4
Risk Assessment Methods for Determining Spacecraft Water Exposure Guidelines
HUMAN exposure guidelines for toxic substances are established through a multiple-step process called risk assessment. The guidelines are set for concentrations that research predicts pose acceptable (usually negligible) risks of adverse health effects to humans under specified conditions of exposure. Quite often, the objective of risk assessment is to establish a daily exposure that is considered safe over a lifetime. For space travel, the anticipated durations are substantially less than a lifetime, but the absolute lifetime risk of adverse health effects is still the focus of the risk assessment. For adverse effects that are transitory and only mildly debilitating, the intent is to ensure that exposure to substances that cause such effects is restricted to amounts that will not impede the normal performance of duties aboard spacecraft.
Although the process of risk assessment uses human data whenever possible, often from epidemiologic studies, it is not aimed at estimating relative risks in the usual epidemiologic sense. Risk assessment frequently involves extrapolation from conditions under which the data are derived by observation to an unobserved or unobservable exposure situation, and it focuses on absolute risk rather than relative risk. More often than not, because of the lack of suitable human data, risk assess-
ment is based on data from experiments with animals. The quality of the data has a major influence on the risk assessment, and meaningful extrapolation of experimental data to an applicable human situation presents significant challenges.
Below is a review of the approaches to conducting risk assessments, as well as the subcommittee's recommended approach to deriving spacecraft water exposure guidelines (SWEGs). A discussion of the exposure conversions and uncertainty factors that should be considered in the calculations is also provided.
HISTORICAL PERSPECTIVE
Risk Assessment for Noncarcinogenic Effects
For toxic effects other than cancer, the practice of risk assessment has been to set acceptable exposure by dividing no-observed-adverse-effect levels (NOAELs) obtained from human studies or animal experiments by a set of uncertainty factors (sometimes called “safety” factors). A NOAEL is the highest experimental dose for which no difference in the occurrence of an adverse effect is observed relative to a control group. The NOAEL-based approach has come to be associated with the presumed existence of threshold doses – doses below which specific toxic effects will not occur, even if exposure continues over a lifetime. The concept of threshold is supported by the observation that many organisms have detoxification mechanisms or repair capacities to compensate for some degree of damage and still maintain normal function (Klaassen and Eaton 1991). Exposure guidance levels that result from reducing NOAELs by uncertainty factors, called acceptable daily intakes or ADIs, are presumed to pose zero risk of the toxic effect in question. In many applications, two uncertainty factors of 10 have been thought to be adequate, the first to allow for possible increased sensitivity of humans to the toxic agent compared with experimental animals and the second to account for variations in susceptibility within the human population (Lehman and Fitzhugh 1954).
In experiments for which a NOAEL is not established, only a lowest-observed-adverse-effect level (LOAEL) will be available for risk assessment. A LOAEL generally corresponds to a response in the range of 1-
10%, and an uncertainty factor of 10 often is used to extrapolate from a LOAEL to a NOAEL, although some investigators indicate that a factor of 3-5 would be more appropriate (Abdel-Rahman and Kadry 1995). Ideally, the selection of the uncertainty factor depends on the slope of the dose-response curve.
Barnes and Dourson (1988) identified two additional uncertainty factors that might be needed for deriving references doses (RfDs), which estimate a daily exposure to the human population that is likely to be without an appreciable risk of harm during a lifetime. These additional factors represent uncertainty with respect to exposure duration and to data quality. The size of each of several uncertainty factors is determined by the best judgment of the risk assessor; however, the U.S. Environmental Protection Agency (EPA) has suggested using a maximum of 3000 for the product of four uncertainty factors and a maximum of 10,000 for five uncertainty factors (Dourson 1994). Uncertainty factors involved in the calculation of SWEGs are discussed later in this chapter.
Risk Assessment for Carcinogenic Effects
It has been assumed traditionally that threshold doses do not exist for carcinogenic effects, particularly those considered to result from genotoxicity. For this reason, it has been considered infeasible to establish low exposure limits that correspond to zero risk. Instead, beginning with the pioneering work of Mantel and Bryan (1961), attempts have been made to estimate carcinogenic risks on a precise, quantitative basis, to estimate exposures that produce very low, but nonzero, cancer risks. These efforts have involved fitting mathematical models to experimental data and extrapolating downward to predict risks at doses well below the experimental range.
The mathematical model most frequently used for low-dose extrapolation is a variation of the multistage model of Armitage and Doll (1960), commonly expressed as
P(d) = 1 − exp(−q_{0} −q_{1} d −q_{2}d^{2} − . . . − q_{k}d^{k})
P(d) is the probability of developing cancer during a lifetime of exposure at a dose d of a carcinogen, and q_{0}, q_{1}, q_{2}, . . . , q_{k} are nonnegative
constants that are estimated via regression analysis of cancer data (usually animal data) at k or more dose levels. Ideally, d is a measure of target tissue dose, but most often in practice d is a measure of external exposure.
According to the multistage theory, a malignant cancer cell develops in stages from a single stem cell through a series of biologic events (mutations) that occur in a specific order. Assuming that the rates of transition between two or more stages in the multistage model are linearly related to target tissue dose, the dose-response curve for the multistage model is linear at low doses (Crump et al. 1976). Low-dose linearity is generally assumed for chemical carcinogens that operate through direct interaction with genetic material. When carcinogenesis occurs by other mechanisms, low-dose linearity might not be applicable. Data developed in recent years suggest that some carcinogens, especially those whose mechanisms of action involve cytotoxicity or disruption of hormonal homeostasis, exhibit practical threshold doses below which the risk of cancer is negligible (Page et al. 1997; Hill et al. 1998). However, if there is a nonzero background cancer risk and if the mechanism of the nongenotoxic carcinogen is the same as the background mechanism, then this “additivity” of cancer risk still implies linearity at very low doses for dose-response relationships that are strictly increasing. Hence, linear extrapolation has been widely used in low-dose cancer risk assessment in the absence of clear information to dictate a different course of action (OSTP 1985; EPA 1996a). Risk assessments that deviate from the use of linear extrapolation require considerable data to ensure that the traditional default approach is not applicable.
Uncertainties in the process of establishing acceptable exposures have been handled differently for carcinogenic and noncarcinogenic effects. For example, instead of a factor of 10 for interspecies uncertainty, a factor derived from a power function of body weight often is used for interspecies conversion (EPA 1992). Also, in general, no uncertainty factor is used for human variation in sensitivity to a substance's carcinogenic effects. However, variation in the experimental data is a source of uncertainty that is recognized for carcinogenic effects through the use of statistical confidence limits instead of central estimates.
RECOMMENDED APPROACH TO RISK ASSESSMENT
Exploiting Similarities of Historical Approaches
Gaylor (1983) was among the first to point out the practical similarities between low-cancer-risk dose based on linear extrapolation and those that would result from the reduction of cancer NOAELs by uncertainty factors. In light of Gaylor's observation that the NOAEL for cancer often corresponds to a central estimate of risk of approximately 10^{−2} (1%), then reducing such a cancer NOAEL by an overall uncertainty factor of 100, 1000, or 10,000 would result in the same dose that would be obtained by linear extrapolation to a risk level of 10^{−4}, 10^{−5}, or 10^{−6}, respectively (Kodell and Park 1995). Conversely, if the true response rate at the NOAEL for a noncarcinogenic effect is acknowledged to be other than zero, say around 1%, then the application of linear extrapolation for a noncarcinogenic effect to estimate a dose with a risk level k orders of magnitude lower would be equivalent to dividing the NOAEL by an uncertainty factor of 10^{k}. The functional equivalence of the two approaches has been highlighted by Wilson (1997).
Despite the apparent practical similarities between the two opposing approaches to risk assessment, little has been done until recently to unify them. Proponents of low-dose linear extrapolation have questioned the presumption by NOAEL proponents that zero-risk limits (thresholds) can be established based on experimental observations; proponents of the NOAEL-uncertainty factor approach have questioned the presumption by modeling proponents that precise risks can be attached to doses below the observed experimental range. Recently, however, proposals have been advanced for the unification of risk assessment procedures for carcinogenic and noncarcinogenic effects (Purchase and Auton 1995; Crump et al. 1996; Gaylor et al. 1999). There is a movement to place less emphasis on numerical estimates of risk of cancer below the data range, and to place more emphasis on estimation of risk of noncancer effects within the data range. The objective is to combine the best features of the two methods into a unified approach to setting safe exposure for all types of toxic effects.
Exploration and development of refined models for low-dose extrapolation is not discouraged. Rather, as biologic processes are better un-
derstood, it is expected that improved mathematical models for risk assessment will evolve. Several promising new approaches are discussed later in this chapter. However, the usual data that are available for risk assessment do not permit precise estimates of risk to be made at doses below the data range. For this reason, the risk assessment methodology presented here emphasizes model fitting within the data range for carcinogenic and noncarcinogenic effects.
Benchmark-Dose Approach to Setting SWEGs
It is important that dose-response data are adequate to establish a NOAEL. However, various authors have documented limitations of the NOAEL as a basis for establishing acceptable exposure (Munro and Krewski 1981; Crump 1984; Kimmel and Gaylor 1988). The determination of the NOAEL is limited by the number and distribution of doses and by sample sizes; using the NOAEL as a basis for setting exposure limits ignores dose-response information. As an alternative to the NOAEL, Crump (1984) proposed use of a benchmark dose (BMD) – a statistical lower confidence limit on a dose that is estimated to correspond to a low level of excess risk above background in the range of 1-10% (ED_{01} to ED_{10}; ED_{p} is an effective dose that yields a response of p). Because of its accounting for experimental variation, the BMD could be lower than the NOAEL and thus could result in lower acceptable exposure limits after it has been reduced by uncertainty factors. Experimental variation, however, is an important source of uncertainty that has been neglected heretofore in risk assessment for noncarcinogenic effects.
The BMD originally was defined as a statistical lower confidence limit on the ED_{p}, for 0.01 ≤ p ≤ 0.10. In addition to the original suggestion of Crump (1984), observations by other investigators also argue for establishing the BMD to correspond to the response range between 1% and 10%. As observed by Gaylor (1992) and Allen et al. (1994a), the incidence of fetal malformation at the NOAEL in typical teratology studies often exceeds 1%. Leisenring and Ryan (1992) argue that the average risk at the NOAEL for quantal data could easily be as much as 10%, depending on the experimental design and the shape of the dose-
response curve. In an analysis of 486 developmental toxicity studies, Allen et al. (1994a) conclude that the average NOAEL approximated a lower 95% confidence limit on the ED_{05}. Several investigators recommend the ED_{01} as an anchor point for risk estimates for carcinogenic effects (Van Ryzin 1980; Farmer et al. 1982; Gaylor et al. 1994). The intent is to avoid dependence on particular mathematical models, which is most apparent at doses below the ED_{01} (Krewski and Van Ryzin 1981). EPA has proposed the ED_{10} as a point of departure for cancer risk assessment (EPA 1996a).
It is recommended that, for chemicals for which there are sufficient dose-response data, a BMD corresponding to a 1% risk (BMD_{01}) be used instead of the NOAEL and that a BMD corresponding to a 10% risk (BMD_{10}) be used instead of the LOAEL. Like the NOAEL and LOAEL, the BMD _{01} and BMD_{10} are merely starting points for establishing safe exposures, but they have more precise definition and determination. Like the NOAEL and LOAEL, they are meant to correspond to very low risk. These BMDs should serve as starting points for setting acceptable human exposures to substances for all types of toxic effects, whether carcinogenic or noncarcinogenic. In the process of setting the exposure levels, BMDs must be modified by appropriate conversion factors and reduced by appropriate uncertainty factors, as will be discussed in subsequent sections. The resulting exposure guidance levels do not have specific risk connotations attached to them, but they are simply expected to reflect adequate safety.
When sufficient data are available, the unified BMD-based method for calculating acceptable human exposures is recommended for determining maximum contamination in water aboard spacecraft – spacecraft water exposure guidelines (SWEGs). The BMD approach is an evolving strategy that will assist in the calculation of SWEGs when adequate data are available. At the current stage of development of the BMD, the recommended method for establishing SWEGs is to determine the lower-confidence, model-based likelihood of a BMD_{01} level and apply appropriate uncertainty factors if necessary. This approach represents a recommended decision process to establish acceptable guidelines, but others, such as the lower confidence limit of a BMD _{10} or central estimates of the BMD, might be more useful for data that are available. Further evolution of BMD methodology should be moni-
tored and appropriate alterations in the approach should be made as warranted. In the absence of sufficient data, or when special circumstances dictate, the recommended default procedure for determining SWEGs is essentially the NOAEL-based procedure currently in use for setting maximum contamination levels in air aboard spacecraft – spacecraft maximum allowable concentrations (SMACs) (NRC 1992; James and Gardner 1996).
BMD CALCULATION
Central Estimate Versus Confidence Limit
Estimating the BMD_{p} involves fitting a dose-response model to observed data and calculating the dose level that corresponds to an excess response, p, above background. Because the estimation of benchmark doses does not stray far from the observed data range, the choice of model might not be critical. As pointed out by Krewski and Van Ryzin (1981), fitted dose-response models for quantal toxicity data do not differ appreciably at responses above 1%. However, as much as possible, knowledge of the biologic mode of action should be used in modeling the dose-response data (Andersen et al. 2000; Wiltse and Dellarco 2000). Clearly, the validity of the observed dose-response data for risk assessment must be ascertained before BMD_{p} estimation begins.
It is desirable that methods used to fit dose-response models to observed data include provisions for calculating statistical confidence limits, because experimental variation is a source of uncertainty that must be considered. Instead of using a formal statistical lower confidence limit on a BMD_{p} as a starting point, one could calculate a central estimate of the BMD_{p}, and reduce it by an uncertainty factor to account for experimental variation. The result would be the same, but the expression of the lower confidence limit on the BMD_{p} via an uncertainty factor for experimental variation provides explicit information regarding the magnitude of this source of uncertainty. The uncertainty factor would be just one of several that would be used to reduce the central estimate to an acceptable exposure guidance level (T.B. Starr, TBS Associates, personal communication, 1997). Thus, the use of confidence
limits instead of central estimates is intended to capture the experimental uncertainty rather than to provide “better” estimates. This topic is discussed further later in this chapter.
Estimating BMD_{p}for Various Toxic Effects
Traditional methods of dose-response modeling for binomially distributed random variables can be used to estimate carcinogenic effects and other quantal toxic responses (lethality, some mutagenic responses) for which subjects are assumed to respond independently from one another. Maximum likelihood estimation is a commonly accepted method of fitting a variety of mathematical dose-response models, including the multistage model, the probit model, or the Weibull model (Crump 1979; Zeise et al. 1987). The maximum likelihood method essentially identifies values of a model's parameters that have the highest likelihood of being correct, given the observed data used to fit the model. Generally, the only data available for modeling will be crude, lifetime incidences. If, however, data on time to occurrence of effects are available, the use of a model that can exploit this additional information is encouraged (e.g., Lensing and Kodell 1995).
Maximum likelihood estimation procedures have been worked out for fitting dose-response models for quantal effects that are assumed to be correlated between subjects. Chen and Kodell (1989), Ryan (1992), Allen et al. (1994a,b), and Krewski and Zhu (1995) all have proposed methodology for calculating BMD _{p} for toxicity data that are overdispersed with respect to simple binomial variation.
For continuous data, such as that arising in neurotoxicity studies, the definition of frank, adverse effects is not straightforward. Such data often are described well by normal (Gaussian) or lognormal distributions. Hence, modeling continuous responses on a probability scale to estimate the dose corresponding to a specified probability, p, of an adverse effect (BMD_{p}) is difficult. However, methods of risk assessment for such data have been developed (Gaylor and Slikker 1990; Kodell and West 1993; Crump 1995; Kavlock et al. 1995; Bosch et al. 1996), including provisions for calculating BMD. Hence, BMD_{p} for continuous, quantitative toxic responses can be calculated.
Appendix B provides examples of BMD estimation.
EXPOSURE CONVERSION
Target Tissue Dose
Toxic substances sometimes require some form of metabolic activation to exert their adverse health effects, which might range from direct, short-term, target tissue toxicity to carcinogenesis. If metabolic activation can be characterized adequately by a pharmacokinetic model, then the dose delivered to the target should be used instead of the administered dose, for purposes of dose-response modeling to estimate BMD. Although the use of delivered dose rather than administered dose can be expected to lead to more accurate predictions of risk, pharmacokinetic modeling could actually lead to additional uncertainty, if physiologically based pharmacokinetic models with many parameters are used for tissue dosimetry (Farrar et al. 1989; Portier and Kaplan 1989). Hence, the question of whether to use target tissue dose or administered dose for dose-response modeling depends on the degree of confidence that can be placed in the pharmacokinetic model.
Differences in Duration
For toxic effects that are believed neither to accumulate nor to increase in adversity over time, a single exposure level for a toxicant can be used for SWEGs of different durations. However, for many toxic end points, an adjustment of the exposure will be required when extrapolating from one duration to another. Whenever possible, such extrapolation should use substance-specific, time response information, which can be in the form of an empirical mathematical relationship between exposure concentration and duration. For example, ten Berge et al. (1986) investigated the relationship between concentration and exposure time based on mortality data from 20 acute studies of locally and systemically acting inhalation toxicants. Using probit analysis, they found that the relationship C^{N}× T = K provided a good explanation of the relationship between concentration and duration. C is the concentration of the agent, T is the duration of exposure, and K is a constant. The value of N was generally greater than 1 and had an average value of approximately 1.8. (In fact, they found that the relation-
ship C^{n} × T^{m} = k described the data well; the average value of n was approximately 3.5 and the average value of m was approximately 2.0. The expression C^{n} × T^{m} = k is actually equivalent to C^{N} × T = K, where N = n/m and K = k1/^{m}. Alternatively, the ten Berge rule could be expressed as C ×T^{M}=K, with M = m/n and K = k1/^{n}.)
The simplest form of ten Berge's formula is C × T = K, commonly known as Haber's rule, for inhalation toxicants. In the absence of chemical-specific information on the relationship between concentration and duration, Haber's rule often has been used as a default approach for making conversions for different (relatively short) durations of exposure. In its guidelines for the establishment of SMACs for airborne contaminants, the NRC (1992) urged caution in the use of this simple approach, and, for noncarcinogenic effects, the NRC subcommittee on SMACs has been reluctant to endorse its use for converting doses derived from longer term exposures to doses that would apply for shorter term exposures (see also James and Gardner 1996). However, like the NRC Subcommittee on Emergency Exposure Guidance Levels (NRC 1986), the subcommittee on SMACs does consider the use of C × T = K appropriate for extrapolating between two exposures that are relatively short term with respect to clearance or repair rate. Also, in the absence of definitive information, the subcommittee on SMACs has endorsed this approach for converting doses corresponding to shorter term exposures to doses corresponding to longer term exposures, although each substance must be considered individually with respect to the applicability of Haber's rule.
A simple comparison of ten Berge's rule to Haber's rule is given in Table 4-1, using N = 2, where the reference concentration is 50 parts per million (ppm) with a duration of 2 days (d). Conversions are made for 1-d and 4-d exposures. For N > 1, ten Berge's rule will give smaller concentrations than will Haber's rule in converting to shorter durations. It will give larger concentrations than will Haber's rule in converting to longer durations.
The National Advisory Committee for Acute Exposure Guideline Levels for Hazardous Substances (NAC/AEGL Committee) (EPA 1997) uses the relationship C^{N}× T = K proposed by ten Berge et al. (1986) to make conversions for different exposure durations in the setting of AEGLs. This relationship should be used whenever possible in making duration conversions when setting SWEGs for water contaminants
TABLE 4-1 Haber's and ten Berge's Rule Compared
Haber's Rule |
ten Berge's Rule |
||||
Concentration |
Time |
K |
Concentration |
Time |
K |
100 ppm |
1 d |
100 |
71 ppm |
1 d |
5000 |
50 ppm |
2 d |
100 |
50 ppm |
2 d |
5000 |
25 ppm |
4 d |
100 |
35 ppm |
4 d |
5000 |
Using 50 ppm as the reference concentration (exposure duration =2 d; N = 2), Haber's rule and ten Berge's rule were used to make conversions for 1 d and 4 d exposures. |
aboard spacecraft. As recommended by ten Berge et al. (1986), the value of N for specific toxicants should be derived empirically from experiments that provide data on various concentrations and various durations of exposure. This can be done by probit analysis. However, even when chemical-specific data are not available for estimating N, it might be possible to choose a default value of N other than N = 1 (Haber's rule), which would be expected to reflect the likely relationship between concentration and duration for a broad range of substances. The NAC/AEGL Committee (EPA 1997) often uses a default value of N = 2 when no exposure-versus-time data are available (e.g., arsine, 1,2-dichloroethane).
The method provided in the SMACs subcommittee's guidelines (NRC 1992) for converting lifetime daily exposure to carcinogens to exposures applicable to the shorter durations associated with spaceflight is based on a multistage model (Kodell et al. 1987) that is equivalent to using a C × T = K adjustment combined with an additional adjustment factor, f (NRC 1992). Murdoch et al. (1992) show that, for typical astronauts, f would not be likely to exceed a value of 2. In fact, for many plausible spaceflight scenarios, f will be about 1 (e.g., 3-stage model, first stage dose-related, and 30-year-old astronaut) for a wide range of exposure durations (e.g., 1-1000 d).
Species Conversions
Conversion of BMD_{p} values derived from data on an appropriately selected test species to comparable values for humans requires experi-
enced scientific judgment. In the best situation, data on metabolism and disposition of the substance of interest in humans and the test species should be used to determine the appropriate conversion factor. For carcinogenic effects, interspecies conversions often are made on the basis of body weight or surface area differences between species (Allen et al. 1988; Travis and White 1988; EPA 1992). Such conversions are intended to correct for metabolic rate based on body size. However, their basis is related more to rates of basal metabolism than it is to xenobiotic metabolism (NRC 1992). Hence, quite often, in the absence of adequate pharmacokinetic or pharmacodynamic information to enable determination of an appropriate conversion factor, an assumption of concentration equivalence between species is made, and extrapolation is done on a straight concentration basis (e.g., parts per million in food, air, or water). Then an uncertainty factor generally is applied to account for unknown and unmeasured species differences.
Different Routes
In most cases, a BMD_{p} for SWEGs will be derived from oral exposure studies in animals. However, some might be based on nonoral routes. Exposures of humans during spaceflight to contaminated water can happen by a variety of routes: inhalation, water consumption, dermal absorption. Where possible, conversions must be made to account for differences between routes of exposure for astronauts and those used in the animal studies from which a BMD_{p} is derived. Assuming that the species-to-species conversion is made separately, all that would be required at this step would be a route-to-route conversion within species. At the very least, differences in rates of absorption for various routes should be considered where possible.
UNCERTAINTY FACTORS
Exposure Duration Uncertainty
When there is insufficient information available on a toxic substance to allow an informed adjustment to be made for differences in exposure duration, and when the rule C^{N} × T = K (ten Berge et al. 1986) cannot be applied even with a default value for N, it might be necessary to
employ an uncertainty factor when extrapolating from one exposure duration to another. In many risk assessment exercises, such as in the derivation of RfDs, an exposure duration uncertainty factor is used when subchronic data must be used to set limits for chronic exposure (Barnes and Dourson 1988). In the past, the default value for this subchronic-to-chronic factor has been 10. In the setting of SWEGs, extrapolating from exposure durations for which data are available to those encountered in spaceflight might require the use of exposure duration uncertainty factors similar to the subchronic-to-chronic factor.
Interspecies Uncertainty
If sufficient pharmacokinetic and pharmacodynamic data are available, then a species-to-species conversion of the BMD_{p}should be made on as quantitative a basis as possible, using experienced scientific judgment. Unfortunately, such data are the exception rather than the rule. In most cases, there is an insufficient quantitative basis for making an informed extrapolation from animals to humans. Thus, it is considered prudent to reduce the BMD_{p} by an appropriate uncertainty factor to account for unknown species differences that might imply a greater sensitivity in humans than in experimental animals. Traditionally, a value of 10 has been used for this species factor, which originally arose in the setting of ADIs for chemicals in the food supply (NRC 1970). Factors greater or less than 10 should be used, depending on the nature of the toxicity. For example, central nervous system effects in most species might be similar to effects in humans, implying a species factor close to 1, whereas the uncertainty factor for other toxic effects might need to be as high as 15 (Calabrese and Baldwin 1995). The choice of a particular interspecies uncertainty factor needs to be justified in each case. A factor of 10 continues to be the default recommended by EPA (1996a), and it should be used as the default factor for SWEGs as well.
Experimental Variation
We recognize that the size of the confidence interval is highly dependent on the number of experimental subjects, often very small at lower doses. Rather than abandoning the BMD_{01} and using the central
estimate, we suggest involving a modification of the uncertainty factor for small numbers. Pragmatically, this uncertainty factor would be used when even small numbers of subjects reflect little experimental variability. Examples include use of primates or other large test species where sample size may be limiting. The uncertainty factor would add 1% to 100% of z_{α}[(1 − p)/(np)]^{½} to the BMD_{01}, depending on certainty; z_{α} is the 100(1 − α)^{th} percentile of a standard normal probability distribution, p is the excess response rate at the BMD_{p}, and n is the sample size on which p is based.
Experimental variation will be an important source of uncertainty in the derivation of SWEGs. Generally, sampling variation will be accounted for by using a statistical lower confidence limit on the BMD_{p} as an anchor point for setting a SWEG. However, it is not absolutely necessary that formal statistical confidence limits be used. Instead, an uncertainty factor for experimental variation could be included as one of several uncertainty factors used to reduce a central estimate of the BMD_{p} – we could call it the CBMD_{p} – to a SWEG. That is, as suggested by T.B. Starr (TBS Associates, personal communication, 1997), a formal statistical lower confidence limit on the BMD_{p} – say the LBMD_{p} – could be calculated and then used to back-calculate the appropriate factor, f= CBMD_{p}/LBMD_{p}, by which to reduce the central estimate of the BMD_{p} to account for experimental variation. Although that approach is exactly the same as calculating a formal statistical lower limit in the first place, it does have the advantage of conveying the size of the uncertainty factor, f, that is used to control for experimental variation. It should be noted that the size of f can be influenced both by the ability of the model to fit the observed dose-response relationship and by any constraints imposed as part of the fitting procedure.
In some situations, the available data might not permit the calculation of a statistical lower confidence limit on the BMD_{p}. This could happen, for example, when a maximum likelihood estimation procedure is used to calculate confidence limits, but, because of poor data, the procedure will not converge for the restricted, lower limit dose-response model. For such situations, R.L Kodell and D.W. Gaylor (National Center for Toxicological Research, Food and Drug Administration, unpublished material, 1998) have shown that an ad hoc uncertainty factor, f, for experimental variation can be calculated by
f = 1 + z_{α}[(1 − p)/(np)]^{½},
where p is the excess response rate at the BMD_{p}, n is the sample size on which p is based, and z_{α} is the 100(1 − α)^{th} percentile of a standard normal probability distribution (z_{α} = 1.645 for 95% confidence).
The application of an uncertainty factor for sampling variation reflects the spirit of the “small-n” factor, 10/√n, which is used by the NRC (1994) in the derivation of SMACs to reflect the uncertainty in NOAELs based on a limited number of human subjects. The small-n factor also could be used in the derivation of SWEGs if insufficient data are available to calculate a specific BMD_{p}.
BMD_{10}to BMD_{01}
If the BMD_{10} rather than the BMD_{01} is chosen as the anchor point for establishing a SWEG, which could happen – for example, if the estimated BMD_{01} is considered too unreliable or unstable – then it is logical to reduce the BMD_{10} by an additional uncertainty factor. An uncertainty factor of 10 is generally used to extrapolate from a LOAEL to a NOAEL. Because, in this document, the BMD_{01} is recommended in place of the NOAEL and the BMD_{10} is recommended in place of the LOAEL, it might be advisable to use the same factor of 10 to establish equivalence when extrapolating from a BMD_{10} to a BMD_{01}. However, some investigators indicate that a factor of 3-5 would be more appropriate for going from a LOAEL to a NOAEL (Abdel-Rahman and Kadry 1995), and the NRC subcommittee on SMACs endorsed the selective use by the National Aeronautics and Space Administration (NASA) of a factor of 2 for short-term irritation, based on available dose-response information (NRC 1994). In the examples of BMD calculation in Appendix B, all ratios of BMD_{10} to BMD_{01} lie between 2 and 10. It could be advantageous to use a BMD_{01} whenever possible rather than start with a BMD_{10} and have to reduce it as much as 10-fold. The recommendation here is that a factor of 3 or 10 be used to reduce a BMD_{10} to an appropriate BMD_{01} equivalent for setting SWEGs.
Environmental Effects
The special conditions of the space environment must be considered in defining SWEGs. Environmental factors that could alter the toxicity
of water contaminants include microgravity, radiation, and stress (Kaplan 1979; Merrill et al. 1990). Astronauts can be physically, physiologically, and psychologically compromised in several ways: decreased muscle mass, decreased bone mass, decreased red-blood-cell mass, depressed immune systems, altered nutritional requirements, behavioral changes, shift of body fluids, altered blood flow, altered hormonal status, altered enzyme concentrations, increased sensitization to cardiac arrhythmia, and altered drug metabolism (NRC 1992). Hence, astronauts in space will be in an altered homeostatic state and might experience increased sensitivity to the toxic effects of contaminated water.
It is important to reduce chemical exposure relative to what would be acceptable on Earth for toxic effects that are influenced by the physiologic changes induced by spaceflight. However, there is generally little definitive information to permit a precise, quantitative conversion that would reflect altered toxicity resulting from spaceflight environmental factors. Hence, the use of an uncertainty factor generally is dictated when available information on a substance indicates that it affects one or more aspects of an astronaut's condition that might be compromised in space. For example, the SMACs subcommittee has agreed with NASA 's practice of applying an uncertainty factor of 3 or 5 to modify allowable exposure concentrations for chemical agents that affect the immune system or that have been demonstrated to sensitize animals to cardiac arrhythmia (NRC 1992, 1994, 1996a,b; James and Gardner 1996).
When data on the effects of microgravity on bodily functions are available, information derived from them might preclude the need for an uncertainty factor. For example, in a study of pulmonary function of astronauts who participated in flights lasting 9-14 d aboard NASA 's Spacelab, West et al. (1997) conclude that, although there were adaptive changes in pulmonary function in microgravity, none of the observed changes would limit spaceflight. Based on this observation, an uncertainty factor for microgravity for pulmonary toxicants under these exposure situations (i.e., as in the West study) might not be required for flights of short duration. For prolonged exposure in space, the practice of using an uncertainty factor might be warranted, but it can be revised, as human data (or suitably predictive animal data) become available.
Other Factors
The conversion factors and uncertainty factors represent the generic modifying factors that must be applied routinely in deriving SWEGs from BMD_{p}s. There could be additional modifying factors that should be applied in specific cases, depending on the substance in question and the nature of the space mission. For example, one important uncertainty factor commonly used in risk assessments that concern general public health is a factor to account for variability among humans in sensitivity to specific substances. Because of the relatively homogeneous, robust health status of astronauts on most space missions, it is not necessary routinely to apply an uncertainty factor for intraspecies variability. To date, biologic diversity (age, sex, toxicogenetic differences) has not presented significant concern. However, there might be special missions or specific substances for which use of an intraspecies factor would be warranted. This issue would be considered case by case, and it is beyond the scope of this document. When the Human Genome Project and the Environmental Genome Project are completed, data will begin to accumulate on genetic polymorphisms. NASA should monitor the progress in the identification of polymorphisms that make certain individuals more susceptible to certain chemicals.
Another modifying factor sometimes applied in general risk assessment practice is one that reflects uncertainty about the quality of data. That is, if the data are considered inadequate or incomplete, an allowable exposure level would be reduced by some factor. However, it has been the practice of the NRC subcommittee on SMACs to recommend that SMACs not be established for substances for which the data are inadequate, rather than set an unreasonably low SMAC that would give the appearance of being data-based when it was not (NRC 1994, 1996a,b). This practice is recommended for setting SWEGs.
Although the intent of the procedure recommended for setting SWEGs is to provide a unified approach that applies to all types of toxic effects, there is one possible point of departure between carcinogenic and noncarcinogenic effects. That is, because of the severity and irreversibility of cancer, some risk assessors recommend that exposure limits based on carcinogenic effects be reduced by an additional factor to take this into account (Renwick 1995; Gaylor et al. 1999). The sub-
committee recommends that this issue be considered on a case-by-case basis, and that any use of an additional uncertainty factor be scientifically justified.
SETTING SWEGs
Overall Uncertainty Factor
Each uncertainty factor discussed earlier accounts for one source of uncertainty for which either the direction of the difference or the magnitude of the difference between an estimated value and the true value is unknown. Each factor is presumed to account for extreme differences that might exist between estimated and true values. Because not all true differences would be expected to be at their extremes simultaneously, reducing a BMD_{p} by a product of uncertainty factors could lead to undue conservatism – in the sense that the resulting SWEG might be lower than necessary to provide the desired protection. Recognizing the compounding of conservatism that occurs in dividing experimental doses by multiple uncertainty factors, as demonstrated by Bogen (1994) and Slob (1994), Gaylor and Chen (1996) suggest using a reduced, combined uncertainty factor to set acceptable limits. Assuming that individual uncertainty factors are lognormally distributed (Dourson et al. 1996), a combined uncertainty factor, F, can be calculated as follows (Kodell and Gaylor 1999; Gaylor and Kodell 2000):
F = exp{∑_{i}avg[In(f_{i})] + z_{α}(∑_{i}s^{2}_{In(}_{f}_{i))}^{½}},
where avg[ln(f_{i})] is an estimate of the mean log_{e}-uncertainty factor for the i^{th} of m sources of uncertainty, s_{ln(}_{f}_{i)} is an estimate of the standard deviation of the distribution of ln(f_{i}), and z_{α} is the 100(1 − α)^{th} percentile of the standard normal distribution.
Implicit in the use of individual uncertainty factors is the assumption that true conversion factors for the various types of extrapolation are random variables, and that the individual uncertainty factors capture a high percentage of the range of variation for each extrapolation. A factor of 10 is the default value for most factors, but some investiga-
tors argue for larger or smaller factors for specific sources of uncertainty. For example, Swartout (1996) observed that a factor as large as 17 might be necessary to cover the uncertainty in estimating chronic effects using subchronic data. Abdel-Rahman and Kadry (1995) argue that a factor as small as 3 could be sufficient to capture LOAEL-to-NOAEL uncertainty (and perhaps also BMD_{10}-to-BMD_{01} uncertainty).
The rationale for the combined uncertainty factor F is that, if estimates of the mean and standard deviation of the individual distributions of uncertainty are available, then statistical techniques for estimating upper tolerance limits of distributions of sums of independent random variables can be used to calculate a reduced overall uncertainty factor (that is less than the product of individual factors) that will still capture a high percentage of the overall range of uncertainty. Specifically, the formula for F is a point estimate of the 100(1 − α)^{th} percentile (e.g., 95^{th} percentile, for α = 0.05) of the combined range of uncertainty. The use of this combined factor would be expected to provide 100(1 − α)% assurance of protection for the combined sources of uncertainty. Recent studies by Baird et al. (1996) and Swartout et al. (1998) have used Monte Carlo simulation techniques to estimate upper percentiles of the distribution of combined uncertainty factors (simulation-based values of F) for consideration in the setting of RfDs. The combined uncertainty factor F is recommended for consideration and use in the process of setting SWEGs.
Table 4-2 contains information derived from the literature on averages and standard deviations of log _{e}-uncertainty factors for sources of uncertainty that are commonly encountered in risk assessment extrapolations.
TABLE 4-2 Estimated Averages and Standard Deviations of Log_{e}-Uncertainty Factors (ln(f_{i})) for Various Sources of Uncertainty
Source of Uncertainty |
Average |
(ln(f_{i})) |
Reference |
Human-to-human |
0 |
1.64 |
Dourson and Stara 1983 |
Animal-to-human |
0 |
1.66 |
Calabrese and Baldwin (1995) |
Subchronic-to-chronic |
0.69 |
1.30 |
Swartout 1996 |
LOAEL-to-NOAEL |
1.25 |
0.60 |
Abdel-Rahman and Kadry 1995 |
Based on NASA's experience in setting SMACs for air contaminants, it appears that practical application of the combined uncertainty factor in setting SWEGs for water contaminants will generally involve only a pair of individual uncertainty factors, one for extrapolating between species and one for extrapolating from a BMD_{10} to a BMD_{01} (or LOAEL to NOAEL). However, an exposure duration uncertainty factor, such as the subchronic-to-chronic factor, might be needed for spaceflights of 1000 d. In general, a factor to account for human variability will not be necessary in the calculation of SWEGs, because variability among individual astronauts would be expected to be small relative to the general population. Based on the NRC formula applied to data from Table 4-2, the upper 95% tolerance limit for animal-to-human variability alone would be
This is near the customary default uncertainty factor of 10. The upper 95% tolerance limit for LOAEL-to-NOAEL (BMD_{10}-to-BMD_{01}) uncertainty alone would be
This is the common default value for this source of uncertainty. Combining the two sources of uncertainty via the above formula gives a combined uncertainty factor (an estimated upper 95% tolerance limit) of
Hence, instead of the product of factors, i.e., 15 × 10 = 150 (or, commonly, 10 × 10 = 100), a reduced factor of 64 can be used with 95% assurance of capturing these two sources of uncertainty. By comparison, the standard product of defaults (10 × 10 = 100) gives about 97% assurance.
The formula for F is intended to apply only to uncertainty factors, and only to those for which estimates of the mean and standard deviation of the distribution are available. It does not apply to exposure conversion factors, factors that account for severity, factors that account for additive or synergistic effects of space flight, factors that ac-
count for data inadequacy, the uncertainty factor for experimental variation, or uncertainty factors for which distributional information is unavailable. All of those factors must continue to be applied separately.
Mixtures of Chemicals
SWEGs for single toxic constituents are set individually, without regard to their occurrence in mixtures with other chemicals in spacecraft water. However, if a substance is present in water, its presence will always be as one component of a complex mixture. Therefore, individual SWEGs must be integrated into group limits to reflect overall water quality conditions judged to be safe for humans in space flight. Those substances that have similar modes of action, or that induce effects in a particular target organ, and might be assumed concentration-additive or perhaps synergistic, could be grouped together, and their respective concentrations, C_{i}, could be determined as follows (NRC 1987a):
C is the measured concentration of a particular chemical in spacecraft water, which is divided by the corresponding SWEG for that chemical.
The group-limit concept has been endorsed by the American Conference of Governmental Industrial Hygienists (ACGIH) for chemical concentrations in air (ACGIH 1991) and by EPA for chemical mixtures that occur in any exposure medium (EPA 1986). For each group with a particular mode of action, a separate group limit calculation should be made for restricting the concentrations of these species in spacecraft water. If it is known or suspected that the action of a mixture of chemicals is greater than additive, then the group limit concept will not guarantee protection. In that case, a further restriction of concentrations is warranted.
Multiple Toxic End Points
In general, toxic substances can affect more than one organ system and have more than one effect within an organ system. In the setting
of SWEGs, all observed toxic effects are considered, including mortality, morbidity (functional impairment), reproductive toxicity, genotoxicity, carcinogenicity, neurotoxicity, immunotoxicity, hepatotoxicity, and respiratory toxicity. For effects that are considered relevant to humans, generally the most sensitive effect determines the SWEG. That is, for a given chemical, potential SWEGs are calculated for a specific exposure duration in space (1, 10, 100, or 1000 d) using data on a variety of toxic end points. Based on the critical significance of the health effect identified, the lowest of those potential SWEGs is then chosen as the SWEG for that substance for that duration of exposure. However, any potential SWEG that is within a factor of 3 of the lowest potential SWEG is also considered a determining factor of that SWEG.
Comparisons with Established Values
All documents used to establish previous industrial or public-health exposure guidance levels for water contaminants should be reviewed before SWEG values are set for NASA. In particular, previous NRC documents on acceptable exposures in drinking water (e.g., NRC 1987b) and water contaminant limits established by EPA, both the maximum contaminant limits and the health advisories (EPA 1996b), provide important reference points for comparison. Such comparisons are not simply to mimic guidance levels set by other entities, but to determine whether the SWEGs that are set in response to NASA's special needs are reasonable in light of previously set concentrations. Finally, the NRC documents on SMACs must be reviewed to ensure compatibility of standards for water with those for air (NRC 1994; 1996a,b; 2000). Any significant differences between exposure levels should be discussed and justified, which would include an evaluation of the approaches and data used to derive the guidance levels.
ALTERNATIVE APPROACHES
This section describes several approaches that are under development for setting acceptable exposure guidelines for toxic substances. These approaches are valid to consider when setting SWEGs, although they generally require more data than are readily available. Progress
in the development of these refined approaches should be monitored, so that they can be included as appropriate in the process of setting SWEGs.
Integrated PBPK and BBDR Models
Just as the multistage model of Armitage and Doll (1960) replaced the earliest susceptibility models, probit and logit, as the primary basis for carcinogenic risk assessment, in recent years more refined biologically based dose-response (BBDR) models have been proposed as replacements for the multistage model. The most popular is the two-stage clonal expansion model of Moolgavkar and colleagues (e.g., Moolgavkar and Luebeck 1990), and extensions thereof (Portier and Kopp-Schneider 1991; Zheng et al. 1995). The BBDR model characterizes the important role of cellular proliferation in cancer. There has been a strong belief that, with sufficient biologic data on the components of the cancer process, complex BBDR models will provide the means for estimating risks below the dose-response range based on biologic knowledge rather than on assumptions.
Concomitant with the refinement of BBDR models has been the development of ever more sophisticated physiologically based pharmacokinetic (PBPK) models, which have been used to obtain better estimates of target tissue doses for risk assessment (Andersen et al. 1993; Kohn et al. 1993). In some cases, this has led to significant modification of risk assessments originally based on the linearized multistage model (e.g., Starr 1990). However, it is only recently that PBPK and BBDR models have been fully integrated into the risk assessment process. Although the results are still the subject of scientific debate, the recent reassessment of TCDD (2,3,7,8-tetrachlorodibenzo-p-dioxin) by EPA (1998) used a fully integrated PBPK-BBDR model to estimate risk of liver cancer in rats. Whether this refined approach will provide more reliable estimates of cancer risk below the experimental dose range is open to question. Nevertheless, the results should be carefully scrutinized to determine whether that is the appropriate direction for risk assessment to take. In the rare case that such data are available for this refined modeling, the exercise certainly should be carried out, and the results should be compared against the general procedure outlined above for setting SWEGs, before SWEG are set.
In addition to more refined biologic models for carcinogenesis, there
have been isolated attempts to develop BBDR models for noncancer effects, specifically for developmental toxicity (Freni and Zapisek 1991; Shuey et al. 1994, 1995; Leroux et al. 1996) and neurotoxicity (Slikker et al. 1998).
Ordinal Regression
A technique called ordinal regression has been proposed as a way to combine data on various toxic effects into a single analysis. The method was first proposed by Hertzberg and Miller (1985) and was later refined by Guth et al. (1991). With ordinal regression, health effects are first assigned to severity categories based on the reported information and consideration of biologic and statistical significance. The aggregate group of subjects at any particular dose and duration of exposure is classified as giving evidence of a specific severity of response. Models such as the logistic regression model are applied with the severity code as the dependent variable and the exposure concentration, duration of exposure, and species as the independent variables. The method allows incorporation of quantal and quantitative data, and it enables the simultaneous analysis of data from many studies. One trade-off is the loss of target-organ toxicity.
The output from ordinal regression is especially useful in that, for any level of severity, it can provide a concentration-by-duration profile (central estimates and confidence limit estimates) for any amount of risk. That capability is particularly useful for making duration conversions, because approaches such as the concentration-by-time conversion are not required. Furthermore, if sufficient human data are available to include in the regression analysis, then interspecies uncertainty is reduced. The ordinal regression method continues to be refined and to be applied to specific toxicants (Simpson et al. 1996), but the complexities of the model fitting appear to make it infeasible for routine use. Nevertheless, whenever possible, the method ought to be applied, and the results should be compared against the general procedure outlined above for obtaining SWEGs before SWEG values are set. Most if not all applications of ordinal regression have been restricted to acute toxic effects, specifically excluding carcinogenic effects. Whether all types of toxic effects, including cancer, can be modeled simultaneously using ordinal regression is still undetermined.
Change-Point Dose-Response Models
In a practical sense, replacing a NOAEL with a BMD_{p} is scientifically justified, in light of the studies reported above that have documented nonzero risk (risk = p > 0, for 0.01 ≤ p ≤ 0.10) associated with NOAELs. In a theoretical sense, however, replacing a NOAEL with a BMD_{p} is inconsistent, because a NOAEL is intended to represent a dose with true zero risk – a threshold. In theory, then, if it were possible to estimate reliably a true threshold dose by way of a mathematical model, it would make sense to replace the NOAEL with this estimated threshold dose, instead of the BMD_{p}. There is research into the use of so-called change-point dose-response models for risk assessment, where the change-point is a dose value that determines where the model changes from a constant response model to a dose response model. Hence, the change-point is a threshold dose, which is a parameter estimated as part of the modeling exercise. It must be emphasized that any estimate of a threshold from current BMD dose-response models is entirely empirical and has no biological basis. If change-point models are shown to be practical, then the guidelines for basing SWEGs on BMD_{p}s should be revisited to evaluate the feasibility of using change-points instead of BMD_{p}s for presumed threshold effects. However, the use of estimated change-points for threshold effects and BMD_{p}s for nonthreshold effects would destroy the unity of the proposed approach for all types of toxic effects, including threshold and nonthreshold effects.
SUMMARY
Using the process of risk assessment to establish SWEGs involves several important steps. Although the intent here has been to provide guidance for implementing this step-by-step process, it must be emphasized that scientific judgment is critical at every step, and it should be the overriding factor throughout the process. Scientific judgment is based on the aggregate of biologic information. Because the process involves a series of extrapolations, each with its own degree of uncertainty, attempting to identify exposures to which specific, low amounts of risk can be attached is not recommended. Instead, emphasis is placed on establishing concentrations that are judged to be reasonably safe for human exposure, based on the best scientific infor-
mation and judgment available. This approach to establishing SWEGs is in line with current thinking on risk assessment, which is moving away from emphasizing numerical estimates of risk for extremely low exposures and is moving toward simply identifying exposures for which the risk of adverse human health effects is judged to be negligible, regardless of whether such effects are carcinogenic or not.
REFERENCES
Abdel-Rahman,M.S., and A.M. Kadry. 1995. Studies on the use of uncertainty factors in deriving RfDs. Hum. Ecol. Risk Assess. 1(5):614-624.
ACGIH (American Conference of Governmental Industrial Hygienists) . 1991. Documentation of the Threshold Limit Values and Biological Exposure Indices, 6th Ed. Cincinnati, OH: American Conference of Governmental Industrial Hygienists.
Allen, B.C., K.S. Crump, and A.M. Shipp. 1988. Correlation between carcinogenic potency of chemicals in animals and humans. Risk Anal. 8(4):531-544.
Allen, B.C., R.J. Kavlock, C.A. Kimmel, and E.M. Faustman. 1994a. Dose-response assessment for developmental toxicity. II. Comparison of generic benchmark dose estimates with no observed adverse effect levels. Fundam. Appl. Toxicol. 23(4):487-495.
Allen, B.C., R.J. Kavlock, C.A. Kimmel, and E.M. Faustman. 1994b. Dose-response assessment for developmental toxicity. III. Statistical models. Fundam. Appl. Toxicol. 23(4):496-509.
Andersen, M.E., J.J. Mills, M.L. Gargas, L. Kedderis, L.S. Birnbaum, D. Neubert, and W.F. Greenlee. 1993. Modeling receptor-mediated processes with dioxin: Implications for pharmacokinetics and risk assessment. Risk Anal. 13(1):25-36.
Andersen, M.E., M.E. Meek, G.A. Boorman, D.J. Brusick, S.M. Cohen, Y.P. Dragan, C.B. Frederick, J.I. Goodman, G.C. Hard, E.J. O'Flaherty, and D.E. Robinson. 2000. Lessons learned in applying the U.S. EPA proposed cancer guidelines to specific compounds. Toxicol. Sci. 53(2):159-172.
Armitage, P., and R. Doll. 1960. Stochastic models for carcinogenesis. Pp. 19-38 in Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics and Probability, Vol. 4, J. Neyman, ed. Berkeley, CA: University of California Press.
Baird, S.J.S., J.T. Cohen, J.D. Graham, A.I. Shlyakhter, and J.S. Evans. 1996. Noncancer risk assessment: A probabilistic alternative to current practice. Hum. Ecol. Risk Assess. 2(1):79-102.
Barnes, D.G., and M. Dourson. 1988. Reference dose (RfD): Description and
use in health risk assessments. Regul. Toxicol. Pharmacol. 8(4):471-486.
Bogen, K.T. 1994. A note on compounded conservatism. Risk Anal. 14(4):379-381..
Bosch, R.J., D. Wypij, and L.M. Ryan. 1996. A semiparametric approach to risk assessment for quantitative outcomes .Risk Anal. 16(5):657-665.
Calabrese, E.J., and L.A. Baldwin. 1995. A toxicological basis to derive generic interspecies uncertainty factors for application in human and ecological risk assessment. Hum. Ecol. Risk Assess 1(5):555-564.
Chen, J.J., and R.L. Kodell. 1989. Quantitative risk assessment for teratological effects.J. Am. Stat. Assoc. 84(408):966-971.
Crump, K.S. 1979. Dose response problems in carcinogenesis.Biometrics 35(1):157-167.
Crump, K.S. 1984. A new method for determining allowable daily intakes. Fundam. Appl. Toxicol. 4(5):854-871.
Crump, K.S. 1995. Calculation of benchmark doses from continuous data.Risk Anal. 15(1):79-89.
Crump, K.S., D.G. Hoel, C.H. Langley, and R. Peto. 1976.Fundamental carcinogenic processes and their implications for low dose risk assessment. Cancer Res.36(9 Pt. 1):2973-2979.
Crump, K.S., H.J. Clewell, and M.E. Andersen. 1996. Cancer and non-cancer risk assessment should be harmonized. BELLE Newslett. 5(2):2-4.
Dourson, M.L. 1994. Methods for establishing oral reference doses (RfDs). Pp. 51-61 in Risk Assessment of Essential Elements, W. Mertz, C.O. Abernathy, and S.S. Olin, eds. Washington, DC: ILSI Press.
Dourson, M.L., and J.F. Stara. 1983. Regulatory history and experimental support of uncertainty (safety) factors. Regul. Toxicol. Pharmacol. 3(3):224-238.
Dourson, M.L., S.P. Felter, and D. Robinson. 1996. Evolution of science-based uncertainty factors in noncancer risk assessment. Regul. Toxicol. Pharmacol. 24(2 Pt. 1):108-120.
EPA (U.S. Environmental Protection Agency). 1986. Guidelines for the Health Risk Assessment of Chemical Mixtures. Fed. Regist. 51(185):34014-34025.
EPA (U.S. Environmental Protection Agency). 1992. Draft report: A Cross-Species Scaling Factor for Carcinogen Risk Assessment Based on Equivalence of mg/kg^{3/4}/day Fed. Regist. 57(109):24152-24173.
EPA (U.S. Environmental Protection Agency). 1996a.. Proposed Guidelines for Carcinogen Risk Assessment Fed. Regist. 61(79):17995-18011.
EPA (U.S. Environmental Protection Agency). 1996b. Drinking Water Regulations and Health Advisories. EPA-822-B-96-002. Office of Water, U.S. Environmental Protection Agency, Washington, DC.
EPA (U.S. Environmental Protection Agency). 1997. National Advisory Committee for Acute Exposure Guideline Levels for Hazardous Substances. Fed. Regist.62(210):58839-58851.
EPA (U.S. Environmental Protection Agency). 1998. Dose-response modeling for 2,3,7,8-TCDD. Chapter 8 in Health Assessment Document for 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) and Related Compounds. EPA/600/ P-92/001C8. [Online]. Available: http://www.epa.gov/ordntrnt/ORD/ WebPubs/dioxin/index.html
Farmer, J.H., R.L. Kodell, and D.W. Gaylor. 1982. Estimation and extrapolation of tumor probabilities from a mouse bioassay with survival/sacrifice components. Risk Anal. 2(1):27-34..
Farrar, D., B. Allen, K. Crump, and A. Shipp. 1989. Evaluation of uncertainty in input parameters to pharmacokinetic models and the resulting uncertainty in output. Toxicol. Lett.49(2-3):371-385.
Freni, S.C., and W.F. Zapisek. 1991. Biologic basis for a risk assessment model for cleft palate. Cleft Palate Craniofac. J.28(4):338-346.
Gaylor, D.W. 1983. The use of safety factors for controlling risk. J. Toxicol. Environ. Health 11(3):329-336.
Gaylor, D.W. 1992. Incidence of developmental defects at the no observed adverse effect level (NOAEL). Regul. Toxicol. Pharmacol 15(2 Pt. 1):151-160.
Gaylor, D.W., and J.J. Chen. 1996. A simple upper limit for the sum of the risks of the components in a mixture. Risk Anal.16(3):395-398.
Gaylor, D.W., and R.L. Kodell. 2000. Percentiles of the product of uncertainty factors for establishing probabilistic reference doses. Risk Anal.20(2):245-250.
Gaylor, D.W., and W. Slikker, Jr. 1990. Risk assessment for neurotoxic effects. Neurotoxicology11(2):211-218.
Gaylor, D.W., R.L. Kodell, J.J. Chen, J.A. Springer, R.J. Lorentzen, and R.J. Scheuplein. 1994. Point estimates of cancer risk at low doses. Risk Anal. 4(5):843-850.
Gaylor, D.W., R.L. Kodell, J.J. Chen, and D. Krewski. 1999. A unified approach to risk assessment for cancer and noncancer endpoints based on benchmark doses and uncertainty/safety factors. Regul. Toxicol. Pharmacol. 29(2 Pt. 1):151-157.
Guth, D.J., A.M. Jarabek, L. Wymer, and R.D. Hertzberg. 1991. Evaluation of Risk Assessment Methods for Short-term Inhalation Exposure . Abstract 91-173.2 in Proceedings of the 84th Annual Meeting of the Air and Waste Management Association. Pittsburgh, PA: Air & Waste Management Association.
Hertzberg, R.C., and M. Miller. 1985. A statistical model for species extrapolation using categorical response data. Toxicol. Ind. Health 1(4):43-57.
Hill, R.N., T.M. Crisp, P.M. Hurley, S.L. Rosenthal, and D.V. Singh. 1998. Risk assessment of thyroid follicular cell tumors. Environ. Health Perspect. 106(8):447-457.
James, J.T., and D.E. Gardner. 1996. Exposure limits for airborne contaminants
in spacecraft atmospheres. Appl. Occup. Environ. Hyg. 11(12):1424-1425.
Kaplan, H.L. 1979. Contaminants. In The Physiological Basis for Spacecraft Environmental Limits. J.M. Waligora, coordinator. NASA-RP-1045. NASA Center for AeroSpace Information, Hanover, MD.
Kavlock, R.J., B.C. Allen, E.M. Faustman, and C.A. Kimmel. 1995. Dose-response assessments for developmental toxicity. IV. Benchmark doses for fetal weight changes. Fundam. Appl. Toxicol. 26(2):211-222.
Kimmel, C.A., and E.W. Gaylor. 1988. Issues in qualitative and quantitative risk analysis for developmental toxicology. Risk Anal. 8(1):15-20.
Klaassen, C.D., and D.L. Eaton. 1991. Principles of toxicology. Pp. 12-49 in Casarett and Doull's Toxicology: The Basic Science of Poisons, 4th Ed., M.O. Amdur, J. Doull, and C.D. Klaassen, eds. New York: Macmillan.
Kodell, R.L., and D.W. Gaylor. 1999. Combining uncertainty factors in deriving human exposure levels of noncarcinogenic toxicants. Ann. N.Y. Acad. Sci. 895:188-195.
Kodell, R.L., and C.N. Park. 1995. Linear extrapolation in cancer risk assessment. Pp. 87-104 in Low-Dose Extrapolation of Cancer Risks: Issues and Perspectives, W. Farland, S. Olin, C. Park, L. Rhomberg, R. Scheuplein, T. Starr, and J. Wilson, eds. Washington, DC: ILSI Press.
Kodell, R.L., and R.W. West. 1993. Upper confidence limits on excess risk for quantitative responses . Risk Anal. 13(2):177-182.
Kodell, R.L., D.W. Gaylor, and J.J. Chen. 1987. Using average lifetime dose rate for intermittent exposures to carcinogens . Risk Anal.7:339-345.
Kohn, M.C., G.W. Lucier, G.C. Clark, C. Sewall, A.M. Tritscher, and C.J. Portier. 1993. A mechanistic model of effects of dioxin on gene expression in the rat liver. Toxicol. Appl. Pharmacol. 120(1):138-154.
Krewski, D., and J. Van Ryzin. 1981. Dose response models for quantal response toxicity data. Pp. 201-231 in Statistics and Related Topics: Proceedings of the International Symposium on Statistics and Related Topcis, M. Csorgo, D.A. Dawson, J.N.K. Rao, and A.K.Md.E. Saleh, eds. Amsterdam: North-Holland.
Krewski, D., and Y. Zhu. 1995. A simple data transformation for estimating benchmark doses in developmental toxicity experiments. Risk Anal. 15(1):29-39.
Lehman, A.J., and O.G. Fitzhugh. 1954. 100-fold margin of safety. Assoc. Food Drug Off., Q. Bull. 18:33-35.
Leisenring, W., and L. Ryan. 1992. Statistical properties of the NOAEL. Regul. Toxicol. Pharmacol. 15(2 Pt. 1):161-171.
Lensing, S.Y., and R.L. Kodell. 1995. Fitting the two-stage clonal expansion model based on exact hazard to the ED_{01} data using SAS NLIN. Risk Anal. 15(2):233-245.
Leroux, B.G., W.M. Leisenring, S.H. Moolgavkar, and E.M. Faustman. 1996. A biologically-based dose-response model for developmental toxicology . Risk Anal. 16(4):449-458.
Mantel, N., and W.R. Bryan. 1961. “Safety” testing of carcinogenic agents. J. Natl. Cancer Inst. 27(2):455-470.
Merrill, A.H., Jr., M. Hoel, E. Wang, R.E. Mullins, J.L. Hargrove, D.P. Jones, and I.A. Popova. 1990. Altered carbohydrate, lipid, and xenobiotic metabolism by liver from rats flown on Cosmos 1887. FASEB J. 4:95-100; erratum FASEB J. (1990) 4(8):2539.
Moolgavkar, S.H., and G. Luebeck. 1990. Two-event model for carcinogenesis: Biological, mathematical, and statistical considerations. Risk Anal. 10(2):323-341.
Munro, I.C., and D.R. Krewski. 1981. Risk assessment and regulatory decision making. Food Cosmet. Toxicol. 19(5):549-560.
Murdoch, D.J., D. Krewski, and J. Wargo. 1992. Cancer risk assessment with intermittent exposure. Risk Anal. 12(4):569-577.
NRC (National Research Council). 1970. Evaluating the Safety of Food Chemicals: Report of the Food Protection Committee. Washington, DC: National Academy of Sciences .
NRC (National Research Council). 1986. Criteria and Methods for Preparing Emergency Exposure Guidance Level (EEGL), Short-Term Public Emergency Guidance Level (SPEGL), and Continuous Exposure Guidance Level (CEGL) Documents. Washington, DC: National Academy press.
NRC (National Research Council). 1987a. Drinking Water and Health, Vol. 7–Disinfectants and Disinfectant By-Products. Washington, DC: National Academy Press.
NRC (National Research Council). 1987b. Drinking Water and Health, Vol. 8–Pharmacokinetics in Risk Assessment. Washington, DC: National Academy Press.
NRC (National Research Council). 1992. Guidelines for Developing Spacecraft Maximum Allowable Concentrations for Space Station Contaminants. Washington, DC: National Academy Press.
NRC (National Research Council). 1994. Spacecraft Maximum Allowable Concentrations for Selected Airborne Contaminants, Vol. 1. Washington, DC: National Academy Press.
NRC (National Research Council). 1996a. Spacecraft Maximum Allowable Concentrations for Selected Airborne Contaminants, Vol. 2. Washington, DC: National Academy Press.
NRC (National Research Council). 1996b. Spacecraft Maximum Allowable Concentrations for Selected Airborne Contaminants, Vol. 3. Washington, DC: National Academy Press.
NRC (National Research Council). 2000. Spacecraft Maximum Allowable Concentrations for Selected Airborne Contaminants, Vol. 4. Washington, DC: National Academy Press.
OSTP (Office of Science and Technology Policy). 1985. Chemical Carcinogens: A Review of the Science and Its Associated Principles. Fed. Regist. 50(50):10371-10442.
Page, N.P., D.V. Singh, W. Farland, J.I. Goodman, R.B. Conolly, M.E. Andersen, H.J. Clewell, C.B. Frederick, H. Yamasaki, and G. Lucier. 1997. Implementation of EPA revised cancer risk assessment guidelines: Incorporation of mechanistic and pharmacokinetic data. Fundam. Appl. Toxicol. 37(1):16-36.
Portier>, C.J., and N.L. Kaplan. 1989. Variability of safe dose estimates when using complicated models of the carcinogenic process. A case study: Methylene chloride. Fundam. Appl. Toxicol. 13(3):533-544.
Portier, C.J., and A. Kopp-Schneider. 1991. A multistage model of carcinogenesis incorporating DNA damage and repair. Risk Anal. 11(3): 535-543.
Purchase, I.F., and T.R. Auton. 1995. Thresholds in chemical carcinogenesis. Regul. Toxicol. Pharmacol. 22(3):199-205.
Renwick, A.G. 1995. The use of an additional safety or uncertainty factor for nature of toxicity in the estimation of acceptable daily intake and tolerable daily intake values. Regul. Toxicol. Pharmacol.22(3):250-261.
Ryan, L. 1992. Quantitative risk assessment for developmental toxicity. Biometrics 48(1):163-174.
Simpson, D.G., R.J. Carroll, H. Zhou, and D.J. Guth. 1996. Interval censoring and marginal analysis in ordinal regression. J. Agri. Biol. Environ. Stat. 1:354-376.
Shuey, D.L., C. Lau, T.R. Logsdon, R.M. Zucker, K.H. Elstein, M.G. Narotsky, R.W. Setzer, R.J. Kavlock, and J.M. Rogers. 1994. Biologically based dose-response modeling in developmental toxicology: Biochemical and cellular sequelae of 5-fluorouracil exposure in the developing rat. Toxicol. Appl. Pharmacol. 126(1):129-144.
Shuey, D.L., R.W. Setzer, C. Lau, R.M. Zucker, K.H. Elstein, M.G. Narotsky, R.J. Kavlock, and J.M. Rogers. 1995. Biological modeling of 5-fluorouracil developmental toxicity. Toxicology 102(1-2):207-213.
Slikker, W., Jr., A.C. Scallet>, and D.W. Gaylor. 1998. Biologically-based dose-response model for neurotoxicity risk assessment . Toxicol. Lett. 102-103:429-433.
Slob, W. 1994. Uncertainty analysis in multiplicative models. Risk Anal. 14(4):571-576.
Starr, T.B. >1990. Quantitative cancer risk estimation for formaldehyde. Risk Anal. 10(1):85-91.
Swartout, J. 1996. Subchronic-to-chronic uncertainty factor for the reference dose. Abstract F2.03. Society for Risk Analysis Annual Meeting, New Orleans, LA
Swartout, J.C., P.S. Price, M.L. Dourson, H.L. Carlson-Lynch, and R.E. Keenan. 1998. A probabilistic framework for the reference dose (probabilistic RfD) . Risk Anal. 18(3):271-282.
ten Berge, W.F., A. Zwart, and L.M. Appleman. 1986. Concentration-time mortality response relationship of irritant and systemically acting vapours and gases. J. Hazard. Mater. 13(3):301-309.
Travis, C.C., and R.K. White. 1988. Interspecific scaling of toxicity data. Risk Anal. 8(1):119-125.
Van Ryzin, J. 1980. Quantitative risk assessment. J. Occup. Med. 22(5):321-326.
West, J.B., A.R. Elliot, H.J. Guy, and G.K. Prisk. 1997. Pulmonary function in space. JAMA 277(24):1957-1961.
Wilson, J.D. 1997. So carcinogens have thresholds: How do we decide what exposure levels should be considered safe? Risk Anal. 17(1):1-3.
Wiltse, J.A., and V.L. Dellarco. 2000. U.S. Environmental Protection Agency's revised guidelines for carcinogen risk assessment: evaluating a postulated mode of carcinogenic action in guiding dose-response extrapolation . Mutat. Res. 464(1):105-115.
Zeise, L., R. Wilson, and E.A. Crouch. 1987. Dose-response relationships for carcinogens: A review. Environ. Health Perspect. 73:259-306.
Zheng, Q., D.W. Gaylor, and W.K. Lutz. 1995. A carcinogenesis model describing mutational events at the nucleotide level. Proc. Biometrics Sec. Am. Statist. Assoc. 349-354.