Using Risk Assessment to Develop Antimicrobial Regulations

Table of Contents

1. Abstract
2. Introduction
3. Risk Assessment
4. Hazard Identification
5. Exposure Assessment
6. Hazard Characterization
7. Risk Characterization
8. Other Issues
9. Quantitative Modelling in Risk Assessment
10. Validation of Risk Assessment
11. Tiered Approach
12. Benefit/Cost Analysis
13. Conclusions
14. References
    
    

Abstract

Risk assessment has only recently evolved as a formal process for public policy formation, however it is gaining acceptance as a valuable tool, especially where there are important elements of complexity or uncertainty involved and where there may be varying or conflicting views or financial interests among affected members of society. The complexities and uncertainties inherent in the antimicrobial resistance issue make it unwise to use an intuitive approach for selecting optimal points of intervention for resistance control and for determining the implications or costs of various risk management options. The difficulty in making sound public health decisions in the face of complexity, uncertainty and varying scientific and public opinion makes a compelling case for a decision-making process that is open and based on scientific evidence, economic analysis and wide consultation with due consideration of societal values.

Introduction

Risk assessment has only recently evolved as a formal process for public policy formation, however it is gaining acceptance as a valuable tool, especially where there are important elements of complexity or uncertainty involved and where there may be varying or conflicting views or financial interests among affected members of society. The complexities and uncertainties inherent in the antimicrobial resistance issue make it unwise to use an intuitive approach for selecting optimal points of intervention for resistance control and for determining the implications or costs of various risk management options. The difficulty in making sound public health decisions in the face of complexity, uncertainty and varying scientific and public opinion makes a compelling case for a decision-making process that is open and based on scientific evidence, economic analysis and wide consultation with due consideration of societal values. Elements of this paper have been presented recently at other meetings (McEwen 1999; McEwen and Robinson, 1999).

Risk Assessment

Risk analysis comprises risk assessment, risk management and risk communication. In the context of antimicrobial resistance in agriculture, risk assessment is the process of estimating the probability and impact of adverse health effects attributable to resistance arising from antimicrobial use on farms. These estimates may be expressed in qualitative terms (e.g. low, medium or high), however quantitative expression of risk is preferred whenever possible (e.g. number of human infections, illnesses or fatalities per year). Risk management is a process that seeks to identify various options for mitigation of risk, and selection of the optimal course of action after consideration of benefits and costs and in light of consultation with interested parties in industry, government, academia and the general public. Risk communication is the process of consultation, discussion and review that seeks to enhance the validity, effectiveness and general acceptance of risk assessment and risk management.

Risk assessment typically comprises hazard identification, exposure assessment, hazard characterization and risk characterisation and in general it should be specific to each hazard (e.g. ampicillin-resistant Salmonella typhimurium), and it should be specific to management system, animal species or food commodity. Risk assessment models should seek to use available scientific information to estimate in quantitative terms the probability and impact of adverse health effects. While quantitative estimates of risk are highly desirable, they are difficult to obtain owing to limitations in expertise, time, data and methodology. A strong research, investigative and surveillance infrastructure is needed to provide the information which underpins the risk assessment process.

In many instances, however, lack of understanding of the biology of hazards and exposure dictate that only qualitative assessments will be possible, nevertheless these can be very useful and informative. It is important to state assumptions, data sources and uncertainties encountered at all stages of risk assessment to facilitate review and enhance credibility of the model structure and the risk estimates obtained. A variety of expert panels have been convened by the National Academy of Sciences (NAS) to provide recommendations on risk analysis practice (National Research Council, 1983; 1989; 1994; 1998). A degree of international credibility of the "NRC model" of risk assessment has also been achieved, as indicated by the adoption of formal risk assessment for development of food safety standards by Codex Alimentarius. The International Commission on Microbiological Specifications for Foods (ICMSF) and the National Advisory Committee on Microbiological Criteria for Foods (NACMCF) have recently published generic principles of risk assessment for illness caused by foodborne biological agents (National Advisory Committee on Microbiological Criteria for Foods, 1998).

The NRC risk assessment model is composed of four main components: hazard identification (HI), dose-response assessment (DRA) or hazard characterization (HC), exposure assessment (EA) and risk characterization (RC). In some cases HC and EA are reversed in sequence, but this is of little practical consequence. This model has been used to assess risks from antimicrobial residues in foods and recently risks from microbial hazards in foods (e.g. Salmonella, E. coli). Other workers have taken a risk assessment approach to the resistance issue (Ferenc, et al, 1998; Report from the Commission on Antimicrobial Feed Additives, 1997). Scientists and industry personnel may not be acquainted with similarities and differences in risk assessment of residues and resistance, consequently there may be merit in comparing the two (Table 1).

The models described in Table 1 differ in at least two important ways. First, drug residues are chemicals and their post-harvest concentrations in edible animal products do not change very much, while bacteria can die, grow and interact with other organisms between harvest and eventual consumption - this has important implications to exposure assessment. Second, drugs are approved for intentional administration to animals and approved uses can be structured to minimize exposure to residues. Conversely, microbial contaminants are naturally occurring and exposure cannot be so readily manipulated. Antimicrobial risk assessment links these two models in the sense that population dynamics of microbes must be considered, and the possibility exists to manipulate approved drug usage in order to minimize risk.

Table 1. Main features of risk assessment models for antimicrobial residues and microbial agents in food.

General Risk Assessment Model Structure (Qualitative or Quantitative)
Model Component	Veterinary Drug Residues	Microbial Agents
Purpose	- determine a) if drug genotoxic, b) establish safe levels in foods, c) drug withholding times - provides for drug use that poses insignificant health risk	- characterize risk in terms of the probability of illness or death from naturally occurring microbial contaminants - assess impact of changes in food production/processing on risk - develop food standards - eventually establish critical limits for HACCP
Hazard Identification - known or potential health effects	- chemical structure of drug identified -evidence of toxicity in animal bioassay and no observed effect level NOEL (dose) derived	-agent identified -evidence of causal role in foodborne disease gathered from outbreak investigation or epidemiological study
Exposure Assessment - degree of intake likely to occur	- assumptions made about consumption of foods derived from treated animals - with ADI calculate maximum residue limits (MRL) in foods - withdrawal time set to ensure MRL not exceeded	- usually complex determinations of prevalence and concentrations of pathogen in foods at consumption - account for growth/death dynamics of microbes - based on surveillance studies, simulation modeling - investigate scenarios to determine effects of processing changes on risk
Hazard Characterisation/Dose Response Assessment - evaluation of likely effects at consumption	- NOEL derived from bioassay - with safety factor calculate acceptable daily intake (ADI)	- data from human volunteer studies, animal models, outbreak investigations used to estimate effect of varying levels of contamination - usually involves complex modeling
Risk Characterisation - integrate previous steps to estimate likely adverse effects with description of uncertainty	- should be negligible risk if regulatory compliance achieved	- risk estimate expressed in terms of probability of illness or death e.g. from one portion of food, expected number of cases/100,000 population, etc., estimates for subgroups of population

Hazard Identification

The principal human health hazards worthy of inclusion in risk assessments of antimicrobial use in agriculture include: resistance in foodborne pathogens and non-pathogens to drugs important to human medicine or ability to select for cross resistance to such drugs; and possible increases in shedding of foodborne agents from animals regardless of resistance status ("pathogen load") (FDA, 1998).

Exposure Assessment

The goal of exposure assessment is to estimate (qualitatively or preferably quantitatively), the frequency (prevalence) and concentration (intensity) of hazard to which people are exposed. If the hazards are foodborne pathogens (resistant or not) the goal is to estimate the prevalence of contamination of various foods with pathogens at the time of consumption and the number of organisms ingested at a sitting. If there is reason to believe that effects could be cumulative (usually not for these bacterial pathogens) then exposure over time would also have to be estimated.

This is one of the two most complex and uncertain aspects of microbial risk assessment (the other is dose-response assessment). In food microbiology, emphasis is placed on estimating the effects of a large number of factors on the dynamics of bacterial populations in foods. These factors may include post-harvest processing, packaging, cooking, storing, food composition, initial contamination levels, hygiene interventions, etc. Other factors to consider include the prevalence of infection in food animals, transportation, slaughter practices, and rates of transmission between animals.

Because the food system is so complex it is difficult to identify the individual, let alone collective importance of these factors to human expose using conventional experiments or observational studies. Consequently, simulation and other types of modeling are increasingly being used in quantitative microbial risk assessment to identify the factors that are especially important to the risk, and to identify the importance of uncertainty on estimates of exposure.

The complexity of the exposure issue, perhaps more than any other factor, has bedeviled past efforts to identify the human health effects of antimicrobial use in animals. The intuitive approach to assessing exposure has been used in past efforts and is no longer sufficient. We must develop models to better characterize exposure. Initially, these should be conceptual, qualitative models that lay out the plausible pathways of exposure, beginning with animal rearing, through drug administration, resistance development, fecal contamination of carcasses, contamination of food products, consumption and human infection. Complexity should be addressed by laying out plausible pathways for transmission between animals, cross-contamination of meat, etc. Cross-selection and exchange of resistance determinant among organisms would add further complexity to exposure models. While valid quantitative models are ideal, qualitative model development is also useful because it forces a step-by-step consideration of plausible routes of exposure that may very well identify important factors that were not initially considered. A major deficiency of the qualitative approach is the inability to know what factors are really important, how variability and uncertainty in the model affect exposure estimates, and what would be the effect on exposure of introducing new interventions (such as prudent use practices, HACCP in slaughter facilities, etc.). Regulatory authorities should encourage development of quantitative models of exposure, at least from birth through to harvest. Final consumption by humans would be the ideal end-point but modeling food processing, etc. is extremely complex. It may eventually be possible to use modules from other quantitative food microbiology risk assessments (e.g. E. coli O157:H7 or Salmonella) with appropriate modifications.

Hazard Characterization

The purpose of this step (also called dose-response assessment (DRA) is to estimate the frequency and severity of illness expected at varying levels of exposure, i.e. for a given number of organisms consumed at a sitting, what is the probability of illness? This is the second step in quantitative microbial risk assessment that is complex and at present very uncertain. An active area of research in the field is development of more appropriate mathematical models for dose-response. Impetus for this work comes from the critical need to specify acceptable levels of pathogens in foods. Similarly, estimation of dose-response relationships is an important issue in risk assessment of antimicrobial resistance and "pathogen load". Identification of surveillance "Threshold levels" of resistance requires consideration of dose-response. Another reason why DRA is important to microbial risk assessment is the range of susceptibility to infectious disease among the human population.

The concept of hazard characterization can be extended to dimensions of risk assessment other than human illness or infection, for example, what dose (and duration, route) of drug treatment in animals will induce or select or cross-select for resistance in animals or the environment? What dose of antimicrobial-resistant Salmonella on the hides of cattle will lead to contamination of meat? What level of resistance must be achieved in a population of bacteria/animals before it becomes established? There is currently very little information on which to base quantitative hazard characterization.

Risk Characterization

The risk characterization step is the main product of the risk assessment and will be used by risk managers to support regulatory policy-making . In the past, undue emphasis has been placed on point estimates of risk and most authorities recommend including a discussion of the uncertainties encountered (statistical, biological, model structure, etc) and ideally, separate analyses of uncertainty (of parameter estimates, distributions used, etc.) and variability would be made. Explicit disclosure of assumptions made throughout the assessment should also be made, along with possible effects of assumptions on risk estimates. Due to the fact that the discipline is just beginning to develop techniques and experience, quantitative techniques that attempt to model microbial dynamics in animal populations and food probably will not be useful to regulatory authorities in risk assessment of microbial new drugs for some time. Therefore, qualitative estimates of risk (e.g. low, medium, high) with narrative description of uncertainties, caveats, defaults used etc. will have to suffice. To aid decision-making, efforts should be made before conducting the risk assessment to determine what constitutes unacceptable risk.

Other Issues

Many risk assessment rely on "default assumptions" when information is lacking, including data needed to specify model parameters or lack of basic biological understanding of underlying processes or mechanisms. There may also be inadequate resources, time or expertise available to carry out full quantitative risk assessments, and there may not actually be a need in the case of most drugs. Examples of the use of defaults include the use of safety factors in calculating acceptable daily intakes (ADI) for drug residues, 10-6 as threshold risk of cancer in environmental chemical risk assessment, and assumption of the similarity of animals and humans when using bioassay data.

When defaults are used in health risk assessment, it is usual to err on the side of human safety to protect public health. One problem with this approach is the tendency to produce risk estimates that are excessively conservative, especially if there are multiple defaults used in the assessment. In any case, a full disclosure of defaults used and their empirical basis (if any) and rationale for use should be made. If the outcomes of risk assessment are acceptable to the public and regulated industries, there may not be a need to conduct additional research or otherwise attempt to refine assumptions. However, if sensitivity analysis shows that defaults have an important impact on risk estimates and the regulatory decision could be unfavorable, the regulator should be willing to consider a challenge to the default if new data or more convincing, detailed analysis is conducted.

Uncertainty is extremely important in risk assessment because its presence is one of the key reasons for doing the assessment in the first place. It is also important because critics of assessment results or the process itself often point to uncertain parameters or assumptions made as major weaknesses. This draws attention to a major risk communication issue that has been identified elsewhere (NRC, 1994) - how to simply and effectively convey the implications of uncertainty to risk managers, the public and other users of risk analysis results? Uncertainty can be an almost impossible problem if attempts are not made to segregate it into manageable components (NRC, 1994). The structured approach to risk assessment facilitates this and indeed this is one of the most important reasons for taking the formal approach.

Quantitative Modeling in Risk Assessment

Use of statistical and mathematical modeling in experiments and epidemiological studies which provide inputs to hazard identification and dose-response assessment guard against errors of inference. Probability-based sampling plans give added confidence that surveillance studies will provide representative data. Most scientists are familiar with these applications because they are commonplace in most disciplines. Statistical and mathematical models used to simulate exposure scenarios and dose-response relationships are generally less familiar to scientists working in microbiology, veterinary medicine and human medicine - food microbiologists may be an exception because modeling of factors affecting growth and survival of bacteria in foods (temperature, ph, water activity, etc.) has been used for years, which may in part explain their comparative enthusiasm and readiness to appreciate quantitative risk assessment in foods. Quantitative risk modeling of microbial pathogens in foods is still in its infancy and is generally quite complex. It also demanding of data at all stages of assessment, including hazard variables, exposure variables and variables that affect illness in people. It is demanding of time and resources and technical skills. For these and perhaps other reasons, it would be unrealistic to think that quantitative modeling could be used immediately and routinely to assess human health risks due to drug use in agriculture. In the future (hopefully not too distant) Monte Carlo and other quantitative methods could be used to model exposure scenarios, dose-response relationships and ultimately provide robust probability-based estimates of risk that reflect the variability in the system and account for uncertainty. They could also be very useful in modeling hypothetical scenarios not amenable to experimentation or observational study. These could for example provide estimates of risk reduction under various scenarios involving interventions. These could be particularly helpful in modeling risk of new drugs before they are on the market and before actual exposure data are available. Eventually they could be used to specify critical limits for control points in HACCP, prudent use or similar process control programs.

Validation of Risk Assessment

Validation of assessments should be conducted whenever it is feasible. In the case of resistance risk, a possible means of validation (or "reality check") is the use of resistance monitoring or surveillance. The results from these efforts can indicate whether model predictions (qualitative or quantitative) were accurate and also provide input to assessment in an iterative fashion. Validation of complex exposure models is problematic for the same reasons that extensive epidemiological studies tracing pathogens from the farm to consumption are rarely done - expense, methodological and logistical difficulties.

Top of Page |

Tiered Approach

In view of the technical and cost difficulties that would be encountered in undertaking exhaustive quantitative risk assessments of all drugs and all hazards, regulatory authorities should consider adopting a tiered approach (NRC, 1994). This could include, as a first tier, a conceptually simple "screening" qualitative risk assessment that perhaps is based, for example, on the drug class/human exposure classification system laid out in the FDA "Framework document" (FDA, 1998) which could include basic default assumptions which protect public health. As quantitative assessment methods develop and experience, expertise and confidence in modeling grows and matures, there could be a second and perhaps additional tiers that comprise increasingly refined, sophisticated but also expensive and data-demanding efforts to assess risk. If screening at the first tier (which because of the defaults is likely to be overly conservative) estimates indicate little or no risk, then there should be no need for further assessment. If screening leads to estimates of risk that preclude or endanger new drug approval, regulated industries should perhaps have the opportunity to invest in more complete assessment that could provide convincing evidence of lower risk.

Benefit/Cost Analysis

Benefit/cost analyses should be incorporated within the risk analysis framework to assist in making and communicating prudent decisions to manage risk. There are currently many barriers to including this type of analysis, not the least of which is the reluctance of public health authorities in many western democracies to openly acknowledge that the public health benefits of interventions may be accompanied by considerable economic costs. This is particularly difficult when the benefits (e.g. reduced incidence of drug-resistant salmonellosis in humans) and costs (e.g. reduced profitability of pig farming because of lack of approved drugs to treat pneumonia) are not borne by the same segments of society.

Conclusions

Adoption of a formal risk assessment approach to characterization of these risks would lend credibility, transparency and presumably validity to the process of science-based resistance regulatory policy development. Current barriers to its widespread use include the lack in most countries of the scientific information, expertise and experience to undertake detailed quantitative analyses, and lack of understanding and confidence in the process. Confidence is being eroded by misuse and misunderstanding of risk assessment. Some view risk assessment as a way to delay implementation of timely public health interventions; it may also be used as a "smokescreen" - to obfuscate a complex issue by the inappropriate use of complex modeling and assumptions. Time will tell whether formal risk assessment has a meaningful impact on resistance regulatory policy in Canada.

| Top of Page |

References

FDA (1998). A proposed framework for evaluating and assuring the human safety of the microbial effects of antimicrobial new animal drugs intended for use in food-producing animals. United States Food and Drug Administration.

Ferenc S, M. McElvaine and M.Miller. (1998). Risk assessment and cost-benefit issues associated with antimicrobial use in food animal production. The Role of Veterinary Therapeutics in Bacterial Resistance Development: Animal and Public Health Perspectives. The American Academy of Veterinary Pharmacology and Therapeutics. 19-21 January.

National Advisory Committee on Microbiological Criteria for Foods. (1998). Principles of risk assessment for illness caused by foodborne biological agents. J Food Prot; 61: 1071-1074.

National Research Council (1983). Risk assessment in the federal government: managing the process. National Academy Press. Washington DC.

National Research Council (1989). Human Health Risks with the Subtherapeutic Use of Penicillin and Tetracyclines in Animal Feed. National Academy Press. Washington DC.

National Research Council (1994). Science and judgement in risk assessment. National Academy Press. Washington DC.

National Research Council (1998). The use of drugs in food animal: benefits and risks. National Academy Press. Washington DC.

McEwen, S. A. (1999). Human health risk assessment of antimicrobial use in food animals. FDA Center for Veterinary Medicine. Advisory Committee Meeting. Gaithersburg, Maryland. 25-6 January.

McEwen, S.A. and R.A. Robinson. (1999). The need for risk analysis, human health indicator monitoring and benefit/cost analyses in veterinary public health. World Health Organization Study Group on Future Trends in Veterinary Public Health, Teramo, Italy. 1-5 March.

Report from the Commission on Antimicrobial Feed Additives. (1997). Antimicrobial Feed Additives. Government Official Reports 1997:132. Ministry of Agriculture, Stockholm.