Introduction for EPA Benchmark Dose Online Training Session

Epa/Ord/Ncea

Sujets

• BMD-1
• Linternaute.com
• Estimated Time of Arrival
• Épaney
• Cancer (genre)
• British Market Research Bureau

Informations

• Publié par: Fioz
• Nombre de lectures: 67
• Langue: English

Extrait

Introduction   Slide 1: Welcome to the EPA Benchmark Dose Online Training Session. This training session is designed to provide risk assessors with guidance on how to use benchmark dose modeling in dose response analysis.   Slide 2: The complete training session consists of five modules. This introduction is the first of the five benchmark dose training modules. The second module will cover dichotomous modeling, the third module will cover cancer modeling, the fourth continuous modeling, and the fifth the nested modeling. In this introduction module, we will present background on the benchmark dose method, and the advantages and limitations of using this method in chemical risk assessment.   Slide 3: The introduction module includes four sections. They are: Background for the benchmark dose (or BMD) approach, Definition of BMD, Pros and Cons for BMD, as well as General procedures in BMD modeling. First, we are going to provide some background for the benchmark dose approach.   Slide 4: Dose response assessment, including dose-response modeling, has long been recognized as a key component of the chemical risk assessment paradigm as is illustrated in this diagram from the 1983 National Academy of Science’s report on risk assessment.     Slide 5: As was shown in the previous NAS diagram, risk assessment is composed of 4 major components: The Hazard Identification or Characterization, which covers what kind of effects one might expect after exposure to this chemical and the conditions under which one might see toxicity. The second aspect is the dose response assessment, which covers how toxic this substance is and what happens at various doses. BMD modeling is one form of dose response assessment. The third component- the Exposure Assessment - covers who is exposed, to how much, how often, and for how long? And finally, the last component is the Risk Characterization, in which the results of the three previous components are combined to determine the overall risk to the exposed population.   Slide 6: The term Benchmark dose modeling was introduced by Dr. Kenny Crump in 1984 as an alternative to existing methods of dose response assessment. Soon thereafter EPA recognized the potential benefits of Benchmark dose modeling and sponsored several workshops. This led to a 1995 publication by EPA’s Risk Assessment Forum on the Use of Benchmark Dose in Risk Assessment and the first application of benchmark

dose methods in an EPA assessment, the 1995 methylmercury assessment. This was later followed by a 2000 publication by EPA’s Ofice of Research and Development presenting Technical Guidance for benchmark dose modeling. In that same year, EPA’s National Center for Environmental Assessment introduced the first version of the agency’s Benchmark dose software.   Slide 7: In the hazard identification component of a risk assessment, we examine what kind of adverse effects are caused by a chemical exposure. This slide shows an example of multiple responses caused by a particular chemical. Each endpoint has its own dose response.   In addition, a dose response analysis will assist in the identification of the most sensitive adverse effect, also called the critical effect, and the point of departure to use in establishing a toxicity benchmark, such as a reference dose (RfD) or reference concentration (RfC). The RfD and RfC are the terms used by the US EPA for safe oral and inhalation exposures levels. In the traditional approach, the point of departure is the no-observed adverse effect-level or NOAEL or Lowest-Observed-Adverse-Effect-Level, LOAEL. But here, we are going to talk about another method of establishing a Point of Departure- that is the benchmark dose method. Benchmark dose methods involve analyzing each endpoint’s dose response separately.    Another approach, categorical regression would allow for an analysis of combined effects that are grouped into severity categories. This categorical approach and EPA’s categorical regression software, CatReg, will be discussed in another training session.   Slide 8: This slide shows how a reference dose or a risk value is calculated in the traditional NOAEL approach. One needs to identify the NOAEL or LOAEL in this approach and then the NOAEL or LOAEL is divided by a composite uncertainty factor to obtain the final reference dose.   The composite uncertainty factor used in this calculation covers variations or uncertainties in the areas of interspecies extrapolation, intraspecies variation, NOAEL to LOAEL extrapolation, subchronic to chronic extrapolation, and database uncertainty. The use of benchmark dose might eliminate the need for NOAEL to LOAEL extrapolation uncertainty factor. This will be discussed later in this presentation.   Slide 9: Here are the steps involved in the dose response assessment using the traditional NOAEL/LOAEL approach. First, one needs to identify the NOAEL or LOAEL for each adverse effect, and then one must identify the most sensitive effect or the critical effect based on the NOAEL or LOAEL for all the adverse effects. The identified NOAEL or LOAEL for the critical effect will then be used as the point of departure, which is divided by uncertainty factors to derive the reference dose or reference concentration.  

Slide 10: This slide represents some of the recognized limitations of the NOAEL approach. First of all, the NOAEL approach is limited to the doses tested in any given study. Secondly, the response levels are not comparable, either biologically, or quantitatively. For example: NOAEL from one study might have caused no effect at all, while a NOAEL from another study might have caused an adverse response in 5% of the animals, but the dose was considered a NOAEL because the effect was not statistically significant from the control group.   Third, the NOAEL does not represent the 0% response level. The NOAEL is a No Observed Adverse Effect Level. It is possible that a better designed study with sufficient power might have observed a statistically significant adverse effect at the NOAEL.   Next, a NOAEL is not always available from a particular study. Adverse effects may have been observed at the lowest dose tested, so that the lowest dose was a LOAEL and there was not NOAEL in the study. Also, the NOAEL approach does not consider the steepness of the dose response curve. For example, a NOAEL with the next dose causing an 80% response will have different confidence level from a NOAEL with the next dose only causing a 10% response. Since the NOAEL approach only focuses on identifying the NOAEL, this difference in the dose response slope is not considered in the risk value derivation.   The NOAEL is also highly dependent on the sample size. All other things being equal, a small study with just a few animals will have less statistical power, and may have a higher NOAEL than a larger study using the same doses, even if the same percent response was observed in both studies. This is the opposite of what is desired from a regulatory viewpoint. From a regulatory viewpoint, one would want to use a more conservative (health protective) approach for the smaller, less certain study. This concept is illustrated in the next two slides.   Slide 11: In this graphic demonstration of the dose response for a study in which 50 animals per dose were exposed to a chemical, we illustrate some of the limitations of the NOAEL approach. In this case, the NOAEL is 50 mg/kg. The fact that the NOAEL approach is limited to the doses tested can be shown if one were to imagine that the 50 mg/kg dose were replaced with a dose of 25 mg/kg. If that were the case, then the NOAEL would be 25 mg/kg instead of 50 mg/kg. So as you can see that the NOAEL must be one of the doses tested in that study.   We noted previously the dependence of the NOAEL on the sample size, which determines the statistical power of the data. For example, if we have a sample size of 50, as in this case, the highest dose that does not cause a statistically significant effect is 50 mg/kg.   Slide 12:

Here in contrast, the same fraction of animals was affected in a study with 10 animals in each dose group. Although the % response is exactly the same as for a sample size of 50 per dose group, the smaller sample size results in wider confidence limits. This means that the exact same response at 100 mg/kg as was reported in the study with 50 animals per dose is no longer statistically significant, and so the dose of 100 mg/kg, which was the Lowest-Observed-Adverse-Effect-Level in the study with 50 animals per group, is now the No-Observed-Adverse-Effect-Level, and the next higher dose of 150 mg/kg is now the LOAEL. Note that despite 30% of the animals being affected at the 100 mg/kg dose, this dose is identified as a NOAEL because the response was not statistically significant relative to the control response given the small sample size used.   Thus, in the NOAEL approach, the smaller sample size will result in the use of a higher NOAEL as the point of departure. This will result in a higher, less health protective risk value. However, because a smaller sample size will provide less confidence in the data, we should use a more conservative or lower value as the point of departure to estimate the risk value instead. Until the benchmark dose approach, this type of uncertainty could only be handled in a rather non-quantitative fashion via the use of uncertainty factors.   In contrast to the NOAEL approach, with the use of benchmark dose, we can more quantitatively assess the uncertainty associated with the use of fewer animals per dose group in a toxicity study, and appropriately reflect that uncertainty in the choice of the point of departure.   Slide 13: In the Benchmark dose approach, the risk assessor will fit a flexible curve to the data , then identify the dose that causes the predetermined benchmark response (BMR). This dose is called the benchmark dose, or BMD. The risk assessor then estimates the lower bound on the BMD (typically the 95% lower confidence limit). This dose is called the Benchmark Dose Lower bound confidence Limit, or BMDL. This BMDL broadens with decreasing sample size, resulting in a lower, more health protective risk value, and appropriately reflecting uncertainty in the study.   Slide 14: When used to derive an RfD or RfC, the benchmark dose lower bound confidence limit or the benchmark concentration in the case of air exposure, will replace the LOAEL or NOAEL in the risk assessment paradigm. As in the NOAEL approach, BMDL is divided by uncertainty factors.   Slide 15: In addition to being used for the derivation of the non-cancer reference dose or reference concentration, the benchmark dose approach can also be applied to derive a cancer slope factor. The cancer slope factor is the benchmark response level divided by the benchmark dose lower bound confidence limit derived at that benchmark response level. For example, if one were using 10% as the benchmark response level, it would be 0.1 divided by the BMDL10.  

Slide 16: Next we will talk about the definition of the Benchmark Dose, and what we mean by Benchmark Dose and Benchmark Dose Modeling.   Slide 17: The goal of Bench mark dose modeling is to define a point of departure (POD) that is relatively independent of study design   The BMD is defined as an estimate of the dose or concentration that produces a predetermined change in response rate of an adverse effect compared to background. The predetermined change in response rate is called the benchmark response or BMR.   For example, the BMD might be defined as the dose that causes a 10% increase in the number of animals developing fatty liver compared with untreated animals.   The BMDL is defined as the 95% lower bound confidence limit on the BMD.   Slide 18:   In this slide, the experimental doses and their respective confidence limits are shown in green. The red line shows the benchmark dose maximum likelihood estimates. The blue lines to the left of it are drawn between BMD lower confidence limit estimates at various response intervals. As was mentioned earlier, the smaller the statistical power, the lower the BMDL, and the farther away the blue line will be from the red line.     Slide 19: As was mentioned previously, once we identify the BMDL, we can use it as the Point of Departure to calculate the RfD or RfC. It is important to understand that the BMDL is not a NOAEL. The BMDL represents an estimate of the low dose (with 95% confidence) associated with the predetermined benchmark response, and it has been shown by some risk assessors to be consistent with the NOAEL when there is adequate sample size in a study. Thus, the BMDL can be used as the NOAEL surrogate in deriving a risk value. As the result, no LOAEL to NOAEL Uncertainty factor is needed when a BMDL from such data is used as the point of departure.   Slide 20: Now we’ll discuss the pros and con sof the benchmark dose approach.   Slide 21: The advantages of the benchmark dose approach include not being limited to the doses tested experimentally, and being less dependent on dose spacing. The BMD or BMDL is an estimate dose of the dose causing a predetermined response.  

The benchmark dose method takes into account the shape of the dose-response curve more explicitly. The slope of the response curve can have a significant impact on the BMD and BMDL estimates.   The benchmark dose approach is flexible in determining biologically significant rates.   The benchmark dose approach is comparable across chemicals and endpoints. All the BMD or BMDL values causing the same BMR can be estimated and compared. This allows a direct comparison of the same response in order to identify the critical effect between different studies and endpoints.   It is consistent with the EPA cancer guidelines. In the new cancer guidelines, BMDL is also used as the point of departure in calculation of the cancer slope factor.   As was illustrated earlier, the BMD approach also appropriately reflects uncertainty in study design. It provides more incentive for industrial laboratories to conduct better studies, resulting in tighter confidence intervals and higher BMDL estimates.   Slide 22: Using the previous example, when a benchmark dose is derived from a study which involves a smaller number of animals per dose group, the lower bound confidence limit on the benchmark dose is broad, accounting for the smaller sample size.   Slide 23: In contrast, when the benchmark dose lower bound confidence limit is calculated for a better study, a study that uses more animals per dose and has higher statistical power, the benchmark dose lower bound confidence limit range is narrower, resulting in a higher point of departure. Note that the estimated NOAEL has moved in the opposite direction, to a lower point of departure.   Slide 24: There are some challenges to performing a benchmark dose modeling exercise. The benchmark dose approach requires that the data are in a certain format. For quantal data (or “count data”), one needs to have informtiaon on the sample size and incidence in each dose group. For continuous data (such as body weight, where the response can be a non- integral number), both the sample size and the means and standard deviations must be available or be able to be calculated from the reported data.   The benchmark dose approach is also more time consuming than the NOAEL approach.  However, the National Center for Environmental Assessment within the EPA is working on improvements to the BMD software, particularly batch processing capabilities that should help to facilitate the analysis of many endpoints and models at one time. As for the NOAEL approach, a number of scientific judgments are required as part of the BMD modeling, as we will be noting throughout this course.   Slide 25:

Next we will cover the general procedures in doing a benchmark dose analysis.   Slide 26: The first step that one needs to follow in the benchmark dose analysis is to determine whether the data are worth modeling. One would evaluate the database much the same as for the NOAEL approach. This includes determining whether Good Laboratory Practices were performed, and whether the appropriate duration and route of exposure were used. It also involves reviewing the measured endpoints and determining whether they identify effects of concern. Finally, all of the potentially sensitive adverse endpoints should be modeled. If we don’t model all of the potentially adversee ndpoints, we would run the risk of not identifying the point of departure that might be important in the final analysis, especially if different uncertainty factors are used for different endpoints.   Slide 27: In analyzing whether a study or database is worth modeling, one needs to review whether there is a significant dose-related trend in the data to be modeled. Secondly, for standard BMD applications, one needs to determine whether we have at least two doses with response levels in excess of the control response so that a curved dose response can be adequately modeled. And thirdly, it is preferred that there is at least one dose with a response in the range of the BMR, so that no extensive extrapolation is needed to obtain a BMDL.   Slide 28: In practice, it may not always be feasible to model all biologically, statistically significant responses. If there are a lot of endpoints in a database, one can model just those endpoints for which the LOAELs are within a 10-fold range of the lowest LOAEL of the database.   In some cases, it may be difficult to obtain a curve that provides a good fit to the entire dose-response for an endpoint, particularly if the response plateaus at high doses. In such cases, the high dose group may be dropped if that improves the data fit at the response close to the BMR. Because we are generally interested in the response at the low dose, it is usually more important for the modeled curve to fit the data in the low dose region than to fit the overall dose-response. However, plateaus or other changes in the dose response at the high dose may reflect changes in toxicokinetics (such as saturation of metabolism) or changes in toxic response that the assessor may wish to model. At the expense of estimating an extra “asymptote” parameter smoe models such as the Hill model are capable of modeling data that “flatten out” in the hgih dose region. Version 1.4.1 of EPA’s benchmark dose software contains a Hill model for the evaulation of continuous data and soon-to-be released version 2.0 of BMDS will contain a Hill model for the evaluation of dichotomous data.   Slide 29:

This is a six step flow chart that represents the benchmark dose analysis process as it is documented in the EPA’s Benchmark Dose Technical Guidance Document of the year 2000.   The first step is to select a benchmark response level. The subsequent steps involve how to do the modeling of the dose associated with that benchmark response level. One first selects a model or modeling approach and model parameters, and runs the model. The next step is to evaluate the fit of the model to the dose response data. Steps 2 and 3 are repeated for all models under consideration.    Next one should collect all of the models that are determined to adequately fit the data. Once the set of models that adequately fit have been identified, the BMDL results can be examined. If the BMDL estimates from those models are not with a factor of three, some model dependence of the BMDL estimate is assumed. Since usually there is no clear remaining biological or statistical basis for choosing among these models, the lowest BMDL is selected as a reasonable conservative or health protective estimate. If the lowest BMDL from the available models appears to be an outlier compared to other results, then an additional analysis and discussion might be necessary. Additional analysis might include the use of additional models, or the examination of the appropriateness of the parameters used in the models that derived this particular set of BMDLs.   On the other hand, if the BMDL estimates from the models with adequate fit are within a factor of three, then this is considered to be a situation where there is no appreciable model dependence and the best fitting model is used. If a single best-fitting BMDL model cannot be determined, one may consider combing the BMDLS by averaging, or some other approach.   In any case, after steps 5 or 6 is completed, one needs to document the results as outlined in the EPA’s 2000 guidance document   Each one of these steps will be reviewed more closely in following slides.   Slide 30:   The first step shown in the previous slide is to select a benchmark response level.     Slide 31: In selecting a benchmark response level, the level should be near the low end of the range of increased risks that can be detected by a standard bioassay. Be careful however not to choose response levels that are well below the range of detected responses. Extrapolation below the range of the data can impart high model dependence, and different models will thereby result in different BMDL estimates. This means that BMD modeling should not be used to extrapolate from the range of the data to doses or responses that are orders of magnitude below the experimental doses, such as might occur with extrapolation to environmental exposure levels. In addition, great care should be used if the lowest

nonzero dose in a study gives a response well above the standard benchmark dose, such as if the low dose gives a 50% response. In such cases, extrapolation below the data involves considerable uncertainty.   Slide 32: This slide graphically illustrates where the benchmark response level should be for a given dose response dataset. In this case the benchmark response level is 10%. The response is very close to the response level observed in the 50 mg/kg dose group. Therefore, no significant extrapolation is needed to estimate the BMD or BMDL.   Slide 33: There are differences in selection of benchmark responses for dichotomous data and continuous data. In selecting a benchmark response level for dichotomous data, an extra risk of 10% is generally used since the 10% response level is near the limit of sensitivity in most cancer bioassays and in some non-cancer bioassays.   Various papers have proposed a benchmark response for quantal responses in the range of 1% to 10%. Response levels lower than 10% can be used if the study has sufficient sensitivity to detect a response in the range of the chosen BMR. This may occur sometimes in the case for developmental toxicity studies, or in studies with unusually large sample sizes. Studies by Allen and colleagues in the mid-1990’s have shown that in developmental toxicity data sets a benchmark response in the range of 5-10% will result in a benchmark dose that is on average similar to a NOAEL.   In any case, the benchmark dose at the 10% level and the lower bound confidence limit on that dose, the BMD10 and BMDL10, should always be presented for comparison purposes   Slide 34: In choosing a BMR for continuous data, if there is consensus across the scientific community regarding what level of change in the endpoint is considered to be biologically significant, leading to an adverse effect, then one should use that amount of change in identifying the benchmark response level. For instance a 10% change in adult body weight, or a doubling of the average serum levels of some liver enzymes are generally considered to be biologically significant and adverse.   In the absence of a scientifically accepted determination, a change in the mean equal to one control standard deviation from the control mean can be used as the default BMR level. This one standard deviation default is based on the fact that when one identifies an adverse response level for which 1% of the animals in the control group are considered to be adversely affected, an exposure that changes the mean response in a population by 1.1 standard deviations has been estimated to result in a 10% increase in the number of animals adversely affected. The 10% increase in the number of affected animals is consistent with the 10% BMR commonly used in BMD analyses of dichotomous data. This is why a BMR of 1 standard deviation should always be reported for continuous data, at least for reference purposes.

  Slide 35: The next step in the Benchmark dose analysis process is to select a model.   Slide 36: Model selection is largely dictated by the type of data being analyzed. Dichotomous models are used to evaluate quantal or dichotomous data, where an effect for an individual may be classified by one of two possibilities, such as death or alive, or in the case of tissue pathology, the existence of a tumor or not.   Slide 37: The EPA BMDS software version 1.4.1 includes five basic types of dichotomous models: the Gamma model; the Logistic model, and within the logistic model, one can use dose or log of dose; the Probit model, and again, one can chose to use dose or log of dose; the multi-stage model; and the Weibull model. There is also a model that is a subset of the Weibull model called the quantal linear model as well, in which the power term for the Weibull model is set to 1. The use of these dichotomous models is covered more extensively in the training module on modeling dichotomous data.   Slide 38: For convenience in modeling cancer data, the BMDS software also provides a preset multistage model, which is called the Multistage-cancer model. The EPA cancer guidelines do not preclude the use of other models, but the Multistage model has historically been the model of choice for EPA cancer risk assessments. A separate training module on the modeling of cancer data will cover this topic more extensively.   The BMDS Multistage-Cancer Model is a modified multi-stage model with several restrictions pre-selected. The beta coefficients within the cancer model are always restricted to be 0 or greater. There is no choice of added risk, only extra risk within the cancer model. The cancer slope factor is always calculated and available in the output. In addition, the model calculates a cancer slope factor by conducting a linear extrapolation between the point of departure for the cancer modeling, the BMDL10, and the background response. This extrapolation always appears on the graphical output, and the upper bound estimate 95% confidence limit on the benchmark dose is printed out in addition to the lower bound 95% confidence limit.   Slide 39: So the cancer slope is calculated based on the linear extrapolation as is graphically shown here from the point of departure, which in this case is the response at the lower bound 95% confidence limit on the benchmark dose to the response at the zero dose.   Slide 40: For continuous data, we need to use models that can evaluate effects that are measured on a continuum, for example, body weight, organ weight, or enzyme levels.   Slide 41:

EPA benchmark dose software provides these 5 continuous models. The polynomial model is a general all-purpose model that is quite flexible and fits many dose-response data sets. Subsets of the polynomial model would be linear, the simplest form of the polynomial model, and non-linear polynomial forms.   A second model type available in BMDS for evaluation of continuous data is the Power model. This model has the capacity to fit L-shaped dose-response curves, and it also can be linear if the power term is 1, or non-linear if the power term is estimated to be anything other than 1.   The Hill model is the only model consistent with the receptor binding as the mechanism of toxicity. It contains a term that predicts a plateau response at the high doses. These models will be discussed more extensively in the training module on the evaluation of continuous data.   Slide 42: A third type of data, nested data, requires special modeling considerations. In the case of nested data, animals are exposed indirectly to chemical, such as through their mothers. The endpoint of concern is measured in the offspring of the mothers, the pups. Thus, all the pups in a litter are related by the mother’s toxicokinetisc and physiological condition. One important aspect of a nested study is that each mother will have a different number of pups, and that litter size may affect the response seen in the offspring, and in addition, each litter may have a particular response attribute, or correlation that needs to be taken into account in the nested data modeling. Currently, the EPA benchmark dose software contains models that address these aspects for nested dichotomous data, but not for continuous response data.   Slide 43: This slide shows an example of some types of nested dichotomous data that one might come across in evaluation of developmental studies. For example, malformations in newborns or neonates might include sternebral defects, vertebral arch defects, and ossification changes. These changes can be identified in neonates and can be modeled as nested data.   Slide 44: The BMDS software contains three nested dichotomous models; the Logistic Nested Model, the NCTR model, and the Rai & Van Ryzin Model. The use of these models is covered more extensively in the BMD training module on modeling nested data.   Slide 45: Beyond the type of data one is analyzing, there are other, principally biological considerations that one needs to take into account before a modeling approach is attempted. Most of the models within BMDS have no biological basis except the Hill model, as mentioned previously, and possible the Multistage cancer model. However, all model predictions must be biologically tenable, and this is why in some cases parameters within the model are restricted to avoid shapes that are not feasible.