Meeting the challenges in the development of risk-benefit assessment of foods

Abstract Background Risk-benefit assessment (RBA) of foods aims to assess the combined negative and positive health effects associated with food intake. RBAs integrate chemical and microbiological risk assessment with risk and benefit assessment in nutrition. Scope and Approach Based on the past experiences and the methodological differences between the underlying research disciplines, this paper aims to describe the recent progress in RBAs, identifying the key challenges that need to be addressed for further development, and making suggestions for meeting these challenges. Key Findings and Conclusions Ten specific challenges are identified and discussed. They include the variety of different definitions and terminologies used in the underlying research disciplines, the differences between the “bottom-up” and the “top-down” approaches and the need for clear risk-benefit questions. The frequent lack of data and knowledge with their consequential uncertainties is considered, as well as the imbalance in the level of scientific evidence associated with health risks and benefits. The challenges that are consequential to the need of considering substitution issues are discussed, as are those related to the inclusion of microbiological hazards. Further challenges include the choice of the integrative health metrics and the potential scope of RBAs, which may go beyond the health effect. Finally, the need for more practical applications of RBA is stressed. Suggestions for meeting the identified challenges include an increased interdisciplinary consensus, reconsideration of methodological approaches and health metrics based on a categorisation of risk-benefit questions, and the performance of case studies to experience the feasibility of the proposed approaches.

Food is a basic requirement for life, providing the essential nutrients and energy required for optimal health. 15 However, food may also be associated with adverse health effects, because it may contain natural toxins, 16 hazardous chemical substances or pathogenic microorganisms that can affect health negatively. 17 Additionally, it is possible that the dietary intake of specific nutrients in foods is either too low or too high, 18 resulting in potential deficiencies or toxicity symptoms. 19 The diverse causes of these health effects associated with food consumption and the demand for advice on 20 safe and healthy diets have led to the development of different research disciplines in food safety and 21 nutrition. The negative health impact of human exposure to chemical substances and pathogenic 22 microorganisms through food is evaluated in two separate disciplines, chemical and microbiological risk 23 assessment. Apart from that, both health risks and health benefits associated with foods and diets have 24 been studied through the discipline of nutrition. However, in the past decade, the joint assessment of risks 25 and benefits associated to hazardous agents, food compounds, nutrients, single foods and whole diets has 26 been taken up, resulting in the establishment of "risk-benefit assessment" (RBA) as a new multidisciplinary 27 and integrated scientific discipline (Boué et al., 2015;Tijhuis et al., 2012;Verhagen et al., 2012a). 28 With the overall aim of exploring how RBA can be further developed, this paper aims to describe the recent 29 progress in RBAs and to identify and discuss key challenges in RBA research. To clarify the fundamentals of 30 RBA and to provide a basic understanding of the background of many of the challenges, the main concepts 31 of the underlying disciplines chemical risk assessment, microbiological risk assessment and nutritional risk 32 and benefit assessment are explained. Following that, the developments in RBA thus far are addressed. The 33 major part of the paper is devoted to a discussion of ten challenges, as well as to suggestions for how they 34 can be met. The conclusion summarizes the authors' vision on the future developments of the research area. 35 1.1. Risk and benefit assessment in food safety and nutrition 36 The use of risk assessments has traditionally been an integrated part of a common risk analysis framework 37 (Figure 1), where risk assessment is done by risk assessors who provide scientific advice to support decision 38 making by risk managers, such as food authorities or food producers, on the potential risks associated with 39 food consumption. Risk communication is an essential part of the risk analysis, both between risk assessors 40 and risk managers, and between assessors, managers and other stakeholders (FAO/WHO 2006a). 41 Risk assessment was first formalised for chemicals by the establishment in 1980 of the International 42 Programme on Chemical Safety (IPCS), which proposed a scientifically based process including four 43 elements: hazard identification, hazard characterization, exposure assessment and risk characterization 44 ( Figure 2). The first step, hazard identification, involves the identification of the inherent toxicological 45 properties of a chemical substance in the food that may affect human health adversely. Depending on the 46 nature of the chemical substance, the information on hazards may stem from in vitro studies (for example on 47 doses, and threshold doses may be derived as cut-off points below which the intake is considered safe, or 125 the associated risk is considered acceptable (Barlow et al., 2015). In contrast, within nutrition, both the 126 situation where there is a risk of nutritional deficiency and the situation where there is a risk of nutrient 127 intoxication are relevant, creating a "window of benefit" (Palou et al., 2009;Tijhuis et al., 2012)). Interestingly, 128 research in situations where the intake is too high (above the upper intake level (UL)) is commonly referred 129 to as toxicology, whereas research considering beneficial intake or too low intake, is part of nutrition. 130 1.2 The development of risk-benefit assessment 131 Although independent risk and benefit assessments have proven to be useful for decision support in food 132 safety and nutrition, their results may be too much focused on one hazard, one food compound or one health 133 effect. When establishing guidelines and advice on food consumption, nutrient intake and diet choices, there 134 is a need for an overarching approach, in which all of the relevant health risks and benefits are included and 135 compared. This need for RBAs has been identified earlier in several publications ( EFSA, 2007;EFSA, 136 2010a; Renwick et al., 2004) and led to the development of RBA of foods as a new research discipline. An 137 RBA is multidisciplinary by nature, and may require expertise from not only toxicologists, microbiologists, and 138 nutritionists, but also from epidemiologists, chemists, librarians, statisticians, and medical scientists. As 139 proposed in the EU-funded project BRAFO (Benefit-Risk Analysis of FOods) (Boobis et al., 2013), it is 140 common to use the risk analysis and risk assessment frameworks (Figures 1 and 2) as the basis for the RBA 141 methodology by applying the established concepts to both risks and benefits. A recent extensive review of 142 studies related to the combined RBA of foods, nutrients and compounds shows that the majority of published 143 studies have been related to fish consumption where the nutritional beneficial effects are compared with the 144 adverse effects from chemicals (Boué et al., 2015). This RBA of fish (e.g. (Hoekstra, Hart, et al., 2013)) is an 145 example of an RBA case where the content of polyunsaturated fatty acids, and in particular 146 docosahexaenoic (DHA), and eicosapentaenoic fatty acids (EPA), recognized for their health benefits, is 147 counterbalanced by the content of pollutants such as methylmercury and dioxins, known to potentially induce 148 adverse health effects. There is also an example of microbiological aspects being added to an RBA of fish 149 (Berjia et al., 2012). The basis is that a number of tiers have to be evaluated before making a decision on the required steps to 155 be taken in the RBA. This approach proposes that a qualitative assessment is sufficient if data are scarce or 156 there is clear evidence that risks outweigh the benefits (or vice versa). If the balance between benefits and 157 risks is unclear, the assessment has to be performed at a higher tier, including quantitative assessment. As 158 part of the BRAFO project, a number of relevant risk-benefit studies that illustrate the usefulness of a tiered 159 M A N U S C R I P T A C C E P T E D 2. Challenges in risk-benefit assessment 163 Although significant progress has been made in the development of methods and terminology in RBA, 164 several challenges remain. Some of these challenges relate to the differences between the underlying 165 research disciplines, which have different use of terminology and different approaches for the assessment of 166 health effects related to the consumption of food. Other challenges relate to the specific objective of RBAs, 167 the scarcity of the required data, or the complexity of the characterization of health effects. Below, we 168 provide a description of ten major challenges that were identified during the course of working with RBAs, 169 with explanations of the challenges and discussion on the way forward for meeting them in the future. 170

Definitions 171
The different approaches used in the disciplines contributing to RBA (Section 1.1) apply different terminology 172 or may apply the same terminology in a different way. Dissimilar definitions can lead to confusion and lack of 173 understanding of the risk-benefit question (Section 2.3). As an example, the central concept of "hazard" is 174 defined differently in various contexts. Published definitions of hazard include "inherent property of an agent 175 or situation having the potential to cause adverse effects when an organism, system, or (sub)population is 176 exposed to that agent" (IPCS, 2004), "the potential of a risk source to cause an adverse effect(s)/event(s)" 177 (Renwick et al., 2003) and "a biological, chemical or physical agent in, or condition of, food with the potential 178 to cause an adverse health effect" (CAC, 2011). In the latter definition, the hazard is the agent (or risk 179 source, that is the pathogen, chemical substance or food compound) and in the others it is an inherent 180 property or the potential of this agent. Due to this difference in definitions, the hazard is usually synonymous 181 to the pathogen(s) of concern in microbiological risk assessment, whereas it usually is the potential health 182 effect caused by the chemical substance or food compound in chemical risk assessment and nutrition 183 (Barlow et al., 2015). 184 Similarly, there are different definitions of "risk", for example "the probability of an adverse effect in an 185 organism, system, or (sub)population in reaction to exposure to an agent" (IPCS, 2004; EFSA, 2010a), or "a 186 function of the probability of an adverse health effect and the severity of that effect, consequential to a 187 hazard(s) in food" (CAC, 2011). So in one definition the risk is a probability, in the other, it is a combination of 188 probability and severity. 189 When mirroring risk assessment to benefit assessment, the benefit is defined at a level comparable to both 190 the hazard and the risk (EFSA, 2006; Boobis et al., 2013), so "benefit" is both the counterpart of "hazard" 191 and the counterpart of "risk". Hence, the term "benefit" can be used for anything between the agent causing 192 the health effect and the probability and magnitude of that effect. Moreover, when used as equivalent of 193 "risk", the benefit is not necessarily interpreted as the probability of a positive effect, but commonly as the 194 The present definitions can be well understood in a historical perspective, given that RBA has evolved from a 197 variety of disciplines. However, for further development, the discipline "risk-benefit assessment of foods" 198 needs a clearer set of definitions and harmonized terminology that is comprehensible for all those involved. 199 To accommodate the fact that some agents or food compounds (i.e. "hazards" of "benefits") can be both a 200 source of positive and negative health effect depending on the exposure (Figure 3), Boué et al. (2015) 201 propose to use the term "health effect contributing factor" (HECF) for "the agent able to cause an adverse or 202 positive health effect in the case of exposure". This is a useful first step in the reconsideration of the 203 terminology used in RBA. Consensus within the international research community is required for clarification 204 and harmonization purposes and definitely when it would be used for regulatory purposes. Obtaining such a 205 consensus is a process that should be led by international authorities, and should include representatives of 206 all relevant disciplines involved in RBA. 207 2.2 Bottom-up versus top-down approach 208 In this paper, we distinguish between two overall approaches to assess health effects in RBA and refer to 209 them as "bottom-up" and "top-down". This terminology is derived from studies in microbiological food safety 210 aimed at ranking microbiological food risks (EFSA, 2015; Cassini et al., 2016). The two approaches are 211 characterised by their different starting point. The typical risk assessment approach, which starts with the 212 hazard identification for the food product or its ingredients and finishes with the human health outcome 213 obtained after combining the exposure assessment with a dose-response model (Figure 2), is referred to as 214 the bottom-up approach. The alternative top-down approach starts with the adverse (or beneficial) health 215 outcomes as obtained from human observational studies, i.e., incidence data and identified risk factors. 216 These are then traced back to the food sources that caused the disease of concern (or benefit of desire), 217 thus linking the health effect to the food product. 218 A similar distinction in approaches can be made in nutritional and chemical risk assessment. The usual risk 219 assessment approach (i.e., bottom -up) is targeted at intake of specific nutrients or food compounds, and 220 the dose-response relation is typically derived from animal experimental data. The alternative top-down 221 approach is an approach where relative risk estimates from human observational studies are used and 222 linked to foods or food compounds that are identified as risk factors. In the review of the BRAFO project, 223 terminology as it is more generic and can also be applied for microbiological risk assessment, which does 226 not apply animal data. 227 Hence, with the bottom-up approach, the assessment starts with the food product, food compound or 228 contaminant, followed by an exposure assessment and a dose-response model used for the risk-benefit 229 characterization. An advantage of this approach is a direct causal link between intake of the food product or 230 M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 8 food compound (or contaminant) of concern and the associated health effect. A disadvantage is that there 231 may be a large uncertainty attending the exposure assessment and (especially) the dose-response. 232 With a top-down approach, the starting point of the analysis is the incidence of a health outcome in the 233 consumer. Typically, data from epidemiological studies (case-control studies, cohort studies, randomized 234 controlled trials) are used to associate human health outcomes with risk factors that are defined in terms of 235 food consumption, allowing for the estimation of metrics such as the odds ratio or the relative risk. These 236 measures of association are then combined with population statistics and incidence data to estimate the 237 actual health risks in the population. The relative risks may also be used to construct a dose-response 238 relation, where the relative risk is a function of the intake as specified in the underlying study. The strength of 239 human observational studies is that they are based on actual health effects, measured in specified 240 populations. Weaknesses are that the observed associations are not a proof of causation, that the studied 241 population may not be representative for the population group of interest and that many data are required if 242 the health effect of interest is small. For microbial pathogens, a top-down approach can be used to estimate 243 the number of cases of disease caused by a pathogen due to its presence in a specific food, a method 244 referred to as "source attribution" (Pires et al., 2009). Here incidence data on a specific health outcome (e.g., 245 gastroenteritis caused by salmonellosis) is traced back to a specific food source (e.g., chicken meat) by the 246 use of subtyping information of isolates of the pathogen in human cases and food sources. case study on Campylobacter in the Netherlands and identified the differences in the underlying 261 uncertainties. They found that the difference in the point estimates of the risks as found by the different 262 approaches can be large, but they still have overlapping uncertainty intervals. This implies that one cannot a 263 priori conclude that one approach is better than the other. It is advisable to aim for evidence synthesis by 264 using an approach that takes advantage of all available data and combines bottom-up and top-down generally a comparison between two, or a series of, choices, alternative policies or courses of action, 272 described in the form of scenarios (Boobis et al., 2013). In these scenarios, both positive and negative health 273 effects have to be taken into consideration. When a series of scenarios is compared, the risk-benefit 274 question can be used to identify the optimum intake (Berjia et al., 2014). An aim of the risk-benefit question 275 is to specify the RBA-task in such a way that it is feasible and will provide useful results. For example, an 276 RBA of fish should indicate what sort of fish (e.g., lean/fatty, farmed/wild), target population group, and in 277 general any other constraint that could narrow the risk-benefit question. In the end, the level of specification 278 of the question will also depend on the data available. 279 As a variety of risk-benefit questions can be asked, it can be helpful to categorise them and to identify 280 specific approaches that can be used to answer these different categories of questions. Here, one type of 281 categorisation is the level of aggregation: the risk-benefit question can be targeted at a food compound level 282 (a nutrient, a chemical or microbiological contaminant), a food product level (e.g., fish) or a diet level 283 (Hoekstra et al., 2008). studies. To assess the total intake of the food compound, it will be necessary to consider the intake of all 290 relevant foods and food products in the diet that contain it, and the concentrations of the compound in these 291 foods and food products have to be known. As this can be rather complicated, one can choose a risk-benefit 292 question that only considers a difference in intake or concentration in one or a few food products, making 293 some assumptions for the background diet. can inform about the health impact of one intake scenario compared with another. Alternatively, a bottom-up 300 approach may be used where all relevant food compounds (and contaminants) in the food product have to 301 be identified and comprised in the RBA to assure that the health effects of interest are included. In that case, 302 a selection of relevant food compounds and contaminants needs to be made based on the associated levels 303 of evidence and the precise risk-benefit question. However, because in some cases only exposure through 10 the selected food product is considered, and not the total exposure from all food products containing the 305 compounds, it is difficult to use a bottom-up approach with a dose-response relation for each compound. 306 When considering a whole diet, the bottom-up RBA approach will usually not be feasible, unless the risk-307 benefit question is clearly delimited: the number of food compounds (and contaminants) and their combined 308 intakes easily get too large for a complete exposure assessment and hazard characterisation. However, a 309 top-down approach using studies on human consumption may be possible if the appropriate data are 310 available, for example from a dietary intervention study. Van Kreyl et al., 2006, performed a study to analyse 311 the health effects of the current diet in the Netherlands that may be regarded as an RBA of diets, but 312 otherwise, to our knowledge, no formal RBAs of whole diets have been performed so far. 313 In each of these three categories of risk-benefit questions, the options for inclusion and exclusion of food 314 compounds and contaminants, food products and health effects are large. To clarify the selected elements in 315 the risk-benefit question, we propose the use of schematic framing of the risk-benefit question, as 316 exemplified in Figure 4 for four published risk-benefit studies for food compounds or food products. A 317 scheme like this is broadly applicable and may offer a transparent way to identify different types of risk-318 benefit questions and clarify how the risk-benefit question is addressed. In the case of an RBA of a whole 319 diet, the scheme would be pretty complex, which stresses the difficulty of doing an RBA of a whole diet. 320 2.4 Lack of data and knowledge and the consequential uncertainties 321 The data needs for an RBA are large and diverse. RBAs frequently face data gaps and lack of knowledge, 322 such as lack of human data, information on dose-response and intake levels for specific population groups. 323 These challenges are also faced in other modelling exercises (such as many risk assessments), and need to 324 be addressed by documentation and discussion of the assumptions made. A consequence of limited data 325 and lack of knowledge is that the uncertainty related to the assessment may be large. Yet, characterising 326 this uncertainty is crucial in the risk-benefit characterisation. relationships, this means that the use of BMD models may be preferred over NOAELs and LOAELs, and that 342 the uncertainty factors used to translate animal data to human guidance values may not be appropriate if the 343 dose-response relationship is to be applied in RBAs. 344 The uncertainty attending the dose-response relations for microbial pathogens is also large. These dose-345 response relations are usually based on human volunteer studies or outbreak data, which means they are 346 based on limited data sets, for specific strains and specific population groups, and generalised thereafter. 347 Dose-response relations based on studies with healthy young volunteers may be expected to underestimate 348 the risk, whereas those derived from outbreaks (with more virulent strains) may overestimate the risk. 349 Further, it is known that immunity plays an important role and may lead to overestimation of the risk, but it is 350 difficult to include this in the modelling (Havelaar & Swart, 2014). Uncertainties are an inevitable intrinsic element of science, risk assessment and RBAs, and it is of utmost 364 importance that they are not ignored. A challenge here is that, as in risk assessment, it is not primarily the 365 objective of an RBA to assure that the uncertainty is small enough (as aiming for a p-value smaller than 366 0.05), but to indicate how large the uncertainty actually is (Nauta, 2007). One should deal with the identified 367 uncertainties by explicitly addressing and characterizing them in the assessment and by clearly 368 communicating them to all stakeholders. By framing the risk-benefit question (Figure 4)  Another consequence of this discrepancy is that different types and levels of uncertainty will be associated to 387 the risk assessment on the one hand and the benefit assessments on the other, which complicates the 388 characterization of the combined RBA even further (Section 2.4). 389 The imbalance in the required level of scientific evidence for risks and benefits demands a paradigm shift 390 from the RBA as a sum of risk and benefit assessment to the RBA as a well-integrated risk-benefit 391 assessment. Such a well-integrated RBA deals not so much with studying a hypothesis about the presence 392 or absence of a health effect associated with the intake of a (certain amount of) food product or food 393 compound or contaminant, but predominantly with the size of the health effects. Even though the strength of 394 evidence for the presence of a health effect is strongly correlated to the size of the effect, these are not the 395 same thing. Stochastic modelling techniques, which include quantification of uncertainty and variability, allow 396 an evaluation of potential health effects, even if the effects themselves are not statistically significant. In 397 doing so, it may be possible to study how the estimated size of the effect, and some alternative scenarios 398 about these effects, may impact public health. From this, one might conclude that the risk or benefit is not 399 very large, even if the evidence would be convincing, or the opposite, that a risk or benefit may be large, 400 even if the level of evidence is low. Findings like this can indicate crucial data gaps (Section 2.4) and may, in 401 an objective way, help identify where further research is needed. 402 2.6 Substitution 403 In general, an RBA compares the health effects of two or more intake scenarios, defined as specified 404 changes in the amount or type of food consumed. As a side effect, these specified changes in intake may 405 also imply a change in the intake of other food products to compensate for the part of the diet that is deleted 406 or added. So far, however, such "substitution" is rarely included in an RBA. The risks and benefits of 407 increasing fish intake are for example frequently studied, but the related decrease in the intake of one or 408 more other foods and the consequential health effects of that decrease are not included in the assessment 409 non-fortified diets are similar except for the content of the specific nutrient. In the sugar-artificial sweetener 420 case, the substitution leads to non-isocaloric diets and this may need to be addressed because it implies that 421 the diet may change in more aspects than just the intended substitution. 422 To meet this challenge, it is a prerequisite that substitution is acknowledged in the RBA, either by specifically 423 addressing it in the intake scenarios that are analysed, or by referring to it in the discussion of the 424 assumptions and in the uncertainty characterization. As simplified substitution scenarios, one can consider 425 replacements in the same food groups (e.g. meat and fish) and isocaloric alternatives (to make sure the 426 energy intake stays similar). Next, the impact of substitution can be analysed in separate scenarios, where 427 different options for substitution are compared. patterns and sensitivity to food hazards. When the RBA is done and the risk-benefit balance for the 450 population is interpreted as the risk-benefit balance for the average consumer, this does not mean that this 451 balance is the same for all individual consumers. It can be that the balance goes in different directions for 452 different subpopulations, e.g., the elderly, pregnant women or children, and because there are differences in 453 intake and sensitivity between individuals. Therefore, the variability between consumers has to be taken into 454 consideration, for example by using a stochastic approach (Hart et al., 2013). are often linked with mild health effects such as short episodes of gastro enteritis. They can also lead to 481 long-term sequelae and severe chronic effects, but these are typically not registered and less often 482 measured (Havelaar et al., 2012). In principle, microorganisms can rather easily be eliminated from foods by 483 application of a heating process, which might suggest that microbiological risks from food can quite easily be 484 prevented. However, microbial contamination of food products and exposure are common, and, to some 485 extent, more easily accepted by consumers (Kher et al., 2013).

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
because the availability of the relevant data is larger, the recent World Health Organization (WHO) study on 488 the global burden of foodborne disease  is primarily focused on microbiological 489 hazards, and only four chemical substances have been considered in the WHO report. The results suggest 490 that the disease burden related to the exposure to microorganisms may be larger than that for chemicals, but 491 more comparable disease burden estimates for chemical substances are required before an overall 492 comparison between the burden of chemical substances and microbiological pathogens can be made. 493 However, the results confirm that risk associated with microbiological hazards can be quantified and that it is 494 important to include microbiological risks in RBAs as well. 495 The inclusion of microbiological risks and benefits in RBAs requires that the specific characteristics of 496 microbiological agents are acknowledged, and that they are included in case studies. As illustrated by Berjia The challenges from differences in approach between chemical and microbiological risk assessment needs 502 further study to allow the development of a more integrated approach towards RBAs (Sections 1.2 and 2.5). 503 Recently developed tools that are increasingly adapted to allow comparisons between chemical and 504 microbiological health risks (e.g. FDA-iRisk; Chen et al., 2013) can help to address these challenges. 505 2.9 The scope of risk-benefit assessments 506 The scope of a risk-benefit question in relation to food may be much wider than direct health impact and can 507 include socio-economic, psychological and/or environmental dimensions (Boobis et al., 2013). When 508 consumers select their food, the health effect is only one of the concerns; others include cost, taste, quality 509 and sustainability of the production. An indicated health risk may be counterbalanced by each of these, for 510 example, if low price and good taste are considered benefits that outbalance the health risk. 511 One may consider widening the scope of RBAs of foods and include some of the aspects mentioned above. 512 Cost is an obvious choice, which is an intrinsic part of the RBA when metrics such as the cost-of-illness or 513 willingness-to-pay are used (Section 2.7). It can also be added to the RBA by means of a cost-utility, cost-514 benefit or cost-effectiveness analysis, as for example done for the costs of intervention strategies that aim to Ultimately, it can be attractive to address all of the relevant aspects in one overall analysis, for example by 525 the use of multi criteria decision analysis (MCDA). This method has for example been applied to the 526 prioritisation of foodborne pathogens (Ruzante et al., 2010), taking into account public health impact 527 (expressed in DALY and cost-of-illness), market impact, consumer perception and acceptance, and social 528 sensitivity to impacts on vulnerable consumer groups and industries. In MCDA, an integrating scoring 529 method is developed, which weighs the importance of different factors that are considered relevant for the 530 decision making, allowing one to come with a final ranking that includes all of these factors. 531 Defining the scope of the RBA is clearly an issue that should be decided upon when the risk-benefit question 532 is formulated. A broader scope includes more relevant issues, but also implies an increasing demand for 533 resources in terms of research efforts, data and method development. Clearly, challenges that complicate 534 RBAs, such as the lack of data and knowledge, and the consequential uncertainties, the imbalance in level 535 of scientific evidence and the use of quantitative metrics, only get more weight when a broader scope is 536 taken. Yet, the ongoing developments show that progress can be made, and with multidisciplinary scientific 537 collaboration and investment in research supporting RBAs, this progress can be strengthened in the future. 538 2.10 The application of risk-benefit assessments 539 So far, several RBAs have been performed, but mainly within research projects that were directed at the 540 development of RBA frameworks and methodology. The aim of these RBAs was primarily to illustrate the 541 potential of the methodology and the risk-benefit question was not posed by independent risk-benefit 542 managers but by the researchers themselves. There is now a need for more experience with the practical 543 application of RBAs and the proposed methodologies. These practical applications of RBAs can fall into two 544 categories: those leading to recommendations or guidelines to food safety and health authorities, and those 545 leading to process and formulation design by industry (Boué et al., 2015). The first application is the one 546 considered most often and typically the request for such an RBA originates from national or international 547 food and health authorities that have a mandate to advise the public on a particular food or diet and have 548 identified a need to establish a scientific basis for this advice. Examples are an RBA on fish and fish 549 products performed in Norway (Skåre et al., 2015) and an RBA on nuts performed in Denmark (Mejborn et 550 al., 2015). Another reason for the authorities to make requests for an RBA is a need for an evaluation of 551 health effects of proposed fortification of foods, as for example with vitamin D, folic acid (Hoekstra et al., 552 2008) or iodine (Zimmermann, 2008). 553 Food producers may have an interest in RBAs when they change their production or the formulation of their 554 products. This is especially of interest when this change is based on a wish to decrease one specific health 555 risk that can go at the expense of another. For example, when a heating step is introduced to decrease 556 microbiological health risks, this can go at the expense of the formation of potentially carcinogenic 557 substances (Havelaar et al., 2000) and/or decreased vitamin levels. In such cases, RBA can be an excellent 558 tool to settle a dispute that cannot be solved on the basis of the identification of risks and benefits alone. This is important to assure the scientific quality, to increase the experience in the research community and to 565 aid the international discussion on the potential and challenges of RBAs. 566 3. Conclusion 567 RBA is an evolving discipline in food safety and nutrition that takes advantage of achievements in a variety of 568 underlying disciplines. As it integrates various health concerns, it is a valuable method to estimate the overall 569 health effects related to food consumption and diet choice, which can be applied both by food and health 570 authorities and the food industry. Recognizing the progress that has been made in the past decade and 571 based on previous work, we have identified a series of challenges that should be met to develop the area 572 further and indicated steps that should be taken for further progress. The challenges and suggested ways 573 forward in meeting them are summarized in Table 1. 574 To meet the challenges of RBA, it is important that researchers in underlying disciplines and stakeholders in 575 food regulation, production, retail and consumption from different regions in the world agree on definitions 576 and concepts that are practical and agreeable for all. Based on relevant risk-benefit questions, a series of 577 risk-benefit studies should be performed, not so much to develop methods, but predominantly to identify the 578 practical challenges that are met when working on RBA case studies. When investigating these practical 579 challenges, steps can be made in categorizing them and in developing and harmonising agreeable methods 580 to address them. 581 Overall, with an increasing demand from different stakeholders for holistic and objective assessments of the 590 health effect of foods, multidisciplinary RBA is a promising research area for the future. Impressive progress 591 has been made and, despite the remaining challenges, we expect that more progress will be made in the 592 next decade. The steps forward proposed in this paper will be useful in taking the research area further, 593 allowing for transparent and reliable RBAs to be performed on a wider scale in the future. and nutrition (right). In toxicology and microbiology, the risk increases with the dose; benefits are not defined. 833 In nutrition, intake of a food compound can be too low or too high; intake between these levels ("the window 834 of benefit") is considered beneficial for health. X: Dose with critical response as used in chemical risk 835 assessment (e.g., LOAEL or BMD); no equivalent metric exists in microbiological risk assessment. LTI: 836 Lower threshold intake, intake below this level represents a deficiency; UL: Upper intake level, intake above 837 this level could give a toxic effect. 838

Highlights
• RBA combines chemical and microbial risk assessment with risk and benefit assessment in nutrition.
• Key challenges in risk-benefit assessment of foods are addressed.
• Challenges relate to interdisciplinarity, methods, data, health metrics and applications.
• Suggestions for meeting the identified challenges are discussed.