rates vary greatly among individual humans, the first approach above is recommended as a primary means of presenting risk assessment results. Actual consump- tion patterns of the exposed population may or may not be estimated (see above. Exposure Assessment). If they are, estimates of average consumption rate (and 95 percent confidence limits) can be identified in a footnote (e.g., Figure 7). Uncertainty in chemical measurements can be illustrated by plotting lines corresponding to the minimum and maximum (or 95 percent confidence limit) values of contaminant concentrations in fishery species, as well as the mean concentration (e.g., each solid Hne in Figure 7). As an interpretive aid, risk assessment results for a reference area can be presented along with those for the study area. Other approaches noted above can be used to supplement plots of risk vs. consumption. Summary maps and histograms may be especially useful for presentation of detailed results of spatial analyses by human subpopulation or by fishery species. Plots of risk vs. contaminant Summary Graphics 69 10-3- OC 10-^- HI CD 5 LU F 10'5- u. 10-6- 10 ,-7 STUDY AREA N- 25 BUTTER CLAMS / ~\^V. REFERENCE AREA N = 25 BUTTER CLAMS 1 (2) 10 (25) 1 00 g/day (250) (meals/yr) CONSUMPTION RATE PCBs Weight-of-evidence classification: PROBABLE HUMAN CARCINOGEN [B2] All cancer risks are plausible-upper- limit estimates of excess risk based on linearized multistage procedure and assumptions suimarized in the text. Solid lines are risks associated with average PCB concentrations in butter clams. Dashed lines are for uncertainty range (e.g., 95 percent confidence limits) for average concentrations of PCBs, not the total uncertainty. Actual risks are likely to be lower than those shown above and may be zero. Figure 7 Example graphic format for display of quantitative risk assessment results for hypothetical study area and reference area. II Risk Comparisons Summary of Assumptions concentration for selected consumption rates and species (e.g., Figure 8) aid in rapid interpretation of tissue contamination data. Interpretation of carcinogenic risk assessment results may be based on comparison of estimated health risks for the study area with: • Estimated health risks for consumption of fishery species from a reference area • Estimated health risks for consumption of alternative foods (e.g., charcoal-broiled steak, marketplace foods). An example of comparison with reference-area risk estimates is shown in Figure 7 above. Comparative risks for alternative foods can be summarized in a table or histogram. Wilson and Crouch (1987) point out the importance of comparing the results of risk assessments with similar assessments of common activities to provide perspective for interpretation of the results by risk managers and the general public. Risk comparisons should be based on consistent exposure analysis and risk extrapolation models. Analogous exposure scenarios should be used for each risk estimate being compared (i.e., either worst case, plausible-upper limit, average, or lower limit). A single model should be applied consistently to calculate exposure and risk. A linear ex- trapolation model, such as Equations 2 and 6 above, is justified in general if the excess risk attributed to the contaminant of concern is regarded as a marginal risk, added to a background of relatively high cancer incidence from all other causes not being modeled (Crump et al. 1976; Omenn 1985). When interpreting the results of risk assessments, risk managers may define an acceptable level of risk to provide a criterion forjudging the significance of potential health effects. The term "acceptable risk" is used to denote the maximum risk considered tolerable by an individual or a regulatory agency. An acceptable risk level has not been strictly determined by EPA. Although acceptable risk levels must be defined on a case-specific basis, some perspective can be gained by examining previous risk management decisions. For example, past regulatory decisions by U.S. federal agencies have allowed environmental risks as high as 10' to 10'" when the exposed population was relatively small (Travis et al. 1987). For exposures of the entire U.S. population, the acceptable risk level has usually been defined as 10 . Assumptions underlying the risk assessment model and estimates of model variables should be summarized in a concise format (see Table 7 for summary of some assumptions and numerical estimates used in the approach presented in this manual). Specific assumptions adopted on a case-by-case basis should be summarized in a similar fashion. 70 10-2 c- w E cc LU o z < o LU I- LU 10-3 10-^ 10-7 Ci* =TISSUE CONTAMINATION GUIDELINE FOR 6.5g.day .0001 I I I I I III I I I I I ml I ^■7 I I I ml I I I I I III! I I I I I III! .001 .01 .10 1.0 10 CHEMICAL CONCENTRATION IN FISH OR SHELLFISH (ppm) Figure 8 Plausible-upper-limit estimate of lifetime excess cancer risk vs. concentration of a chemical contaminant in fish or shellfish (ppm wet wt.) at selected ingestion rates. TABLE 7. Summary of Assumptions an d Numerical Estimates Used in Risk Assessment Approach Reterence Assumption/Eslimales Parameter Exposure Assessment: Worst case for No effect on cooking. Contaminant concentrations in parent compounds. tissues of indicator species Net effect on risk is uncertain. Low, moderate, and 6.5 g/day, 20 g/day, 165 Average consumption rate' high values for U.S. g/day population (see text). U.S. EPA 1980b; 1.0, Assumes efficiency of Gastrointestinal absorption 1986a,b absorption of con- taminants is same for humans and bioassay coefficient animals. , U.S. EPA 1980b; 70 years Exposure duration 1986a,b U.S. EPA 1986a,b 70 kg (= average adult male) Human body weight Risk Characterization: U.S. EPA 1980b, Linearized Multistage, At risks less than 10 : Carcinogenic risk model 1986a, 1987a Risk = Exposure x Potency U.S. EPA 1987a Potency factors are based on low-dose extrapola- tion from animal bioas- say data. Upper bound of 95 per- cent confidence interval on potency is used. Carcinogenic potency U.S. EPA 1987a RfDs for noncarcinogens compared with estimated exposure. Noncarclnogenic risk Estimates of consu mption for local population should be used in place of values shown for U.S. population whenever possible. Other assumptions, such as general approaches or assumptions under- lying models that are commonly used to estimate risk, can be sum- marized in the text of a risk assessment document. Some additional assumptions involved in applying the risk assessment approach described in this manual include the following: • Adverse effects in experimental animals are indicative of ad- verse effects in humans (e.g., Hfetime incidence of cancer in humans is the same as that in animals receiving an equivalent dose in units of mg per surface area) • Dose-response models can be extrapolated beyond the range of experimental observations to yield plausible-upper-bound estimates of risk at low doses • A threshold dose does not exist for carcinogenesis • A threshold dose (e.g., NOAEL) exists for noncarcinogenic effects • The most sensitive animal species is appropriate to represent the response of humans • Cumulative incidence of cancer increases in proportion to the third power of age (this assumption is used to estimate lifetime 71 Uncertainty Analysis Sources of Uncertainty incidence when data are available only from less-than-lifetime experiments) • For carcinogens, average doses are an appropriate measure of exposure dose, even if dose rates vary over time • In the absence of pharmacokinetic data, the effective (or target organ) dose is assumed to be proportional to the administered dose • Risks from multiple exposures in time are additive • For each chemical, the absorption efficiency of humans is equal to that of the experimental cmimal • If available, human data are preferable to animal data for risk estimation • For chemical mixtures, risks for individual chemicals are addi- tive. However, the total sum of individual chemical risks is not necessarily the total risk associated with ingestion of con- taminated fish or shellfish because some important toxic com- pounds may not have been identified and quantified. Uncertainty analysis is an integral part of risk assessment. The EPA guidelines on exposure assessment describe general approaches for characterizing uncertainty (U.S. EPA 1986b). Methods for uncertainty analysis are discussed by Cox and Baybutt (1981), Morgan (1984), and Whitmore (1985). A detailed discussion of procedures is beyond the scope of the present effort. General approaches to uncertainty analysis will be described briefly after a discussion of sources of uncertainty. Uncertainties in the risk assessment approach presented in this manual arise from the following factors: 1. Uncertainties in the determination of the weight-of-evidence classification for potential carcinogens. 2. Uncertainties in estimating Carcinogenic Potency Factors or RfDs, resulting from: • Uncertainties in extrapolating toxicologic data obtained from laboratory animals to humans • Limitations in quality of animal study • Uncertainties in high- to low-dose extrapolation of bioas- say test results, which arise from practical limitations of laboratory experiments and variations in extrapolation models 3. Variance of sitespecific consumption rates and contaminant concentrations 4. Uncertainties in the selection of 6.5 g/day, 20 g/day, and 165 g/day as assumed consumption rates when site-specific data are not available 5. Uncertainties in the efficiency of assimilation (or absorption) of contaminants by the human gastrointestinal system (assumed 72 to be the same as assimilation efficiency of the experimental animal in the bioassay used to determine a Carcinogenic Potency Factor or RfD) 6. Variation of exposure factors among individuals, such as: • Variation in fishery species composition of the diet among individuals • Variation in food preparation methods and associated changes in chemical composition and concentrations due to cooking. Variance in estimates of carcinogenic potency or RfDs (#1 above) account for one major uncertainty component in most risk assessments. Chemical potencies are estimated only on an order-of-magnitude basis, whereas analytical chemistry of tissues is relatively precise (on the order of Jl20 percent). The choice of a low-dose extrapolation model greatly influences estimates of the Carcinogenic Potency Factor and calculated risks. This uncertainty contributed by the model is substantial when predicting risks below 10' . For example, the plausible-upper limit to lifetime cancer risk associated with 50 /e«lcal froftla Effect* Aaaea—ent Accnaphthene 1 Acenaphthylcnr X Acetic arid X Acetone X X Acrolein X Acnrlonltrlle X Aldrin X Anthracene X AntlBony X Araenlc X X Aabeato* X X Barlua X X Benzene X X Beoaldlntr X Benio(a)anthracene X &enzo(a)pyrene X tenzothlazole X BcryllluB X alpha-BHC X beta-BHC j Xaaaa-BHC (lindane) X X delta-BHC X Butane] X Butvl acetate X Cadeiur X X Carbon tetrachloride X X cla-Chlordane X X trans-Chlordane X X Chlorine X Chlorobenzene X X Chlorobcnzliatc X Chloroethane X Cliloroforv X X p-Chloro-»-cresol X l-Ctiloro-3-nltrobenzene X blt(2-Chloroethoxy)ethane X Chroalua (total) X ChroaluB (hezavalent) X ChrooluB (trlvaleot) X Chryccne X Coal tars X Cobalt X Copper X X Creaol X X Cyanides X X Cyanurlc acid X p.p'-DDD X O.p'-DDD X p.p'-DDE X p.p'-DDT X X o.p'-DOT X X Olbrofsochloropropane X 1 .Z-Dlchlorobenzene X 1 ,3-DIchlorobeniene X 1 ,4-Dlchlorobentene X 1 , l-Dlchlorocthane X X 1 .2-Dlchloroethane X X 1 ,1-Dlchloroethylene X X 1 ,2-cl*-Dlchloroethylene X 1 ,2-trans-Dlchloroethylenc x X 2,ft-Dlchloropheool X 2.4-DlchloropheDoxyacetlc acid X t ,2-Dlchloropropanc X TABLE C-1. (Continued) Chealcal OWPE Chealcal rrofllc OEU Health Effects Aaaeiaaant I ,3-Olchloropropane I ,]-01chloropropcnc Dlcofol Dlcldrlo X Diethyl benzene X Dlethvlene glycol X Diethyl phthalate X Dllsobutyl ketone X DlBethyUnlnoethyl aethacrylatc X DlBetbyl aolllne X Dlaethylnltroaaalne Z 2,4-Dlaethyl pentane X 2.4-DlBethylphenol X n-Dloctyl phthalate X 1 ,4-Dloxane X Dlphenyl ethane X Endrin Z Ethanol X blc(2-Chloro«thyl) ether X Ether X Ethyl acetate X Ethylbenzenc X Z Ethylene glycol Z Ethyl hexanedlol Z bls-?-Eth.Tlhexyl phthalate X Ethyl toluene X Fluoranthene X forealdehyde X Glycol ctherc X Heptachlor X Heptane X Hexachlorobenzcne X X Bcxachlorobutadlcne X X Hexachlorocyclohexane I Hexachlorocyclopentadlene X Hexachloroethane X Hexacblorophene X Rexanc X Iron X X Iiobutyl alcohol X Isopropyl benzene X laopropyl ether X Lead X X Llthius Z Kagneslua X Manganese I X Hercury X X Methacryllc acid X Methanol X Methyl chloride Z 2-Methyl dodecanc Z Methylene chloride X X Methyl ethyl benzene X Methyl ethyl ketone X X 3-M«thyl hexane X Methyl Isobutyl ketone X Methyl aethacrylate X Methyl parathlon Z 2-Methyl pentane Z 3-Methyl pentane X 2-Methyl 2-Methyl 2-Methyl 1-pentene tetradecane trldecane TABLE C-1. (Continued) OWPE OEU Health Ch«Blc«l Chflctl Profile Efftctt Astc«»»gnt Koncthanolaalnc j lUphchalcnc X Z o-Xylen« X p-Xylcne X Zinc X Rltrocclluloi* 2-Nltrophcnol Pencachlorophcnol X Pcntadacanr n> (0 4-1 4-* s i- (0 "8 M 8. .c "? 4-> o 0) 1 (A ? (0 s. 1 flj to (A X 01 (A U) >. >^ 4-' o L. L. Oi 4-' c 0) u V ^ "E 4-* o 4-' (D £ 4-» 10 10 1- 8- (A 4-» (- 3 £ a g o> 0) 0) o ? o 4-' o c tA 10 < (A ^ 0) (A > > c c f £ c_ 10 o (D (0 > c- s > t-f OJ g ■o c ? t^ 01 0) (A ID o u j: 4-» 4-» (J U f (A f_ 0) "S c 0) ■8 j: ^-i ♦^ 3 H- *j 10 'i > ^ 0) H- 01 c 10 4^ o 10 .c UJ 1- H- c o 01 0< 2 .c o c a> to U) 'i w £ 0> 10 10 a; ^ § ? 4-» .c Ui s ^ s 0) J3 ^ c <0 > 0) <^ 01 0) t_ « JZ *-* o> »*- »^ 4-* u o "S o c 01 l_ u (/) H- % a 0) t-l 4-* o 10 c 3 u c 3 o o 8 3 10 c *j *4- t/} c o *♦- ♦J ^ 01 o o (_ o t- ■§ 4m> o ^ (J (D 4-> >- ^ s^ 41 10 c 4-» 8 1 ? o 4-> ^ o £ 4-' 10 c 4^ o X S ^ ^ t_ 8 O ^ 4-1 u tj « c_ 01 o o ■? 10 •£ 2 (A s 2L s 01 10 c 1 a *-» g X O) n s. lA g to * (A c 10 >- i-f o > m O L. 4-* 4-> t-l c 10 L> *-* o c ^ & 10 u (D (0 c 0) 4-' >. 1 JJ (J £ D *-» "8 1- l_ l_ o ^-l 10 jC 0) 8. •o (0 c 1 8- ^ u *-» S. F o o o (D 0) (0 x: t- 01 »♦- g H- >^ o o t. H- o £ *4- a o (J L ^J o s. o u a> >. >. >~ »♦- 19 8 m >^ i_ L. t. E a> [_ 3 t_ 0) ^ (A ? Cfl H- c [_ X CO 3 CO o — ' a to Crt 0» O t- *" -mm <0 ^ § ^ I o « "^ CD ^-i 0; o >•- S ■□ o *-> c c 10 4^ 8- - -8 U 0) — t_ O CA o 8. a. 0) > ^ 0> tf) c o — ' i "8 > a> 1- tf> 0) o w T3 -e ^ ? s. U) c >. 0) ^ 4-' « 10 01 3 *-• 10 10 3 > t g - 4-' — ' ro 4_> a> £ -S < UJ O 0) 4-* Oi o (A s. 3 4-> -g ^ , (0 3 g CO IT ^ (- o 01 l_ *D 4-< Q. 10 C 3 s a> ? '4- O 4-' 01 4_, tA .c tA QC *-* QC s c s ^ "" "g (. o s g o *•- 4-» 4>J <1> 0) lA i (A L_ (A s. — • (- o < 0) V H- ■g 1- £ c a 10 4-* < ? 10 4) .- o c •- c at 3 '^ O) o 4-» O 10 91 o ■z -i o ••- ^^ c a 3: 01 ■o t_) (- 10 **• M 0) o 10 l_ z H- Q. o 10 1- ID o a. U s c o 1 £ j: CJ 4-' u — 01 > — c — ID a. •»- 4) •». *-> (ft — 0) 10 in o> Ifl < 00 W on < < < Q < U o o -] o u >< o Q Z < < i w u Q u H U u u w CQ < ^ V) 5 5 "o o . u U) £ (TJ 1- O o V 1 1- ^ o s ,^ o v> E « lA •4- *— O Jr to t- L. £ "g —J u > (D V g in 5 4'' ? (0 z «-• re o o CO < "S _J O LU z o ^5 O il l_ t- o o c **- •♦- (0 c o *-• u 1 4-* (/) *-» CA (/) ra >^ *CTI "(O g- CO a> u < CJ flC •— (D c - i UJ > 0) i/i JZ a: 1 u U t- LU 3 4-* 0} u QC O < a: IS fM i N (A U) OsJ < •□ r- t- s s R 3 a£ CO O" ^ U s. u 1- L. t_ ^ c m o fVJ z >- ,= «, c in 4_> CD 00 s o fvj ra o u. f^ ffi N-* M u a V* c o u X ■M 44 ^ 0) "S :q (0 ^ o 3 (0 o a> "O (- (- l_ c *-» «) 1_ o o (t) yi (U u o u JI « s- o 0> O 2 1 c •M (- ;_ o o £ UJ o L. u u u o o e £1 o c o U to 4) (0 tl] a o JZ 4^ o" t- o ♦J t_ o o !5 — ' Q LA (0 M s o >s c o *-» ro 4) ^ u w H- t u 1 at **- 5 o s i 1 . t_ 0) o *-» u 4-* 10 > nj (A c u 4-> u o (U u o >- 1 ^ 'i o c 1 ■g (- CJ o <0 o £ (fl u 0) (J (- *-» V o u o iA > t-t in 1 u •? H- —1 £ o *" (0 0) 4-« -3 I- ■8 >< o 4J s c 1 3 u y) >- •S ^ ^ o (0 (D «-* crt ♦^ u 10 g u u c 1 ni 3 U) )< CO (0 c ro Ji o < 0) > < CD u y— u u o CI 5 g 4-« 3 CO CO > 9 UJ > .s 10 z u «J « 1 « u 1 VI f 9 u (/) a UJ « o M o o \r\ >. & 10 H- o o C Z i "8 - ^ "8 o re o c £ i; o - "8 (JO ■•- .2 ^ c m *> re E JS 1- O 3 o ]c (A o Cft s U) ^ . ^ (D re N «4- 4-* 1 to o re JZ T) u + u >. o 4-' o O i o 8. D X H- V- o o 1- a 4-* o ^ **- o Ul « re 4-' «-* u 3 s « i S" LU **- ^ o o "S « "E Z) m >. c 8 J3 u (A (A to ? [_ gi f- o 1- v 4^ g 4-* re § 0) 4-* 3 in U) in >- C re < ~ o CO g re < 0) re z M r CO re 3: TO re 3= O z Cl 5 i 4-> o -o 1 I u <-• rj ^ a> in "D lA « o <-> a. >- 3 O ^ s o Y , m tA O 03 o (D ■M (/) C a; c QC O 5 5 c UJ g l- ra ID UJ 4-* >. ^ ID a "2 _§• u u (D 0) >■ to "o £ 10 O) L. 4-« u *-» yi rtj in c u ^ JS S ro UJ < en »— c — — O u UJ Appendix D Evaluation of the Effects of Composite Sampling on Statistical Power of a Sampling Design Tetra Tech (1986b) used simulation methods to make a direct com- parison of grab and composite-sampling strategies. Simulation refers to the use of numerical techniques to generate random variables with specified statistical properties. For the analyses described below, Tetra Tech (1986b) developed computer programs to: 1) produce individual random samples from populations with normally distributed con- centrations of contaminants, and other statistical properties similar to those of historical bioaccumulation data sets described in Tetra Tech (1986b), 2) construct composite samples, and 3) calculate statistical power of sampling designs using individual or composite samples. Two sets of analyses were performed by Tetra Tech ( 1986b) . In the first set, simulation methods were used to show the effect of sample com- positing on the estimate of the population mean. Power analyses were used in the second set of analyses to demonstrate the effect of increas- ing the number of subsamples in a composite sample on the probability of detecting specified levels of differences among stations. The first set of analyses demonstrated that the confidence in the estimate of the mean increases as the number of subsamples in the composite increases (Figure D-1). The simulated sampling consisted of randomly selecting 10,000 composite samples from two populations exhibiting two different levels of variability in the sampling environ- ment. The mean value in both populations was fixed at 18.52, but the population variances were set at 70.90 or 354.19, corresponding to coefficients of vju-iation of 45.5 and 101.6, respectively. These popula- Analysis 1 Mean(n) = 18.52 Coefficient of Variation = 45.5 Variance { o^) = 70.90 30-1 25. Z < o UJ 15. < o o o - -] r - \^ 1 1 1 1 4 6 10 20 NUMBER OF SUBSAMPLES IN COMPOSITE Reference: Jet ra Tech (19£ 56b) Figure D-i Effects of increasing composite sample size on confidence in the estimate of the mean. don characteristics were selected as representative of the range of values for the coefficient of variation observed in the historical data sets for selected metals and organic compounds in marine organisms (TetraTech 1986b). For a series of individual fish samples taken from the corresponding populations used in Analysis 1, the 95 percent confidence intervals would range from 1.7 to 35.4 concentration units (e.g., ppm). To demonstrate the effect of sample compositing on the power of the statistical test of significance, Tetra Tech (1986b) performed statistical power analyses using a one-way Analysis of Variance (ANOVA) model. In these analyses (Figure D-2), the number of stations (5), number of replicate composite samples at each station (5), significance level of the test (0.05), residual error variance level, and level of minimum detectable difference (100 percent of overall mean) were fixed. The power of the test (i.e., the probability of detecting the specified minimum difference) was then calculated as a function of the number of subsamples constituting each replicate composite sample. Power analyses were conducted for three levels of sample variability. All design parameters except the residual error variance were identical in each set of analyses. Values of the residual error variance were selected to represent the range of values found in the historical data sets described by Tetra Tech (1986b). The coefficients of variation selected for these three sets of analyses were 45.5, 101.6, and 203.5. As shown in Figure D-2, the probability of statistically detecting a difference equal to the overall sample mean among stations increases with the collection of replicate composite samples at each station and as the number of subsamples constituting the composite increases. The results of both sets of analyses shown in Figure D-2 also demonstrate the phenomenon of diminishing returns for continued increases in the number of subsamples per composite. In Analysis Set 1, for example, virtually no increase in the power of the statistical test was achieved with increasing the subsample size above three. In the second analysis set, substantial increases in statistical power were achieved by increas- ing the number of subsamples in each composite from 2 to 10. However, with each successive increase in subsample size, the relative benefit was reduced until very little was gained by increasing the subsample size above 10. For moderate levels of variability, 6-10 subsamples within each of 5 replicate composite samples may be adequate to detect a treatment difference equal to 100 percent of the mean among treat- ments. At the highest level of variability analyzed, the collection of replicate composite samples composed of 25 subsamples each is re- quired to obtain a testing power of 0.80 (Figure D-2). Analysis 1 2 3 Coefficient of Variation 45.5 101.6 203.5 LU 1.0 n 0.8 0.6 0.4 0.2 0.0 «^-« T 1 1 1 r- 9 10 11 12 13 3 4 5 6 7 NUIVIBER OF SUBSAMPLES (a) -I — I — I 14 15 16 1.0 0.0 -1 1 1 1 1 1 1 1 1— I 1 10 12 14 16 18 20 22 24 26 28 30 NUtVlBER OF SUBSAMPLES (b) Reference: Tetra Tech (1 986b) Figure D-2 Power of statistical tests vs. number of subsamples in composite replicate samples. Fixed design parameters; number of stations 5, number of replicates = 5, significance level = 0.05, minimum detectable difference = 100 percent of overall mean value. Appendix E Evaluation of the Effects of Sample Replication on Statistical Power of a Sampling Design Statistical power analysis can be used to evaluate alternative sampling designs with varying levels of replication (Cohen 1977; Gordon et al. 1980; Tetra Tech 1986b). In statistical power analysis, relationships among the following study design parameters are evaluated: • Power - Probability of detecting a real difference among treat- ments (e.g., species, stations, times) • Type I error (a ) - Probability of wrongly concluding that there are differences among treatments • Minimum detectable difTerence - Magnitude of the smallest difference that can be detected for given power and Type I error • Residual error - Natural variability • Number of stations • Numlier of replicate samples. The analyses presented below were conducted with the objective of providing guidance in selecting levels of sampling replication. This objective was addressed by determining the magnitudes of difference among variables that can be reliably detected with varying levels of sampling effort. A one-way ANOVA model was used to evaluate statistical sensitivity relative to level of sample replication. Tetra Tech (1986b,d) provides details of the ANOVA model and results of the analyses. All power analyses were conducted using the Ocean Data Evaluation System (ODES) maintained by EPA's Office of Marine and Estuarine Protection (Tetra Tech 1986d). The measure used to evaluate the statistical sensitivity of the monitoring design was the minimum detectable difference between two mean values. To general- ize the results of the power analysis, the minimum detectable difference was expressed as a percentage of the grand mean among treatments. The power of the test was fixed at 0.80. Predicted values of minimum detectable difference are shown for various levels of sample replication m Figures E-1 and E-2. For these analyses, the Type I error was fixed at 0.05. Minimum detectable difference was plotted vs. number of replicate samples for the following cases: • Number of stations (or sampling times) equal to 4, 6, 8, and 16 stations (or times) • Data Variability Coefficient (across treatments) equal to 30, 50, 70, and 90 percent. The Data Variability Coefficient is equal to the within-groups mean square divided by the grand mean among groups (and multipHed by 100 to convert to a percentage). In designing a bioaccumulation study, the Data Variability Coefficient can be estimated by performing an ANO VA on available data from the literature or on a preliminary data set. If such data cannot be obtained, the average Coefficient of Varia- tion (within groups) can be used as a rough estimate of the Data Variability Coefficient. The effect of setting a different value for Type I error is shown in Figure E-3. The effect of changes in Type I error is greater for higher levels of data variability. Note that substantial increases in sensitivity (i.e., decreases in minimum detectable difference) are achieved only for the case of three replicate samples in Figure E-3. o LU o LU QC LU LI. LU _l CD < H O LU I- LU O 550 500- 450 400 350- 300 it 250- 200 150- 100- 50 - Data Variability Coefficient Number of Stations 4 6 90 70 50 30 4 6 8 10 12 NUMBER OF REPLICATES 14 16 Reference: Tetra Tecti (1986b) Figure E-i Minimum detectable difference versus number of replicates at selected levels of unexplained variance for 4 and 6 stations. Power of test = 0.80, significance level = 0.05. 550 n 500- 450 400 350 300 < o LU o z LU DC LU Li. — 250 CI LU CO < I- o LU h- LU Q 200- 150 100- 50 Data Variability Coefficient Number of Stations 8 16 90 70 50 30 4 6 8 10 12 NUMBER OF REPLICATES 14 16 Reference: Tetra Tech (1 986b) Figure E-2 Minimum detectable difference versus number of replicates at selected levels of unexplained variance for 8 and 16 stations. Power of test = 0.80, significance level = 0.05. o z cc Ul 11. 5" O lU -J u. m o < H I- O Z UJ HI H O LU £E Q UJ Q. S^ 3 -• 3 REPLICATES -O 5 REPLICATES -, _ . (T3 ■— ■ O XI Q.I — QJ E > O fO 3 X> "D , ' N QJ ■^ > -c: W I- CZ CO 3 QJ -^ Q) Crt 1- U- ■— Tl ■— CL y) ' — ■ — E - — OJ (O ^ ^ ^ CO t— u y) ■ O T3 QJ I TJ I- CD T3 — ■ J= OJ ■" ■* a « o .^'^ ■=-^ ^ • i * := o ° -o -r-i OJ OJ — -□ "O ^ .^ — TO C\J (J -C "O QJ >i u N QJ ,^ a. in > — B — "a o M .? E TO <; QJ t/i 1- =3 CO -Q W WJ QJ QJ "O CD T) -*-» 4-J cn >. tz o S- QJ -.- 03 E ■M > QJ Q. TO E >- E -D E 3 4-> 3 -^ X +-> fO Q.Q E 3 >. 00 Q) c > O i- (_) 3 C/5 0) — iTJQ.ro t- E " Ol Q) E >» 1— • CO I—I r^ U- __^ O) oo LTi r-H o CO 1 O) r-- .-H rv. c o -D XI tu fO -M E u E >> o =3 o QJ QJ t_ C TJ QJ ^- e — O -C J u O) -Q >r* 1 (U 0) > ■!-> u > 0) E o ^ QJ UO « ^ — QJ LT) . 4- JZ '- -*-< "D -Q TO TO C Li_ O , OO E — C < 0) — O -C U_ C/1 w ^ oi^ in ■ ^ C TO CO TO 3 "O ^ .H > t < +J — o O C "D CL CO O C OJ ^ (_> — ' Qi - to LO 00 -in CO 00 «. ?^ <1^ '^ o s.^ > »- TO TO 5 « ro-S ^ : QJ QJ o XI . 3 (J >> ™ : <— C TO 3 XI QJ ' O^— -l-i- O t- ^-.-^, yi 1— +J J-i i<_ 1- — ' -o 3 ■ • ■— TO 4-. 3 Q_TO.£3 — DIOO o — (0 QJ ^ O) C '- -O 5 M- — < Sfi -Q Q. -M -*-■ QJ CJ (_ QJ C n3 ' — TJ Q. 0) +J C I_ -C QJ -D U .C Q) i 2 ° ?i i £ TJ QJ - 0) 00 •MB)"' ■ fc " i ■ o '^ 5 — • o >-e: m en > S- n> E E 3 ■M O E c =J TD ITS W (U ■'- O 1- Q) O "D E " * m ,„ N^ -M O t/1 +J 03 (J fO >> 3 <— 4-> o fO OJ en-— fO E O. > -.- to -»-> C o (_ •4- c c TJ 3 o OJ t/l c — yi 0) Q O 4J 0) O CI 2 E oi ■"a- Q. L 0) to E u ct oo 3 QJ V) (0 U OJ "d- C o u I- -C -c II " fc ■D o O X- u 0) V} <4- o « '*" 14- >. 0) 0) c t/) > o (- 3 ■M (A Q. c E 3 o 00 (/) c c -^ £ o -^ T3 t4_ OJ ^ c +-> — 4_i 0} rO 0) a. (d I- E T3 CX> lO CM —t CO 05 CO CNJ •*-> Q. E 3 (/) e O (0 «_> 4-> n3 -C Q (/I •-- >. > cz o s_ OJ -^ •• Z3 10 Q. > '~.U-> t- E *o (U =J -D o u (U <0 >1 OJ (U .— 0) ' ro ;;; ^ QJ E y) c >1 4^ tJ (- o t- c yi u <0 3 Q) (1> 0] OJ > Q QJ a: 0) X (/J c r— ^— r— r— LkJ ,-^ ' — ■M o OJ o ■M -M -M -M E (/) to ■4-J E o (- 4- c C C C C 03 c C U o c (U OJ QJ QJ u x: o QJ QJ c . o U U U 1- 4- u U 1/) (1) (1) OJ QJ -Q 3 fO OJ T3 0) C QJ T3 .1- ^ x: QJ (X c: QJ X X +-» ■*-> -«-> 4-> TJ m -4-' 4-> *N O D o o in cn c c LD 4-1 (0 QJ 5 « C iT) cn cn cn XI o cn -D 4-J o -»-» . . O o QJ *o to a: c O E UJ to XI u -D (U ■M o z LT, TJ 0) 4-> O 3 O 3 O CO - c in o c r^ c C\J u t/> , 4-> 4-> I- 0) c (0 X u 0) u o -o Q) fO -C 1— ^^ cn QJ (/) ■f- *" x- !2 CO to X 4-» ■o c X u c Ll_ c o ■M U (0 oo S- o to ■o QJ 10 QJ ^ u QJ 3 . QJ X c >, 0 -- ^- §5 £5 ££• ^ \^ SI cn 4_i cn cji QJ cn cn Ol c E 4- QJ 1- ■o (W LU o QJ 3 QJ u i*- (_ -Q Ol 4-» QJ > QJ E QJ >-i 14— U fO X O u ■«-" c QJ >> OJ 14- •0 fO cn QJ X (J c Q. ro > ►— OJ i_ t- "O CO u QJ 3 OJ Q) "O u > V) > U QJ CO (O Q) 3 ■M 4) ■M u 01 to T3 Q) E O QJ X 13 c -o u- ' — fO 4-» (/> O O i- Q. 3 -M ■— -C ^ U Ol (0 X Z3 Qi 03 O J^ tn 1 c ^ tn o » E E a c o a. E 3 (/> c O nJ O -M ^ O I/) •-- >i Ll_ 01 > i- •• 3 '— .OO *- O 10 Q) t; °- Ol M X « .X 3 •u ^ j: nj S O w c cn t- CO O **- to E 4-> > -*-> > 10 a. (Q 1- e -D (u a ^. > tn o < c o o E *-• ' — x; "o 3 yi ro CD CD fO O) rs to LT) ^r o CO 00 CO to CJl CD CD CD >1 0) ■^2 UJ o u c •— o z — gj > ^ C iO t- V) m CT o C GJ -r- O -!-> U « W fO U C -M Q> O C Q_ t/> O -M -C Crt « ^— 4- 2 +-» ca t/) O c o 3 CO O CO 0) S U 8 ^ 00 3 >» ae 8- u O 8 ~ eo O X 2 — INI 8: KJ o 3 ?J S S S >» •o in fc CO ift •o fc S S e s § fSI :s 3 <>• IM CO 3 o •o 2 i rj ■» IM Si! § o >o B^ 3 o = 5 5 8 § 5i 8 ^ o o 8 it. ^ 8 1^ ^ X 13 s ■o o O) ■<-> 3 00 c o 3 c o s — to 5-^ 3 ^ 8^ 8 ^ CO < • -J o o — o on c o 3 c o X5 s 3 -* 4/» •- u. O o 8' ~ oo 2 — t\t < St! 8 S S <3 O O o o 55 —I o^ S8 o C3 a. 8 5i 8 5 o CD 3 3 o 8 8 ►^ to 5 -o o u. O o g' a CM rg o » o o. 3 •O Kl •* :$ ■o in in 3 r^ ;5 ;s S I s IC f2 S § s eo S ■o 8 8 8 - « a. L-l. 3 ^» o CO m c in w~ Z3 UJ 1 -C rg o "" Ll_ <-» M- O o 3 O 8 c ^ eo t>- o < 'O .^ d ■4-> j? Q. 1 5 E ^ -o •o 3 kk. d i/l o c X o ^* . o 1/1 d a> UJ ^ o o o (/) .^^ s o 'I' >, S r— ',^ "» « E •r- O (TS <+- s- - — o o > 00 o CM o 00 in > to o CO o 00 ■o o •a T3 "O c ta ■(-> .c a> >> ■o o C x> O) o CO O) a cn-'- in ^ 0) CVJ \. s >> o (T3 > ■o -o ^-, o cn.^ CM C3 CO (/) CI as S- O) > (B r— <: +J n- -,- a> LT) O..CZ I— <-> ^ m <: +-> -1- 4-> 00 a. fo et n3 0) h- O E •a: s- -o Q- QJ tU LlJ Q- Qi >. S- -t-> 0) ^— CT> =3 (O o s- o. 0) > OS + < -•-> •-- +-> 00 Q. nS <: OS QJ 1— o 2: < s- -o Q. (U a. c o Q. £ c o u 0) a> to i~ 0) > to i- o > > oo r; a) CVJ CO CO ™ s- o ^ 4- Appendix G EPA Office of Research and Development ENVIRONMENTAL RESEARCH LABORATORIES Region 1 Environmental Research Laboratory/ORD South Ferry Road Narragansett, RI 02882 FTS: 8-838-5087 DDD; (401) 789-1071 Environmental Research Laboratory/ORD Sabine Island Gulf Breeze, FL 32561 FTS: 8-686-9011 DDD: (904) 932-5311 Environmental Research Laboratory/ORD College Station Road Athens, GA 30613 FTS: 8-250-3134 DDD: (404) 546-3134 Environmental Research Laboratory/ORD 6201 Congdon Boulevard Duluth, MN 55804 FTS: 8-780-5550 DDD: (218) 720-5550 Region 4 Region 5 Region 6 Region 10 Environmental Ecological and Support Laboratory/ORD 26 W. St. Clair Street Cincinnati, OH 45268 FTS: 8-684-7301 DDD: (513) 569-7301 Center for Environmental Research Information/ORD 26 West St. Clair Street Cincinnati, OH 45268 FTS: 8-684-7391 DDD: (513) 569-7391 Robert S. Kerr Environmental Research Laboratory/ORD P.O. Box 1198 Ada, OK 74820 FTS: 8-743-2011 DDD: (405) 332-8800 Environmental Research Laboratory-Corvallis/ORD 200 S.W. 35th Street Corvallis, OR 97333 FTS: 8-420-4601 DDD: (503) 757-4601 Pacific Division - Environmental Research Lab/ORD Hatfield Marine Science Center Marine Science Drive FTS: 8-867-4040 DDD: (503) 867-4040 ]32 Appendix H Compilation of Legal Limits for Chemical Contaminants in Fish and Fishery Products TABLE R-1. COMPILATION OF LEGAL LIMITS FOR HAZARDOUS METALS IN FISH AND FISHERY PRODUCTS Metals (ppm) Country As Cd Cr Cu Hg Pb Sb Se Zn Australia 1.0,1.5" 0.2-5.5 10-70 0.5,1.0 1.5-5.5 1.5 1.0,2.0 40-1,000 Brazil 0.5^ Canada 3.5 0.5 0.5 Chile 0.12,1.0 0.5 10 2.0 0.05,0.3 100 Denmark 0.5 Ecuador 1.0 10 1.0 5.0 Finland 5.0 1.0 2.0 France 0.5,0.7 Germany 0.5 1.0 0.5 Greece 0.7 Hong Kong 1.4-10 2.0 1.0 0.5 6.0 1.0 India 1.0 10 0.5^ 5.0 50 Israel 0.5 Italy 0./ 2.0 Japan 0.3,0.4' Korea 0.5 Netherlands 0.5-1.0 1.0"^ 0.5,2.0 New Zealand 1.0 1.0 30 0.5' 2.0 1.0 2.0 40 Philippines 30 0.5 0.5 Poland 4.0 10-30 1.0-2.0 30-50 Spain 0.5 Sweden 1.0' 1.0-2.0 Switzerland 0.1 0.5 1.0 Thailand 2.0 20 0.5 1.0 United Kingdom 1.0 20 2.0-10 50 United States 1.0' U.S.S.R. 0.2-1.0 Venezuela 0.1 0,0.1 10 0.1-0.5 2.0 Zambia 3.5-5.0 100 0.2-0.3 0.5-10 100 Range Minimum 0.1 0 1.0 10 0.1 0.5 1.0 0.05 30 Maximum 10 5.5 1.0 100 1.0 10 1.5 2.0 1,000 ^ Limit varies among states. Inorganic. ' Total. References: Nauen (1983); U.S. Food and Drug Administration (1982, 1984). > ai u s Q S Z CD u Q H U CL. Q 1/3 -J 0£ O a. U >1 o 2 £ X a o -S *i E E a OS o o :^ .J O z o I Qu o u s o. 1/3 H u Q O a. u ^ c ■c 2- < 15 Is o o o .-> o — o o o o " O li-l — • o o o o o p p o o — o o ri o to o o o d o o p p o o p p o o I § 5 2 ,.Q . ^ CO o o o pop wS ci ^ o o 6 C e C V *^ ** U O u u > s Q O ii Z (^ 1^ •o U1 1^ O O O O d o 3 3 E E &2I 2 g c -w b£ j:: .y E e 5 3 £f •&I « o E < SO a Q D S. := <2 5 "^ T a .= "c — 1/1 w u See i; g:^:j ^^ VO •• <» CO I-- c <-*> V) £ J' ja* do i-Hg 9 DOCUMENT LIBRARY Wools Hole Oceano-jn-thic i.i,4rjtion rf^j ^ >r 1.1 ^_ -•t 00 CD 3-= 3 ffl -^ 3 w s* 3< -> 7 5- 5