Update project Chapter 5: Dose–response assessment Draft May 2008

1 PRINCIPLES AND METHODS FOR THE RISK ASSESSMENT OF CHEMICALS IN FOOD

2
3 CHAPTER 5: DOSE–RESPONSE ASSESSMENT
4
5
6 CHAPTER 5: DOSE–RESPONSE ASSESSMENT.................................................................. 1
7 5.1 INTRODUCTION .................................................................................................................. 1
8 5.2 DOSE–RESPONSE ASSESSMENT: BASIC CONCEPTS .............................................................. 3
9 5.2.1 Dose........................................................................................................................... 3
10 5.2.2 Response.................................................................................................................... 4
11 5.2.3 Models ....................................................................................................................... 4
12 5.2.4 Dose–response assessment and dose–response modelling ....................................... 5
13 5.2.5 Issues of extrapolation............................................................................................... 8
14 5.3 DOSE–RESPONSE MODELLING ............................................................................................ 8
15 5.3.1 General perspectives ................................................................................................. 8
16 5.3.2 The NOAEL approach to deriving health-based guidance values .......................... 10
17 5.3.3 Benchmark dose approach to deriving health-based guidance values ................... 11
18 5.3.4 Uncertainty.............................................................................................................. 13
19 5.3.5 Study design............................................................................................................. 13
20 5.3.6 Biological information ............................................................................................ 14
21 5.4 PRINCIPLES OF DOSE–RESPONSE MODELLING .................................................................. 14
22 5.4.1 Data ......................................................................................................................... 14
23 5.4.2 Mathematical models .............................................................................................. 14
24 5.4.2.1 Dose–response models for continuous data ..................................................... 14
25 5.4.2.2 Dose–response models for quantal data ........................................................... 15
26 5.4.2.3 Modelling with covariates ................................................................................ 16
27 5.4.2.4 Biologically based dose–response models ....................................................... 17
28 5.4.3 Model fitting and estimation of parameters ............................................................ 17
29 5.5 REFERENCES .................................................................................................................... 17
30
31
32 5.1 Introduction
33 Risk assessment approaches generally take one of two forms: analyses that provide a
34 quantitative (or sometimes just qualitative) estimation of risk, and analyses that establish
35 health-based guidance values, such as an ADI or TDI, which are levels of human exposure
36 considered to be “without appreciable health risk” (IPCS, 1987, 1990). The latter approach,
37 which is often described as “safety assessment”, is used more often in cases where exposure
38 can be controlled, such as for food additives and residues of pesticides and veterinary drugs in
39 foods.
40 One of the primary criteria of a risk assessment is determination of the presence or
41 absence of a cause–effect relationship. If there is sufficient plausibility for the presence of
42 such a relationship, then dose–response data are essential, and dose–response analysis is a
43 major part of the hazard characterization within the risk assessment paradigm.
44 Dose–response data may be derived from in vivo studies in animals or humans, which
45 usually provide the basis for risk characterization, and in vitro studies, which are often related
46 to investigations of mode of action. In each case, interpretation of the effect data usually
47 requires recognition of the levels of exposure that do not produce a measurable effect and the
48 relationship between the increase in incidence, severity or nature of the effect with increase in
49 exposure.

1
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 Toxicological or epidemiological data have been used in hazard characterization by

2 JECFA and JMPR in three main ways (see chapter 7):
3
4 1) derivation of a health-based guidance value, such as an ADI, TDI or ARfD;
5 2) estimation of the margin of exposure (MOE) between a defined point on the dose–
6 response relationship and the level of human exposure; and
7 3) quantification of the magnitude of the risk at a prespecified level or levels of human
8 exposure.
9
10 In addition, it is possible to use dose–response data to define the exposure that
11 theoretically would be associated with some prespecified level of risk, such as a 1 in 10−6
12 increase in lifetime risk of cancer.
13 Having established that there is a statistically significant, treatment-related or
14 exposure-related effect that is relevant to human health, the calculation of a health-based
15 guidance value or MOE requires definition of a reference point or point of departure (POD)
16 on the dose–response relationship. There have been two basic approaches to dose–response
17 assessment applied to data from studies in animals:
18
19 1) Pairwise comparisons of the findings in different groups in order to define experimental
20 doses that cause statistically significant effect(s) and the highest experimental dose that
21 does not produce a measurable effect, the NOAEL. The NOAEL is then used as the POD
22 to estimate a health-based guidance value, after allowing for uncertainties such as species
23 differences and human variability.
24 2) Fitting a model(s) to the dose–response data for all groups in order to define the
25 relationship in the observed range; the model can then be used to define the exposure
26 associated with a prespecified level of response. This value can then be used as the POD
27 to estimate a health-based guidance value or calculate an MOE or extrapolated to estimate
28 the risk at the levels of human exposure that are relevant to problem formulation and risk
29 characterization.
30
31 These approaches and variants on them are discussed in this chapter, which is based
32 on a draft EHC on Principles for Modelling Dose–Response for the Risk Assessment of
33 Chemicals that has been developed as part of the IPCS Harmonization Project on Approaches
34 to the Assessment of Risk from Exposure to Chemicals. The Dose–Response Modelling
35 Planning Group met in October 2002 in Geneva to develop an outline, and a first draft
36 working paper was discussed at the Task Group Meeting in September 2004 and subsequently
37 revised and edited into the draft EHC (IPCS, 2008).
38 The draft EHC (IPCS, 2008) covers toxicants with threshold effects and those for
39 which there may be no practical threshold, such as substances that are genotoxic and
40 carcinogenic. The draft EHC focuses primarily on experimental animal studies, but dose–
41 response relationships are also critical to the assessment of human experimental studies and
42 epidemiological data. Dose–response assessment is also important for studies that attempt to
43 define the relationships of different steps in a postulated mode of action. The draft EHC also
44 includes areas that are not of direct relevance to this chapter, such as the basic risk analysis
45 paradigm and the consequences of dose–response modelling (DRM) for the advice provided
46 by risk assessors to risk managers.
47

2
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 5.2 Dose–response assessment: basic concepts

2 5.2.1 Dose
3 It is critical when performing dose–response analyses to have a clear concept of what type of
4 “dose” has been used in the available dose–response data. There are three basic types of
5 “dose” that arise from scientific investigations; they are inter-related, and each of them can be
6 used to express dose–response relationships. They are 1) the administered or external dose, 2)
7 the internal (absorbed) dose and 3) the target or tissue dose.
8 External dose denotes the amount of an agent or chemical administered to an
9 experimental animal or human in a controlled experimental setting by some specific route at
10 some specific frequency. In the terminology used by JECFA, the external dose is often
11 referred to as exposure or intake (see chapter 6). External dose, or external exposure, is
12 frequently the dose metric that is used in observational epidemiological studies.
13 Internal dose is the amount that is systemically available and can be regarded as the
14 fraction of the external dose that is absorbed and enters the general circulation. It is a
15 consequence of absorption, distribution, metabolism and excretion of the chemical and can be
16 derived from suitable toxicokinetic mass balance studies. The analytical method used in the
17 toxicokinetic studies will determine whether the “dose” refers to the parent compound alone
18 or the parent compound plus first-pass metabolites (see section 4.3). Biomarkers of internal
19 dose, such as plasma concentrations or urinary excretion, are sometimes available in
20 epidemiological studies.
21 The tissue dose is the amount that is distributed to and present in a specific tissue of
22 interest. As for internal dose, the analytical method used in the toxicokinetic studies will
23 determine whether the “dose” refers to the parent compound alone or the parent compound
24 plus first-pass metabolites (see section 4.3). An additional consideration for tissue dose is
25 whether the dose metric is the peak concentration or a time-weighted average, such as the
26 AUC.
27 Two parameters are important determinants of dose: the dose frequency and duration
28 of dosing. Dosing can be acute, subchronic or chronic; the term “dose” can apply to any of
29 these, and the principles of dose–response assessment apply to all three forms. The
30 description of dose should reflect the magnitude, frequency and duration over which it applies.
31 Dose can be expressed in a variety of metrics, including a simple single external dose (e.g.
32 mg/kg body weight), daily intake (e.g. mg/kg body weight per day), total body burden (e.g.
33 ng/kg body weight), body burden averaged over a given period of time or tissue concentration.
34 In epidemiological studies, exposure (the external dose) is rarely known precisely, and
35 its estimation often requires various assumptions. Sometimes exposure is measured by the
36 biomonitoring of blood or tissue concentrations; dose–response assessment for such data
37 usually raises the issue of conversion of the biomarker of internal exposure into an external
38 dose. An additional problem that arose (e.g. with the dioxin database) was that measurements
39 of the biomarker were made many years after what was believed to be the period of highest
40 exposure (WHO, 2002).
41 Sometimes the doses used in an animal study are transformed to the equivalent human
42 exposures prior to dose–response modelling (DRM). In this situation, models of internal
43 exposure linked to the response data may be used to develop a dose–response model.
44 However, such models need knowledge for animals and humans of the events controlling
45 absorption, tissue distribution, metabolism, excretion and the other molecular and biochemical
46 processes that ultimately lead to particular responses. Interspecies extrapolation of such a dose
47 metric may be possible by the use of a PBTK model. While this more sophisticated approach
48 can refine DRM, incomplete data will add uncertainty to the output of the modelling. The
49 issue of interspecies extrapolation is usually addressed separately and subsequent to DRM
50 using the unadjusted animal data and application of an uncertainty factor (section 5.3.2).

3
 Update project Chapter 5: Dose–response assessment Draft May 2008

1
2 5.2.2 Response
3 Response, in this context, generally relates to an observation or effect seen following exposure
4 in vivo or in vitro. Possible end-points cover a broad range of observations, from early
5 responses such as biochemical alterations to more complicated responses such as cancer and
6 developmental defects.
7 Responses can be either adaptive or adverse. Adverse effects are defined as a change
8 in the morphology, physiology, growth, development, reproduction or life span of an
9 organism or subsystem (e.g. subpopulation of cells) that results in an impairment of functional
10 capacity, an impairment of the capacity to compensate for additional stress or an increase in
11 susceptibility to other influences (IPCS, 2004). The responses are sometimes species and/or
12 tissue specific and have different degrees of variation across individuals. DRM can address
13 each response, provide insight into their quantitative similarity across species and tissues and
14 link responses in a mechanistically reasonable manner.
15 Response is generally considered to vary across experimental units (animals, humans,
16 cell cultures) in the same dose group in a random fashion. This random variation is usually
17 assumed to follow some statistical distribution describing the frequency of any given response
18 for a population. In general, statistical distributions are characterized by their central tendency
19 (usually the mean or average value) and their effective range (usually based on the standard
20 deviation).
21 Most responses of interest in the context of dose–response assessment fall into one of
22 four basic categories:
23
24 • Quantal responses: These generally relate to an effect that is either observed or not
25 observed in each individual subject (laboratory animal or human); for each dose, the
26 number of subjects responding out of the number of subjects available is reported (e.g. the
27 proportion of animals with a tumour in a cancer bioassay).
28 • Counts: These generally relate to a discrete number of items measured in a single
29 experimental unit (e.g. number of papillomas on the skin).
30 • Continuous measures: These generally relate to a quantitative measurement that is
31 associated with each individual subject and can take on any value within a defined range
32 (e.g. body weight).
33 • Ordinal categorical measures: These generally take on one value from a small set of
34 ordered values (e.g. tumour severity grades); ordinal data are an intermediate type of data
35 and reflect (ordered) severity categories—i.e. they are qualitative data but with a rank
36 order (e.g. histopathological severity data) in each individual. When the categories are
37 non-ordered, they are called categorical data, but these are rare for response data.
38
39 Sometimes it is useful for DRM purposes to convert continuous data into proportions
40 (e.g. number of animals outside a clinically relevant range for an immune system marker) or
41 categories (e.g. measured degree of liver necrosis converted to minimal, moderate or
42 extensive).
43 There are some differences in how each of these different types of data are handled for
44 DRM, but as a general rule, the goal of DRM is to describe the mean and variance of the
45 response as a function of exposure and/or time.
46
47 5.2.3 Models
48 Dose–response models are mathematical expressions fitted to scientific data that characterize
49 the relationship between dose and response. Mathematical models consist of three basic

4
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 components: 1) assumptions used to derive the model, 2) a functional form for the model and
2 3) parameters that are components of the functional form.
3 The simplest dose–response model is a linear model to describe a continuous response
4 for which the key components are:
5
6 • assumptions: the mean added response is proportional to dose;
7 • functional form: R(D) = α+β•D where R(D) is the mean response as a function of dose D;
8 and
9 • parameters: α is a parameter describing the mean response in the control (unexposed)
10 group and β is a parameter describing the mean change in response per unit dose.
11
12 Dose–response models range from very simple models, such as the linear model
13 described above, to extremely complicated models for which the eventual functional form
14 cannot easily be expressed as a single equation (e.g. biologically based dose–response
15 models).
16 Models can also be linked, meaning that one model could describe part of the dose–
17 response process while another describes the remainder of the process. For example, for
18 chemical carcinogenesis, in most cases tissue concentration is more closely linked to cancer
19 risk than is administered dose. Given data on dose, tissue concentration and tumour response,
20 a toxicokinetic model may be able to relate external dose to tissue concentration, and a
21 multistage cancer model may be able to relate tissue concentration to response. The two
22 models need to be combined in order to describe the dose–response relationship.
23 Dose–response models may incorporate other information into the model form. Age
24 and time-on-study are commonly used in DRM, but other factors, such as
25 species/strain/human ethnicity, sex and body weight, have also been used to expand the utility
26 of dose–response models.
27
28 5.2.4 Dose–response assessment and dose–response modelling
29 DRM can be described by six basic steps, with a variety of options at each step (Table 5.1).
30 The first four steps relate to the analysis of the dose–response data, which is referred to as
31 dose–response analysis (IPCS, 2008). Dose–response analysis provides the linkage of a model
32 to dose–response data for the purposes of predicting response to a given dose or predicting the
33 dose causing a given level of response. The last two steps deal with implementation and
34 evaluation of the results of the analysis.
35
36 Table 5.1. Basic steps in dose–response assessment/modelling (adapted from IPCS, 2008)
37
Step Description Options
1. Data selection Determine the response to be modelled and End-point, quality, sample
select appropriate data size, utility, availability
2. Model selection Choose the type of model to be applied to the End-point, data availability,
data purpose
3. Statistical linkage Assume what statistical distributions describe End-point, data type, model
the response choice, software availability
4. Parameter Combine the first three steps in an Linkage function, software
estimation appropriate computer program to obtain availability, variance
estimates of the model parameters
5. Implementation Use the estimated model parameters and the Outputs, target selection,
model formula to predict response/dose as model predictions, BMD,
needed direct extrapolation

5
 Update project Chapter 5: Dose–response assessment Draft May 2008

Step Description Options

6. Evaluation Examine the sensitivity of the resulting Model comparison,
predictions to the assumptions used in the uncertainty
analysis.
1
2 Step 1 involves selection of appropriate data for dose–response assessment. The
3 criteria applied to assess whether the data are suitable for risk characterization purposes are
4 similar whether hazard characterization is based on pairwise analyses of groups or modelling
5 using all dose groups.
6 Step 2 involves the choice of an appropriate model. The type of data available can
7 have a marked impact on the complexity of the model that can be used. For example, while
8 two points can be used to identify the slope of a line, it takes at least three points to identify
9 the shape of a more complex dose–response relationship. The issue of whether there are
10 enough data to support a given model is complex (see IPCS, 2008). Models may be divided
11 into two categories, empirical and biologically based models. Most DRM to date has used
12 empirical models—i.e. mathematical descriptions of the data that are not based on a
13 mechanism of action. Biologically based models generally are based on basic principles about
14 the onset and progression of disease in a biological system, are functionally complex and have
15 far greater data requirements than empirical models.
16 Step 3 requires the choice of a statistical linkage between the data and the model. The
17 most common linkage method is to assume a statistical distribution for the response and use
18 that distribution to derive a mathematical function describing the quality of the fit of the
19 model to the data. However, a considerable amount of DRM has been done by simpler linkage
20 functions, such as drawing a straight line through the data points. The advantage of choosing a
21 formal statistical linkage is the ability to test hypotheses and derive confidence intervals for
22 model predictions.
23 Step 4 is the fitting of the selected model(s) to the data. Since the primary components
24 of a model are the parameters that define the model, curve fitting simply involves choosing
25 values for the parameters in the model. If a formal statistical linkage has been developed for
26 linking the data to the model, then the parameters are chosen such that they “optimize” the
27 value of the linkage function. A common choice is to link the data to the model by minimizing
28 the sum of the squares of the differences between the predicted value from the model and the
29 observed value. Simpler methods can also be used to estimate model parameters. Formal
30 optimization is a better choice for modelling than ad hoc procedures, which often do not meet
31 the criterion of transparency.
32 Step 5 is to make the inferences necessary to address the risk assessment questions
33 developed at the problem formulation stage. Usually, when making a prediction, the emphasis
34 is on the change in response seen in treated animals compared with response (baseline data) in
35 the control animals. The different types of data (quantal, count, continuous, categorical)
36 require different methods for predicting changes in response beyond the normal response. In
37 general, treatment-related responses may be described by added response (treated minus
38 control response), relative response (fold change relative to control response) and extra
39 response (added response scaled to range from zero to the maximum possible response). Each
40 of these choices can impact the final decision, so care should be taken to understand why a
41 specific choice is made. Development of risk assessment advice usually requires extrapolation
42 of results from the specific responses seen for the experiment being modelled to other
43 exposure scenarios and other doses. This step can also involve an extrapolation from a
44 laboratory species to humans.
45 DRM may be used to develop risk assessment advice in a variety of ways:
46

6
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 1) DRM may be used to define levels of exposure where the response measurement is
2 assumed to be virtually unchanged relative to the control measurement. In this approach, a
3 simple functional model such as a straight line can be used. An alternative and simpler
4 approach is the use of pairwise comparisons of the data for different dose levels to define
5 the NOAEL, which is used as a POD for the observed dose–response data.
6 2) The dose–response model(s) may be used to identify a dose with a known level of
7 response at or slightly below the observable range. The prespecified level of response is
8 known as the benchmark response (BMR), and the dose associated with that response the
9 BMD. The lower one-sided confidence limit on the BMD (the BMDL) can be used as the
10 POD for the derivation of a heath-based guidance value or for calculation of an MOE.
11 Alternatively, the BMDL may be the starting point for linear low-dose extrapolation (see
12 below).
13 3) The model(s) may be used to find the dose associated with a negligible (e.g. 1 in a million)
14 response over control. In general, this requires extrapolation far beyond the range of the
15 data, which creates considerable uncertainty.
16
17 Approach 1 is used currently by JECFA and JMPR to derive health-based guidance
18 values in order to protect against effects that are considered to show a threshold.
19 Approach 2 was used by JECFA at its sixty-fourth meeting (WHO, 2006) to define
20 MOEs for a number of genotoxic carcinogens. The same meeting also considered the use of
21 linear extrapolation from the BMDL to estimate the risk of cancer at relevant levels of human
22 exposure and concluded that “calculation of the intake associated with an incidence of 1 in 1
23 million from the BMDL for a 10% incidence using linear extrapolation is simply equivalent to
24 dividing the BMDL by 100,000, and this approach is therefore no more informative than
25 calculation of a MOE”.
26 Approach 3 was considered by JECFA at its sixty-fourth meeting (WHO, 2006), and
27 the Committee concluded that
28
29 In order to provide realistic estimates of the possible carcinogenic effect at the estimated exposure for
30 humans, mathematical modelling would need to take into account the shape of the dose–response
31 relationship for the high doses used in the bioassay for cancer and for the much lower intakes by
32 humans. Such information cannot be derived from the available data on cancer incidence from studies in
33 animals. In the future, it may be possible to incorporate data on dose–response or concentration–
34 response relationships for the critical biological activities involved in the generation of cancer (e.g.
35 metabolic bioactivation and detoxication processes, DNA binding, DNA repair, rates of cell
36 proliferation and apoptosis) into a biologically based dose–response model for cancer that would also
37 incorporate data on species differences in these processes. However, such data are not currently
38 available. At present, any estimate of the possible incidence of cancer in experimental animals at intakes
39 equal to those for humans has to be based on empirical mathematical equations that may not reflect the
40 complexity of the underlying biology. A number of mathematical equations have been proposed for
41 extrapolation to low doses. The resulting risk estimates are dependent on the mathematical model used;
42 the divergence increases as the dose decreases and the output by different equations can differ by orders
43 of magnitude at very low incidences.
44
45 In step 6, the basic steps of DRM shown in Table 5.1 are repeated to consider other
46 options in the process in order to understand the impact of choices on the health-based
47 measures derived from DRM. This final step is aimed at understanding the sensitivity of the
48 analysis to specific choices and to judge the overall quality of the final predictions. Depending
49 on the degree of difference between choices, there could be value in performing a formal
50 analysis of the quality of the fit of the model to the data. Other methods can also be used to
51 assess the impact of choices used in the modelling on the eventual outcome, such as
52 uncertainty analysis and Bayesian mixing.
53

7
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 5.2.5 Issues of extrapolation

2 Extrapolation is a necessary part of all risk assessments, except in those rare cases where
3 DRM uses data from studies in sufficient numbers of humans who are representative of the
4 potential exposed population and who have had a level of exposure similar to that which is of
5 concern.
6 Most of the methods used to implement the results of a dose–response analysis (step 5)
7 address these extrapolation issues. The strategies used for extrapolation basically fall into two
8 categories: those aimed at providing estimates of risk for exposures outside of the range of the
9 data used in the dose–response analysis, and those aimed at establishing health-based
10 guidance values, such as the ADI, without quantification of risk. The methods that have been
11 used for extrapolation are diverse and sometimes contentious, with different countries, and
12 even different agencies within a given country, using different approaches.
13 Even when human data are available and suitable for dose–response analysis, they are
14 generally from selected populations or study groups, such as workers in occupational settings,
15 whose exposures differ from the general population. Thus, dose–response analyses normally
16 need to be extrapolated from the observed conditions where scientific support is available to
17 conditions where scientific support is weaker or non-existent. For dose–response analyses
18 based on human studies, extrapolation is generally a downward extrapolation to different
19 levels of exposure, but can also be to different life stages (e.g. fetus, child) or different
20 populations with different environmental factors that might affect exposure (e.g. dietary
21 differences).
22 In most cases considered by JECFA and JMPR, the data used for DRM come from
23 experiments in laboratory animals administered doses significantly exceeding the potential
24 human exposure. For such dose–response analyses, there are two issues of extrapolation:
25 extrapolating from the test species to humans and allowing for possible human differences in
26 response. The methods employed for these extrapolation issues are varied, ranging from the
27 use of uncertainty factors (see sections 5.3.2 and 5.3.3) to more complicated modelling
28 schemes based upon differences in toxicokinetics and toxicodynamics between humans and
29 animals and between different human individuals.
30 Extrapolation issues are most commonly dealt with by using uncertainty factors that
31 have been developed over years of experience. Alternatively, an MOE may be calculated as
32 the numerical ratio of the point of departure on the dose–response data divided by the
33 estimated human exposure; however, such a ratio does not allow for issues of extrapolation,
34 and these will need to be discussed as part of the MOE analysis.
35
36 5.3 Dose–response modelling
37 5.3.1 General perspectives
38 The NOAEL approach identifies the POD as a single dose that is assumed to be without
39 appreciable effect, whereas the BMD approach provides a POD based on data from the entire
40 observed dose–response range for the critical effect. The BMD approach thus reflects the
41 characteristics of the dose–response curve, particularly in providing estimates of the slope. In
42 the case of a regression framework, it provides the standard error and confidence interval for
43 the model parameters. The NOAEL approach relies on the concept that the NOAEL is
44 sufficiently distant from the lowest dose associated with an effect within the observable range
45 that it will be without appreciable risk under the conditions of the experiment giving the dose–
46 response data, but the magnitude of any possible effect at the NOAEL is not defined. In
47 contrast, the BMD is a dose for which the size of the effect has been predefined, and thus it is
48 under the control of the risk assessor.
49 Selection of the data used to calculate a health-based guidance value is similar for both
50 the NOAEL and BMD approaches, although the different methods mean that optimum data

8
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 sets will have different characteristics. The NOAEL approach has been used for over 50 years,
2 and testing guidelines (chapter 4) have been developed to ensure that toxicological data are
3 suitable to identify the adverse effect of concern and also to define a NOAEL. An important
4 criterion of the NOAEL approach is that the group sizes should be sufficient to ensure that
5 any undetected effect at the NOAEL would be toxicologically insignificant. In the BMD
6 approach, a NOAEL does not have to be established, but doses with graded responses are
7 needed to provide optimum model output.
8 Calculations of a health-based guidance value based on the NOAEL or BMD approach
9 are summarized in Table 5.2.
10
11 Table 5.2. Comparisons of methods to derive heath-based guidance values based on NOAEL
12 and BMD approaches (using the Weibull model for illustrative purposes) for the case of quantal
13 data (adapted from IPCS, 2008)
14
Step NOAEL-derived ADI BMD-derived ADI
1. Data selection Sufficient sample sizes, at least one dose with Sufficient number of doses with
“no” effect and one dose with effect. Relevant different response levels and a
end-points in a relevant species are important sufficient number of total
for any approach. subjects.
2. Model selection Statistical method Fit dose–response model (e.g.
Weibull model).
⎧ if response at dose D
⎪0 is not significantly different
⎪
⎪ from control response
R( D) = ⎨
⎪ if response at dose D
⎪1 is significantly different
⎪
⎩ from control response

3. Statistical Pairwise statistical tests between dose groups Predicted fractions are linked to
linkage and control group. observed fractions, and their
“distance” is minimized by
optimizing some fit criteria
function (e.g. likelihood function
based on assumed
distribution).
4. Parameter Assessment of point of departure Choose an appropriate
estimation NOAEL = DNOAEL response, p, in the range of
experimental response.
where R (D) = 0 for all D ≤ D NOAEL and Estimate BMDLp, the 95%
lower confidence bound on the
R (D) = 1 for all D > D NOAEL BMDp, where
This procedure presupposes that all doses R(BMDp ) − R(0)
below the NOAEL are non-significant and all =p
doses above the LOAEL are significant. This is 1− R(0)
often not the case.
5. Implementation NOAEL BMD p L
ADI = ADI =
UFs UFs
where UF is uncertainty factor.
6. Evaluation Statistical power analysis should be performed Sensitivity of BMD to model
to check if the test was sensitive enough to choice can be checked by
detect relevant effects. fitting various models.
15

9
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 The table shows calculation of an ADI, but the methods are applicable to any health-
2 based guidance value.
3
4 5.3.2 The NOAEL approach to deriving health-based guidance values
5 The critical steps in this approach are selection of the appropriate data and determination of
6 the NOAEL. Historically, JECFA has used the term NOEL, which was defined in EHC 70
7 (IPCS, 1987) as “The greatest concentration or amount of an agent, found by study or
8 observation, that causes no detectable, usually adverse, alteration of morphology, functional
9 capacity, growth, development, or lifespan of the target”. In contrast, JMPR has used the term
10 NOAEL, which was defined in EHC 104 as “The highest dose of a substance at which no
11 toxic effects are observed” (IPCS, 1990). In reality, both terms have similar meaning, and the
12 NOAEL/NOEL has been used similarly to set health-based guidance values by both
13 Committees. At its sixty-eighth meeting (WHO, 2007), JECFA decided to differentiate
14 between the terms NOEL and NOAEL in future risk assessments with the following
15 definitions:
16
17 • No-observed-effect-level (NOEL): Greatest concentration or amount of a substance, found
18 by experiment or observation, that causes no alteration of morphology, functional capacity,
19 growth, development, or lifespan of the target organism distinguishable from those
20 observed in normal (control) organisms of the same species and strain under the same
21 defined conditions of exposure.
22 • No-observed-adverse-effect-level (NOAEL): Greatest concentration or amount of a
23 substance, found by experiment or observation, which causes no detectable adverse
24 alteration of morphology, functional capacity, growth, development, or lifespan of the
25 target organism under defined conditions of exposure.
26
27 The main difficulty with this approach is that it is based on demonstrating the absence
28 of an effect, and the result is critically dependent on the sensitivity of the test method. The
29 statistical linkage (step 3) determines whether or not there is a statistically significant effect
30 (e.g. at the 5% level) compared with background (e.g. the control group) for each dose level
31 separately. When the response is not statistically significant, it is considered that this level of
32 intake is without significant adverse health effects, but the power of the study to detect an
33 adverse effect is not analysed. Given the typical animal studies used in toxicology, the effect
34 size that can be detected by a statistical test may be larger than 10% (additional risk).
35 Therefore, the NOAEL may be expected to be a dose where the effect is in reality somewhere
36 between 0% and 10% or more. The selection of the NOAEL (step 4) identifies the highest
37 dose level that does not produce a statistically significant effect. The NOAEL approach tends
38 to give lower health-based guidance values for studies with a higher power to detect adverse
39 effects, which in effect “penalizes” better designed studies. This emphasizes the importance of
40 adherence to testing guidelines in order to ensure that the data are suitable for risk assessment
41 purposes.
42 The value of the NOAEL depends strongly on the following characteristics of the
43 study design:
44
45 • Group size. The power to detect a NOAEL at some dose level is directly dependent on the
46 sample sizes chosen at those dose levels. The larger the group size, the smaller the
47 possible undetected effect size at the NOAEL.
48 • Dose selection. The NOAEL can only be one of the doses actually applied in the study. If
49 the true threshold were higher than the NOAEL, the distance between the two can be

10
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 expected to be limited (related to the dose spacing used), whereas if the true threshold
2 were lower than the NOAEL, the distance between the two is potentially unlimited.
3 • Experimental variation. Experimental variation comprises biological (e.g. genetic)
4 variation between subjects, variation in experimental conditions (e.g. time of feeding,
5 location in experimental room, time of section or interim measurements) and measurement
6 errors. Larger experimental variation between subjects will result in lower statistical
7 power, and hence higher NOAELs.
8
9 Calculation of the health-based guidance value (HBGV, e.g. ADI, TDI) from NOAEL-
10 based DRM (step 5 above) is given by the equation:
11
NOAEL
12 HBGV =
UFs
13
14 Uncertainty factors (UFs) are default factors used to account for both uncertainty and
15 variability. Historically, an uncertainty factor of 100 has been used to convert the NOAEL
16 from an animal study into a health-based guidance value (IPCS, 1987). Additional uncertainty
17 factors may be used to allow for database deficiencies such as the absence of a chronic study
18 or when effects are detected at all experimental dose levels and a NOAEL therefore cannot be
19 defined. In such cases, a LOAEL may be used for establishing a health-based guidance value
20 (IPCS, 1994).
21 The default 100-fold uncertainty factor may be seen to represent the product of two
22 separate 10-fold factors that allow for interspecies differences between the experimental
23 animal species used and humans and the variability in responses in humans (IPCS, 1987). The
24 recognition that the original 100-fold uncertainty factor could be considered to represent two
25 10-fold factors allowed some flexibility, because different factors could be applied to the
26 NOAEL from a study in humans as well as from a study in animals.
27 The concept of CSAFs (IPCS, 1994, 2005) has been introduced to allow appropriate
28 data on species differences and/or human variability in either toxicokinetics (fate of the
29 chemical in the body) or toxicodynamics (actions of the chemical on the body) to modify the
30 relevant default 10-fold uncertainty factor. The strategy used by IPCS involves replacing the
31 original 100-fold uncertainty factor with CSAFs (IPCS, 1994, 2005). For example, JECFA
32 used comparative body burden data rather than external dose data in its calculation of a
33 provisional tolerable monthly intake (PTMI) for dioxin-like substances, allowing the usual
34 100-fold uncertainty factor to be subdivided and replaced with lower values, since there was
35 no need for an uncertainty factor for toxicodynamics differences between species (WHO,
36 2002).
37 Step 6 could be undertaken to analyse the power of the dose group identified as
38 representing the NOAEL to detect the adverse effect found at higher dose levels. For example,
39 DRM could be used to determine with 95% confidence intervals the magnitude of effect that
40 would be predicted to occur in the NOAEL group. In addition, step 6 could be used for both
41 the NOAEL and BMD approaches to evaluate the sensitivity of the calculated health-based
42 guidance value to the values of the uncertainty factors chosen.
43
44 5.3.3 Benchmark dose approach to deriving health-based guidance values
45 As an alternative to the NOAEL approach, the BMD concept has been introduced (Crump,
46 1984; Kimmel & Gaylor, 1988). In contrast to the NOAEL approach, this method defines a
47 level of exposure producing a non-zero effect size or level of response as the POD for risk
48 assessment. The BMD method has a number of advantages, including the use of the full dose–
49 response data in the statistical analysis, which allows the quantification of the uncertainty in

11
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 the data. Higher uncertainty in the data—for example, due to small group sizes or high
2 variation within a group—would be reflected in lower health-based guidance values.
3 In choosing the data (step 1) for BMD modelling, similar basic considerations apply as
4 for the NOAEL method. Group sizes are less critical, because the POD is not based on
5 identifying a level of exposure at which the adverse effect was not detected. Studies showing
6 a graded monotonic response with a significant dose-related trend provide the best
7 experimental data for modelling.
8 The main difficulty with this approach is that it requires the preselection of a level of
9 response, the BMR. In general, the level chosen is such that it is close to the LOD of the study,
10 or a level that would generally be considered as representing a negligible health effect. A
11 generic form of the BMD and BMDL is presented in Table 5.2. A variety of response levels,
12 such as 1%, 5% and 10%, may be selected as the BMR; differences in selection of the BMR
13 could lead to discrepancies in health-based guidance values between different regulatory
14 bodies.
15 Choosing a model (step 2) for the BMD method is dependent upon the types of data
16 available and the characteristics of the response being modelled. Complicated models will
17 require a larger number of dose groups than simpler models, and several models have been
18 proposed for each type of data. In the USEPA BMD software program, a number of routinely
19 used models are cited (http://cfpub2.epa.gov/ncea/cfm/recordisplay.cfm?deid=20167).
20 Caution should be used in interpreting the results of DRM if widely varying BMDL estimates
21 are given when multiple models are applied to the same data, because this could indicate
22 inadequate data for modelling.
23 At its sixty-fourth meeting, JECFA calculated the MOEs for a number of genotoxic
24 and carcinogenic food contaminants using BMDL values derived by fitting a range of models
25 to the available experimental dose–response data (WHO, 2006). Annex 3 of the report of that
26 meeting (WHO, 2006) provides useful background information on the use of the BMD
27 approach for risk assessment purposes.
28 The statistical linkage (step 3) between the data and the model can assume a number
29 of different forms. For quantal data, it is appropriate to assume that the data are binomially
30 distributed for each dose group.
31 Selection of the POD (step 4) for the BMD method is in reality selection of the BMR,
32 because the model outputs simply report the BMD and BMDL values for the selected BMR. It
33 is often not clear what level of response (BMR) can be considered as representing a negligible
34 health effect. Selection of the BMR requires discussion among toxicologists and clinicians.
35 Although an explicit statement on the BMR is an improvement compared with the generally
36 unknown response level that may be associated with a NOAEL, choices of a BMR need
37 consensus building and will remain a subjective “expert” judgement in what is essentially a
38 mathematical approach. An alternative approach to selection of the BMR is to choose an
39 excess response, often 10% response, that is close to the LOD of the study, below which there
40 was insufficient support from the experimental data; however, this simply leaves open the
41 issue of the possible health consequences of the resulting level of response at that BMR.
42 Further information on the selection of the BMR is given in IPCS (2008).
43 The health-based guidance value can be calculated as follows:
44
BMDL
45 HBGV =
UFs
46
47 In this calculation, the values of the uncertainty factors could be the same as used for
48 the NOAEL or adjusted to account for a slightly different interpretation for the BMDL
49 relative to the NOAEL. Unlike with the NOAEL approach, an extra uncertainty factor would

12
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 not be necessary if all dose levels resulted in significant levels of adverse effect (indeed, such
2 data would be more suitable for modelling). Empirical investigations showed for a large and
3 representative set of compounds that the BMDL may be regarded as an analogue to a NOAEL,
4 and substituting one by the other would result in similar health-based guidance values (Crump,
5 1984; Barnes et al., 1995).
6 Unlike the NOAEL approach, the BMD method includes the determination of the
7 response at a given dose, the magnitude of the dose at a given response and their confidence
8 limits. By extrapolation of the dose–response model below the biologically observable dose
9 range, the response at specified (lower) dose levels can be estimated, as well as the dose
10 corresponding to a specific response level. It should be noted, however, that extrapolation
11 from a single model that fits the data in the observed range cannot be justified, since other
12 models fitting the data equally well may result in substantially different estimates of low-dose
13 risk. Linear extrapolation from a BMD for a 10% response (BMD10) has been applied as a
14 simple method for low-dose extrapolation, but the sixty-fourth meeting of JECFA considered
15 and concluded that “Linear extrapolation from a point of departure offers no advantages over
16 an MOE and the results are open to misinterpretation because the numerical estimates may be
17 regarded as quantification of the actual risk”.
18
19 5.3.4 Uncertainty
20 A disadvantage with the NOAEL as a POD (“starting point”) for formulating advice to risk
21 managers is that it is not possible to quantify the degree of variability and uncertainty that
22 may be present. The NOAEL is assumed to be a dose without biologically significant effects.
23 This assumption is more likely to be valid in toxicological studies with larger sample sizes.
24 While uncertainty factors are amenable to uncertainty analysis (Slob & Pieters, 1998), the
25 NOAEL approach is not readily amenable to quantitative estimation of uncertainty or to a
26 sensitivity analysis.
27 A modelling approach facilitates both sensitivity and uncertainty analyses. Uncertainty
28 can be expressed numerically when the doses and responses are linked by a model (see Table
29 5.2). Such numerical analyses can also be subject to sensitivity analyses to test the
30 contribution of different aspects of the database or of model characteristics to the overall
31 uncertainty. The uncertainty that arises from aspects of study design, such as dose spacing,
32 sample size and biological variability, on the risk estimates can be assessed using DRM.
33
34 5.3.5 Study design
35 A design optimal for the NOAEL approach could limit the use of DRM, and vice versa. While
36 the NOAEL approach requires sufficient sample sizes within dose groups (to provide
37 statistical power), the BMD approach requires a sufficient number of dose groups (to provide
38 a description of the whole dose–response).
39 The BMD approach can be used on studies carried out in the past and based on the
40 traditional designs (with three dose groups and a control). Although these may not be optimal
41 for model fitting, the BMD approach retains the advantages outlined above.
42 Both the BMD and NOAEL approaches may prove inadequate when the number of
43 animals per dose group is too small. For example, when the critical effect is seen in an
44 experimental animal, such as the dog, with few animals per dose group, the NOAEL may be
45 high because of the insensitivity of the test. Although the BMD approach is better for
46 evaluating sparse dose–response data, it may also provide very uncertain estimates, although
47 unlike the NOAEL approach, the inherent uncertainty is more explicit.
48

13
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 5.3.6 Biological information

2 The NOAEL approach incorporates biological information through the application of
3 subjective “expert” judgement. More complicated forms of DRM have the potential for a
4 more “science-rich” analysis through the more formal quantitative inclusion of
5 factors/covariates into the models, in the case of both human epidemiological and animal data.
6 Such an approach can lead to more certain estimates centred on a toxicologically based
7 concept of estimating the dose–response relationship on the basis of all available biological
8 knowledge using empirical data and applying statistical inference.
9
10 5.4 Principles of dose–response modelling
11 5.4.1 Data
12 When considering which data to use from a set of available toxicity studies on a particular
13 compound, it is not necessary to undertake DRM for each observed end-point in each study.
14 Whether the NOAEL or BMD approach is used for risk assessment, the aim is to define the
15 adverse effect that is produced at the lowest levels of exposure. Therefore, a first step would
16 be to exclude studies that have obviously larger NOAELs compared with the other studies. In
17 addition, end-points not showing a clear dose–response on visual inspection can be omitted.
18 Then, based on the toxicological impact together with the apparent magnitude of the response,
19 a selection of end-points can be made as candidates for DRM. After selecting the potentially
20 relevant end-points, the suitability of each dose–response data set for dose–response analysis
21 is considered. For the NOAEL approach, there should be an adverse effect detected at one or
22 more dose levels and no response detected at one or more dose levels. For the BMD approach,
23 it is generally desirable to have at least three or four different doses (including controls) and
24 different levels of effect associated with different doses.
25 There are various types of response data (see section 5.2.2), and the main distinction is
26 between quantal and continuous data. Although the type of data is important for statistical
27 reasons, the distinction between quantal and continuous data also has a crucial impact on
28 interpretation of results and their use in risk assessment. In the case of quantal dose–response
29 data, information on the change of incidence with dose is available at one particular degree of
30 effect. For example, the incidence of inflammation as judged from histopathology may
31 increase as dose increases, but under the quantal categories “no inflammation” or
32 “inflammation”, there is no indication of the degree of effect. In ordinal and continuous data,
33 in contrast, information on both the degree of effect and the incidence is available as a
34 function of dose. So, for example, inflammation might be categorized into an ordinal variable
35 using the categories “none”, “mild”, “moderate” and “severe”, or might be quantified in a
36 continuous variable, for example, from measurements on biochemical markers of
37 inflammation.
38
39 5.4.2 Mathematical models
40 A number of mathematical models have been or can be used to describe dose–response data.
41 These are outlined below, but their application and interpretation require specialized expertise.
42 Further details of the main models are provided in the report of the sixty-fourth meeting of
43 JECFA (WHO, 2006) and in IPCS (2008).
44
45 5.4.2.1 Dose–response models for continuous data
46 The models listed in Table 5.3 are some of the forms that may be used to describe the
47 relationship between dose and the magnitude of a response on a continuous scale in an
48 individual. When combined with a statistical distribution (e.g. normal or lognormal), these
49 equations can also be used to describe the relationship between dose and a continuous
50 response in a population, where the continuous model corresponds to the central estimate.

14
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 Dose–response data are often adjusted by subtracting the (mean) control value from
2 each individual observation. However, this procedure does not account for the fact that the
3 background response level in the controls is, like the experimental groups, subject to sampling
4 error and individual variability. A better approach is to account for the background response
5 in the model with a parameter that needs to be estimated from the data (see IPCS, 2008).
6
7 Table 5.3. Dose–response models for continuous data
8
Parameter
Name(s) Notes Equation for response explanations
Michaelis-Menten A theoretical account of RMax is the
[S ]
law of mass
action
enzyme- or receptor-
based activity where = RMax maximum rate of
the reaction, [S] is
the rate of action is a
function of the rate of
K M + [S ] the substrate
concentration, and
association (ka) and the KM is the Michaelis-
rate of dissociation (kd). Menten constant,
which is equal to
ka/kd.
Hill equation log- A modification of the RMax is the
logistic Michaelis-Menten Dn maximum
equation that supposes =RMax response, D is the
KD + D
that the occupation of n n dose, KD is the
multiple sites or reaction constant
receptors is required for for the drug–
the production of an receptor
effect. interaction, and n is
the number of
(hypothetical)
binding sites.
First-order
exponential
If the interaction
between a chemical
and a target site is
(
= RMax 1 − e − rD ) RMax is the
maximum
response, D is the
irreversible, then the dose, and r is the
rate of the reaction is exponential rate
determined by the rate constant.
of association (ka) only.
Power Simple exponential = βDα D is the dose, α is
model. the shape
parameter, and β is
the scale
parameter.
Linear Although there is = mD D is the dose and
usually no biological m is the slope.
theory to suggest it,
linear models are often
justified by their
simplicity; linear models
have but a single
parameter.
9
10 5.4.2.2 Dose–response models for quantal data
11 Quantal dose–response functions describe the relationship between dose and the frequency of
12 a particular outcome in a population (see Table 5.4). For a group of homogenous or nearly
13 identical individuals, the relationship between dose and frequency could be described with a
14 step function where all subjects either respond or fail to respond at any given dose. However,
15
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 because variability is ubiquitous in living organisms, quantal dose–response data typically

2 show gradually increasing incidence with dose. One interpretation of this is that individual
3 subjects differ in tolerance to the agent, which can be described by a statistical tolerance
4 distribution. Hence, any cumulative distribution function (CDF) may be used as a quantal
5 dose–response function. Other models have been derived from statistical assumptions about
6 how the agent might exert its effect in an organism, such as the gamma multi-hit model.
7
8 Table 5.4. Dose–response models for quantal data
9
Name(s) Theoretical basis Equation for frequency (F) Parameter explanations
Step function No variability. If D < T, F = 0 D is the dose and T is the
If D ≥ T, F = 1 threshold parameter.
One-hit Hit theory models D is the dose, e is Euler’s
(single-hit) employ the use of a rate = 1 − e−(α+βD) constant, α is a location
to describe the parameter, and β is the
interaction between a slope parameter.
group of causal agents
(e.g. molecules) and a
group of targets (e.g. a
human population).
Gamma multi- An expansion of the = Γ(gamma*D, k) Γ() is the incomplete
hit one-hit model, which is gamma CDF, D is the
based on the notion that dose, gamma is a rate
multiple hits or events parameter, and k is the
are required to produce number of hits required to
a particular effect. produce the effect.
Probit normal A descriptive model = Φ(α + D*β) Φ() is the normal CDF, D
based on a normal or is the dose, α is a location
Gaussian distribution. parameter, and β is the
slope parameter.
Logistic The statistical logistic 1 D is the dose, α is a
model is also a = −α − D× β location parameter, and β
descriptive tool with no 1+ e is the slope parameter.
theoretical basis.
Weibull A flexible descriptive D is the dose, α is the
− (α + ( β × D )γ )
model originally
developed to describe
= e background parameter, β
is the slope parameter,
survival data in and γ is an exponent.
demography.
10
11 Background response rates should be accounted for by incorporating an additional
12 parameter in the dose–response model (see IPCS, 2008).
13
14 5.4.2.3 Modelling with covariates
15 In some circumstances, it is desirable to include variables in addition to just an exposure
16 variable in dose–response models. For example, in epidemiological studies, it is common to
17 model disease risk in terms of not only exposure, but also age, sex, socioeconomic status,
18 smoking status and other measurements that may be relevant to the disease state. These other
19 factors may not themselves be directly affected by the exposure, but they may be correlated
20 with exposure status because of the way the sample was taken. Unless the proper covariates
21 are included in a model for the relationship between exposure and the health end-point, the
22 effect of exposure will be incorrectly estimated.

16
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 In principle, this sort of confounding cannot occur in bioassay studies in which

2 animals are randomized to treatment groups, but it may be useful to include a covariate such
3 as sex or body weight to account for some of the variability in a related measure.
4
5 5.4.2.4 Biologically based dose–response models
6 While biological considerations may motivate the choice of one or several empirical models,
7 the level of biological detail in such models is minimal. Thus, their credibility for
8 interpolating and extrapolating a data set derives mainly from their fit to the data, as evaluated
9 statistically. Another class of model, the biologically based dose–response models, are much
10 more complicated and are explicitly designed to model the biological details that lead from
11 initial exposure to a toxicant to the ultimate pathological outcome. Typically, such a model
12 includes a PBTK model to describe the distribution and metabolism of the parent compound
13 and toxic metabolites, as well as other mechanistic, or toxicodynamic, models that link target
14 tissue concentration to the ultimate response. The toxicodynamic part of the model may be
15 relatively simple or may be as complicated as a fully elaborated stochastic model for
16 carcinogenesis.
17 Such a model is really a quantitative expression of a set of biological hypotheses and,
18 when rigorously tested against critical experiments, becomes a credible tool for extrapolating
19 from experimental results into exposure realms that are difficult or expensive to reproduce in
20 controlled experiments. Such models are quite expensive both in resources and time to
21 construct and thus would be expected to be developed fully only for exposures and toxicities
22 of the highest concern.
23
24 5.4.3 Model fitting and estimation of parameters
25 Details of mathematical methods used to optimize the fitting of the model to the data and
26 methods for comparisons of the outputs of different models are given in the report of the
27 sixty-fourth meeting of JECFA (WHO, 2006) and in IPCS (2008). IPCS (2008) also provides
28 details on incorporation of sampling errors, study errors and model errors into uncertainty
29 analysis, together with detailed information on selection of the BMR and interpretation of the
30 BMD and BMDL.
31
32 5.5 References
33 Barnes DG, Daston GP, Evans JS, Jarabek AM, Kavlock RJ, Kimmel CA, Park C, Spitzer HL (1995)
34 Benchmark dose workshop: criteria for use of a benchmark dose to estimate a reference dose. Regulatory
35 Toxicology and Pharmacology 21, 296-306.
36 Crump KS (1984) A new method for determining allowable daily intakes. Fund. Appl. Toxicol., 4, 854-871.
37 IPCS (1987) Principles for the Safety Assessment of Food Additives and Contaminants in Food, Environmental
38 Health Criteria 70, International Programme on Chemical Safety, Geneva.
39 IPCS (1990). Principles for the toxicological assessment of pesticide residues in food. WHO Environmental
40 Health Criteria, No. 104). Available at: http://www.who.int/pcs. World Health Organization, Geneva.
41 IPCS (1994) Assessing Human Health Risks of Chemicals: Derivation of Guidance Values for Health-based
42 Exposure Limits. Environmental Health Criteria 170, International Programme on Chemical Safety, Geneva.
43 IPCS (2004) IPCS Risk Assessment Terminology, International Programme on Chemical Safety, Geneva.
44 IPCS (2005) Chemical-Specific Adjustment Factors for Interspecies Differences and Human Variability:
45 Guidance Document for Use of Data in Dose/Concentration–Response Assessment (Harmonization Project
46 Document No. 2; http://whqlibdoc.who.int/publications/2005/9241546786_eng.pdf).
47 IPCS (2008) Principles for Modelling Dose–Response for the Risk Assessment of Chemicals. Geneva, World
48 Health Organization, International Programme on Chemical Safety (Environmental Health Criteria, in press).
49 Kimmel CA, Gaylor DW (1988) Issues in qualitative and quantitative risk analysis for development toxicology.
50 Risk Anal 8, 15-20.
51 Slob W, Pieters MN (1998) A probabilistic approach for deriving acceptable human intake limits and human
52 health risks from toxicological studies: general framework. Risk Analysis 18, 787-798.
53 WHO (2002) Evaluation of Certain Food Additives and Contaminants, Fifty-seventh report of the Joint
54 FAO/WHO Expert Committee on Food Additives, Technical Report Series 909, Geneva.

17
 Update project Chapter 5: Dose–response assessment Draft May 2008

1 WHO (2006) Evaluation of certain food contaminants. Sixty-fourth report of the Joint FAO/WHO Expert
2 Committee on Food Additives. Geneva, World Health Organization (WHO Technical Report Series No. 930;
3 http://whqlibdoc.who.int/trs/WHO_TRS_930_eng.pdf).
4 WHO (2007) Evaluation of certain food additives and contaminants. Sixty-eighth report of the Joint FAO/WHO
5 Expert Committee on Food Additives. Geneva, World Health Organization (WHO Technical Report Series
6 No. 947; http://whqlibdoc.who.int/trs/WHO_TRS_947_eng.pdf).
7

18