2.3. Exposure Guideline Derivation – Consideration of Uncertainty and Variability

Generally, scientifically sound, peer-reviewed assessment-specific data are preferred when deriving exposure guidelines. However, if such data are not available, default values are applied when deriving both cancer and noncancer exposure guidelines (see Sections 2.3.1 and 2.3.3). These default values are designed to err on the side of being health protective. Any effort to reconsider these values during the process of developing clearance goals should involve an experienced toxicologist and should ensure that both cancer and noncancer protection is maintained for the site-specific exposure considerations.

2.3.1. NONCANCER EXPOSURE GUIDELINES

Agencies that develop noncancer exposure guidelines, such as EPA and the Agency for Toxic Substances and Disease Registry (ATSDR), utilize an approach that is intended not to underestimate risk in the face of uncertainty and variability. When there are gaps in the available information, uncertainty factors are applied to derive exposure guidelines that are intended to be protective against appreciable risk of deleterious effects. Uncertainty factors are commonly default values⁸⁹ (e.g., factors of 10 or 3), used in the absence of compound-specific data. However, when data are available, uncertainty factors may also be developed using compound-specific information.

EPA begins the development of a toxicity value (dose-based exposure guideline) by evaluating all of the available peer-reviewed literature to determine noncancer endpoints of concern, evaluating the quality, strengths, and limitations of the available studies. EPA typically chooses the relevant endpoint that occurs at the lowest dose, often using statistical modeling of the available data, and then determines the appropriate point of departure (POD) for derivation of the toxicity value. A POD is determined by: (1) a statistical estimation using the benchmark dose (BMD) approach [preferred method]; and (2) use of the dose or concentration at which the toxic response was not significantly elevated (no observed adverse effect level [NOAEL]); or by (3) use of the lowest observed adverse effect level (LOAEL).⁹⁰

A series of downward adjustments using uncertainty factors is then applied to the POD to estimate the toxicity value.90 While collectively termed “uncertainty factors,” these factors account for a number of different quantitative considerations when utilizing observed animal (usually rodent) or human toxicity data in a risk assessment. The uncertainty factors are intended to account for: (1) variation in susceptibility among the members of the human population (i.e., inter-individual variability such as the elderly and children); (2) uncertainty in extrapolating from experimental animal data to humans (i.e., interspecies differences); (3) uncertainty in extrapolating from data obtained in a study with less-than- lifetime exposure (i.e., extrapolating from subchronic to chronic exposure); (4) uncertainty in extrapolating from a LOAEL in the absence of a NOAEL; and (5) uncertainty when the database is incomplete or there are problems with applicability of available studies. When scientifically sound, peer-reviewed assessment- specific data are not available, default adjustment values are selected for the individual uncertainty factors. For each type of uncertainty (when relevant to the assessment), EPA typically applies an uncertainty factor value of 10 or 3 with the cumulative uncertainty factor value leading to a downward adjustment of 10-3,000-fold from the selected POD. If an extrapolation step or adjustment is not relevant to an assessment (e.g., if applying human toxicity data and an interspecies extrapolation is not required) the associated uncertainty factor is not used. The major adjustment steps are described more fully below.

1. Heterogeneity among humans is a key source of variability as well as uncertainty. Uncertainty related to human variation is considered in extrapolating doses from a subset or smaller-sized population, often of one sex or of a narrow range of life stages (typical of occupational epidemiologic studies), to a larger, more diverse population. In the absence of pollutant-specific data on human variability, a 10- fold uncertainty factor is used. The actual degree of human variability may be larger or smaller; however, data to examine the potential magnitude of human variability are often unavailable. In some situations, a smaller uncertainty factor of 3 may be applied to reflect a known lack of significant variability among humans.
2. Extrapolation from results of studies in experimental animals to humans is a necessary step for the majority of chemical risk assessments. When interpreting animal data, the concentration at the POD (e.g., NOAEL) in an animal model (e.g., rodents) is extrapolated to estimate the equivalent human dose. While there is long-standing scientific support for the use of animal studies as indicators of potential toxicity to humans, there are uncertainties in such extrapolations. In the absence of data to the contrary, the typical approach is to use the relevant endpoint from the most sensitive species, strain, and sex in assessing risks to the average human. However, because the most commonly available data for an assessed compound are usually from rodent species, the extent of interspecies variability is often unclear.
3. Pharmacokinetic models are useful to examine species differences in pharmacokinetic processing and associated uncertainties; however, such dosimetric adjustments are not always possible. Information may not be available to quantitatively assess toxicokinetic or toxicodynamic differences between animals and humans, and in many cases a 10-fold uncertainty factor (with separate factors of 3 for toxicokinetic and toxicodynamic components) is used to account for expected species differences and associated uncertainty in extrapolating from laboratory animals to humans in the derivation of an exposure guideline. If information on one or the other of these components is available and accounted for in the cross-species extrapolation, an uncertainty factor of 3 may be used for the remaining component.
4. In the case of developing toxicity values for chronic exposures when data from only shorter duration studies are available (e.g., 90-day subchronic studies in rodents), or when such data are judged to be the most appropriate for development of an inhalation reference concentration (RfC), an additional uncertainty factor of 3 or 10 is typically applied unless the available scientific information supports use of a different value.
5. Toxicity data are typically limited as to the dose or exposure levels that have been tested in individual studies; in an animal study, for example, treatment groups may differ in exposure by up to an order of magnitude. The preferred approach to arrive at a POD is to use BMD analysis; however, this approach requires adequate quantitative results for a meaningful analysis, which is not always possible. Use of a NOAEL is the next preferred approach in determining a POD for deriving a health-based exposure guideline. However, many studies lack a dose or exposure level at which an adverse effect is not observed (i.e., a NOAEL is not identified). When using data limited to a LOAEL, an uncertainty factor of 10 or 3 is often applied.
6. The database uncertainty factor is intended to account for the potential for deriving an underprotective value due to an incomplete characterization of the chemical’s toxicity. In the absence of studies for a known or suspected endpoint of concern, an uncertainty factor of 10 or 3-fold is typically applied.

2.3.2. ACUTE NONCANCER EXPOSURE GUIDELINES

Many of the uncertainty factors used to account for variability and uncertainty in the development of acute exposure guidelines are quite similar to those developed for chronic durations, but more often using individual uncertainty factor values that may be less than 10. Uncertainty factors are applied based on chemical-specific or health effect-specific information (e.g., simple irritation effects do not vary appreciably between individuals, hence a value of 3 is typically used). The uncertainty factors generally applied in the derivation of acute toxicity values include: (1) heterogeneity among humans, (2) uncertainty in extrapolating from animals to humans, (3) uncertainty in LOAEL to NOAEL adjustments, and (4) uncertainty in accounting for an incomplete database on toxic effects of potential concern. Additional adjustments are often applied to account for uncertainty in extrapolation from observations at one exposure duration (e.g., 4 hours) to arrive at a POD for derivation of an acute toxicity value at another exposure duration (e.g., 1 hour).

2.3.3. CANCER EXPOSURE GUIDELINES

For cancer endpoints EPA usually derives an oral slope factor for ingestion and a unit risk value for inhalation exposures. These values allow estimation of an upper bound lifetime probability of developing cancer given long-term exposures to a pollutant. Depending on the pollutant being evaluated, EPA relies on both animal bioassays and epidemiological studies to characterize cancer risk. There is long-standing scientific support for the use of animal cancer bioassays as indicators of potential human risk when other human cancer risk data are unavailable. Extrapolation of study data to estimate potential risks to human populations is based upon EPA’s assessment of the scientific database for a pollutant using EPA’s guidance documents and other peer-reviewed methodologies. The EPA “Guidelines for Carcinogen Risk Assessment” describes the Agency’s recommendations for methodologies for cancer risk assessment.⁹¹ EPA believes that cancer risk estimates developed following the procedures described in the Cancer Guidelines and outlined below generally provide an upper bound estimate of risk. That is, EPA’s upper bound estimates represent a “plausible upper limit to the true value of a quantity” (although this is usually not a true statistical confidence limit).⁹² In some circumstances, the true risk could be as low as zero; however, in other circumstances the risk could also be greater. When developing an upper bound estimate of risk and to provide risk values that do not underestimate risk, EPA generally relies on conservative default approaches.⁹³ EPA also uses the upper bound (rather than lower bound or central) estimates in its assessments, although it is noted that this approach can have limitations for some uses (e.g., priority setting, expected benefits analysis).