State of the art in benefit–risk analysis: Food and nutrition

This page is a nugget. The page identifier is Op_en5568
Moderator:Essi Vuorinen (see all)
Upload data {{#opasnet_base_link:Op_en5568}}	Unlike most other pages in Opasnet, the nuggets have predetermined authors, and you cannot freely edit the contents. Note! If you want to protect the nugget you've created from unauthorized editing click here

This page (including the files available for download at the bottom of this page) contains a draft version of a manuscript, whose final version is published and is available in the Food and Chemical Toxicology 50 (2012) 5–25. If referring to this text in scientific or other official papers, please refer to the published final version as: M.J. Tijhuis, N. de Jong, M.V. Pohjola, H. Gunnlaugsdóttir, M. Hendriksen, J. Hoekstra, F. Holm, N. Kalogeras, O. Leino, F.X.R. van Leeuwen, J.M. Luteijn, S.H. Magnússon, G. Odekerken, C. Rompelberg, J.T. Tuomisto, Ø. Ueland, B.C. White, H. Verhagen: State of the art in benefit–risk analysis: Food and nutrition. Food and Chemical Toxicology 50 (2012) 5–25 doi:10.1016/j.fct.2011.06.010 .

Title

Editing State of the art in benefit–risk analysis: Food and nutrition

Authors and contact information

M.J. Tijhuis, correspondence author

(mariken.tijhuis@rivm.nl)

(National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands)

(Maastricht University, School of Business and Economics, Maastricht, The Netherlands)

N. de Jong

(National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands)

M.V. Pohjola

(National Institute for Health and Welfare (THL), Kuopio, Finland)

H. Gunnlaugsdóttir

(Matís, Icelandic Food and Biotech R&D, Reykjavík, Iceland)

M. Hendriksen

(National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands)

J. Hoekstra

(National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands)

F. Holm

FoodGroup Denmark & Nordic NutriScience, Ebeltoft, Denmark

N. Kalogeras

(Maastricht University, School of Business and Economics, Maastricht, The Netherlands)

O. Leino

(National Institute for Health and Welfare (THL), Kuopio, Finland)

F.X.R. van Leeuwen

(National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands)

J.M. Luteijn

University of Ulster, School of Nursing, Northern Ireland, United Kingdom

S.H. Magnússon

(Matís, Icelandic Food and Biotech R&D, Reykjavík, Iceland)

G. Odekerken

(Maastricht University, School of Business and Economics, Maastricht, The Netherlands)

C. Rompelberg

(National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands)

J.T. Tuomisto

(National Institute for Health and Welfare (THL), Kuopio, Finland)

Ø. Ueland

(Nofima, Ås, Norway)

B.C. White

(University of Ulster, Dept. of Pharmacy & Pharmaceutical Sciences, School of Biomedical Sciences, Northern Ireland, United Kingdom)

H. Verhagen i,j

(National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands)

(Maastricht University, NUTRIM School for Nutrition, Toxicology and Metabolism, Maastricht, The Netherlands)

(University of Ulster, Northern Ireland Centre for Food and Health (NICHE), Northern Ireland, United Kingdom)

Article info

Article history: Available online 12 June 2011

Abstract

Benefit–risk assessment in food and nutrition is relatively new. It weighs the beneficial and adverse effects that a food (component) may have, in order to facilitate more informed management decisions regarding public health issues. It is rooted in the recognition that good food and nutrition can improve health and that some risk may be acceptable if benefit is expected to outweigh it. This paper presents an overview of current concepts and practices in benefit–risk analysis for food and nutrition. It aims to facilitate scientists and policy makers in performing, interpreting and evaluating benefit–risk assessments.

Historically, the assessments of risks and benefits have been separate processes. Risk assessment is mainly addressed by toxicology, as demanded by regulation. It traditionally assumes that a maximum safe dose can be determined from experimental studies (usually in animals) and that applying appropriate uncertainty factors then defines the ‘safe’ intake for human populations. There is a minor role for other research traditions in risk assessment, such as epidemiology, which quantifies associations between determinants and health effects in humans. These effects can be both adverse and beneficial. Benefit assessment is newly developing in regulatory terms, but has been the subject of research for a long time within nutrition and epidemiology. The exact scope is yet to be defined. Reductions in risk can be termed benefits, but also states rising above ‘the average health’ are explored as benefits. In nutrition, current interest is in ‘optimal’ intake; from a population perspective, but also from a more individualised perspective.

In current approaches to combine benefit and risk assessment, benefit assessment mirrors the traditional risk assessment paradigm of hazard identification, hazard characterization, exposure assessment and risk characterization. Benefit–risk comparison can be qualitative and quantitative. In a quantitative comparison, benefits and risks are expressed in a common currency, for which the input may be deterministic or (increasingly more) probabilistic. A tiered approach is advocated, as this allows for transparency, an early stop in the analysis and interim interaction with the decision-maker. A general problem in the disciplines underlying benefit–risk assessment is that good dose–response data, i.e. at relevant intake levels and suitable for the target population, are scarce.

It is concluded that, provided it is clearly explained, benefit–risk assessment is a valuable approach to systematically show current knowledge and its gaps and to transparently provide the best possible science- based answer to complicated questions with a large potential impact on public health.

Abbreviations

ADI, acceptable daily intake; AICR, Association for International Cancer Research; ALARA, as low as reasonably achievable; AR, Average Requirement; ATBC, alpha-tocopherol, beta carotene cancer prevention (trial); BCRR, Benefit Cancer Risk Ratio; BENERIS, benefit–risk assessment for food: an iterative value-of-information approach; BMD, benchmark dose; BMDL, lower one-sided confidence limit on the BMD; BNRR, Benefit Noncancer Risk Ratio; BRA, Benefit–risk analysis; BRAFO, benefit risk analysis of foods; CARET, beta-carotene and retinol efficacy trial; DALY, disability adjusted life years; EAR, Estimated Average Requirement; EFSA, European Food Safety Authority; FAO, [United Nations] Food and Agriculture Organization; FOSIE, food safety in Europe: risk assessment of chemicals in the food and diet; FUFOSE, Functional Food Science in Europe; ILSI, International Life Sciences Institute; JECFA, Joint FAO/WHO Expert Committee on Food Additives; LLAB, lower level of additional benefit; LOAEL, Lowest Observed Adverse Effect Level; MOE, margin of exposure; NOAEL, No Observed Adverse Effect Level; OECD, Organisation for Economic Co-operation and Development; PAR, population attributable risk; PASSCLAIM, process for the assessment of scientific support for claims on foods; POD, point of departure; PRI, population reference intake; RDA, recommended dietary allowance; RfD, Reference Dose; RIVM, [Dutch] National Institute for Public Health and the Environment; RR, relative risk; SCF, Scientific Committee on Food; (P)TDI/WI/MI, (provisional) tolerable daily/weekly/monthly intake; UL, tolerable upper intake level; ULAB, upper level of additional benefit; WHO, World Health Organization; QALIBRA, quality of life – integrated benefit and risk analysis; QALY, quality adjusted life years; WCRF, World Cancer Research Fund.

Keywords

Keywords: Benefit–risk, Benefit, Risk, Food, Nutrition

Introduction to the scope of this paper

Food and nutrition are in essence necessary and beneficial to, but may also have adverse effects on, human health. Public health professionals have realized, while acknowledging the success of the current food safety system, that the health loss due to unhealthy food and nutrition is many times greater than that attributable to unsafe food; and that the health gains to be made through the consumption of certain foods are many times greater than the health risks involved (van Kreijl et al., 2006). The beneficial and adverse effects may occur concurrently through a single food item (for example fish or whole grain products) or a single food component (for example folic acid or phytosterols), within the same population range of dietary intake. This means that any policy action directed at the adverse effects also affects the degree of beneficial effects. The challenge in the relatively new field of benefit–risk assessment in food and nutrition is to scientifically measure and weigh these two sides.

Historically, the assessments of benefits and risks have been separate processes, where by far most attention has been directed at the assessment of risks, as required by regulation (EU, 2000, 2002b). Focus has been on food safety, with precaution as a risk management option (EC, 2000a) surrounded by much debate as to when and how it should be applied. Worded differently, it is a political decision whether or not to include benefits. From a public health perspective, however, a decision-making process which strives for the lowest risk does not by definition lead to the optimal population health outcome. If the benefits are high enough, some risk may be acceptable. As such, there is a broader picture to be assessed.

The benefit–risk assessment approach entails a paradigm shift from traditional risk analysis, as known from toxicology, to benefit– risk analysis. Essentially, it brings together different scientific disciplines, which have their own history, perspective, tools and uncertainties:

• Toxicology, the discipline that traditionally investigates the risk

of ‘too much’ (acute or chronic adverse effects), as required by the regulatory process.

• Nutrition, a younger discipline that investigates the risk of ‘too

little’ (nutritional deficiencies), the risk of ‘too much’ (nutritional intoxications or affluence-associated problems) and the benefits of optimal nutrient intakes.

In benefit–risk assessment, there is an important role for epidemiology, the methodology-oriented discipline that investigates associations between (real-life human) exposures and health outcomes, adverse or beneficial.

In order to weigh the risks and benefits of foods or their components, they must be evaluated and expressed in a comparable way. Methodologies have been developed and applied to actual cases. New grounds have been explored, yet consensus on the general principles or approaches for conducting benefit– risk analysis for foods and food components still needs to be reached.

This paper presents an overview of current approaches to an integrated weighing of benefits and risks, a state of the art in benefit– risk analysis, in the field of food and nutrition. In this, most attention is directed at the assessment phase, but benefit–risk management and benefit–risk communication are also recognized as an integral part of the benefit–risk paradigm. The perspective is broad, rather than technical. The aim is to facilitate scientists and policy makers in carrying out and judging benefit–risk assessments, to eventually come to better informed decisions about food-related health issues.

We start with an overview of key terms in the field. We then describe risk assessment and benefit assessment separately to build up to a description of current approaches to combine the two. This is illustrated by an overview of benefit–risk case studies and followed by a brief description of benefit–risk management and communication. We end with conclusions and recommendations.

Key terms in benefit–risk assessment of food and nutrition

As benefit–risk assessment of food and nutrition involves different disciplines using different scientific jargon, we first define the key terms as they are used in this review: food, nutrition, risk, benefit and benefit–risk assessment.

In this review, food is defined as any substance or product, whether processed, partially processed or unprocessed, intended to be, or reasonably expected to be ingested by humans (EU, 2002a). Foods are made up of many different components affecting body functions. ‘‘Food’’ and ‘‘food component’’ are viewed as a broad concept: macronutrients, micronutrients, bioactive nonnutrients, additives, contaminants, residues, phyto/phyco/mycotoxins, micro-organisms, allergens, supplements, whole foods and novel foods. The total of these components that are ingested via food, naturally occurring or added, can be seen as nutrition. Notably, ‘nutritious’ has the connotation of ‘healthy’. As a science, ‘‘nutrition’’ studies the total of processes by which food or its components are ingested, digested, absorbed, metabolized, utilized, needed and excreted, and the interactions between these components and between these processes, and their effect on human health and disease.

In the food and nutrition context, we see risk as the probability and severity of an adverse health effect occurring to humans following exposure (or lack of exposure) to a food or food component. In risk assessment, a distinction is made between ‘‘risk’’ and ‘‘hazard’’. In this context, the intrinsic potential of a food (component) to result in adverse health effects is called a hazard. A hazard, then, becomes a risk if there is sufficient exposure. Risk can result from both action and lack of action and both can be analyzed (Wilson and Crouch, 2001). An ‘‘adverse health effect’’ is defined by the WHO as ‘a change in morphology, physiology, growth, development or life span of an organism, which results in impairment of functional capacity or impairment of capacity to compensate for additional stress or increase in susceptibility to the harmful effects of environmental influences’ (WHO, 1994). A ‘‘beneficial health effect’’ can analogously be seen as ‘an improvement of functional capacity or improvement of capacity to deal with stress or decrease in susceptibility to the harmful effects of environmental influences’ in the above definition (Palou et al., 2009). On the benefit side, no parallel words exist for ‘‘hazard’’ and ‘‘risk’’. In order to make the parallel distinction on the benefit side but also have similarity in nomenclature, we will use ‘‘adverse health effects’’ and ‘‘beneficial health effects’’ instead of hazards and benefits. In this review we use benefit analogously to ‘‘risk’’, as the probability and degree of a beneficial health effect occurring to humans following exposure (or lack of exposure) to a food (component). Beneficial health effects may postpone the onset of disease and thus benefit can also be measured as a reduction of risk.

Benefit–risk assessment is seen here as a science-based process intended to estimate the benefits and risks for humans following exposure (or lack of exposure) to a particular food or food component and to integrate them in comparable measures, thus facilitating better informed decisions by decision-makers.

Risk assessment

Risk assessment is an established field, that is mainly addressed by toxicology (Faustman and Omenn, 2008). The traditional risk assessment paradigm (EC, 2000b) consists of

• hazard identification (what effect?),
• hazard characterization (at what dose? how?),
• exposure assessment (how much is taken in?) and
• risk characterization (what is the probability and severity of the effect?).

A risk assessment framework specifically for food has been developed in the context of the FOSIE project (Food Safety in Europe, www.ilsi.org/Europe) (Smith, 2002). The basic concepts and approaches to risk assessment in toxicology will be addressed in Section 3.1. A separate paragraph is dedicated to micronutrients, which is placed under toxicology for practical reasons. There is an additional (not yet clearly defined) role for epidemiology, which will be addressed in Section 3.2. Focus will be on those characteristics where the most pronounced differences exist between toxicology, epidemiology and later benefit assessment, and which need to be considered when integrating the different disciplines. Table 1 presents a summary of some of the characteristics of toxicological and epidemiologic risk assessment.

Table 1. Characteristics of toxicological and epidemiological risk assessment^a, that need to be integrated, adapted from van den Brandt et al., 2002.

Characteristic	Tox	Epi
Initiation	Mainly regulatory	Mainly academic
Starting point	Usually suspicious substance	Both substance and endpoint
Study design	Mainly experimental	Mainly observational
Formal procedures defined	Yes	No
Study population	Selected, genetically homogeneous animals	Mainly free-living, genetically heterogeneous humans
Exposure under study	Controlled (administered), mostly one agent, supraphysiological doses	Uncontrolledb (measured), multiple agents likely, physiological doses
Extrapolation	To lower doses (‘actual exposure’) using default value of 10 or chemical specific adjustment value	No, actual exposure
Presentation of exposure and risk	Usually absolute	Usually relative (ranking of study subjects)
Quantification of effect (with chronic exposure)	Incidence (proportion of animals with adverse effect over duration of study; usually lifetime risk;) or continuous measure	Incidence rate (proportion with adverse effect per person-year, standardized to a particular age distribution; lifetime risk rare) or continuous measure
Comparison	Difference	Usually ratioc
Evaluation of evidence	Selection of high-quality study with the most sensitive effect in the most sensitive subgroup (worst-case scenario or positive evidence approach)	All high-quality studies (weight of evidence approach, calculating overall effect)
Some sources of uncertainty	Threshold assumption, low-dose extrapolation Interspecies differences Human interindividual variation Imprecision	Exposure measurement error Residual confounding Selection bias Human interindividual variation Imprecision

^a With focus on chronic exposures.

^b Unless randomized clinical trial (RCT).

^c Necessary for case-control study; for cohort studies calculation of risk differences is also possible.

Food toxicology

Classical food toxicology focuses on the absence of risks, viz. on safety (EU, 2000, 2002b). Risk assessment may be initiated for a variety of purposes, most commonly it is triggered by a legal requirement (EC, 2000b).

Basic principles

The basic tenet of quantitative risk assessment is that data on health effects detected in small populations of animals exposed to relatively high concentrations of an agent can be used to predict health effects in large human populations exposed to lower concentrations of the same agent. Generally, risk is measured as the fraction of the population exceeding a defined upper intake or ‘guidance’ level. This level is established by working down from the most sensitive hazardous effect (the so-called critical effect) in animals and applying uncertainty factors for extrapolation to the most sensitive human subgroup. The toxicological approach differs between compounds that are assumed to show a threshold effect and those that are not, between compounds that are avoidable in the diet and those that are not, and depends on the amount of scientific information available and the policy of decision-makers towards acceptance of risk. First we briefly address the identification of adverse health effects, then we address their characterization and translation into risk.

Identification of adverse effects

The better toxicity studies have an appropriate number of dose levels, sufficient sample sizes and focus on relevant endpoints in a relevant species (WHO-Harmonization-Project, 2005–2003). These are mostly animal toxicity tests, with an experimental design following OECD guidelines for the testing of chemicals (OECD). Tests of acute toxicity, i.e. the toxicity that manifests itself immediately or within 14 days after exposure to a single administration of a chemical by ingestion, inhalation or dermal application, are often not useful for hazard identification and risk assessment in relation to foods and food chemicals (Barlow et al., 2002), because human exposure is generally of chronic nature and is usually much lower than the dose that is identified in these tests. Tests of subacute or subchronic toxicity, i.e. repeated dose toxicity studies mostly performed in rodents for a period of 28 or 90 days, should reveal the major toxic effects. The effects may include changes in body and organ weight, organ histopathology, hematological parameters, serum and urine clinical chemistry and sometimes extensions to screen for neurotoxicity or immunotoxicity. Tests of chronic toxicity cover a larger part of the animal’s life span, e.g. 12 or 24 months in rodents. There are specific tests and procedures for establishing reproductive and developmental toxicity, for neurotoxicity, for genotoxicity, for carcinogenicity, for immunitoxicity and for food allergies (Barlow et al., 2002).

Characterization of adverse effects and risk

The standard approach to characterize a dose–response relationship, for substances assumed to show non-carcinogenic threshold effects, is the derivation of the No Observed Adverse Effect Level (NOAEL). This is the largest amount of a substance that the most sensitive animal model can consume without adverse health effects (see Fig. 1a). Usually, the most robust datasets are chosen with adverse effects occurring at the lowest levels of exposure from studies using the most relevant exposure routes (Faustman and Omenn, 2008). If there are no adequate data demonstrating a NOAEL, then a LOAEL (the lowest amount at which adverse effect are observed) may be used. Based on the NOAEL (or LOAEL), health-based guidance levels are established, under the rationale that these will ensure protection against all other adverse health effects which may be caused by the compound considered. This means that although different endpoints may yield different NOAEL’s, one NOAEL is chosen as a point of departure from the experimental dose–response range for setting acceptable exposure levels and this is the one for the most relevant sensitive endpoint in the most sensitive species; Based on this NOAEL, a health based guidance value is established by means of uncertainty factors. By default (i.e. unless specified data are available to do otherwise) the NOAEL is divided by 10 to take into account differences between animals and humans (interspecies), and again by 10 to take into account differences in sensitivity between humans (interindividual) (Dorne and Renwick, 2005; Renwick, 1993).

The guidance level for nutrients (see also Section 3.1.4) is the tolerable upper intake level (UL), for additives is termed the ADI (acceptable daily intake) and for contaminants in food the terminology in Europe is tolerable daily (or weekly/monthly) intake (TDI/TWI/TMI). In the USA, Reference Dose (RfD) is preferred as a less value-laden term for the latter. All these terms refer to an estimate of the intake of a substance over a lifetime that is considered to be ‘without appreciable health risk’ (WHO, 1994). They are meant to protect 98% of the population. To most toxicologists, guidance levels are ‘‘soft’’ estimates and not a matter of marking acceptability from non-acceptability (EPA, 1993). Exposures somewhat higher than the guidance level are associated with increased probability of adverse effects, but that probability is not a certainty. Similarly, the absence of a risk to all people cannot be assured at the level below the guidance level (EPA, 1993). As a consequence, more and more dose–response modeling is being advocated as a more science-based way to go. Different mathematical and physiological approaches are being explored. Edler et al. (2002) describe a number of statistical and mechanistic modeling approaches with differing data requirements, degree of complexity, applicability and type and quality of resulting risk estimates, including: structure–activity relationships and the threshold of toxicological concern; the benchmark dose (BMD); probabilistic risk assessment; and physiologicallybased pharmacokinetic modeling. The latter fits in the development towards a ‘systems toxicology’; in vivo dose–response curves could be predicted by combining in vitro toxicity data and in silico kinetic modeling (Louisse et al., 2010; Punt et al., 2008). Other initiatives exist, e.g. the Key Events Dose–Response Framework (KEDRF) incorporates information on the substance’s mode of action and examines critical events that occur between the initial dose of a bioactive agent and the effect of concern (Julien et al., 2009).

By default, the threshold approach is not applied to genotoxic carcinogens, because theoretically (although the organism is equipped with some degree of cytoprotection) one molecule can initiate the cancer process. The Joint FAO/WHO Expert Committee on Food Additives (JECFA) and in Europe the SCF (the predecessor of EFSA, the European Food Safety Authority) advised that the presence of carcinogenic contaminants in the diet should be reduced to ‘irreducible levels’ by current technological standards (Edler et al., 2002), which is complicated by the fact that detection limits are increasingly more sensitive. This approach is captured in the ALARA (as low as reasonably achievable) management option to regulate avoidable genotoxic carcinogens. These committees considered that there was no adequate science basis for low-dose extrapolation to quantify a ‘‘virtual safe dose’’, an acceptable additional cancer risk for lifetime exposure to the compound of interest which can be set at, for example, one additional cancer case in a population of one million people. Current methods that do quantify a virtual safe dose are based on linear extrapolation from a point of departure on the dose–response curve (Boobis et al., 2009). It is argued that for genotoxic carcinogens, both the threshold concept and the low-dose linearity concept are in need of refinement (Boobis et al., 2009; Rietjens and Alink, 2006; Waddell, 2006). This will require better mechanistic insight and risk–benefit analysis (Boobis et al., 2009; Rietjens and Alink, 2006; Waddell, 2006). It can help overcome the critique with regard to the precautionary principle that it does not provide a transparent way to weigh and account for different risks, or in other words, that it does not help regulators to decide which risks to regulate (EFSA, 2005; Post, 2006).

EFSA recommends using the margin of exposure (MOE) approach as a harmonised methodology for assessing the risk of substances in food and feed, which are genotoxic and carcinogenic (EFSA, 2005). The MOE approach uses a reference point, that is often taken from an animal study and that corresponds to a dose that causes a low but measurable response in animals. This reference point is then compared with various dietary intake estimates in humans, taking into account differences in consumption patterns in the population (EFSA, 2005). The MOE essentially shows whether the human levels of exposure are close to effect levels and enables the comparison of the risks posed by different genotoxic and carcinogenic compounds. It is for the risk manager to decide if the magnitude of the MOE is acceptable, and if further action is needed taking into consideration additional aspects, such as the feasibility of removing the substance from the food supply. In general a MOE of 10,000 or higher, if it is based on a bench mark dose lower confidence limit for a 10% change in effect (BMDL10, see below) from an animal study, might be considered of low concern from a public health point of view and as a low priority for risk management actions. However, this number is arbitrary and the judgment is ultimately a matter for decision-makers. Moreover an MOE of that magnitude in itself should not preclude the application of risk management measures to reduce human exposure. Recently, an outline framework for calculating an MOE was proposed in order to help to ensure transparency in the results (Benford et al., 2010).

The MOE approach could also be applied for non-carcinogenic substances. EFSA recommends the use of the benchmark dose (BMD) to estimate the MOE, for instance where the human exposure is close to the ADI (EFSA, 2005, 2009b). The BMD approach estimates the dose that causes a low but measurable critical effect, e.g. a 5% increase in the incidence of kidney toxicity or a 10% percent change in the level of liver weight. By calculating the lower confidence limit of the estimated dose (BMDL), the uncertainty and variability in the data is taken into account (EFSA, 2009b). In this BMD approach more information is used than in the NOAEL approach.

Micronutrients

For micronutrients, a somewhat different approach needs to be followed, as risk can come from intakes that are too high, but also from intakes that are too low. Micronutrients, home in the field of nutrition, are addressed here because toxicology-related measures are used in their risk assessment. The higher end guidance level, the UL (see Fig. 1a), differs in two principal aspects from the ADI and TDI. First, micronutrients, as opposed to additives or contaminants, are subject to homeostatic control. This means that the body may adapt its functioning to intake and deal with some level of excess or shortage. Second, UL’s may be set for different life stage groups.

The other often used guidance level, the recommended dietary allowance (RDA, see Fig. 1b) (also known as population reference intake, PRI) (EFSA, 2010a), is by default set for different life stage groups and by sex and represents the nutritional needs of most healthy individuals (thereby exceeding the minimal requirements for almost all). It is calculated from the Estimated Average Requirement (EAR, see Fig. 1b) or Average Requirement (AR) (EFSA, 2010a), which is the daily intake value that is estimated to meet the requirement, as defined by a specified indicator of adequacy, in 50% of the individuals in a life stage or sex group (Renwick et al., 2004). Often, a normal distribution of requirements and a 10– 15% coefficient of variation (CV) for the EAR are assumed. The RDA must be considered when deriving the UL, as it is possible that default uncertainty factors would yield a UL below the RDA. The intake range between deficiency and toxicity varies greatly between nutrients. For example, it is large for vitamin C and narrow for zinc. In addition, it may overlap between different populations. Related to this given, there is critique on current methods of establishment of the upper level for micronutrients and maximum permitted levels of vitamins and minerals in food supplements working from precaution, and a call for using ‘decision science’ is made (Verkerk, 2010). Here, the point is that dosages that induce risks in a few might actually at the same time induce benefits in many (Verkerk, 2010).

Nutritional epidemiology

In the traditional risk assessment of foods, the use of epidemiologic data has been limited. There is no mandatory role (and there are no clear criteria) for epidemiologic data in risk assessment for regulatory purposes, despite the fact that (nutritional) epidemiologists are conceptually and practically already used to describe effects at physiological intake levels and despite the limitations of animal studies due to the requirement of both interspecies and high-to-low dose extrapolations.

Basic principles

Epidemiology is about identifying and quantifying associations between (real-life) human exposures and health outcomes, adverse or beneficial. The exposure and the health outcome, usually a disease or marker for a disease, are most often a priori specified and hypothesis-driven. If the association between exposure and outcome is found to be inverse, i.e. higher exposure leads to lower disease occurrence, this signifies a reduction in risk. In other words, in epidemiology, an ‘exposure’ is a factor or condition that may increase or decrease the risk of disease (WCRF/AICR, 2007). This work often serves the quest for knowledge and understanding more than a practical regulatory goal, but it is also possible to quantify the effect on (public) health to directly aid in decision making. Reflecting the above, three types of measures can broadly be distinguished in epidemiology: measures of disease occurrence, measures of association and measures of effect (Rothman et al., 2008). Disease occurrence can be expressed as cases that are prevalent (current) or incident (new). The measures of association are most often expressed as relative differences (or ‘ratio’ measures), for example: the relative risk (RR) to acquire disease X among those consuming food (component) Y in the 5th quintile of the intake distribution is 1.5 compared to those in the 1st quintile (the reference, set to 1); but often absolute difference measures, such as the ‘excess risk’ that consuming food (component) Y contributes to acquiring disease X, can also be calculated. The measures of effect have impact on those exposed or on the population. An example of the latter is the population attributable risk (PAR), which can be seen as the reduction in occurrence of disease X that would be achieved if the population had been entirely unexposed to food (component) Y, compared with its current exposure pattern (Rothman et al., 2008). In the risk assessment context, epidemiologic studies are mainly observational in design.

Study design and dose–response characterization in epidemiology

The great advantage of epidemiology is the fact that it actually studies the species it wants to make inferences about. Yet this is also its hurdle, as humans are free-living and subject to many influences. Both exposure and outcome are difficult to measure accurately and subject to systemic bias and measurement error (Kroes et al., 2002; Rothman et al., 2008; Willett, 1998). These are not insurmountable problems, however. The possibility of bias occurring can be reduced through good study design. Study design can be experimental or observational. Experimental design in humans, encompassing randomized controlled trials (RCT’s), is not ethical or feasible in case of (severe) risks or long latency (van den Brandt et al., 2002). In some cases, unintended risks turn up in trials, such as in the ATBC and CARET trials (ATBC-Study-Group, 1994; Omenn et al., 1996) and in the Norwegian Vitamin Trial and Western Norway B Vitamin Intervention Trial for folate (Ebbing et al., 2009), where an increased percentage of cancer cases was seen where a decrease was expected. These trials were, however, limited to one dose, and thus dose–response relationships could not be characterized. Observational studies include cohort studies (starting with exposure assessment and following up on disease development) and case-control studies (starting with diseased and non-diseased and assessing past exposure). They allow for long latencies between exposure and health outcome of interest. These studies do not provide direct evidence for causal relationships, but a number of methodological considerations have been proposed to judge whether an observed association is a causal relationship and to help in making decisions based on epidemiologic evidence (Hill, 1965; Phillips and Goodman, 2004; Rothman et al., 2008). All factors suspected to be related to the factor of interest need to be measured accurately and in unbiased samples of the target population. Statistical modeling techniques can then control for their influence on the outcome (confounding) or detect associations that are different in subgroups (effect modification or interaction). In studies that do not show an effect, it is often not clear whether the effect is null or simply smaller than can be detected by the studies (insufficient power). Combining data from different studies will increase the power and estimate dose–response functions with more precision. However, gathering the appropriate dose–response data will not always be possible. An example of successful use of epidemiologic studies is the case of aflatoxins and liver cancer (van den Brandt et al., 2002). Animal models had been found to be inappropriate, showing a large degree of variability in the carcinogenic potential of aflatoxins across species. From human studies looking into Hepatitis B and liver cancer, which included aflatoxin intake, the aflatoxin potency to induce liver cancer (cancers/ year/100.000 people/ng aflatoxin/kg body weight/day) could be calculated. Another example, in which dose–response relations have been estimated from human observational studies, is the relation between methyl mercury in fish and cognitive development, as recently summarized by the FAO/WHO expert consultation (FAO/WHO, 2010).

Proposed frameworks for use of epidemiologic data in risk assessment

The need for a more structured use and presentation of epidemiologic data is risk assessment is long recognized (Hertz-Picciotto, 1995). Hertz-Picciotto (1995) has proposed a classification framework for the use of individual epidemiologic studies in dietary quantitative risk assessment, with special focus on dose– response assessment, and this has been modified by van den Brandt et al. (2002). They identified three classification categories for epidemiologic studies: (1) the study can be used to derive a dose–response relationship; (2) the study can be used to check on plausibility of an animal-based risk assessment; (3) the study can contribute to the weight-of-evidence determination of whether the agent is a health hazard. For the classification of a study into the three categories a number of criteria are used addressing its validity and utility. Recently, a three tier framework has been proposed through which human observational studies can be selected for use in quantitative risk assessment, with specific focus on exposure assessment (Vlaanderen et al., 2008). General recommendations for reporting of observational studies have been provided by STROBE, which stands for Strengthening the Reporting of Observational Studies in Epidemiology (http:// www.strobe-statement.org/ (Vandenbroucke et al., 2007)).

Scope of risk assessment

In contrast to benefit assessment, the scope of risk assessment is fairly well established (see Section 4.4). It deals with the assessment of adverse health effects caused by physical or chemical agents. These can be occurring naturally in foods, resulting from food preparation or manufacturing processes or environmental contaminants. At the moment there is further development towards a better understanding of the nature and mechanisms of toxic effects caused by these agents, the inclusion of data other than the traditional toxicological data (Eisenbrand et al., 2002; van den Brandt et al., 2002), and consideration of the shape of the dose–response relationships (linear, threshold, J-shape, Ushape) (Calabrese et al., 2007; Zapponi and Marcello, 2006).

Benefit assessment

Benefit assessment is not an established field. Although beneficial effects conferred by food have been the topic of investigation for a long time (for example within nutritional epidemiology, see Section 3.2), the structured characterization of benefits has only recently begun to develop. Benefit assessment has recently gained more attention because of the developments in the form of health claims on foods. In December 2006, the European Union published its Regulation 1924/2006 on nutrition and health claims made on foods, which, several decades after risk assessment, places a legal perspective also on benefit assessment. Below we will address some concepts and approaches in benefit assessment.

Basic principles and concepts

In the past decades, the goal of ‘‘adequate’’ nutrition, as defined by the absence of nutritional deficiencies, has shifted to the ambitious goal of ‘‘optimal nutrition’’. This change goes beyond the concept of ‘risk reduction’ and does not only apply to traditional foods and food components, but has also led to the concept of ‘‘functional foods’’. Food is seen as a possibility to improve health. Taking beneficial effects into account, may also uncover a discrepancy. The RDA (see Section 3.1.4 micronutrients) allows a small percentage of the population to be at inadequate intake for the specified indicator of adequacy; however, if the intake needed to gain a theoretical benefit (as indicated by the dashed line in Fig. 1b) is higher than this level, then a substantial part of the population may not have the advantage of this theoretical benefit. In the benefit assessment context, epidemiologic studies can be both observational and experimental, but experimental (‘intervention’) studies are preferred.

Study design and strength of evidence

As described in Section 3.2 both observational and experimental methodologies can be used to assess risks and benefits. From observational studies, there is ample experience in describing associations between human exposure and health outcomes. The level of evidence they provide is most commonly assessed by using

• the WCRF/AICR criteria, resulting in the judgment ‘‘convincing’’,

‘‘probable’’, ‘‘limited-suggestive’’, ‘‘limited-no conclusion’’ or ‘‘substantial effect on risk unlikely’’ (WCRF/AICR, 2007); or

• the WHO criteria, resulting in the judgment ‘‘convincing’’,

‘‘probable’’, ‘‘possible’’ or ‘‘insufficient’’ (WHO, 2003).

For ethical reasons, human experiments (or ‘‘intervention studies’’) are generally pre-eminently carried out to study beneficial and not adverse health effects, once observational data have shown that the benefit is reasonably probable and/or toxicological screening has given reasonable proof of safety. However, it is not uncommon that these trials are short-term and limited to only one or two doses, and these doses are generally rather high to yield the best chance of ‘success’ (van den Brandt et al., 2002). This hampers the construction of dose–response curves that are relevant for long-term human exposure. Short-term trials may not always be appropriate to characterize effects of food and nutrition, because after long-term exposure response mechanisms may adapt to a different state (tolerance) or wear out (threshold). Guidelines for the design and reporting of human intervention studies and for a standardized approach to prove the efficacy of foods and food constituents are being developed (ILSI, 2010). The created knowledge and methodology can be applied in a broader context, and this includes answering a benefit–risk question.

Proposed framework

There is agreement in the field to consider benefits according to standardized assessment procedures and specific definitions (Palou et al., 2009). Benefits should be scientifically proven. It is advocated to structure benefit assessment in the same framework as risk assessment; For example, the EU advisory working group on harmonization of risk assessment procedures identifies a 4-stage approach consisting of

• ‘value identification’,
• ‘value characterization’,
• ‘use assessment’, and
• ‘benefit characterization’(EC, 2000b).

EFSA uses the terminology

• ‘identification of positive / reduced adverse health effect’,
• ‘characterization of positive / reduced adverse health effect’,
• ‘exposure assessment’, and
• ‘benefit characterization’ (EFSA, 2010b).

Examples of initiatives specifically focusing on benefit assessment, in the form of claims made by industry, are the FUFOSE and PASSCLAIM projects, coordinated by ILSI Europe (www. ilsi.org/Europe). One of the goals for the FUFOSE project was to assess critically the science base required to provide evidence that specific nutrients and food components beneficially affect target functions in the body (Bellisle and et al., 1998; Diplock et al., 1999). Its follow-up, the PASSCLAIM project, aimed to define a set of generally applicable criteria for the scientific substantiation of claims on foods (Aggett et al., 2005; Asp and Bryngelsson, 2008). The PASSCLAIM criteria cover: (1) the characterization of the food or food ingredients for the health claim, (2) the necessity of human data for the substantiation of a health claim (by preference data coming from human intervention studies but also accepting human observational data), (3) the use of valid endpoints or biomarkers as well as statistically significant and biological relevant changes therein, and (4) the requirement for the totality of the data and weighing of the evidence before making a judgment whether a health claim is or is not substantiated (Aggett et al., 2005; Asp and Bryngelsson, 2008). Addressed in the claims regulation 1924/2006 is the particular beneficial nutritional property of a food (in so-called nutrition claims) or the relationship between a food or food category and health (in so-called health claims) (Verhagen et al., 2010). Examples of the latter are the relationship between calcium and development of bones and teeth, the relationship between plant sterols and coronary heart disease through reduced blood cholesterol and the relationship between iron and cognition in children. A state of the art in nutrition and health claims in Europe is given by Verhagen et al. (2010).

Scope of benefit assessment

Claims are a driver of, but not the only, development in benefit assessment. There are other initiatives to increase benefit from food, such as voluntary food reformulation by industry (for example reducing the amount of salt, sugar or trans fatty acids) (van Raaij et al., 2009). With respect to traditional foods there has been a development in research to look at whole foods or whole diets (dietary patterns), for example the Mediterranean diet.

Benefits are often measured on a disease scale, either directly as a reduction in disease risk or indirectly as a change in disease marker. Currently, markers of optimal health are also being explored (Elliott et al., 2007). This is a fundamentally different process, as subjects are healthy individuals and effects are expected to typically be small. Two bottlenecks have been identified in the development of the newly explored biomarkers to quantify health optimization (van Ommen et al., 2009): the robustness of homeostasis and inter-individual differences in what appear to be normal values. Many markers are maintained within a limited range and effects are masked unless homeostasis is challenged. Nutritional challenge tests (e.g., a high-fat diet) are being developed in this context. These will bring imbalance to normal physiology. Quicker recovery then signifies more optimal health. A combination of perturbation of homeostasis and systems biology and the -omics technologies have been proposed as a means to quantify the system’s robustness and to measure all the relevant components that describe the system and discriminate between individuals (Elliott et al., 2007; Hesketh et al., 2006; Keijer et al., 2010; van Ommen et al., 2009). This development also signifies a shift from a population orientation to a more individual based orientation of benefit– risk analysis.

So far, benefits are based in the biological health domain. Decision- makers are bound by a limited amount of (public) money and (thus) require management options based on physiological mechanisms, not on for example individual ‘preferences’. However, certain perceptions and behaviors may well influence the quality of life in such a way that they are relevant for benefit–risk weighing (for a detailed description of consumer perception, see Ueland et al., 2011). Also, long-term benefits, in the form of ecological considerations, could be relevant. Thus, it might be reasonable to accept a broad scope of what entails a benefit.

Benefit–risk assessment

From a legal perspective, benefit assessment is about the permission to carry a claim on a product and risk assessment is about defining a maximum intake dose below which the population intake is safe (without appreciable adverse effects). From a public health perspective, benefit assessment and risk assessment are not yes/no or safe/unsafe decisions, but describe a range and continuum of doses and likelihoods of effect. Benefit–risk assessment is about evaluating the whole scope of human health and disease. It involves more than adding an assessment of benefits to an assessment of risks; the scientific processes to be integrated are not symmetric. Below, we will first briefly address the integration of the science. Next, we will discuss a number of frameworks for performing a benefit–risk analysis, the integration of outcome measures, uncertainty in the assessment and other (problematic) issues.

Integration of disciplines

Until recently, benefit assessment and risk assessment were approached as separate processes and the domain of different underlying scientific disciplines. In benefit–risk analysis, all inevitably comes together. There are differences in the process and in the output which need to be integrated. Table 2 presents some of these differences. The integration requires adaptations from all disciplines involved.

There is increasing interest to know which effects occur on the entire range of human exposure to be able to characterize benefits and risks. Toxicologists have suggested that toxicology should redirect its focus from looking at adverse effects at high levels of experimental exposure, to characterizing the complex biological effects, both adverse and beneficial, at low levels of exposure, the so-called ‘low-dose toxicology’ (Rietjens and Alink, 2006). This will help risk quantification of realistic intakes and is also interesting from the idea that some low (physiologic) doses actually turn out to be beneficial (Son et al., 2008). Epidemiologists and nutritionists have urged all who are generating human data to improve the quality of their data presentation and expand it, by presenting more detailed information focusing on dose–response necessities, and/or by sharing primary data (de Jong et al., submitted for publication). Some also stress the important role for early, quantifiable biomarkers of exposure and of effect as well as for the -omics technologies (Palou et al., 2009). This is expected to improve exposure assessments and provide early response profiles.

Ideally the (old and new) disciplines could complement each other within the benefit–risk assessment process, with their roles differing depending on e.g. the type of food (component) and its history of use, but always from the aim to end up with data most relevant to humans.

Table 2. Characteristics of (regulatory) benefit and risk assessment^a, that need to be integrated.

Characteristic	Benefit assessment	Risk assessment
Study design	Experimental (observational may also be accepted)	Experimental (observational may also be accepted)
Study population	Human	Animal (Human is possible)
Widely accepted methodology	In development	Yes
Widely accepted endpoints	In development; also measured as reduction in adverse endpoint	Yes
Exposure	Usually few (supraphysiological) doses in experimental study; physiological doses/observational design may also be accepted	Usually few (supraphysiological) doses in experimental study
Study duration	Usually short (days/weeks) in experimental design, years/decades can be covered in observational design	Usually lifetime of animal until adverse endpoint occurs, years/decades can be covered in observational design
Evidence required	Grading ‘convincing’ (sometimes ‘probable’ admitted) in weight of evidence approach	Strongest effect/steepest slope in positive evidence approach
Desired outcome in benefit–risk assessment setting	Benefit characterization, including low dose–response information	Risk characterization, including low dose–response information
Desired outcome in regulatory setting	Benefit/no benefit	Safe/unsafe or acceptable/non-acceptable

^aWith focus on chronic exposures.

Benefit–risk assessment approaches

A number of research efforts have been dedicated specifically to the development of benefit–risk methodology in the field of food and nutrition. There have been three recent Europe-based projects, each with its own focus: BRAFO (2007–2010), QALIBRA (2006– 2009) and BENERIS (2006–2009); in addition, the topic has been subject of investigation by the European Food Safety Authority (EFSA, 2007, 2010b), by national food and health related institutes (Fransen et al., 2010; van Kreijl et al., 2006) and by individual groups.

Generalities in approaches

Conceptually, those exploring the field of benefit–risk assessment in food and nutrition agree that much health gain can come from investing in a better diet, more than from further improving food safety (see Table 3). Practically, there is consensus about the importance of a well formulated and well described benefit– risk question. This prior narrative needs address the exposure, target population, health effects and scenarios. As there is more experience with risk assessment than with benefit assessment, it is generally suggested that the steps taken in benefit–risk assessment should mirror the traditional risk assessment paradigm of hazard identification, hazard characterization, exposure assessment and risk characterization, with a risk arm and a benefit arm running parallel until comparison (see Fig. 2). A fully quantitative benefit–risk analysis requires a large amount of data (see Fig. 3) and time. Assumptions about missing or incomplete data are inevitable. A tiered (stepwise) approach is advocated, allowing transparency and an early stop in the analysis in case the question is answered or in case there is substantial lack of data. In most approaches there is not only room for in-between scientific evaluation, but also for interim consultation with the decision-maker. This means that if a risk is found which is acceptable to the decision- maker, then the benefit–risk assessment may also come to a stop (i.e., a management judgement). In fact, policy makers are most often the intended user and initiator of the benefit–risk assessment. In general, a distinction can be made between a qualitative comparison and a quantitative comparison of benefits and risks. In a qualitative comparison the benefits and risks are compared in their own currency. This is sufficient when there is clear dominance of either the benefits or the risks, e.g. low gastritis incidence versus high cancer incidence, or when comparison with guidance values indicates that either the benefits or the risks are negligible and there is no true benefit–risk question left. In cases where benefits or risks do not clearly dominate, quantitative comparison may answer the benefit–risk question. Quantitative comparison of benefits and risks is seen as an (optional) end stage and can be achieved by expressing the effect of each endpoint in the same metric (e.g. quality or disability adjusted life years). In calculating these measures, the nature of the estimate may be deterministic, i.e. using point estimates, or probabilistic, i.e. using distributions. In general, some kind of screening is incorporated in the assessment to evaluate whether the benefits clearly prevail over risks or vice versa, and can remain qualitative. This may entail that exposure assessment is performed before the dose–response characterization (see Fig. 2). For transparency and best insight in who receives the benefits and risks, all separate health effects need to be communicated apart from the net health effect. In the process of developing the methodology, case studies have been performed to test and refine the approaches.

Specific approaches

Below we describe specific initiatives to approach the benefit– risk assessment field.

The ILSI Europe project BRAFO (www.brafo.org) follows up the projects FOSIE (see under ‘risk assessment’), FUFOSE and PASSCLAIM (see under ‘benefit assessment’). The primary aim of BRAFO has been to develop a framework that allows for the quantitative comparison of human health risks and benefits in relation to a wide range of foods and food compounds. The BRAFO approach consists of 4 tiers, preceded by a pre-assessment and problem formulation (Hoekstra et al., 2010). The problem is worded as a comparison in net health effect between a reference scenario and an alternative scenario. In tier 1, the individual risks and benefits are assessed. Should only risks or only benefits be identified, the assessment stops. When both risks and benefits are identified, the assessment proceeds to tier 2, which entails a qualitative integration of risks and benefits. In this, four main dimensions are considered: the incidence, the severity/magnitude of the health effects, their duration and the additional (reduction in) mortality caused by the effect and consequent (reduction in) years of life lost. Should either the risks or the benefits clearly dominate, then the assessment stops. Special table formats are suggested to present the data and information relevant in tier 1 and 2, for clarity and transparency. In tier 3, the benefits and risks are weighed quantitatively. Tier 3 is closely related to a 4th tier, which also entails a quantitative assessment. These tiers are actually a continuum, starting fully deterministically, using fixed estimates, and ending more and more probabilistically, using probability distributions. The approach has been tested and further developed with several case studies.

The tiered approach developed by the National Institute for Public Health and the Environment (RIVM) of the Netherlands also distinguishes steps separated by ‘‘stop’’-moments to allow for a timely evaluation to conclude whether the gathered information is sufficient to answer the initial benefit–risk question (Fransen et al., 2010). As compared with the BRAFO-approach, the exposure assessment has been moved upward and the dose–response modeling as part of hazard and benefit characterization is moved to a later stage. The characterization step includes an extensive literature search and modeling to estimate dose–response functions for the selected effects. The advantage of giving priority to the exposure assessment is that one gets early information of the population distribution of the exposure. In case of no or very limited exposure in risk groups, one can terminate the risk–benefit assessment at an early stage, before going to a benefit–risk characterization. The decision tree has been tested with case studies.

QALIBRA (www.qalibra.eu) has developed methods for quantitative assessment that integrate the risks and benefits of dietary change into a single measure of net health impact (See Fig. 3), and that allow quantification of associated uncertainties. It is a typical example of the 3-rd and 4-th tier in BRAFO (Hoekstra et al., 2010). QALIBRA also follows a stepwise approach, the consecutive steps consisting of: (1) Problem formulation including specification of the dietary scenarios and the population to be considered. (2) Identification of the adverse and beneficial health effects to be assessed, including influential factors and affected population. (3) Estimation of the intakes or exposures that cause those health effects. (4) Modeling of the dose–response relationship for each effect, including the probability of onset at the current age and, if the effect is continuous, the magnitude of the effect. (5) Estimation of the probabilities of recovery and mortality for affected individuals. (6) Selection of a common currency (e.g. DALY or QALY, see Section 5.3). (7) Specification of the severity and duration of the effect. (8) Calculation of the net health impact. (9) Evaluation of (quantifiable and unquantifiable) uncertainties and variabilities.

These methods are implemented in the QALIBRA tool, which is available to registered users on the QALIBRA website. The user is responsible for supplying the data such as intake scenarios, dose–response functions and if needed uncertainty intervals. The tool can be used in BRAFO’s quantitative tiers 3 and 4. QALIBRA recommends that practitioners follow the BRAFO tiered approach, resolving benefit–risk questions without quantitative integration where possible, and reserve the QALIBRA framework for those cases where assessment at tiers 3 and 4 is necessary. The tool works according to the annual directly attributable health impact method (van Kreijl et al., 2006). This approach considers a single year and only health effects that have their onset during that year. Their direct potential impact is considered, without taking into account interactions with other health effects. It is a less datademanding, but less realistic, alternative for the ‘simulation approach’, which takes into account a longer period, in which a proportion of the population will change class with respect to age, risk factor and disease status. QALIBRA advises to start with the simpler approach (addressing in a narrative how the overall impact might be affected by the way individual effects combine) and reserve the simulation of health over whole lifetimes as a higher tier option should the direct attributable health impacts method not answer the benefit–risk question sufficiently. The RIVM Chronic Disease Model is an example of a simulation approach; it has been used in (a part of) a large Dutch case study (Hoogenveen et al., 2010; van Kreijl et al., 2006).

The BENERIS (http://en.opasnet.org/w/Beneris) project has made a significant contribution to the methodological foundation of ‘open assessment’, which considers assessments as open collaborative processes of creating shared knowledge and understanding (see also Pohjola et al., this issue). The open process brings scientific experts, decision-makers, and any other stakeholders to the same collaborative process of tackling public health problems. Collaboration is facilitated by the Opasnet web-workspace (http:// en.opasnet.org), which is the major output of the project and consists of a wiki-interface, a modeling environment, and a database. In the BENERIS approach, open criticism has a crucial role. The objective in each tier is to convince a critical outsider (‘an impartial spectator’) about the current conclusions. Thus, the assessment work is directed towards the aim to gain acceptability in the eyes of a critical outsider. Openness is a critical feature of an assessment to make it possible in practice for outsiders to join and criticize the current content of the assessment. Another characteristic of BENERIS is the use of ‘value-of-information’, a scientific method for estimating the expected benefit that occurs when an uncertainty is resolved, to give guidance about the critical uncertainties that should be focussed on in the next tier. The main tools are probability distributions and Bayesian nets.

EFSA has organized a colloquium on ‘risk–benefit analysis of foods: methods and approaches’ in 2006 (EFSA, 2007) and has recently published a guidance document with respect to methodology, approaches, tools and potential pitfalls in the risk–benefit assessment (EFSA, 2010b). It proposes a three step approach, after the problem formulation: (1) initial assessment, addressing the question whether the health risks far outweigh the health benefits or vice versa, (2) refined assessment, aiming at providing semiquantitative or quantitative estimates of risks and benefits at relevant exposure by using common metrics, and (3) comparison of risks and benefits using a composite metric such as DALYs or QALYs (see Section 5.3) to express the outcome of the risk benefit assessment as a single net health impact value (EFSA, 2010b).

Along a slightly different line, the ‘‘window of benefit assessment’’ has been proposed as a concept to integrate nutritional optimization and safety considerations, see Fig. 4 (Palou et al., 2009). In this concept, the threshold model and a score model are combined as a framework for weighing risks and benefits. The traditional threshold model works from the guidance values for each (non-) nutrient that must not be exceeded or that must be reached, for example the UL and RDA in the case of nutrients (see Section 3.1.4 nutrients). Two new thresholds are introduced: the upper level of additional benefit (ULAB) and the lower level of additional benefit (LLAB). These are defined as the upper and lower levels of daily nutrient intake that can provide detectable additional benefits, apart from benefits derived from avoiding excessive intake or deficiency. Intake, then, should be within UL and RDA (absence of excess or shortage), but preferably within ULAB and LLAB (improved health). LLAB and ULAB will not be necessarily within the window defined by the RDA and UL; it is possible, for example, that the UL is based on a small and mildly adverse effect but that a larger beneficial effect occurs at a higher dose. Thus, type and severity of effect are relevant. It is postulated that, based on the robustness of homeostasis, LLAB and ULAB will delimit the optimal intake, that is needed when the biological system is challenged, while RDA and UL indicate levels of intake that are not permissible on a population- wide scale (Palou et al., 2009). It is proposed to score nutritionally beneficial and adverse characteristics of foods and combine them into a global score (Palou et al., 2009). The use of a score model allows the weighing of risks and benefits. A tool like this is still under development, for example by the PASSCLAIM project (see Section 4). The window of benefit is targeted to the needs of specific population subgroups (Palou et al., 2009). The same intake of a nutrient leads to an individually different windows of benefit depending on age, sex, physiological or pathological conditions, individual genetic constitution and lifestyle history (Hesketh et al., 2006; Palou et al., 2009).

Similarly, Verkerk describes a conceptual model which shows how beneficial effects may arise above the UL and that benefit–risk evaluation is required across a ‘zone of overlap’ (Verkerk, 2010; Verkerk and Hickey, 2010). Significant benefits occurring at intake levels above the UL are postulated to be the norm rather than the exception (Verkerk, 2010). It is argued that current management by regulatory prohibition and using a single most sensitive risk factor in the most sensitive population denies benefits to large parts of the population (Verkerk, 2010; Verkerk and Hickey, 2010). Also, it is argued that different management strategies are required for different molecular forms of the micronutrients, for example in the case of folates, carotenoids and niacin, as adverse effects differ between them (Verkerk, 2010; Verkerk and Hickey, 2010).

Renwick et al. (2004) have been one of the first to recognize that the difference in approach between establishing an RDA and a UL means that any benefit–risk analysis is based on different underlying criteria and approaches. They have provided a basis of how different aspects of the intake–incidence data for either the benefit or toxicity can be taken into account by suitable modeling, and how the output of the model can be used to provide usable advice for risk managers.

Integrated measures

One important question in weighing benefits and risks is which unit of measurement to use. A distinction can be made between single outcome health metrics and ‘‘composite’’ or ‘‘integrated’’ health measures. Single outcome health measures reflect a single measurement dimension and unit, for example mortality, disease incidence or functioning (quality of life). Composite health measures capture a combination of single outcome health metrics into one currency using a valuation function to weight the different health states or outcomes. It then becomes possible to unite mortality and morbidity and also to weigh different health states or different outcomes such as a reduced incidence of angina against an increased breast cancer incidence. Currently, the most used integrated health measures in food-related benefit–risk analysis are the DALY (disability adjusted life years) and the QALY (quality adjusted life years), measures which use the value, or preference, that people have for health outcomes, along a continuum between 0 and 1 (Gold et al., 2002).

In the QALY concept, 1 represents full health and 0 represents death, see Fig. 5a. To generate QALY values, (often theoretical) health states reflecting physical, social and emotional well-being (not associated with a particular disease or particular condition) are valued by members of the general public. The values (also termed ‘‘utilities’’) are derived either directly, using methods such as standard gamble, time trade-off and visual analog scales (Gold et al., 2002); or indirectly, using generic quality of life/health status questionnaires, in which some health states can be used to estimate the value of (a large number of) other health states, varying in sensitivity, administrative burden, cost, bias, etc. QALY’s are traditionally mostly used to measure health gains at micro scale, for example to compare two interventions. QALY’s of all individuals are summed, and the scenario with the highest number of QALY’s represents the highest health maximization. Per individual, QALY’s are calculated by multiplying the duration of the disease (for example the last 17 years of one’s life) by a quality weight (for example quality is reduced from 1 to 0.7) and adding this to the age of onset of disease (for example age 45), in other words: the life years reached at age of death (for example age 62 years) or at an earlier reference point are adjusted for their lack of quality and result generally in a number which is lower than the true age of death (for example age 57 years) or age at an earlier reference point, see Fig. 5a. When comparing two treatment groups or alternative scenarios, the average areas under the curve are compared and this difference is the number of QALY’s that the ‘‘better’’ intervention or scenario may gain over the ‘‘worse’’ intervention or scenario.

In the DALY context, 0 represents no disability and 1 represents death (Gold et al., 2002), see Fig. 5b. DALYs tend to be based on a universal set of standard weights covering specific diseases, based on expert judgments using ‘trade-off’ methods, the most wellknown being those defined by Murray (Murray, 1994). For many diseases, disability weights, which are currently being revised in the Global Burden of Disease 2010 study, are available at the WHO website (WHO, 2004). For example, the disability weight for uncomplicated diabetes mellitus is 0.015, for angina pectoris 0.124, for cancer in the metastatic phase 0.75 and first ever stroke 0.92. Some countries have established national weighing factors (Schwarzinger et al., 2003). DALY’s are based on disease years, but for short-term discomforts, such as (minor) food poisoning of influenza, there is also a solution (though the weight itself can be hard to find); the condition can either be weighed as a full year and then corrected to the actual duration or weighing can be adapted to represent a year which includes an episode of disease (the ‘year profile approach’) (van Kreijl et al., 2006). DALYs were introduced to primarily communicate a population-aggregate measure of (loss of) health – the Burden of Disease used by WHO. DALY’s of all individuals are summed; the scenario with the lowest number of DALY’s represents the highest health maximization or lowest health loss. One DALY represents the loss of the equivalent of one year of full health. It counts down from a standard ‘‘ideal’’ age of life expectancy at birth (for all population subgroups, but separate for men and women). From this reference point (for example 82 years for women, see Fig. 5b), years spent with a disease (for example 17) are multiplied by a disability weight (for example 0.3, where 0 is healthy) and added to the years that death (for example at age 62) prevented one from reaching full life expectancy. In other words, the years of life lost (YLL, in this example 20 years) and the years lived with a disease (YLD, which count more with higher disease burden) are added and thus result in a number which is higher than the true years lost (in this example 25 years). When comparing two intervention groups or alternative scenarios, the average areas above the curve are compared and this difference is the number of DALY’s that the ‘‘worse’’ intervention or scenario loses over the ‘‘better’’ intervention or scenario.

Often, the choice for DALY or QALY is a pragmatic one, based on data availability or experience of use rather than on a fundamental choice. Ethical and equity issues are not accounted for in QALY’s and DALY’s. For example, when calculating QALY’s, the elderly can be at a disadvantage, because they by definition have fewer life years left and thus have less opportunity to increase quality adjusted life years. Also, benefits and risks incorporated in one integrated measure may occur in different population subgroups. Folic acid is a good example, where prevention of neural tube defects and masking of B12 deficiency or possibly increased cancer risk apply to different subgroups. Decision-makers need to be aware of this and thus it is important to always accompany the DALY or QALY with the distribution of separate risks and benefits in subgroups.

In the area of benefits and risks of fish consumption some less ‘intensive’ integrating measures have been explored. Gochfeld and Burger (2005), have developed a composite benefit and risk by dose curve to identify a zone of benefit, above the benefit threshold and below the harm threshold. Separate dose–response curves for benefit and harm are combined into net benefit–harm composites. Foran et al. (2005) have calculated the Benefit Cancer Risk Ratio (BCRR) and the Benefit Noncancer Risk Ratio (BNRR). These ratios represent the rates of consumption of n-3 fatty acids (g/d) while controlling for the cumulative level of acceptable carcinogenic or noncarcinogenic risk of contaminants in fish. The risks are assessed via the increased probability of death from cancer (compared to a target risk of 1 in 10�5 and using a substance-specific cancer slope factor) and the RfD (compared to a target noncancer risk of 1), respectively. Their population values represent the long run trade-off of benefits and risks over repeated consumption. If the distribution of the estimates is not expected to be normal, upper and lower 95% CI’s can be constructed by bootstrap methods. Foran et al. have calculated the BCRR and BNRR to compare wholesale farmed, retail farmed and wild salmon. Along a somewhat similar line, Ginsberg and Toal (2009) have developed a risk–benefit equation, calculated separately for each endpoint, using dose– response functions and exposure information per fish meal. The increased risk (for example on coronary heart disease, through methylmercury) is then subtracted from the decreased risk/benefit (on coronary heart disease, through n-3 fatty acids). Different fish species can be compared, values >1 indicate a net benefit and values <1 indicate an increased risk.

Dealing with uncertainties

In general, those involved in the benefit–risk process stress the need to (at least) document and (if possible) assess uncertainties, in every tier or stage of the analysis. A generally applicable approach of evaluating uncertainties is a ‘Monte Carlo simulation’. In this approach each uncertain variable is represented by a probability distribution. Then multiple (plausible) values are generated by repeated sampling values from the probability distributions for each uncertain variable, and these values are used for calculating the parameter of interest. The variation in the resulting outcomes then represents the overall uncertainty for the parameter of interest. In situations where uncertainty can not or hardly be described in a probability function, for instance when no data are available and experts find it difficult to make an estimate, the uncertainty will be explored with sensitivity analyses ‘‘what-if’’ scenarios). Here, the uncertain parameter is simply varied within a range of most likely values. The models will then be run using different values of the uncertain parameter. The baseline benefit or risk model with default values of the parameters will then be compared to the alternative models with different parameter values for the uncertain variables. Furthermore, a sensitivity analysis allows one to investigate which uncertainties and assumption are important for the final benefit–risk result and which uncertainties can be ignored (Saltelli and Annoni, 2010). It should be noted that the concept of uncertainty needs to be discerned from the concept of variability (van der Voet and Slob, 2007) and from the concept of error (Wilson and Crouch, 2001).

Issues yet to be solved

Some issues in benefit–risk assessment still need to solved and below some of them will be addressed.

Evaluation of evidence

The evaluation of evidence for the identification of beneficial and adverse health effects is asymmetrical: the presence of benefits needs to be guaranteed and the evidence for this is reached differently than for risks, for which the absence needs to be guaranteed (Fig. 6). Information sources may differ, as some of the information, mostly on the risk side, may be found in the so-called ‘‘gray literature’’ (for example government reports), which likely presents good quality research but which is not as easily accessible as the common scientific literature. The WCRF/WHO strength of evidence grading (addressed under Section 4) is useful to identify large effects, or small effects that have been investigated in many studies. The same problem arises for adverse or beneficial effects for which sufficient evidence exist but for which a decision needs to be taken whether it is an adverse or beneficial effect of sufficient magnitude. A well-known example is flushing with niacin excess. The grading criteria also present some difficulties in case of new effects or in cases where small effects are suspected but not sufficiently proven on a broad range of outcomes (e.g. with consumption of sugar). Subtle changes, especially for substances affecting human behaviour, are difficult to identify. A large number of possible effects may in reality have a more detrimental or beneficial effect on public health than one convincing effect and ways need to be found to take this into account to avoid a suboptimal decision. Also, they by definition do not apply to animal data. In this context, animal studies can go into category 3 of the classification categories for epidemiologic studies mentioned under Section 4.3 (Hertz-Picciotto, 1995; van den Brandt et al., 2002), i.e. the study can contribute to the weight-of-evidence determination of whether the agent has an effect on health.

Animal to human translation

The cut-off values ADI, TDI, UL or RDA provide a safety assessment. A benefit–risk assessment in which the 95th percentile of the population distribution has an estimated intake of e.g. 6.5 mg/kg, where the ADI is 7, has to conclude that there is no appreciable risk. Yet this is subject to different uncertainties from when a human study on benefits concludes there is no benefit. Guideline values can be used in a (first) qualitative screening whether (substantial) risks are to be expected or not, from where it can be decided whether a quantitative benefit– risk assessment is needed. However, when the ADI is close or exceeded, it is currently very difficult to quantify what the risks are to humans. In some cases, for example for certain additives, the management action at this point will be to lower the amount that is allowed in food, thereby reducing the population intake levels to below the ADI. In other, less avoidable cases, risk quantification and integration into a common currency is desired, but conversion of the measured pathological effect in an experimental animal to a health state for which a quality of life weight exists or to a clinical endpoint for which disease weights exist is not straightforward; this is especially true when an endpoint is measured that has no direct clinical endpoint, e.g. involuntary weight loss. Another problem is that in animal experiments, age of onset and age at death are seldom measured, whereas in the directly attributable health loss calculation the probability of onset per year for specified ages or age groups is required. It will then be necessary to convert lifetime probabilities to annual probabilities, which will often vary with age. But whether age of onset has been measured or not, it would still be difficult to translate this to the human situation. When the appropriate treatment for age of onset is very uncertain, sensitivity analysis should be performed to explore the impact of the different assumptions.

Integrated measures

Apart from animal-related issues, not for all related health outcomes (diseases affected) DALY weights are available or elicited using similar methodology. Also, disability states in DALYs do not take account of coexisting diseases. The weights are culture/country specific (as are intake and disease occurrence). Because the QALY is grounded in domains of health rather than descriptions of specific diseases, it is at least theoretically possible to describe and therefore value combinations of illness. When using QALYs, health status profiles of those affected by the diseases under study need to be accessible. QALY values vary with the actual questionnaire used, often depend on framing of the questions, are not very sensitive and differ between general practitioner, patient, family or general public. Thus, the data requirements for calculation into the common currency are high and there are many stumbling blocks.

Exposure assessment

Another issue is that food composition tables do not routinely contain data on contaminants, additives or specific fractions such as added sugars. To obtain this information, the target food should be sampled and analytically analyzed or in some cases concentration data are collected in special databases. The resulting values should then be added to food consumption data manually, on a one-on-one basis or by subgroups. In the interpretation of the results it is important that there is also an idea about the background exposure. But even for nutrients, for which much information is available, estimation of intake is difficult (Kroes et al., 2002; Willett, 1998). Development of biomarkers may help here (Jenab et al., 2009), but need to be applicable on a large scale to be really useful.

Lack of data

Lack of data is a common reason for authors initially aiming to perform benefit–risk analysis, to quit or restrict their analysis. For example, in the UK, an expert group of nutritionists and toxicologists was formed to review the benefits and risks associated with fish consumption. They concluded that a ‘‘formal risk–benefit analysis was not possible because the nutritional data were not sufficiently quantitative’’ (Maycock and Benford, 2007). EFSA expressed a problem with the data on one of the disease outcomes: ‘‘There is currently insufficient data to allow full quantitative risk assessment of folic acid and cancer’’ (EFSA, 2009a). Recently it was argued that ‘‘quantitative risk–benefit analyses of cardiovascular effects of consuming specific fish species, based on joint contents of fatty acids, methyl mercury, and selenium, cannot currently be performed until the cardiovascular effects of methyl mercury and selenium are established’’ (Park and Mozaffarian, 2010). The lack of data problem can sometimes be circumvented by adjusting the initial problem definition (and accepting that data availability influences the problem definition) or by using best case/worst case scenarios. In general, lack of data introduces uncertainties that depending on the problem formulation may result in the inability to provide a useful assessment.

Alternative actions and scenarios

The scenarios to be compared are usually focused on the food (component) of interest; the effects of substitution are usually not taken into account. It is conceivable that for example the benefits of eating fish will be different if not-eating fish in a western diet entails eating more processed meat or if fish is replaced by e.g. soy and walnuts. It is important to take other aspects of the diet into account, even if only in a narrative, as there are many possible alternatives which cannot all be quantified. Also, analyses and resulting consumer recommendations can be refined. Fish, for example, can be differentiated into types with general higher or lower contamination levels (Ginsberg and Toal, 2009; Loring et al., 2010).

Benefit–risk assessment in food and nutrition: case studies

So far, several dozens of case studies have been performed to test the methodological developments and expose their possible shortcomings (‘‘learning by doing’’). These (completed and ongoing) case studies, summarized in Table 4, deal with foods that may confer both risks and benefits (fish, human milk, vegetables, soy, whole grain cereals), with food components that may confer both risks and benefits (folic acid, other micronutrients, phytosterols), with substitution scenarios (added sugar/artificial sweeteners, SAFA/MUFA, SAFA/carbohydrates) and with effects of alternative food processing strategies (acrylamide, benzo(a)pyrene/HAAs). It can be seen from Table 4 that the most investigated topic is fish consumption. Also, many of the case studies following the newly developed tiered approaches are yet to be completed. Microbiology-related studies are addressed in Magnússon et al. (this issue). It is worth mentioning that the concept of weighing the pros and cons of foods has also been mentioned in relation to additional topics, such as coffee consumption (Guenther et al., 2007) and dairy consumption in specific subgroups such as vegetarians (Lanou, 2009; Weaver, 2009) or the elderly (van Staveren et al., 2008). Recently, a plea was made for benefit–risk analysis of the use of conjugated linoleic acid supplements in cancer patients (Rastmanesh, 2010). The importance of considering both risks and benefits is increasingly recognized.

In the Netherlands, the National Institute for Public Health and the Environment (RIVM) has performed analyses of what can be seen as an early and broad case study with regard to healthy diet and food safety as a whole in the Dutch situation (van Kreijl et al., 2006). It showed that the benefits of several foods outweigh their risk. Benefit–risk analyses were performed for fruits and vegetables, wholegrain products, fish and human milk (breastfeeding) see Table 4. The associated risks are mostly 100 times lower than, and sometimes negligible compared to, the benefits. Taking this to a broader picture, in terms of potential health gain, the report has made clear that the benefits of food are not fully utilized. The annual health gain to be achieved through improvements to the diet is many times greater than that possible through further improving food safety, see Table 3. The assessment includes a scenario comparison between actual consumption and recommended intake and takes into account the major chronic diseases diabetes, cardiovascular diseases and cancer. This was placed in a food safety perspective. Not including overweight as an outcome, the number of healthy years lost (DALY’s) from the actual diet composition (taking into account saturated fatty acids, trans fatty acids, fish, fruit and vegetables) was 40–100 times greater than those lost for food safety issues (van Kreijl et al., 2006). This reflects a maximum theoretical health gain, because the whole population adhering to the dietary recommendations is not realistic. Including the food contribution to overweight in the comparison, the health loss expressed in DALY’s was 60–140 times greater.

The obesity epidemic could also be seen as a case in the field of food and nutrition. The benefits of calories are obvious, but so are the risks of ‘too much’; an optimal intake range associated with optimal health exists.

Benefit–risk management of food and nutrition

Benefit–risk management entails the process of weighing policy alternatives in light of the result of a benefit–risk assessment and other relevant evaluation.

Currently, the interaction between risk assessment and risk management on several occasions during the assessment process is generally advocated and incorporated in the benefit–risk frameworks (EFSA, 2010b; Fransen et al., 2010; Hoekstra et al., 2010). Active interaction is necessary to ensure that the assessment will meet the needs and answer the concerns of the risk manager, and the analysis is seen as an iterative process (FAO/WHO, 1997). As a principle, however, the appropriate functional separation, i.e. different teams and preferably different agencies, is also advocated (FAO/WHO, 1997; Palou et al., 2009). Separation is important to ensure and show independency and objectivity of the scientific process of risk assessment (FAO/WHO, 1997; Palou et al., 2009). This is thought to enhance consumer trust in the system (Palou et al., 2009). In the European Union, these ideas and a number of food crises have resulted in the founding of EFSA (www.efsa.europa. eu), the independent risk assessment body that provides scientific opinions and advice as a foundation for European risk management decisions taken by the European Commission, the European Parliament and European Union Member States. Separation is, however, thought not to be sufficient to create trust; transparency of the scientific advice is thought to be vital or and a key issue in many approaches (EU, 2000; Frewer and Salter, 2007; Jensen and Sandoe, 2002).

Historically, the approach of regulatory agencies has been to work from the food safety side and control the health risk of chemicals to zero or negligible levels. There has been ample debate over the degree to which this so-called precautionary principle should be adhered to (Post, 2006). In more recent years, it has been advocated that there are probably many other food-related health risks that need to be preferentially addressed rather than keeping minor toxicants to zero or negligible level. An ad libitum food intake, for example, has considerably more impact on health risk (Kasamatsu and Kohda, 2006; van Kreijl et al., 2006). Also, it is increasingly recognized that accepting some risk in order to achieve more benefits is as much a risk management decision as is focusing on food safety and disregarding benefits. The food safety system controlling contaminants in food has been very successful in recent decades, but needs to continue to adapt to new scientific insights and developments in food supply and food consumption (van Kreijl et al., 2006), as should management.

In the end, the decision-maker decides what degree of risk is acceptable. The benefit–risk assessment is placed in a larger context; it is weighed against issues such as cost-efficiency and equity (Anand and Hanson, 1998).

The benefit–risk assessor must present his/her results in a form which is useful to the decision- maker, who acts on behalf of the public, so that some consideration of public perception is essential (Jensen and Sandoe, 2002); if a (benefit-)risk analysis cannot be understood, it cannot be used (Wilson and Crouch, 2001).

Benefit–risk communication of food and nutrition

Benefit–risk communication entails the interactive exchange of information and science-based opinions concerning the comparison of benefits and risks among assessors, managers, consumers and other actual or potential stakeholders.

Communication on risks and benefits of food consumption is not by default an integrated process. However, the importance to provide consumers with information about both the risks and benefits of food consumption in integrated messages is being more and more advocated (FDA, 2009b). Yet, consumer responses to such risk-benefit communications need to be better understood. Understanding perception is important to promote better health and avoid undesirable consequences (for a more detailed discussion on consumer perception, see Ueland et al., 2011).

Many general psychological mechanisms should be taken into account when presenting the assessment (interim-) results. Important factors which influence the perception of the ‘objective’ assessment data are effect catastrophality, controllability, voluntariness, reversibility, information framing, individual’s state of mind and knowledge (understanding, familiarity) and trust (van Kreijl et al., 2006). It has long been recognised that in general, the acceptability of risk is much lower than that of benefits (Starr, 1969). This is related to theories which explain that people tend to value their own property higher (‘endowment effect’) than that of others and that they have higher value for what they lose than what they may gain (‘loss aversion’). A consequence of this is the preference for risk-reduction instead of (health) gain. Beneficial effects as expressed in claims are required to be understood by the average consumer. Early research, however, shows that consumers have trouble differentiating between graded levels of evidence and make little or no distinction between nutrition and health claims (Verhagen et al., 2010). Furthermore, in people’s minds risk and benefit are often inversely correlated. To what degree depends not only on knowledge (the ‘analytic’ system) but also how they feel affectively (the ‘experiental’ system) about those risks and benefits (Alhakami and Slovic, 1994). This so-called ‘‘affect-heuristic’’ also influences the processing of numerical information and as such also influences the interpretation of dose-response information (Peters et al., 2007). Difficulty of working with probabilities, especially dose-response insensitivity, occurs most when the outcomes are highly emotional. Furthermore, there is widespread underestimation of the role of the unconsciousness in decisionmaking and behaviour (Custers and Aarts, 2010; Dijksterhuis et al., 2005; Slovic et al., 2004).

So far, benefit–risk communication case studies have focussed on fish consumption. In a study on consumption of sports-caught fish in the US, respondents appeared to be influenced more strongly by risk–risk comparisons (i.e., risks from other foods vs. risks from fish) than by risk–benefit comparisons (i.e., risks from fish vs. benefits from fish) (Knuth et al., 2003). The authors conclude that risk assessors and risk communicators should improve efforts to include risk–risk and risk–benefit comparisons in communication efforts, and to clarify to whom the health risks and benefits from fish consumption may apply (Knuth et al., 2003). In a Belgian study, comparing risk-only, benefit-only, benefit–risk and risk–benefit communication strategies with respect to fish consumption, it was found that the latter two did not yield a significant change in behavioural intention, though they did negatively affect attribute perception (Verbeke et al., 2008). Within the QALIBRA framework, consumer perceptions of the adequacy of the current information provided about health risks and benefits associated with food consumption have been explored, and consumer perceptions and responses to several integrated health metrics have been examined. Furthermore, preferences of assessors and managers for the communication of benefit-risk assessment outputs have been examined. Results from these studies are now being published (van Dijk et al., 2011).

Conclusions and recommendations

The concept of benefit–risk analysis stems from the recognition that food has both benefits and risks, that there are situations in which the benefits outweigh the risks, and that focus on food safety alone does not do justice to a well informed public health system, where much health can be gained by prevention of disease occurrence through better food and nutrition.

Both in concept and practice, benefit–risk assessment entails more than adding a benefit assessment to a risk assessment and requires adaptations in both benefit and risk assessment to provide a weighed estimate. On the risk side, it is recognized that dose–response relationships directly relevant to humans are needed. Thus, attention needs to be redirected from the description of safe levels to the description of dose–response functions covering also the lower levels. This means adaptation of design and use of modeling techniques in animal studies as well as adaptation of design and better data presentation in human studies. With regard to the establishment and scope of benefits, more experience is needed. There is agreement, however, that the relation between a food (component) and its benefit(s) needs to be scientifically characterized. Along with the movement towards considering both benefits and risks, there is also a movement away from assessing solely the needs of the most susceptible individuals and applying the results to the whole population, towards considering more the individual differences.

In addition to integrating risk and benefit assessment approaches, there also needs to be special attention for the way the outcome of this process is expressed. The possibilities and limitations of integrated measures need to be further examined and brought under the attention of a wider scale of users. A benefit– risk assessment conclusion should not merely state that ‘the benefits outweigh the risks’, or vice versa, for average or susceptible persons, but it should designate continued ways to maximize benefits while minimizing risks. Public health (promotion) is not a status quo but a dynamic process.

Even with developing benefit assessment, risk assessment and benefit–risk assessment fields, a truly quantitative benefit–risk assessment will often not be possible because of lack of data. If quantification is possible, it should not yield a false sense of security or insecurity and thus it is recommended that a benefit–risk conclusion be always accompanied by a narrative of the major uncertainties and assumptions in the assessment. Furthermore, the benefit–risk assessment should be contextualised by issues outside the assessment (such as equity).

Overall, benefit–risk assessment, provided it is carefully explained and interpreted, is a valuable approach to systematically show the current knowledge and its gaps and to transparently provide the best possible science-based answer to complicated questions with large potential impact on public health.

Conflict of Interest

The authors declare that there are no conflicts of interest.

Acknowledgements

The preparation of this manuscript was funded through Safefoodera project BEPRARIBEAN (project ID 08192) by the Dutch Food and Consumer Product Safety Authority (VWA), the Research Council of Norway (RCN) and the Nordic Council of Ministers (NCM) and supported by Matís, The National Institute for Health and Welfare (THL), the University of Ulster and the National Institute for Public Health and the Environment (RIVM).

References

Aggett, P.J., Antoine, J.M., Asp, N.G., Bellisle, F., Contor, L., Cummings, J.H., Howlett, J., Muller, D.J., Persin, C., Pijls, L.T., Rechkemmer, G., Tuijtelaars, S., Verhagen, H., 2005. PASSCLAIM: consensus on criteria. Eur. J. Nutr. 44 (Suppl. 1), i5–i30.

Alhakami, A.S., Slovic, P., 1994. A psychological study of the inverse relationship between perceived risk and perceived benefit. Risk Analysis 14, 1085–1096.

Anand, S., Hanson, K., 1998. DALYs: efficiency versus equity. World Dev. 26, 307– 310.

Asp, N.G., Bryngelsson, S., 2008. Health claims in Europe: new legislation and PASSCLAIM for substantiation. J. Nutr. 138, 1210S–1215S.

ATBC-Study-Group, 1994. The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. The alpha-tocopherol, beta carotene cancer prevention study group. N. Engl. J. Med. 330, 1029–1035.

Barlow, S.M., Greig, J.B., Bridges, J.W., Carere, A., Carpy, A.J., Galli, C.L., Kleiner, J., Knudsen, I., Koeter, H.B., Levy, L.S., Madsen, C., Mayer, S., Narbonne, J.F., Pfannkuch, F., Prodanchuk, M.G., Smith, M.R., Steinberg, P., 2002. Hazard identification by methods of animal-based toxicology. Food Chem. Toxicol. 40, 145–191.

Becker, W., Darnerud, P.O., Petersson-Grawé, K., 2007. Risks and Benefits of Fish Consumption. A Risk–Benefit Analysis Based on the Occurrence of Dioxin/PCB, Methyl Mercury, n-3 Fatty Acids and Vitamin D in Fish. National Food Administration, Livsmedelsverket, Upsala, Sweden.

Bellisle, F. et al., 1998. Functional food science in Europe – Theme papers. Br. J. Nutr. 80, S1–193.

Benford, D., Bolger, P.M., Carthew, P., Coulet, M., DiNovi, M., Leblanc, J.C., Renwick, A.G., Setzer, W., Schlatter, J., Smith, B., Slob, W., Williams, G., Wildemann, T., 2010. Application of the Margin of Exposure (MOE) approach to substances in food that are genotoxic and carcinogenic. Food Chem. Toxicol. 48 (Suppl. 1), S2– S24.

Boobis, A.R., Daston, G.P., Preston, R.J., Olin, S.S., 2009. Application of key events analysis to chemical carcinogens and noncarcinogens. Crit. Rev. Food Sci. Nutr. 49, 690–707.

Bottex, B., Dorne, J.L., Carlander, D., Benford, D.J., Przyrembel, H., Heppner, C., Kleiner, J., Cockburn, A., 2008. Risk–benefit health assessment of food – Food fortification and nitrate in vegetables. Trends Food Sci. Technol. 19, S113–S119.

Büchner, F.L., Hoekstra, J., van Rossum, C.T.M., 2007. Health gain and economic evaluation of breastfeeding policies. Model simulation. National Institute for Public Health and the Environment (RIVM), Bilthoven, pp. 113.

Calabrese, E.J., Bachmann, K.A., Bailer, A.J., Bolger, P.M., Borak, J., Cai, L., Cedergreen, N., Cherian, M.G., Chiueh, C.C., Clarkson, T.W., Cook, R.R., Diamond, D.M., Doolittle, D.J., Dorato, M.A., Duke, S.O., Feinendegen, L., Gardner, D.E., Hart, R.W., Hastings, K.L., Hayes, A.W., Hoffmann, G.R., Ives, J.A., Jaworowski, Z., Johnson, T.E., Jonas, W.B., Kaminski, N.E., Keller, J.G., Klaunig, J.E., Knudsen, T.B., Kozumbo, W.J., Lettieri, T., Liu, S.Z., Maisseu, A., Maynard, K.I., Masoro, E.J., McClellan, R.O., Mehendale, H.M., Mothersill, C., Newlin, D.B., Nigg, H.N., Oehme, F.W., Phalen, R.F., Philbert, M.A., Rattan, S.I., Riviere, J.E., Rodricks, J., Sapolsky, R.M., Scott, B.R., Seymour, C., Sinclair, D.A., Smith-Sonneborn, J., Snow, E.T., Spear, L., Stevenson, D.E., Thomas, Y., Tubiana, M., Williams, G.M., Mattson, M.P., 2007. Biological stress response terminology: integrating the concepts of adaptive response and preconditioning stress within a hormetic dose-response framework. Toxicol. Appl. Pharmacol. 222, 122–128.

Cohen, J.T., Bellinger, D.C., Connor, W.E., Kris-Etherton, P.M., Lawrence, R.S., Savitz, D.A., Shaywitz, B.A., Teutsch, S.M., Gray, G.M., 2005. A quantitative risk–benefit analysis of changes in population fish consumption. Am. J. Prev. Med. 29, 325– 334.

Combes, R.D., 1987. Brown FK and the colouring of smoked fish – A risk–benefit analysis. Food Addit. Contam. 4, 221–231.

Custers, R., Aarts, H., 2010. The unconscious will: how the pursuit of goals Operates outside of conscious awareness. Science 329, 47–50.

de Jong, N., Verkaik-Kloosterman, J., Verhagen, H., Boshuizen, H., Bokkers, B., Hoekstra, J., submitted for publication. An appeal for the presentation of detailed human derived data for dose-response calculations in nutritional science.

Dewailly, E., Rouja, P., Dallaire, R., Pereg, D., Tucker, T., Ward, J., Weber, J.P., Maguire, J.S., Julien, P., 2008. Balancing the risks and the benefits of local fish consumption in Bermuda. Food Addit. Contam. Part A Chem. Anal. Control Expo. Risk Assess. 25, 1328–1338.

Dijksterhuis, A., Smith, P.K., van Baaren, R.B., Wigboldus, D.H.J., 2005. The unconscious consumer: effects of environment on consumer behavior. J.Consum. Psychol. 15, 193–202.

Diplock, A.T., et al., 1999. Scientific Concepts of Functional Foods in Europe – Consensus Document British Journal of Nutrition 81, S1–S27.

Dorne, J.L., Renwick, A.G., 2005. The refinement of uncertainty/safety factors in risk assessment by the incorporation of data on toxicokinetic variability in humans. Toxicol. Sci. 86, 20–26.

Ebbing, M., Bonaa, K.H., Nygard, O., Arnesen, E., Ueland, P.M., Nordrehaug, J.E., Rasmussen, K., Njolstad, I., Refsum, H., Nilsen, D.W., Tverdal, A., Meyer, K., Vollset, S.E., 2009. Cancer incidence and mortality after treatment with folic acid and vitamin B12. Jama 302, 2119–2126.

EC, 2000a. Communication from the Commission on the precautionary principle. COM/2000/001.

EC, 2000b. First report on the harmonisation of risk assessment procedures. Part 1: The report of the Scientific Steering Committee’s Working Group on Harmonisation of risk assessment procedures in the scientific committees advising the European Commission in the area of human and environmental health.

Edler, L., Poirier, K., Dourson, M., Kleiner, J., Mileson, B., Nordmann, H., Renwick, A., Slob, W., Walton, K., Wurtzen, G., 2002. Mathematical modelling and quantitative methods. Food Chem. Toxicol. 40, 283–326.

EFSA, 2005. Opinion of the scientific committee on a request from EFSA related to a Harmonised approach for risk assessment of substances which are both genotoxic and carcinogenic. EFSA J. 282, 1–31.

EFSA, 2007. EFSA’s 6th Scientific Colloquium Summary Report. Risk–benefit analysis of foods. Methods and approaches. 13-14 July 2006. EFSA, Parma, Italy.

EFSA, 2008. Nitrate in vegetables. Scientific opinion of the panel on contaminants in the food chain. EFSA J. 689, 1–79.

EFSA, 2009a. ESCO report prepared by the EFSA Scientific Cooperation Working Group on analysis of risks and benefits of fortification of food with folic acid.

EFSA, 2009b. Guidance of the scientific committee on a request from EFSA on the use of the benchmark dose approach in risk assessment. EFSA J. 1150, 1–72.

EFSA, 2010a. Scientific opinion on principles for deriving and applying dietary reference values. EFSA J. 8, 1458.

EFSA, 2010b. Scientific opinion. Guidance on human health risk–benefit assessment of foods. EFSA J. 8, 1673.

Eisenbrand, G., Pool-Zobel, B., Baker, V., Balls, M., Blaauboer, B.J., Boobis, A., Carere, A., Kevekordes, S., Lhuguenot, J.C., Pieters, R., Kleiner, J., 2002. Methods of in vitro toxicology. Food Chem. Toxicol. 40, 193–236.

Elliott, R., Pico, C., Dommels, Y., Wybranska, I., Hesketh, J., Keijer, J., 2007. Nutrigenomic approaches for benefit–risk analysis of foods and food components: defining markers of health. Br. J. Nutr. 98, 1095–1100.

EPA, 1993. Reference dose (RfD): description and use in health risk assessments. Background Document 1A. United States Environmental Protection Agency.

EU, 2000. White Paper on Food Safety. Commission of the European Communities, Brussels.

EU, 2002a. Regulation (EC) No 178/2002 of the European Parliament and of the Council of 28 January 2002 laying down the general principles and requirements of food law, establishing the European Food Safety Authority and laying down procedures in matters of food safety. Chapter 1 ‘Scope and definitions’. Official Journal of the European Communities 1.2.2002.

EU, 2002b. Regulation (EC) No 178/2002 of the European Parliament and of the Council of 28 January 2002 laying down the general principles and requirements of food law, establishing the European Food Safety Authority and laying down procedures in matters of food safety. Chapter 11 ‘General Food Law’, article 6 ‘Risk analysis’ and article 14 ‘Food safety requirements’. Official Journal of the European Communities 1.2.2002.

FAO/WHO, 1997. Risk Management and Food Safety. Report of a Joint FAO/WHO Consultation. Rome, Italy, 27 to 31 January 1997. Issued by the food and agriculture organization of the United Nations in collaboration with the World Health Organization, Rome.

FAO/WHO, 2010. Joint FAO/WHO Expert Consultation on the Risks and Benefits of Fish Consumption. Food and Agricultural Organization of the United Nations and World Health Organization, Rome, Italy.

Faustman, E.M., Omenn, G.S., 2008. Risk assessment. In: Klaassen, C.D. (Ed.), Casarett and Doull’s Toxicology. The basis Science of Poisons, Seventh ed. McGraw-Hill. Medical Publishing Division, New York.

FDA, 2009a. Draft risk and benefit assessment report. Report of quantitative risk and benefit assessment of consumption of commercial fish, focussing on fetal neurodevelopmental effects (measured by verbal development in children) and on coronary heart disease and stroke in the general population.

FDA, 2009b. FDA’s strategic plan for risk communication. US Department of Health and Human Services. Food and Drug Administration.

Foran, J.A., Good, D.H., Carpenter, D.O., Hamilton, M.C., Knuth, B.A., Schwager, S.J., 2005. Quantitative analysis of the benefits and risks of consuming farmed and wild salmon. J. Nutr. 135, 2639–2643.

Fransen, H., de Jong, N., Hendriksen, M., Mengelers, M., Castenmiller, J., Hoekstra, J., van Leeuwen, R., Verhagen, H., 2010. A tiered approach for risk–benefit assessment of foods. Risk Anal. 30, 808–816.

Frewer, L.J., Salter, B., 2007. Societal Trust in Risk Analysis: Implications for the Interface of Risk Assessment and Risk Management. In: Siegrist, M., Earle, T.C., Gutscher, H. (Eds.), Trust in Cooperative Risk Management. Uncertainty and Scepticism in the Public Mind, Earthscan, London.

Ginsberg, G.L., Toal, B.F., 2009. Quantitative approach for incorporating methylmercury risks and omega-3 fatty acid benefits in developing speciesspecific fish consumption advice. Environ. Health Perspect. 117, 267–275.

Gochfeld, M., Burger, J., 2005. Good fish/bad fish: a composite benefit–risk by dose curve. Neurotoxicology 26, 511–520.

Gold, M.R., Stevenson, D., Fryback, D.G., 2002. HALYS and QALYS and DALYS, Oh My: similarities and differences in summary measures of population Health. Annu. Rev. Public Health 23, 115–134.

Guenther, H., Anklam, E., Wenzl, T., Stadler, R.H., 2007. Acrylamide in coffee: review of progress in analysis, formation and level reduction. Food Addit. Contam. 24 (Suppl. 1), 60–70.

Guevel, M.R., Sirot, V., Volatier, J.L., Leblanc, J.C., 2008. A risk–benefit analysis of French high fish consumption: a QALY approach. Risk Anal. 28, 37–48.

Hansen, J.C., Gilman, A.P., 2005. Exposure of Arctic populations to methylmercury from consumption of marine food: an updated risk–benefit assessment. Int. J. Circumpolar Health 64, 121–136.

Hendriksen, M.A., Tijhuis, M.J., Fransen, H.P., Verhagen, H., Hoekstra, J., 2010. Impact of substituting added sugar in carbonated soft drinks by intense sweeteners in young adults in the Netherlands: example of a benefit–risk approach. Eur. J. Nutr., 10.1007/s00394-010-0113-z.

Hertz-Picciotto, I., 1995. Epidemiology and quantitative risk assessment: a bridge from science to policy. Am. J. Public Health 85, 484–491.

Hesketh, J., Wybranska, I., Dommels, Y., King, M., Elliott, R., Pico, C., Keijer, J., 2006. Nutrient-gene interactions in benefit–risk anal.. Br. J. Nutr. 95, 1232– 1236.

Hill, A., 1965. The environment and disease: association or causation. Proc. R. Soc. Med. 58, 295–300.

Hoekstra, J., Verkaik-Kloosterman, J., Rompelberg, C., van Kranen, H., Zeilmaker, M., Verhagen, H., de Jong, N., 2008. Integrated risk–benefit analyses: method development with folic acid as example. Food Chem. Toxicol. 46, 893–909.

Hoekstra, J., Hart, A., Boobis, A., Claupein, E., Cockburn, A., Hunt, A., Knudsen, I., Richardson, D., Schilter, B., Schutte, K., Torgerson, P.R., Verhagen, H., Watzl, B., Chiodini, A., 2010. BRAFO tiered approach for benefit–risk assessment of foods. Food Chem. Toxicol. (epub ahead of print). <http://www.ncbi.nlm.nih.gov/pubmed/20546818>.

Hoogenveen, R.T., van Baal, P.H., Boshuizen, H.C., 2010. Chronic disease projections in heterogeneous ageing populations: approximating multi-state models of joint distributions by modelling marginal distributions. Math. Med. Biol. 27, 1– 19.

Husoy, T., Mangschou, B., Fotland, T.O., Kolset, S.O., Notvik Jakobsen, H., Tommerberg, I., Bergsten, C., Alexander, J., Frost Andersen, L., 2008. Reducing added sugar intake in Norway by replacing sugar sweetened beverages with beverages containing intense sweeteners – A risk benefit assessment. Food Chem. Toxicol. 46, 3099–3105.

ILSI, 2010. Beyond PASSCLAIM – Guidance to substantiate health claims on foods. Summary of a workshop held in december 2009 in Nice, France. ILSI Europe Functional Food Task Force.

Jenab, M., Slimani, N., Bictash, M., Ferrari, P., Bingham, S.A., 2009. Biomarkers in nutritional epidemiology: applications, needs and new horizons. Hum. Genet. 125, 507–525.

Jensen, K.K., Sandoe, P., 2002. Food safety and ethics: the interplay between science and values. J. Agric. Environ. Ethics 15, 245–253.

Julien, E., Boobis, A.R., Olin, S.S., 2009. The key events dose-response framework: a cross-disciplinary mode-of-action based approach to examining dose-response and thresholds. Crit. Rev. Food Sci. Nutr. 49, 682–689.

Kasamatsu, T., Kohda, K., 2006. Balancing risks. Regul. Toxicol. Pharmacol. 46, 100– 104.

Keijer, J., van Helden, Y.G., Bunschoten, A., van Schothorst, E.M., 2010. Transcriptome analysis in benefit-risk assessment of micronutrients and bioactive food components. Mol. Nutr. Food Res. 54, 240–248.

Knuth, B.A., Sheeshka, J., Patterson, J., 2003. Weighing health benefit and health risk information when consuming sport-caught fish. Risk Anal. 23, 1185–1197.

Kroes, R., Muller, D., Lambe, J., Lowik, M.R., van Klaveren, J., Kleiner, J., Massey, R., Mayer, S., Urieta, I., Verger, P., Visconti, A., 2002. Assessment of intake from the diet. Food Chem. Toxicol. 40, 327–385.

Lanou, A.J., 2009. Should dairy be recommended as part of a healthy vegetarian diet? Counterpoint. Am. J. Clin. Nutr. 89, 1638S–1642S.

Loring, P.A., Duffy, L.K., Murray, M.S., 2010. A risk–benefit analysis of wild fish consumption for various species in Alaska reveals shortcomings in data and monitoring needs. Sci. Total Environ. 408, 4532–4541.

Louisse, J., de Jong, E., van de Sandt, J.J., Blaauboer, B.J., Woutersen, R.A., Piersma, A.H., Rietjens, I.M., Verwei, M., 2010. The use of in vitro toxicity data and physiologically based kinetic modeling to predict dose-response curves for in vivo developmental toxicity of glycol ethers in rat and man. Toxicol. Sci. 118, 470–484. Magnússon, S.H., Gunnlaugsdóttir, H., van Loveren, H., Holm, F., Kalogeras, N., Leino, O., Luteijn, J.M., Odekerken, G., Pohjola, M.V., Tijhuis, M.J., Tuomisto, J.T., Ueland, Ø., White, B.C., Verhagen, H., 2011. State of the art in benefit-risk analysis: Food micro biology, Food and Chemical Toxicology, this issue.

Maycock, B.J., Benford, D.J., 2007. Risk assessment of dietary exposure to methylmercury in fish in the UK. Hum. Exp. Toxicol. 26, 185–190.

Mozaffarian, D., Rimm, E.B., 2006. Fish intake, contaminants, and human health: evaluating the risks and the benefits. Jama 296, 1885–1899.

Murray, C.J., 1994. Quantifying the burden of disease: the technical basis for disability-adjusted life years. Bull World Health Organ 72, 429–445.

OECD, Guidelines for the testing of chemicals. Section 4: health effects.

Omenn, G.S., Goodman, G.E., Thornquist, M.D., Balmes, J., Cullen, M.R., Glass, A., Keogh, J.P., Meyskens Jr., F.L., Valanis, B., Williams Jr., J.H., Barnhart, S., Cherniack, M.G., Brodkin, C.A., Hammar, S., 1996. Risk factors for lung cancer and for intervention effects in CARET, the beta-carotene and retinol efficacy trial. J. Natl. Cancer Inst. 88, 1550–1559.

Palou, A., Pico, C., Keijer, J., 2009. Integration of risk and benefit analysis – the window of benefit as a new tool? Crit. Rev. Food Sci. Nutr. 49, 670–680.

Park, K., Mozaffarian, D., 2010. Omega-3 Fatty acids, mercury, and selenium in fish and the risk of cardiovascular diseases. Curr. Atheroscler. Rep. 12, 414–422.

Peters, E., Hibbard, J., Slovic, P., Dieckmann, N., 2007. Numeracy skill and the communication, comprehension, and use of risk–benefit information. Health Aff. (Millwood) 26, 741–748.

Phillips, C.V., Goodman, K.J., 2004. The missed lessons of Sir Austin Bradford Hill. Epidemiol. Perspect. Innov. 1, 3.

Pohjola, M.V., Leino, O., Kollanus, V., Tuomisto, J.T., Gunnlaugsdóttir, H., Holm, F., Kalogeras, N., Luteijn, J.M., Magnússon, S.H., Odekerken, G., Tijhuis, M.J., Ueland, Ø., White, B.C., Verhagen, H., 2011. State of the art in benefit-risk ananlysis: Environmental Health. Food and Chemical Toxicology, this issue.

Ponce, R.A., Bartell, S.M., Wong, E.Y., LaFlamme, D., Carrington, C., Lee, R.C., Patrick, D.L., Faustman, E.M., Bolger, M., 2000. Use of quality-adjusted life year weights with dose-response models for public health decisions: a case study of the risks and benefits of fish consumption. Risk Anal. 20, 529–542.

Post, D.L., 2006. The precautionary principle and risk assessment in international food safety: how the world trade organization influences standards. Risk Anal. 26, 1259–1273.

Punt, A., Freidig, A.P., Delatour, T., Scholz, G., Boersma, M.G., Schilter, B., van Bladeren, P.J., Rietjens, I.M., 2008. A physiologically based biokinetic (PBBK) model for estragole bioactivation and detoxification in rat. Toxicol. Appl.

Pharmacol. 231, 248–259.

Rastmanesh, R., 2010. An urgent need to include risk–benefit analysis in clinical trials investigating conjugated linoleic acid supplements in cancer patients. Contemp. Clin. Trials..

Renwick, A.G., 1993. Data-derived safety factors for the evaluation of food additives and environmental contaminants. Food Addit. Contam. 10, 275–305.

Renwick, A.G., Flynn, A., Fletcher, R.J., Muller, D.J., Tuijtelaars, S., Verhagen, H., 2004. Risk–benefit analysis of micronutrients. Food Chem. Toxicol. 42, 1903–1922.

Rietjens, I.M., Alink, G.M., 2006. Future of toxicology–low-dose toxicology and risk– benefit analysis. Chem. Res. Toxicol. 19, 977–981.

Rothman, K.J., Greenland, S., Lash, T.L., 2008. Modern Epidemiology. Little Brown Publishers.

Saltelli, A., Annoni, P., 2010. How to avoid a perfunctory sensitivity analysis. Environmental Modeling and Software. 25, 1508–1517.

Schwarzinger, M., Stouthard, M.E., Burstrom, K., Nord, E., 2003. Cross-national agreement on disability weights: the European Disability Weights Project. Popul. Health Metr. 1, 9.

Seal, C.J., de Mul, A., Eisenbrand, G., Haverkort, A.J., Franke, K., Lalljie, S.P., Mykkanen, H., Reimerdes, E., Scholz, G., Somoza, V., Tuijtelaars, S., van Boekel, M., van Klaveren, J., Wilcockson, S.J., Wilms, L., 2008. Risk–benefit considerations of mitigation measures on acrylamide content of foods – A case study on potatoes, cereals and coffee. Br. J. Nutr. 99 (Suppl. 2), S1–S4.

Slovic, P., Finucane, M.L., Peters, E., MacGregor, D.G., 2004. Risk as analysis and risk as feelings: some thoughts about affect, reason, risk, and rationality. Risk Anal. 24, 311–322.

Smith, M., 2002. Food safety in Europe (FOSIE): risk assessment of chemicals in food and diet: overall introduction. Food Chem. Toxicol. 40, 141–144.

Son, T.G., Camandola, S., Mattson, M.P., 2008. Hormetic dietary phytochemicals. Neuromolecular Med. 10, 236–246.

Starr, C., 1969. Social benefit versus technological risk. Science 165, 1232–1238.

Tuomisto, J.T., Tuomisto, J., Tainio, M., Niittynen, M., Verkasalo, P., Vartiainen, T., Kiviranta, H., Pekkanen, J., 2004. Risk–benefit analysis of eating farmed salmon.

Science 305, 476–477 (author reply 476–477).

Ueland, Ø., Gunnlaugsdottir, H., Holm, F., Kalogeras, N., Leino, O., Luteijn, J.M., Magnússon, S.H., Odekerken, G., Pohjola, M.V., Tijhuis, M.J., Tuomisto, J.T., White, B.C., Verhagen, H., 2011. State of the art in benefit–risk analysis: Consumer perception. Food Chem. Toxicol., this issue, doi:10.1016/j.fct.2011.06.006 .

van den Brandt, P., Voorrips, L., Hertz-Picciotto, I., Shuker, D., Boeing, H., Speijers, G., Guittard, C., Kleiner, J., Knowles, M., Wolk, A., Goldbohm, A., 2002. The contribution of epidemiology. Food Chem. Toxicol. 40, 387–424.

van der Voet, H., Slob, W., 2007. Integration of probabilistic exposure assessment and probabilistic hazard characterization. Risk Anal. 27, 351–371.

van der Voet, H., de Mul, A., van Klaveren, J.D., 2007. A probabilistic model for simultaneous exposure to multiple compounds from food and its use for risk– benefit assessment. Food Chem. Toxicol. 45, 1496–1506.

van Dijk, H., Fisher, A.R.H., Frewer, L.J., 2011. Consumer responses to integrated risk–benefit information associated with the consumption of food. Risk Anal. 31, 429–439.

van Kreijl, C.F., Knaap, A.G.A.C., van Raaij, J.M.A., 2006. Our food, our health – Healthy diet and safe food in the Netherlands. RIVM report 270555009. National Institute of Public Health and the Environment, Bilthoven, The Netherlands.

van Ommen, B., Keijer, J., Heil, S.G., Kaput, J., 2009. Challenging homeostasis to define biomarkers for nutrition related health. Mol. Nutr. Food Res. 53, 795– 804.

van Raaij, J., Hendriksen, M., Verhagen, H., 2009. Potential for improvement of population diet through reformulation of commonly eaten foods. Public Health Nutr. 12, 325–330.

van Staveren, W.A., Steijns, J.M., de Groot, L.C., 2008. Dairy products as essential contributors of (micro-) nutrients in reference food patterns: an outline for elderly people. J. Am. Coll. Nutr. 27, 747S–754S.

Vandenbroucke, J.P., von Elm, E., Altman, D.G., Gotzsche, P.C., Mulrow, C.D., Pocock, S.J., Poole, C., Schlesselman, J.J., Egger, M., 2007. Strengthening the reporting of observational studies in epidemiology (STROBE): explanation and elaboration. Epidemiology 18, 805–835.

Verbeke, W., Vanhonacker, F., Frewer, L.J., Sioen, I., De Henauw, S., Van Camp, J., 2008. Communicating risks and benefits from fish consumption: impact on Belgian consumers’ perception and intention to eat fish. Risk Anal. 28, 951–967.

Verhagen, H., Vos, E., Francl, S., Heinonen, M., van Loveren, H., 2010. Status of nutrition and health claims in Europe. Arch. Biochem. Biophys. 510, 6–15.

Verkerk, R.H., 2010. The paradox of overlapping micronutrient risks and benefits obligates risk/benefit analysis. Toxicology 278, 27–38.

Verkerk, R.H., Hickey, S., 2010. A critique of prevailing approaches to nutrient risk analysis pertaining to food supplements with specific reference to the European Union. Toxicology 278, 17–26.

VKM, 2006. Fish and seafood consumption in Norway – Benefits and risks. English summary. Norwegian Scientific Committee for Food Safety.

Vlaanderen, J., Vermeulen, R., Heederik, D., Kromhout, H., 2008. Guidelines to evaluate human observational studies for quantitative risk assessment. Environ. Health Perspect. 116, 1700–1705.

Waddell, W.J., 2006. Critique of dose response in carcinogenesis. Hum. Exp. Toxicol. 25, 413–436.

WCRF/AICR, 2007. Food, Nutrition, Physical Activity, and the Prevention of Cancer: a Global Perspective. AICR, Washington DC, p. Chapter 3: Judging the evidence.

Weaver, C.M., 2009. Should dairy be recommended as part of a healthy vegetarian diet? Point. Am. J. Clin. Nutr. 89, 1634S–1637S.

WHO, 1994. International Programme on Chemical Safety. Assessing human health risks of chemicals: derivation of guidance values for health-based exposure limits. Environmental Health Criteria 170 World Health Organization, Geneva.

WHO, 2003. Diet, nutrition and the prevention of chronic diseases: report of a joint WHO/FAO expert consultation. WHO, Geneva.

WHO, 2004. Global burden of disease 2004 update: disability weights for disease and conditions. <http://www.who.int/healthinfo/global_burden_disease/GBD2004_DisabilityWeights.pdf>.

WHO-Harmonization-Project, 2005-2009. Harmonization of approaches to the assessment of risk from exposure to chemicals.

Willett, W., 1998. Nutritional Epidemiology, Second ed. Oxford University Press.

Wilson, R., Crouch, E.A.C., 2001. Risk–Benefit Analysis. Harvard University Press.

Zapponi, G.A., Marcello, I., 2006. Low-dose risk, hormesis, analogical and logical thinking. Ann. NY Acad. Sci. 1076, 839–857.