Section III. Chemical Risk Assessment
- Quantitative Chemical Risk Assessment Methods
- CHEMICAL RISK ASSESSMENT
- Hazard Identification
- Hazard Characterization
- Exposure Assessment
- Risk Characterization
- RISK ASSESSMENT METHODS
- Conceptual Model of Risk Assessment
- Deterministic Approach
- Probabilistic Approach
- Tiered Approach
- Uncertainty Analysis
Quantitative Chemical Risk Assessment Methods
Risk assessment has changed considerably since its first steps in the 1980s. There have been significant efforts in terms of funding and human resources to produce representative and suitable data to refine exposure assessments (EFSA/FAO/WHO 2011). Mathematical modelling and computational capacities have also helped to introduce probabilistic methods to account for population behaviour and chemical level variability instead of using mean or maximum point estimates (Voet H and Slob 2007). Tiered exposure and risk assessment strategies have been developed to guide risk assessors in identifying priorities for risk management, taking into account available information and uncertainty (Meek et al. 2011). New data, such as biomonitoring data from human cohorts, require the use of toxicokinetic models to be linked with external exposure assessment and to be interpreted in risk management. Moreover, regulation pressures such as the REACH regulation and risk assessment recommendations have prompted risk assessors to develop high-throughput methods to assess exposure to a wide range of chemicals while accounting for all sources and routes of exposure.
This chapter aims to describe the different steps that are part of a chemical risk assessment as well as the different data and methods used to perform it. New developments and future challenges are also addressed.
CHEMICAL RISK ASSESSMENT
Chemical risk assessment is a scientifically based process that aims to quantify the risk of an adverse health effect resulting from exposure of humans to chemicals present in food or the environment over a specified period. It is generally associated with the identification of risk factors and the measurement of their impact on the risk in order to propose management options. Risk assessment is initiated by risk managers, who will decide whether to apply the proposed management options. As a result, risk assessment is preceded by a first step describing the purpose of the risk assessment and the safety questions to be answered, setting time schedules and providing the resources necessary to carry out the work. The risk assessment is divided into four main steps defined by the Codex Alimentarius (Commission 2003) and the United States National Research Council (National Research Council 1983).
This first step is the identification of chemical hazards that could have a negative effect on health. Depending on the various definitions, the hazard could be a chemical with the potential to cause an adverse health effect (FAO/WHO 2004) or a property associated with a chemical rather than the chemical itself (WHO/IPCS 2004). Hazard identification describes the nature and the type of effects that could be caused by the hazard and identifies the affected target organs or target tissues as well as the circumstances under which the effects may be expressed. It must be based on accurate data to evaluate the weight of evidence for adverse health effects (EFSA 2017a). Hazard identification can be done in a prospective way, using in vivo and in vitro toxicological tests, or in a retrospective way by reporting incidents or by examining epidemiological surveys.
Hazard characterization consists of quantitatively defining, when possible, or alternatively qualitatively defining the relationship between the dose and the probability of the occurrence of an effect and/or its seriousness. Combined with the previous one, this step forms the hazard assessment part of the risk assessment.
There are two types of effects: (1) effects with a threshold under which there is no observed effect and (2) effects without a threshold (mutagenic, genotoxic and carcinogenic effects). For the latter, the effect is observed continuously with a greater or lesser probability or severity depending on the dose. When the toxic effect is assumed to have a threshold, the purpose of hazard characterization is to establish health- based guidance values, when the available data make this possible. For example, this threshold value is called an acceptable daily intake (ADI) for additives and pesticide residues or a tolerable intake (TI) for contaminants. These threshold values are established by international organizations (WHO, EFSA, JECFA, JMPR, etc.) or national agencies (US-EPA, RIVM, Health Canada, ANSES, etc.). Exposure to chemicals below such health-based guidance values is generally considered to be safe, and conversely, exposure above them is generally thought of as unsafe. These values are derived from points of departure such as the no observed adverse effect level (NOEAL), the lowest observed adverse effect level (LOAEL) or the benchmark dose (BMD), applying uncertainty factors for possible inter-species and/or inter-individual variability. The BMD is used when a percentage effect or a probability of effect in a given population is considered. This approach is based on modelling of experimental data using a dose-response curve (EFSA 2009). Generally, it is the lower bound (BMDL) of the confidence interval of the BMD that is used to set up the health-based guidance value. Most of the time, these points of departure for a given compound are estimated from toxicological studies conducted in animals. When available, retrospective analyses of human data from epidemiological surveys can be used, integrating the evaluation of inter-individual variability. In this case, no inter-species variability factor is needed. With modern toxicology and efforts to reduce in vivo studies, researchers began to use in vitro testing to estimate points of departure, applying in vitro to in vivo extrapolation methods (IVIVE). However, these methods are adversely affected by the current lack of defined validation protocols and by the complexity of the in vitro to in vivo extrapolation process (Zgheib et al. 2017). Other uncertainty factors than those used for inter-species and/or inter-individual variability can also be applied: an uncertainty factor for the use of the LOAEL, for the extrapolation from subchronic to chronic exposure and for low data quality and/or quantity. Health-based guidance values are specific to a substance on the basis of the administered doses, a duration, a route of exposure (inhalation, ingestion and cutaneous contact), a population and a type of effect. To account for all exposure routes, it is possible to set internal health-based guidance values using absorption factors and/or PBTK models (see Sections 15.4.2 and 15.4.3).
Exposure assessment consists of estimating the dose of a chemical to which an organism or a population may be exposed. The organism or the population can be exposed to a chemical from different exposure sources/ media (diet, air, dust, consumer products, veterinary usages, etc.) and via different routes (ingestion, inhalation and dermal contact). Exposure, also called intake, is generally estimated by combining different types of data depending on the sources of exposure. In the case of food chemicals, dietary exposure assessment takes into consideration the consumed quantities of foods that may contain the chemical, and the amount and frequency of the chemical in those foods. Usually, a range of intake or exposure estimates are provided (e.g. for average consumers and for high consumers), and estimates may be broken down by subgroup of the population (e.g. infants, children or adults). Considering the chemical and the studied effect, exposure can be "acute" and thus calculated over a short period of time using a meal or a day, or "chronic" with an estimate over the entire lifespan. For chronic exposure, the mean values of consumed quantities and of concentrations are used, whereas for acute exposure, the consumed quantity over a meal or a day is combined with the highest observed concentration in the media.
Generally, different scenarios are studied considering different subpopulations, times of exposure, and assumptions regarding censored data and calculations.
The risk is defined as the probability of occurrence of known and potential adverse health effects in a given organism and/or population. Risk characterization consists in combining the hazard characterization step with the exposure assessment step in order to propose suitable advice for decision-making in risk management. For example, a comparison of the relative risks obtained with different risk management options is useful for risk managers. In practice, quantitative risk assessment consists in comparing the health-based guidance values with the estimated levels of exposure under different exposure scenarios. In this way, the risk in a population can be estimated by determining the number of individuals with a higher exposure level than the health-based guidance value divided by the total number of individuals in the population. It is possible to focus on consumers only, divided by the total number of consumers or users. Other indicators such as the hazard index (exposure divided by the health-based guidance value) or the margin of exposure (МОЕ, exposure divided by the NOEAL) have been developed and are commonly used to characterize the risk.
The risk assessment should include all key assumptions, the different sources and types of uncertainty and should describe the nature, relevance and magnitude of any risk to human health. It should also include, where relevant, information on susceptible sub-populations, including those with greater potential exposure or those with specific predisposing physiological conditions or genetic factors.
RISK ASSESSMENT METHODS
Conceptual Model of Risk Assessment
A conceptual model is a written description and visual representation of predicted relationships between the organism or the population and the hazard to which they may be exposed. It aims to represent the sequence of events that can lead to risk and to evaluate this risk qualitatively or quantitatively (Figure 15.1). A risk assessment model is composed of input variables and output variables.
The input variables of a risk model for chemical exposure are those related to:
- • The medium/media (e.g. dust, soil, diet) where the chemical(s) can be found, expressed as the frequency and the level of the chemical(s) in these media.
- • The behaviours of the organism or the population, for example food consumption quantity when dietary exposure is considered,
Figure 15.1 Conceptual model of general population exposure to four pyre- throids (cyfluthrin, deltamethrin, cypermethrin and permethrin). The aggregate exposure to each pesticide was estimated from different sources and cumulated using relative potency factors to calculate the margin of exposure with a BMDL value. Also, internal exposure was estimated using a PBTK model of five metabolites and compared with biomonitoring data. (From the article by Vanacker, M. et al., 2020.) or the applied quantity of cream for cosmetic exposure assessment, or time spent outdoors and indoors when considering air exposure.
- • The physiological characteristics of the individuals in the population, which are needed to estimate the exposure, for example body weight, individual size or respiratory volume.
- • The health-based guidance value, if it is available, or a chosen point of departure and its corresponding uncertainty factors. Dose-response curve modelling can also be used in some cases, such as in food allergies.
The output variables are external and/or internal exposure and the associated risk. The model can become more complex, depending on the information available, by adding different sources and routes of exposure, using toxicokinetic models, and other input variables that can influence the risk. The conceptual model can be reproduced considering different exposure scenarios and sub-populations.
A deterministic risk assessment consists in setting point values for the input variables of the risk assessment model. The output variables of the model are then also expressed as a point value. The values used can be the average or the median, but most often they are extreme values (the 95th, 97.5th or 99th quantiles or the maximum) and are chosen in order to propose "worst case" scenarios. The use of extreme scenarios is called the conservative approach, and the result is considered a maximized risk. This approach requires few data and technical means. As a result, it is used in data-poor situations or as a first-line approach to study whether in the "worst case", the risk can be ruled out or not. In the case where the risk cannot be ruled out, a so-called probabilistic approach is generally proposed in order to produce more precise estimates of the risk. If, on the other hand, the risk is zero with this conservative approach, it is often considered unnecessary to perform more calculations. Before concluding, however, special attention must be paid to the fact that the conservative scenario incorporates all possible uncertainties. For example, if the uncertainty related to the variability of consumption is high, then the extreme value used for this variable can be undervalued and thus skew the associated risk. This is why it is important to adopt an uncertainty analysis approach, as explained in Chapters 7 and 16.
The probabilistic approach involves the use of the set of possible values that the input variables of the risk assessment model can take. The variables are then described by probability distributions determined from observed data or from expert reports if the data are lacking (elicitation of experts' statements).
There are different approaches to determining and using the distribution of a variable from observed data. The most intuitive approach is to use the empirical distribution of data, i.e. the distribution is defined only by the observed values. In this case, it is not necessary to define beforehand the form of the distribution. In contrast, the parametric approach consists in defining a form of the distribution by the use of known probability distributions, such as the Gaussian, exponential, lognormal or Weibull distribution, etc. In this case, it is necessary to determine the values of the parameters of the distribution from the observed data. Two approaches are then possible: the frequentist approach, based solely on the maximization of the likelihood from the observed data, and Bayesian inference, which takes into account both the likelihood and also a priori information on the parameters, which may come from expert reports or results of previous studies. In the case of Bayesian inference, the parameters also take the form of a distribution. Once the probability distributions are defined, they are integrated into the model using Monte Carlo simulation methods. These methods involve randomly drawing multiple values in the different probability distributions of the variables. Thus, the variability and uncertainty are propagated throughout the model and make it possible to produce output variables that also take the form of probability distributions. Variability and uncertainty can be propagated separately using two-dimensional (2-D) Monte Carlo simulations. The combination of 2-D Monte Carlo with Bayesian inference as proposed in Rimbaud et al. (2010) for food allergies easily allows this separation.
Intermediate methods between the deterministic and probabilistic approaches have been developed. For example, intervals around a quantitative value or quantitative uncertainty tables are ways of integrating variability and uncertainty without going as far as a probabilistic approach. It should also be noted that using a probabilistic approach for all input variables is often not possible. In this case, a deterministic approach for some variables and a probabilistic approach for others are often combined (as proposed in Figure 15.1).
The tiered approach is a widely used approach in risk assessment. Its objective is to organize scientific knowledge in a situation of uncertainty in order to help the decision-maker in the implementation of management actions. This approach is of an iterative nature and can be defined as the setting up of a series of questions whose answers require the mobilization of more and more precise knowledge in terms of both hazard and exposure. It therefore provides a graduated level of response that will depend on: basic knowledge (how far is it possible to go?), the aspects that the manager needs in order to act (do we know enough to act?), the constraints of implementation in terms of means (human, financial, etc.) and a scientific approach proportionate to the needs of the question asked.
In risk assessment, the tiered approach allows gradual recognition of uncertainty. WHO/IPCS (2008), EFSA (2017b) and Meek et al. (2011) recommended a tiered approach at three levels (qualitative, deterministic and probabilistic). At the first level, simple and fast methods based on conservative hypotheses (overestimation of risk) are favoured. The purpose of these methods is to help the decision-maker to identify priority issues that require either specific regulation, data collection or specific research work. In other words, this first level draws attention to worrying situations. In this case, higher levels are implemented. These levels rely on the use of mathematical models, which are increasingly complex and/or more and more probabilistic.
It is increasingly recommended that any risk assessment should be accompanied by an uncertainty analysis (NRC 2013, WHO/IPCS 2008,2014, EFSA 2017b); see Chapters 7 and 16 for more details. The purpose of the uncertainty analysis is to identify and quantify the different sources of uncertainty associated with the risk being studied. The risk is then expressed as a function of its associated degree of uncertainty, thereby increasing the reliability and robustness of the result. An uncertainty analysis is broken down into several stages: 
concerning the population studied and the question asked, the tools and methods of data collection, the level of complexity, the adaptability of the chosen model, the studied scenarios, etc. Tools such as uncertainty matrices make it possible to classify the various sources of uncertainty according to their dimensions and characteristics such as their origin (lack of knowledge, variability, etc.), and their location (data, model, etc.). It is not possible to address all identified uncertainties in the following steps. In general, the most important uncertainties in relation to the question asked will be considered in the assessment steps.
- • Evaluate uncertainties and their impact on risk. The step for evaluating the uncertainties consists in estimating the range of the possible values of the various uncertainties relating to the problem posed. In the best case, the evaluation is carried out using data, which can come from the literature or other studies, or by referring to experts who will be able to describe the amplitude of the uncertainties. Probability distributions, whose parameters have been determined on the basis of expert data or statements, or the empirical distribution of the observed values, are then used to quantify the uncertainties. In the case where there are no data or experts who can be consulted, qualitative tools such as verbal adjectives or ordinal scales can be used to characterize their amplitude. Once the uncertainties have been assessed, they are generally propagated using Monte Carlo simulations throughout the various stages of the model in order to evaluate their impact on risk. It is important to simultaneously vary all the uncertainties to take into account their dependence on the final result. It is also beneficial to vary only one source of uncertainty at a time and to set others so as to assess the individual impact of the various uncertainties. Indeed, this step makes it possible to identify the most important sources of uncertainty and thus, to decide to put in place measures to reduce them.
- • Represent and communicate uncertainties. This step consists in communicating the results of the uncertainty analysis to the managers in the form of graphs or tables in order to take into account the uncertainties relating to the risk assessment when making the decision.
-  Identify and describe the uncertainties. These two steps consist in listing all the uncertainties about the risk being studied.The sources of uncertainty are related to the incompletenessof knowledge, the representativeness and quality of data