Previous Page Table of Contents Next Page


1.1 Scope

This chapter summarizes the results of the FAO/WHO Risk Assessment of Salmonella in Eggs and Broiler Chickens, noting the sources used and techniques applied, and drawing out the main conclusions. Following an itemized response to the questions posed by the 33rd CCFH, a number of recommendations are made. Attention is drawn to areas where further research and data collection are required for extending the risk assessment to provide a more comprehensive and reliable tool for risk management.

1.2 History

FAO and WHO have attempted to meet the expressed needs of their member countries and the Codex Alimentarius Commission (CAC) for guidance in conducting risk assessments of Salmonella in eggs and broiler chickens. On the one hand, CAC requested scientific advice as the basis for the development of guidelines and recommendations for the management of risks posed by these microbiological hazards. On the other hand, member countries required adaptable models to use in conducting their own national assessments. A series of topic-specific modules would satisfy a great need, and particularly dose-response modules that could be adapted and used with exposure assessments within national or regional boundaries.

Figure 1. Year 1 of the FAO/WHO process for microbiological risk assessment of Salmonella in eggs and broiler chickens.

FAO and WHO developed a process for conducting microbiological risk assessments at the international level. The process incorporated essential principles regarding functional separation of risk assessment from risk management; transparency; and freedom from bias (Figures 1 and 2). At its 32nd Session, in December 1999, the Codex Committee on Food Hygiene (CCFH) prioritised pathogen-commodity combinations of concern to public health and international trade in food. CCFH identified Salmonella in eggs and in poultry as the top two priorities on their list of 21 pathogen-commodity combinations of concern in food. FAO and WHO were requested to convene ad hoc expert consultations to provide risk assessment advice.

Figure 2. Year 2 of the FAO/WHO process for microbiological risk assessment of Salmonella in eggs and broiler chickens.

In January 2000, to conduct these risk assessments, FAO and WHO mobilized an international team of scientists with documented expertise in microbiological risk assessment. The team prepared technical documentation on the hazard identification, exposure assessment and hazard characterization components of the risk assessment. These documents were reviewed and evaluated during a Joint Expert Meeting on Microbiological Risk Assessment (JEMRA) held at FAO, Rome, 17-21 July 2000. The consultation identified two important issues for discussion with the CCFH: the lack of clear risk management questions; and limitations in the usefulness of a global risk estimate. FAO and WHO presented the draft risk assessment and the report of the expert consultation to CCFH at its 33rd Session (Washington DC, 23-28 October 2000) in order to inform risk managers of the progress of the risk assessment and to seek more precise guidance on the needs of the Committee. In response to the request for further elaboration of the risk managers' questions, the CCFH provided a list of risk management questions (Tables 6 and 7).

In the next year, the team of scientists developed the risk characterization element of the assessment. A second expert consultation was held at FAO headquarters (Rome, 30 April - 4 May 2001) to review the work. The report of the Expert Consultation, which included preliminary answers to the questions posed by the CCFH, was presented to the 34th session of CCFH at its meeting in Bangkok (8-13 October 2001). The draft risk assessment was then made available for public comment and also sent for peer review by scientists in a number of countries. The risk assessment was subsequently revised and finalized.

The development of the risk assessments thus took two years, and the peer review required an additional year. Draft documents were reviewed twice by the CCFH and by two joint FAO/WHO expert consultations. In addition, peer review was employed to gather technical comments, and comments were solicited from the general public. This rigorous review process promoted transparency, as well as the involvement of all stakeholders in the process.

1.3 Objectives

The objectives of the risk assessments of Salmonella in eggs and broiler chickens were:

1. To develop a resource document of all currently available information relevant to risk assessment of Salmonella in eggs and broiler chickens and also to identify the current gaps in the data that need to be filled in order to more completely address this issue.

2. To develop an example risk assessment framework and model for worldwide application.

3. To use this risk assessment work to consider the efficacy of some risk management interventions for addressing the problems associated with Salmonella in eggs and broiler chickens.

Although a cost-benefit analysis of potential mitigations would assist risk managers in determining what measures to implement, it was not within the scope of this work and is not considered here.

1.4 Hazard identification

Salmonellosis is one of the most frequently reported foodborne diseases worldwide. International data (where available) indicate an estimated incidence of salmonellosis from 14 to 120 per 100 000 people in 1997 (Table 1). The US Centers for Disease Control and Prevention (CDC) estimate 1.4 million cases, 16 430 hospitalizations, and 582 deaths in the United States of America annually (Mead et al., 1999). Of the total number of cases, 96% are estimated to be caused by foods. Costs of foodborne salmonellosis for the United States of America population are estimated to be as high as US$ 2 329 million annually (in 1998 US dollars) for medical care and lost productivity. Over 2 000 serotypes of Salmonella have been identified, the most prevalent of which are S. Enteritidis, S. Typhimurium and S. Heidelberg.

Table 1. Estimated annual incidence of salmonellosis


Cases per 100 000 population







The Netherlands




SOURCE: Thorns, 2000.

Salmonellosis is characterized by diarrhoea, fever, abdominal pain or cramps, vomiting, headache and nausea. The incubation period ranges from 8 to 72 hours. Symptoms can last up to a week. Salmonella infections vary from mild to severe, and are occasionally fatal. Fatalities are more often seen in susceptible populations, which include infants, the elderly and the immuno-compromised. In the United States of America between 1985 and 1991, there were 54 reported S. Enteritidis outbreaks occurring in hospitals or nursing homes, accounting for 90% of all Salmonella-associated deaths, but only 12% of all cases. A small proportion of infected individuals may develop Reiter's syndrome, an arthritic disease characterized by symptoms of joint pain, eye irritation and painful urination.

Poultry have a major role as vehicles of transmission in human cases of salmonellosis. An assessment of factors affecting the prevalence and growth of Salmonella on broiler chicken carcasses would be useful to risk managers in identifying the intervention strategies that would have the greatest impact on reducing human infections. Broiler chicken is the main type of chicken consumed as poultry in many countries. A large percentage of poultry is colonized by salmonellas during grow-out, and the skin and meat of carcasses are frequently contaminated by the pathogen during slaughter and processing.

Since the late 1970s, S. Enteritidis has emerged as the major cause of salmonellosis in North America, Europe and South America. A significant increase in the incidence of S. Enteritidis infection has also been reported in Yugoslavia, Finland, Sweden, Norway and the United Kingdom. Hen eggs have become a principal source of the pathogen. The emergence of S. Enteritidis as the leading cause of human salmonellosis in many countries is attributed to this serovar's unusual ability to colonize the ovarian tissue of hens and to be present within the contents of intact shell eggs.

Most foodborne S. Enteritidis infection is associated with the consumption of raw eggs and foods containing raw eggs, such as homemade egg nog, biscuit batter, homemade ice cream, mayonnaise, Caesar salad dressing and Hollandaise sauce. In fact, 77% to 82% of S. Enteritidis outbreaks have been associated with grade A shell eggs, or egg-containing foods Undercooked eggs and products containing undercooked eggs, such as soft custards, French toast, soft-fried and poached eggs, are also significant sources of S. Enteritidis. According to a recent USFDA report, between 128 000 and 640 000 Salmonella infections are annually associated with the consumption of S. Enteritidis-contaminated eggs, and the CDC estimates that 75% of all Salmonella outbreaks are due to raw or inadequately cooked Grade A whole shell eggs.

Salmonella is transmitted to eggs by two routes: transovarian (vertical transmission) or trans-shell (horizontal transmission). In vertical transmission, Salmonella are introduced from infected ovaries or oviduct tissue to eggs prior to shell formation. Horizontal transmission is usually derived from faecal contamination on the eggshell. It also includes contamination through environmental vectors, such as farmers, pets and rodents. Vertical transmission is considered to be the major route of Salmonella contamination and is more difficult to control, while horizontal transmission can be effectively reduced by cleaning and disinfection of the environment.

1.5 Hazard characterization

1.5.1 Sources of data

FAO and WHO requested data from member countries through Codex Circular Letters. Data on salmonellosis outbreaks were obtained from a variety of sources, including published literature, national reports and unpublished data. The Ministry of Health and Welfare in Japan provided unpublished data on 16 outbreaks that the agency had investigated since 1997. This information was especially useful because it contained data on the number of organisms present in food implicated in human illness.

1.5.2 Description of the database

Of the 33 outbreak reports that FAO and WHO received, 23 contained sufficient data on the number of people exposed, the number of people that became ill, and the number of organisms in the implicated food to enable calculation of a dose-response relationship. Three of the 23 outbreaks were excluded because the immune status of the persons exposed could not be determined. The remaining 20 outbreaks formed the database used to calculate a dose-response relationship.

Of the 20 outbreaks in the database, 11 occurred in Japan and 9 occurred in the United States of America. The number of people exposed in Japanese outbreaks (»14 037; 52%) was about the same as that in United States of America outbreaks (»12 728; 48%). These numbers are approximate because in some cases the number of people exposed had to be estimated from the outbreak report. The overall attack rate in the data was 21.8% (26 765 exposed, 5 636 ill). The attack rate among Japanese outbreaks (27.4%) was higher than that of United States of America outbreaks (15.6%). This was due in part to one large outbreak in the United States of America (8 788 people exposed) with an attack rate of 11.7%, and one large outbreak in Japan (5 102 people exposed) with an attack rate of 26.9%. Several serotypes were associated with the outbreaks, including Enteritidis (12), Typhimurium (3), Heidelberg, Cubana, Infantis, Newport and Oranienburg. Several vehicles were implicated, including food (meat, eggs, dairy products and others), water, and a medical dye capsule (carmine dye).

Reports provided by the Ministry of Health and Welfare of Japan represented a valuable source of information on the real-world dose-response relationship and considerably expanded the database of Salmonella pathogenicity. The data in these reports were generated as part of the epidemiological investigations that take place in Japan following an outbreak of foodborne illness. In accordance with a Japanese notification (from March 1997), large-scale cooking facilities that prepare more than 750 meals per day or more than 300 dishes of a single menu at a time are advised to save food for future possible analysis in the event of an outbreak. The notification is also applicable to smaller-scale kitchens with social responsibility, such as those in schools, daycare centres and other child-welfare and social-welfare facilities. Fifty-gram portions of each raw food ingredient and each cooked dish are saved for more than 2 weeks frozen at a temperature lower than -20°C. Although this notification is not mandatory, the level of compliance is high. Some local governments in Japan also have local regulations that require food saving, but the duration and the storage temperature requirements vary.

1.5.3 Description of the dose-response relationship

The availability of a reasonably large data set representing real-world observations for the probability of illness upon exposure to Salmonella (outbreak data) provided a unique opportunity to develop a data-based dose-response relationship. A beta-Poisson model was used as the mathematical form for the relationship, and this was fitted to the outbreak data.

The maximum likelihood technique was used to generate the curve best fitting the data. The fit was optimized using an iterative technique that minimized the deviance statistic, which is based upon a binomial assumption.

The uncertainty in the outbreak data set was incorporated into the fitting routine by reviewing the outbreak information and assigning an uncertainty distribution on observed variables that were potentially uncertain. A detailed summary of the assumptions associated with each outbreak and the estimation for the range of uncertainty for each of the variables were described. A summary of the data set, with uncertainty for the variables, is given in Table 2.

In order to fit the dose-response model to the uncertain outbreak data, the data were re-sampled based on the uncertainty distributions, generating a new data set at each sample. The dose-response model was then fitted to each of the re-sampled data sets. This procedure was repeated approximately 5000 times, generating 5000 dose-response data sets, to which 5000 dose-response curves were fitted. The fitting procedure used places a greater emphasis on fitting the curve through the outbreaks with larger numbers of people exposed compared to the smaller outbreaks. This is primarily a result of the binomial assumption and the greater variance associated with data from a small observation compared with a large one.

Table 2. Uncertainty ranges assigned to the variables in reported outbreak data



Log Dose

Response [Attack Rate]






S. Typhimurum






S. Heidelberg






S. Cubana






S. Infantis






S. Typhimurium






S. Newport






S. Enteritidis






S. Enteritidis






S. Typhimurum






S. Enteritidis






S. Enteritidis






S. Enteritidis






S. Enteritidis






S. Eteritidis






S. Enteritidis






S. Enteritidis






S. Enteritidis






S. Enteritidis






S. Enteritidis






S. Oranienburg





NOTE: "Outbreak" refers to the number of the outbreak as listed in the main report..

It was not possible to get a statistically significant single "best fitting" curve to the expected value of all the outbreak data points. However, the characterization of the observed outbreak data by the fitted dose-response model was better than that of other published dose-response models. It is important to note that the range of possible responses at any one given dose shown in the background of Figure 3 do not represent the statistical confidence bounds of the dose-response fit, but rather the best fit of the beta-Poisson model to different realizations of the observed data, given its uncertainties.

Figure 3. Uncertainty bounds for dose response curves superimposed on the dose-response curves generated by fitting to samples from uncertain outbreak observations

Figure 3 compares the fitted curves and the expected values. The upper bound, lower bound, expected value, 97.5th percentile and 2.5th percentile for the dose-response curves fitted to the 5000 data sets are also shown. The fitted dose-response range captures the observed outbreak data quite well, especially at the lower and mid-dose ranges. The greater range at high dosages is due to the existence of several large-scale outbreaks at the lower- and mid-dose levels through which the curves attempt to pass, while the two high-dose data points are for relatively small-scale outbreaks that allow greater "elasticity" in the fit.

Table 3. Beta-Poisson dose response parameters that generate the approximate bounds shown in Figure 3.



Expected Value



Lower Bound



2.5th Percentile



Since the fitting procedure generated a dose-response curve for each of the 5000 data sets, there are also 5000 sets of beta-Poisson dose-response parameters (alpha & beta). In order to apply the dose-response relationship in a risk assessment, the ideal approach would be to randomly sample from the set of parameters that are generated, thereby recreating the dose-response curves shown. As an alternative, it is also possible to use the upper, lower, expected value, 2.5th percentile or 97.5th percentile to represent the uncertainty ranges in the dose-response relationship, as opposed to a full characterization resulting from the sampling of the parameter sets. The parameters that generate dose-response curves that approximate the bounds shown in Figure 3 of the dose-response relationship are summarized in Table 3.

In dose-response analysis, the critical region is the lower dose region, as these are the doses that are most likely to exist in the real world. Unfortunately, this is also the region where experimental data is least available. The outbreak data extend to a much lower doses than is common in experimental feeding trials, and consequently may offer a greater degree of confidence in the lower dose approximations generated by the outbreak dose-response model.

1.5.4 Analysis of the dose-response relationship

Some strains of S. Enteritidis, particularly the phage-types isolated from the increased number of egg-related outbreaks seen in recent years, may be more infectious than other serotypes of Salmonella. Twelve sets of data were evaluated for S. Enteritidis, against 8 sets of data for other serotypes. From the outbreak data used to examine the dose-response relationship, it could not be concluded that S. Enteritidis had likelihood of producing illness different from other serotypes. However, increased severity of illness once infected was not evaluated.

An attempt was made to discern whether separate dose-response curves could be justified for different subpopulations, defined on the basis of age and 'susceptibility'. Comparing the attack rates of Salmonella for children less than five years of age with those for the rest of the population did not reveal an increased risk for this subpopulation. It is important to note that the database of outbreak information may lack the power to reveal the existence of true differences. Severity could potentially be influenced by patient age or Salmonella serotype. However, the current database of information was insufficient to derive a quantitative estimate of these factors.

The dose-response model fitted to the outbreak data offers a reasonable estimate for the probability of illness upon ingestion of a dose of Salmonella. The model is based on observed real-world data, and, as such, is not subject to some of the flaws inherent in using purely experimental data. Nevertheless, the current outbreak data also have uncertainties associated with them, and some of the outbreak data points required assumptions to be made. Overall, the dose-response model generated in the current exercise can be used for risk assessment purposes, and generates estimates that are consistent with those that have been observed in outbreaks.

1.6 Salmonella in eggs

For Salmonella in eggs, the risk assessment estimates the probability of human illness due to Salmonella following the ingestion of a single food serving of internally contaminated shell eggs, either consumed as whole eggs, egg meals, or as ingredients in more complex food (e.g. cake). This work addressed selected aspects of egg production on farms, further processing of eggs into egg products, and retail and consumer egg handling and meal preparation practices.

1.6.1 Exposure assessment

The exposure assessment for Salmonella in eggs consists of a production module, a module for the processing and distribution of shell eggs, a module for the processing of egg products, and a module for preparation and consumption. The production module predicts the probability of a S. Enteriditis-contaminated egg occurring. The shell egg processing and distribution, and preparation and consumption modules predict the probability of human exposures to various doses of S. Enteriditis from contaminated eggs. The model developed combines existing models that have been elaborated at the national level. The output of the exposure assessment, in general, feeds into the hazard characterization to produce the risk characterization output. This output is the probability of human illness per serving of an egg-containing meal.

Figure 4. Schematic diagram showing the stages of the risk assessment for Salmonella in eggs.

1.6.2 Risk characterization of Salmonella in eggs

This risk characterization of Salmonella in eggs was intentionally developed so as not to be representative of any specific country or region. However, some model inputs are based on evidence or assumptions derived from specific national situations. Caution is therefore required when extrapolating from this model to other, country-specific situations.

The exposure assessment included consideration of yolk-contaminated eggs and growth of Salmonella in eggs prior to processing for egg products. These issues have not been previously addressed by exposure assessments of Salmonella in eggs. Yolk-contaminated eggs might allow for more rapid growth of Salmonella inside such eggs compared with eggs that are not yolk-contaminated.

The output of the shell egg model is the probability that a serving of an egg dish results in human illness. This probability is determined as the weighted average of all egg servings (contaminated and not contaminated) in a population. Clearly, the risk per serving is variable when we consider individual egg servings (e.g. a serving containing 100 organisms is much more likely to result in illness than a serving containing just a single organism), but the meaningful measure is the population likelihood of illness. This risk per serving can be interpreted as the likelihood of illness given a person consumes a randomly selected serving.

The range in risk of illness predicted by this model extends from at least 0.2 illnesses per million shell egg servings to 4.5 illnesses per million servings. The scenarios considered represent a diversity of situations that approximate some countries or regions in the world. Nevertheless, no specific country is intentionally reflected in this model's inputs or outputs.

Three values for flock prevalence (5%, 25% and 50%) were considered, with three levels of egg storage time and temperature (reduced, baseline and elevated).

The lowest risk of illness is predicted when flock prevalence is 5% and storage times and temperatures are reduced (Table 4). In this scenario, the calculated risk is 2 illnesses in 10 million servings (0.00002%). The highest risk is predicted when flock prevalence is 50% and storage times and temperatures are elevated. In this case, the calculated risk is 4.5 illnesses in each million servings (0.00045%).

Table 4. Predicted probabilities of illness per egg serving, reflecting different flock prevalence settings and egg storage time and temperature scenarios.

Flock prevalence

Time-temperature scenarios
















Changes in risk are approximately proportional to changes in the flock prevalence. For example, 5% flock prevalence is one-fifth of 25%. Correspondingly, the risk of illness for scenarios with 5% flock prevalence is one-fifth that of scenarios with 25% flock prevalence. Similarly, doubling flock prevalence from 25% to 50% also approximately doubles the risk of illness, if all other inputs are constant.

Although the degree of change in risk reflects change from baseline conditions, these simulations demonstrate, for example, that changing storage times and temperatures from farm to table implies disproportionately large effects on risk of illness. In addition. the calculated probability of illness per serving can be used to estimate the number of illnesses in a population. For example, a region with 100 egg production flocks of 10 000 hens each could expect about 1300 cases per year.

The output of the egg products model is a distribution of the numbers of Salmonella remaining in 4500 litre containers of liquid whole egg following pasteurization. The Salmonella considered in this output are only those contributed by internally contaminated eggs. This output serves as a proxy for human health risk until the model is extended to consider distribution, storage, preparation - including additional processing - and consumption of egg products. Figure 5 shows the output for the 25% flock prevalence, baseline scenario. About 97% of the pasteurized lots are estimated to be S. Enteritidis-free, and the average level is about 200 Salmonella remaining per lot.

Figure 5. Predicted distribution of S. Enteritidis (SE), contributed by internally contaminated eggs, remaining in 4500 litre containers of liquid whole egg after pasteurization. This distribution is predicted based on an assumed 25% flock prevalence and the baseline egg storage times and temperatures used in the model.

The risk of human illness per serving appears to be insensitive to the number of Salmonella in contaminated eggs across the range considered at the time of lay. For example, whether it is assumed that all contaminated eggs had an initial number of 10 or 100 Salmonella organisms, the predicted risk of illness per serving was similar. This may be because the effect of Salmonella growth is greater than the initial contamination level in eggs under the storage conditions considered in this model.

It should be noted that the data available upon which to base this risk assessment was limited. For example, evidence regarding enumeration of the organism within eggs was based on only 63 Salmonella-contaminated eggs, and in part on estimates of the concentration of the organism in contaminated eggs. It is difficult to represent uncertainty and variability with such limited data. In addition, neither statistical nor model uncertainty was fully explored.

1.6.3 Discussion

Although this model was deliberately configured and parameterized so that it did not reflect any specific country or region, its results might be indicative of many country situations. A generic risk assessment such as this one provides a starting point for countries that have not developed their own risk assessment. It can serve to identify the data needed to conduct a country-specific risk assessment, as well as to provoke thinking in policy development and analysis.

Control of prevalence, either proportion of flocks infected or proportion of hens that are infected within flocks, has a direct effect in reducing probability of illness per serving. On the whole, egg storage times and temperatures can disproportionately influence the risk of illness per serving. The numbers of organisms initially in eggs at the time of lay seems less important.

Testing flocks, combined with diversion of eggs from positive flocks, is predicted to reduce public health risk substantially. In the scenarios considered, diversion of eggs from test-positive flocks also reduced the apparent risk from egg products. Vaccination might reduce risk of illness by about 75%, but is typically less effective because producers would only vaccinate test-positive flocks.

Biological inputs may be constant between models for different countries or regions, yet little else is likely to be similar. The predictive microbiological inputs, the distribution of within-flock prevalence, and the frequency at which infected hens lay contaminated eggs are examples of biological inputs that might be constant (although not necessarily), and the effect of uncertainty regarding these biological inputs to the model was considered. Nevertheless, there are many aspects of uncertainty not fully considered (e.g. alternative statistical distributions were not evaluated for the predictive microbiology equations or within-flock prevalence distributions). Furthermore, many of the inputs are both highly uncertain and variable within and between countries (e.g. times and temperatures of egg storage may vary considerably). It is difficult for any country to know precisely its distributions for storage times and temperatures.

The model introduces two new concepts not included in previous exposure assessments of Salmonella in eggs. First, it considers the possibility of eggs being laid with Salmonella already inside the yolk. Such eggs defy previous models' description of the time and temperature dependence of Salmonella growth in eggs. Although predicted to be uncommon, yolk-contaminated eggs can support rapid growth of Salmonella in much shorter times than eggs contaminated in the albumen.

Second, this model considers the role of Salmonella growth in eggs destined for egg products. While most eggs are modelled as being shipped very quickly to egg products plants (i.e. nest run eggs), some eggs can experience moderate or high levels of growth before being broken and pasteurized.

Many of the results generated by this model are contingent on epidemiological assumptions, namely:

These may be reasonable default assumptions, but more research is needed to determine their appropriateness. Changing these assumptions can generate different results from the model, although the model can be adapted to consider such changes.

1.7 Salmonella in broiler chickens

For Salmonella in broiler chickens, the risk characterization estimates the probability of illness in a year due to the ingestion of Salmonella originating from fresh whole broiler chicken carcasses with the skin intact and which are cooked in the domestic kitchen for immediate consumption. Due to a lack of suitable data, particularly enumeration data, this work commenced at the conclusion of slaughterhouse processing (i.e. at the end of stage 2 in Figure 6) and considers in-home handling and cooking practices. The effects of pre-slaughter interventions and the slaughter process are not currently included in this model.

Figure 6. Modular pathway to describe the production-to-consumption pathway for broiler chickens. Each step describes the changes to prevalence (P) and numbers of Salmonella (N) that occur within that specific module.

1.7.1 Exposure assessment

The exposure assessment of Salmonella in broiler chickens mimics the movement of Salmonella-contaminated chickens through the food chain, commencing at the point of completion of the slaughter process. Based on an assumed level of infection, a chicken carcass was allocated an infection status and those carcasses identified as contaminated were assigned a number of Salmonella organisms on each iteration of the model, using available data. From this point until consumption, changes in the size of the Salmonella population on each contaminated chicken were modelled using equations for growth and death. The growth of Salmonella was predicted using inputs for storage time in retail stores, transport time, storage time in homes, and the temperatures the carcass was exposed to during each of these periods. Death of Salmonella during cooking was predicted using inputs describing the probability that a carcass was not adequately cooked, the proportion of Salmonella organisms attached to areas of the carcass that were protected from heat, the temperature of exposure of protected bacteria and the time for which such exposure occurs. The number of Salmonella consumed were then derived using an input defining the weight of chicken meat consumed per serving and the numbers of Salmonella cells in meat as defined from the various growth and death processes. Exposure via cross contamination was modelled as well as exposure as a result of consuming undercooked chicken. In particular ingestion of organisms transferred from the raw poultry and onto hands and uncooked food was described using transfer and frequency rates. Outputs from the model relate to exposure via undercooked chicken and exposure via cross-contamination. In both cases, the probability of the event and the numbers of organisms are output.

The exposure assessment is defined in terms of a number of parameters that describe the processes of broiler chicken carcass distribution and storage, preparation, cooking and consumption. Some of these parameters can be considered general in that they can be used to describe the situation in many countries. In contrast, some parameters are country specific, such as the prevalence of carcasses contaminated with Salmonella at the completion of processing. Predictions of risk for a particular country are best obtained from data relevant to that country.

It should also be noted that, in the course of the work, efforts were made to identify features that have an impact on the validity of findings and the appropriateness of extrapolating findings to scenarios not explicitly investigated in the risk assessments. These are identified and discussed in the risk assessment document.

1.7.2 Risk characterization of Salmonella in broiler chickens

In the risk characterization stage, the outputs from the exposure assessment were combined with the dose-response model to produce two estimates of risk: the risk per serving and the risk via cross-contamination. The risk estimates for probability of illness were first derived using a set prevalence for the presence of Salmonella in chilled, raw broiler chickens. At a prevalence of 20% contaminated carcasses (the baseline case), and based on the other model parameters, including the probability that the product would be undercooked, approximately 2% of the broilers prepared for consumption in the home could potentially contain viable cells of Salmonella. Figure 6 shows the distribution of average doses (colony-forming units (CFUs)) per serving for contaminated chicken that is subsequently undercooked.

Since the data shown in Figure 7 represent the average dose per serving, the interpretation of values of less than 1 CFU per serving is 1 CFU per multiple servings, e.g., an average dose of 0.01 cells per serving translates to one in 100 servings contains a single cell.

Figure 7. Average dose (colony-forming units (CFUs) of Salmonella) per serving in meals prepared from contaminated broiler chickens.

Figure 8. Distribution of average risk per serving.

Assuming a 20% prevalence of contaminated broilers, the estimated frequency and cumulative distribution of average risk per serving are shown in Figure 8. The expected risk per serving is 1.13E-5, or 1.13 illnesses per 100 000 servings. This value represents the average risk for all individuals in the population that consume servings of chicken that are stored, transported and prepared in the manner described in the model, and also accounts for the probabilities that the serving was from a chicken contaminated with Salmonella, and that the meal was undercooked. It should be recognized that some individuals consuming a serving on certain occasions would experience a much higher risk than others who might be consuming servings with no salmonellosis risk at all, since the serving would be free of the pathogen.

The expected risk per serving can be extended to the expected risk over multiple servings, such as meals consumed in a year. If it is assumed that the risk posed by one exposure (serving) is statistically independent of any other exposure (serving), then the overall risk of infection following a series of exposures (annual risk) can be estimated from risk of infection per exposure (daily or serving risk). In order to estimate the annual risk of infection, two items of information are required: the risk of infection per serving, and the number of servings consumed in a year. The calculation of annual risk based on the estimated average per serving risk and the assumptions for this baseline scenario are shown in Table 5.

Table 5. Calculation of expected annual risk.

Prevalence of contaminated carcasses


Expected risk per serving

1.13E-05 (1.13 illnesses per 100 000 servings)

Number of servings in year


Annual expected risk

2.94E-04 (2.94 illnesses per 10 000 servings)

Rate of illness per 100 000


Illustrative calculation for annual expected number of illnesses for a country or region with this annual expected risk:


20 000 000

Proportion of population that eats chicken


Potentially exposed population

15 000 000

Expected number of cases in the year


The assumptions inherent in the calculation in Table 5 are that each of the servings consumed during the year have the same expected risk per serving, and that the risk from each exposure is independent of every other exposure. The annual risk was estimated using the assumption that 26 chicken meals were consumed in a year, i.e. chicken was consumed once every 2 weeks. For illustration, a population risk for 20 million individuals was considered, with 75% of that population eating chicken. In this case, the total expected number of salmonellosis cases arising from the model assumptions is estimated to be 4400, equivalent to a rate of 29 cases per 100 000 population. Obviously these statistics need to be tailored for a specific country or region.

In addition to estimating the risk per serving based on consumption of undercooked poultry, the assessment also modelled the risk from cross-contamination. The sequence and nature of events that need to occur in order for the bacteria on raw chicken to be disseminated and ingested via other pathways is complex and difficult to model completely. There is a lack of information to adequately describe cross-contamination, but it is acknowledged that this is an important route for foodborne illness. The following estimates offer an approximation of the magnitude of the problem, although not all potential pathways that could result in exposure and illness were modelled.

In the baseline scenario, the expected risk from cross-contamination (transfer from raw chicken to hands to non-cooked foods, or from raw chicken to cutting board to non-cooked foods) was estimated to be 6.8E-4, or 6.8 illnesses per 10 000 exposures to contaminated material. This is more than an order of magnitude higher than the expected risk from a serving. This estimate is a function of two factors (conditional probabilities) in the current model: first, the expected risk when the event occurs, and, second, the expected probability that the event occurs.

The expected probability that the event occurs is driven by the prevalence of contamination plus the probability of undercooking in the case of consumption, versus the prevalence of contamination plus the probability of not washing hands or not washing cutting boards in the case of cross-contamination. Given the assumptions made in the model, the expected risk from this cross-contamination pathway is equivalent to approximately 60 chicken consumption exposures. Although the frequency with which people do or do not wash their hands can be debated, the ultimate risk from cross-contamination could in fact be even higher than that estimated here, since there are multiple cross-contamination opportunities in the home preparation environment.

The risk assessment of Salmonella in broiler chickens does not consider all parts of the production-to-consumption continuum and this limits the range of control options that can be assessed. This is primarily due to the lack of representative data to analyse how much change in either the prevalence or level of Salmonella, or both, in poultry could be attributable to any specific treatment or action. However, the establishment of a baseline model provided a means to compare the effects on risk when prevalence and cell numbers were changed. The model parameters can be modified to evaluate the efficacy of risk mitigation strategies that target those parameters. For example, the parameter describing prevalence of Salmonella-contaminated broiler chickens exiting processing can be modified to evaluate the effectiveness of a processing measure such as chlorination of the chilling water to reduce the prevalence of Salmonella-contaminated carcasses.

Reduction in the prevalence of Salmonella-contaminated chicken was associated with a reduction in the risk of illness. A one-to-one relationship was estimated, with a percentage change in prevalence - assuming everything else remains constant - reducing the expected risk by a similar percentage. For instance, a 50% reduction in the prevalence of contaminated poultry (20% to 10%) produced a 50% reduction in the expected risk of illness per serving. Similarly, a large reduction in prevalence (20% to 0.05%) would produce a 99.75% reduction in the expected risk of illness, a risk reduction that perhaps might be obtained by pre-slaughter risk management actions.

If management strategies are implemented that affect the level of contamination, i.e. the numbers of Salmonella on chickens, the relationship to risk of illness is estimated to be greater than a one-to-one relationship. A shift of the distribution of Salmonella cell numbers on broiler chickens exiting the chill tank at the end of processing, such that the mean number of cells is reduced by 40% on the non-log scale, reduces the expected risk of illness per serving by approximately 65%.

A small reduction in the frequency of undercooking and the magnitude of the undercooking event results in a marked reduction of the expected risk of illness per serving. The important caveat here is that altering cooking practices does not address the risk of illness through the cross-contamination pathway. Any strategy to change consumer cooking practices needs to be tempered by the fact that cross-contamination may in fact be the predominant source of risk of illness, and it must be remembered that the nature of cross-contamination in the home is still a highly uncertain phenomenon.

1.7.3 Summary and recommendations

To date, no full exposure assessments have been undertaken of Salmonella in broiler chickens. This report has therefore considered:

The following recommendations are made for directing future work:

(i) Reporting of prevalence at different steps of the full exposure pathway should be encouraged in all regions of the world.

(ii) Reported data should give full details of study methodology, including sampling site, sampling time, how the sample relates to the overall population, and microbiological methods.

(iii) Determination of quantitative data should be encouraged and, if it becomes available, then full exposure assessments could be developed to investigate mitigation strategies (e.g. use of chlorine in chill water) or to compare alternative practices (e.g. air chilling versus immersion chilling).

(iv) Cross-contamination during processing and handling operations should be studied quantitatively, and methodologies for modelling this process developed. Cross-contamination during these stages is a critical factor, which is often associated with outbreaks.

(v) At the national level, the collection of consumption data should be promoted. The design of such studies should accommodate the data requirements for exposure assessments. These requirements include population variability, portion size, and frequency of consumption.

(vi) In predictive microbiology, the area of survival has been less well studied than growth or death. There are few predictive models that describe survival at chill and frozen temperatures, and so further development of such models is essential.

1.8 Overall key findings

One of the important outcomes of the risk assessment work was the compilation and collation of a wealth of information on Salmonella and broiler chickens, and Salmonella associated with eggs. The organization of these data in the structured risk assessment format has led to the identification of significant gaps in the existing data. This provides a guide for future research work to help ensure that such research targets the generation and collection of the most useful and relevant data.

In general the following conclusions could be drawn from the hazard characterization work.

Previous Page Top of Page Next Page