A Review of the Biomonitoring of Environmental Status and Trends Program: The Draft Detailed Program (1995)

This chapter addresses the question: Will the program be able to identify causes (including identification of pathways and sources) of observed changes (trends and threats) in trust resources, whether they are human causes or natural fluctuations?

Two sections of the detailed plan address causality: Section 2 on the strategy of data collection and Section 5 on the strategy of impact assessment and prediction (FWS, 1993). Subsection 2.1 lists four general categories of bioassessment methods that will be used to determine the presence and effects of toxic substances on local to global scales (see Figure 2-1). Subsection 2.2 describes a two-tiered approach to bioassessment. Tier 1 (Table 2-3 in the detailed plan) is to serve as a general screening for existing and potential contaminants using inexpensive methods, and Tier 2 (Table 2-4 in the detailed plan) is to detect toxic substances and diagnose their effects. Subsection 2.4 explicates the concept of validation, which planners intend to implement by rigorous quality assurance.

Section 5 describes the strategy for impact assessment and prediction based on toxic-substance data collection. Five general methods are proposed for impact assessment and prediction. Ecological action levels (EALs) would be used to define thresholds above which deleterious effects have been identified. Figure 3-1 illustrates the program's four-step approach for determining the relative potential of a contaminant's concentration to cause adverse biological effects. An ecological index would be used to express the “global ecological health” of a site in terms of a single measure. Ecological risk assessment would be used to relate toxic substances and observed effects through the use of dose-response studies when possible. Trend analysis would be used to evaluate observed temporal and spatial changes in toxic-substance concentrations. Mathematical models of transport and partition dynamics of toxic sub-

stances—including their cumulative impacts on organism population dynamics, genetic diversity, and behavior—are to be used to supplement impact assessment.

The objective is then to describe the exposure of key trust species to toxic substances and to measure changes in exposure and response over time. What toxic substances are causing changes in populations or species? Can the toxic substances be linked to changes in the populations of other species? The proposed method involves choosing a species or population of interest and collecting information about it. Objectives will be developed on the basis of that information. Then a sampling program will be designed to address the objectives for the species. Mathematical modeling will be used to aid in making the assessments.

Although a number of the elements of the plan are desirable, the committee has some reservations regarding the current approach. The reservations and suggestions for alleviating them are outlined here.

Section 5 contains the strategy for impact assessment and prediction. The committee expected some detailed discussion of the assessment method to be used in the Biomonitoring of Environmental Status and Trends Program, especially its statistical procedures. Instead, only a vague description is provided with a simplistic flow diagram ( Figure 3-1).

Although ecological risk assessment appears to be an important component of the plan, the approach to risk assessment is not detailed and not consistent with the state of the art. Models are alluded to whose integration into the plan is not well developed. If models are to play as important a role as they should, then their use needs to be explained. Models should be used as a tool, in addition to other means of assessment, with the recognition that models alone cannot predict consequences at the level of accuracy needed for risk assessment.

The cause-effect portion of the detailed plan should be framed in the context of ecological risk assessment, and the plan should use a relevant paradigm and terminology based on the Environmental Protection Agency framework (e.g., EPA, 1992; or see Suter, 1993). This would involve

• Problem formulation. Preliminary characterization of exposure and effects, integration of existing data, assessment of data needs, development of conceptual models and of policy and regulation issues, and determination of site-specific factors. An important component is the view that strong inference about causation can be obtained through the setting of testable hypotheses. Elimination of unsupported hypotheses will lead to more direct assignment of cause.

• Analysis. Characterization of exposure and effects. Important components include choice of end points, determination of the importance of other stressors, and development of models to relate toxic substances and responses. Exposure, fate, and transport models would be developed, evaluated, and used. The effects assessment would determine the relationship between the exposure to toxic substances and its effect on a given end point.

• Risk characterization. Determination of the likelihood of adverse effects through comparison of exposure and effects information. The purpose of risk characterization is to integrate the results of the exposure and effects assessments and to estimate the magnitude of risk.

• Verification and monitoring. Validation of the process and components, confirmation of predictions, and identification of additional components to incorporate into the process.

• Uncertainty assessment. Listing of assumptions and checking of their importance and implications to address weaknesses in the analytical approach, data, and knowledge and to assess the influence of natural variation. Uncertainty assessment is important because it describes the ability to distinguish between competing toxic substances, to quantify limitations in predictive ability, to identify important assumptions, and in general to describe the limitations of the risk assessment. Of course, ecological risk assessment has its own shortcomings. For example, because comprehensive data are usually not available, risk assessments can depend too heavily on assumptions about critical pathways and exposures. In addition, many of the parameters in a risk assessment must be modeled; this procedure has its own set of constraints and potential problems. Finally, risk assessments can involve linking the results of many analysis and modeling steps, each with its own associated degree of uncertainty. The resulting uncertainty is often difficult to evaluate.

No method is perfect; however, the committee believes that an ecological-risk framework in combination with attempts to incorporate principles of experimental design into observational studies will best fit the needs of the program. Additional National Research Council reports —including Issues in Risk Assessment, Managing Troubled Waters, Review of EPA 's Environmental Monitoring and Assessment Program: Interim Report, Review of EPA's Environmental Monitoring and Assessment Program: Forests and Estuaries, Review of EPA's Environmental Monitoring and Assessment Program: Surface Waters, and Ecological Knowledge and Environmental Problem-Solving: Concepts and Case Studies—provide guidance on these issues.

To produce the strongest causal arguments, the framing of the assessment should be based on scientific inference (Platt, 1964). Holland (1986) suggested that in causal analysis it is important to determine the results or effects of the contaminant, not to describe the causes of effects. Ideally, determination of the causes of effects would take place in an experimental setting. However, this is difficult to do on the time and space scales in interest of contaminant studies. Thus, although it is difficult to confirm causal relationships with observational data and multiple

causes, comparisons can be designed to obtain some strength of causal inference (Cook and Campbell, 1979; Cochrane, 1983; Eberhart and Thomas, 1991).

Assessments based on simple correlations of high toxicant concentration and adverse biological effect, as will be the case in the Biomonitoring of Environmental Status and Trends Program, should be avoided where possible. Contaminants might result in a sample of adverse and indirect biological effects that do not correlate with the cause. Correlation does not prove causation. In actuality, relationships can involve other stressors (e.g., flooding, high temperatures, interactions with other organisms, and habitat alterations), can be nonlinear, and can be confounded by other factors. When changes occur in trust populations, many interacting causal factors might be involved, all of which change in time. Delineating the effects of multiple factors on populations is difficult.

This approach to assessment can be expected to be iterative and to involve development of testable hypotheses, selection of end points and statistical models, and design of studies to focus on assessing the hypotheses. The approach can be iterative as hypotheses are eliminated, and it allows for the use of formal procedures for selection of sample size. Details are in Chapter 7 of Rope and Breckenridge (1993). Elements of the approach in this chapter include goals, selection of end points and potential causes, and design of monitoring programs.

Section 2 of the detailed plan is intended to define the plan for data collection. However, the bioassessment portion of the plan is vague in its terminology and does not detail explicitly what will be done, how it will be done, where it will be done, and when it will be done. No concrete examples are presented to illustrate successful programs in the Fish and Wildlife Service (FWS) with similar data-collection objectives.

The design of programs, the management and manipulation of the data, and the interpretation of the data must be linked. The data-management aspect still seems to be in the planning stage (there is no database program), and there are no specifics about the statistical design of data collection. A causality component cannot be carried out without the linkage. The data-management program needs to be given higher priority. The program should be in place by the time data are collected in pilot studies. Testing of the key aspects of the data-management program (e.g., usability and interpretation with other data-management programs) should be a part of the pilot studies.

A comparison of Table 2-3 and Table 2-4 in the detailed plan reveals that Tier 2 has not yet been fully worked out. To the extent that it has, it is partly similar to Tier 1. Methodologically, this “ maximally cost-effective” approach to bioassessment appears to be one-tiered, rather than two-tiered.

An important element of causation is the linkage of toxic substances with morbidity or mortality of species or ecosystem dysfunction or the linkage of toxic substances with specific sources in the vicinity of trust lands. Linkage is very valuable and is the most challenging theme of the Biomonitoring of Environmental Status and Trends Program. To be successful, this component of the plan must be based on a high-quality toxic-substances monitoring program. Toxic substances on which well-established dose-response data are available are more likely to have defensible cause-effect relationships than those with only weak dose-response data. Toxic-substance stress has to be separated from other anthropogenic and natural stressors. The program relies on external information to make this separation. Without strong participation of other agencies, the causal aspect of the plan might be difficult to implement. Partnerships for planning the initial stages of a study should be established to ensure high-quality, relevant data.

The purpose of validation is to determine whether a method is used in a manner supported by scientific evidence. The criteria put forth on p. 20 of the detailed plan for judging such support are tautologous; i.e., a method is valid if it has been adequately documented, if a response measured by it correlates quantitatively with the supposed cause at the desired level of significance, if it is applied within the range of its intended use, and if “acceptance criteria and other quality control measures” have been established. The committee suggests that rather than focusing on all criteria on p. 20 of the document, FWS should focus on whether a method has been shown to be acceptable by the rigorous testing by the scientific community. In the laboratory, a response should have a relationship with dose(s) of several known and potential contaminant stressors.

A detailed example of a standardized approach to bioassessment was to be included in Appendix H of the detailed plan, but Appendix H was never developed, and there are no plans to do so. However, although it is not included in the draft version of this plan, the committee believes that it would be valuable to include it in the final version because it could relate the proposed methods to successful past efforts to use toxic-substance monitoring to establish cause-effect relationships and source identification. Recent examples, such as the assignment of causes of deformities in birds in the Great Lakes (Tillitt et al., 1993), indicate that cause-effect studies can be successfully implemented. Other examples of studies that satisfied the rules of assigning causality are given by Suter (1993, p. 337); many of these studies involved FWS personnel.

The detailed plan only marginally addresses the use of models in cause assignment. The proposal in the detailed plan is not to develop models. Rather, personnel will use models developed by others and will make use of partnerships if changes in models are required. This approach must be carried out with great care because computer models can be based on assumptions that are not reasonable for sites and might be designed for particular species or populations.

However, models must be an integral component. For example, fate and transport modeling is essential to relating potential exposure and later need for exposure assessment. In addition, models can be valuable in defining field-collection techniques. Finally, modeling is useful in connecting fate and transport of substances, exposure as indicated by assessment of biological markers,¹⁴ and responses at the species and population levels (see Kendall and Lacher, 1994, for example). Models are best used by researchers who are familiar with the site characteristics, the toxic substances present, and the species or populations involved. The studies suggested in the detailed plan will require much time and knowledge to obtain useful results. Furthermore, it is important to describe the uncertainties associated with the models. Partnerships with other federal agencies, whether external or within the Department of the Interior, might be difficult as well. The committee recommends a stronger commitment to the modeling component of risk assessment and a clearer plan on how partnerships between the Biomonitoring of Environmental Status and Trends Program and other agencies would be arranged.

Like problem identification, pilot studies are particularly important for the overall success of the causal component: they can be used to test the program and to find redundancies in end points and weaknesses in sampling and information. The plan does little to describe how pilot studies will be used to test the causality component of the program.

The design of pilot studies needs higher priority. Specific goals of pilot studies should be set. A second set of pilot studies should not be carried out until the analysis and interpretation of the first set are complete (including assessment of the data-management program). A sharper focus is needed for the pilot studies. These studies are needed to identify weaknesses in the proposed program. They should be designed to test the link between monitoring data, biological effect, and toxic-substance source. If EALs constitute an essential component of the decision to implement more costly studies, knowledge of the error rates associated with the decision process would be valuable.

The proposed use of external information to assess the importance of confounding risk factors that are not measured directly but that are an integral part of the Biomonitoring of Environmental Status and Trends Program presents a difficulty. These risk factors are not only various contaminants but other potential human-induced effects as well. The plan presents no details as to how external data are to be collected and integrated. The use of external information is problematic because it can include different scales of measurement in both time and space, poor quality of data and data collection, inappropriate measurement, and errors in data transfer. The committee would prefer a detailed plan on how this needed information would be integrated.

Addressing causality with EALs based on biological markers might not be adequate. Consideration should be given to exposure to chemicals that have delayed, long-term effects (e.g., effects on immune systems, behavior, and fertility) and that do not build up in animal tissue. Timing of exposure is more critical than dose in such cases. Seasonal application of chemicals and intermittent industrial releases can be important. Markers that address causality and are useful for prediction need to be considered (Huggett et al., 1992; McCarthy and Shugart, 1990). Toxicology theory and results of previous studies need to be used to develop hypotheses about observed effects and to help in designing causality studies.

Box 3-1 illustrates some causality study tools. Most studies on risk reveal numerous potential causative factors. To eliminate factors when experiments are not feasible, the weight-of-evidence approach must be used. A weight-of-evidence approach considers all information from the field, laboratory, and models and assesses the adequacy and quality of each factor before eliminating those which do not contribute to determining causality. A final set of possible agents can then be ranked and assessed in terms of

likelihood as causative agents. More generally, the weight of evidence can be used to describe the quality of information and whether it is sufficient for assessing causality.

On the basis of the objectives presented and research plans outlined for the causality portion of the Biomonitoring of Environmental Status and Trends Program in the detailed plan, the committee believes that the program as described probably would not be implementable in a scientifically defensible manner. FWS should rethink and rewrite the plan to clarify what is to be measured, how, where, and how frequently.

To establish strong causal relationships between toxic substances and biological effects, it is important to measure the relevant ecological characteristics and show that the observed effects are due to the toxic substances.

The committee believes that the program will not be able to address issues of causality of changes on trust lands unless the strategy and the methods for investigating cause-effect relationships are scientifically constructed.

Recommendation 3-1: The cause-effect portion of the detailed plan should be framed in the context of ecological risk assessment and should use a relevant paradigm and terminology based on the EPA framework.

Recommendation 3-2: To demonstrate causality, analytical procedures need to be used to distinguish between alterations resulting from changes in natural conditions, changes due to toxic substances, and other anthropogenic changes.

Recommendation 3-3: The strategy for assessing causal relationships should follow the well-developed strategy for ecological risk assessment and should involve hypothesis-testing.

Recommendation 3-4: Toxicological theory should be used to choose which substances to investigate and to establish causal hypotheses.

Recommendation 3-5: Monitoring programs should be designed to address hypotheses and not merely be “surveillance” programs.

Recommendation 3-6: Pilot studies should be used to assess whether the methods for establishing causality will succeed; they should be designed on the basis of lessons gleaned from previous studies of causality.