What is the difference in the per-portion microbiological risk to UK consumers from minced meat and meat preparations intended to be eaten cooked, from UK cutting plants producing less than two metric tonnes per week, compared to those producing more than two metric tonnes per week?

This risk assessment assesses the risk to consumers posed by minced meat and meat preparations produced from animal species including beef, lamb, pork, venison, chicken and game, where data are available. The focus is on establishments with an average throughput of less than two tonnes per week (“low throughput”), which are currently exempt from the specified microbiological sampling in assimilated Regulation (EC) No 2073/2005 (Regulation (EC) No 2073/2005 in Northern Ireland) (Gov.UK, 2005a).

Regulation 2073/2005 sets out the sampling frequencies for undertaking microbiological testing against food safety criteria (FSC) for Salmonella spp., and process hygiene criteria (PHC) for E. coli and aerobic colony counts (ACC), with exemptions permitted in some cases. Chapter 3.2 of the regulation sets out that when justified on the basis of risk analysis and consequently authorised by the competent authority, small establishments producing minced meat and meat preparations in small quantities may be exempt from the specified sampling frequencies (Gov.UK, 2005b).

Exemptions and flexibilities from the specified sampling frequencies have been applied to smaller Food Business Operators (FBOs) in the UK, since at least January 2006, when the first version of the FSA’s Meat Industry Guide (MIG) was published. The MIG was withdrawn in November 2020; however, these sampling frequencies are now set out in Chapter 4.3 of the Manual for Official Controls (MOC) (FSA, 2024a). One of the exemptions applies to establishments producing an average of less than two metric tonnes per week of combined minced meat and meat preparations intended to be eaten cooked, which are currently not required to undertake any sampling.

As a result of an internal audit by Operational Assurance in 2023, it was agreed to review the rationale for the exemptions. This assessment compares the per portion risk from low-throughput establishments with that associated with the same product types produced in establishments averaging more than two tonnes throughput per week, which are required to undertake microbiological sampling.

This risk assessment aims to establish the evidence base to assess whether the exemption for establishments producing an average of less than two tonnes per week of combined minced meat and meat preparations intended to be eaten cooked remains justified.

The assessment focuses on comparing the relative risk between throughput categories, rather than characterising the absolute risk within each category. The lack of comprehensive information on the total throughout per establishment size prevented a more comprehensive category-specific risk characterisation and limited the extent to which conclusions could be generalised to a population basis.

In scope:

Out of scope:

Key Assumptions

For the purposes of this risk assessment, it is assumed that post-production handling of the meat products is the same irrespective of the size of the establishment in which they were produced. Post-production handling includes transport, storage, pre-cooking preparation, and cooking.

Microbiological criteria provide benchmarks for assessing the microbiological quality of food products and verifying the correct functioning of manufacturing processes. Although microbiological testing on its own cannot guarantee that a tested food is safe, these criteria serve as key targets and reference points for FBOs and competent authorities to manage and monitor food safety (European Commission, 2005). In the UK, the FSA is the national authority responsible for food safety and food hygiene across England, Wales and Northern Ireland, while Food Standards Scotland (FSS) is the food safety authority in Scotland.

Assimilated Regulation (EC) No 2073/2005 (as amended) establishes the microbiological criteria applicable to specific micro-organisms, along with the implementing rules that FBOs must comply with when applying the general and specific hygiene requirements set out in Article 4 of Assimilated Regulation (EC) No 2073/2005 (Gov.UK, 2005b).

Any establishment that cuts meat and then places it on the market as fresh meat must be approved as a cutting plant and is therefore subject to official veterinary controls, with certain exemptions for retail premises selling directly to the final consumer, e.g. retail establishments that supply cut meat directly to the consumer (FSA, 2024a). In practice, that means butcher shops or supermarkets that cut meat on site and sell directly to consumer are exempt from a cutting plant approval requirement. Regulation (EC) No 853/2004 (EC) sets out the hygiene requirements that apply to all approved establishments where meat cutting takes place (FSA, 2024a; FSS, 2018).

When implementing general and specific hygiene measures described in (EC) No 853/2004, the FBO must follow assimilated Regulation (EC) No 2073/2005 (as amended) which sets out the microbiological criteria for specific microorganisms. This regulation defines two distinct categories of criteria: food safety criteria (FSC) and process hygiene criteria (PHC) (Gov.UK, 2005a).

Food safety criteria (FSC) evaluate the safety of the final product. Failure to meet an FSC requires that the affected batch must not be placed on the market (withdrawal) or, if already distributed, must be removed from the market (recall). Any FSC established must be relevant to protecting consumer health and practical to implement. They are developed in line with internationally recognised principles, such as those of the Codex Alimentarius (Codex, 2025), and are based on formal risk assessments whenever possible.

Process hygiene criteria (PHC) are set to assess the hygienic performance of the production process. They indicate whether slaughter, dressing and/or processing operations are carried out hygienically at the time of sampling and whether the food safety management systems in place are effective. If PHC is not met, the FBO must review and improve their production processes (FSA, 2024a).

During the production of minced meat, meat preparations and meat products, any bacteria, including pathogenic bacteria, present on the surface of the carcase meat can be spread throughout the product. Additionally, contamination may be introduced from the environment, or through handling and processing failures (FSS, 2018). Compliance with PHC is therefore critical for food safety. PHC results are used to verify whether FBOs’ production processes and their Hazard Analysis and Critical Control Points (HACCP) systems are functioning effectively, and ensuring compliance with regulatory requirements (FSA, 2024a). Where PHC results are unsatisfactory the FBO must take action to improve production hygiene and selection of raw materials.

In the UK, PHC are set for faecal and other forms of bacterial contamination in food. PHC apply only to food businesses involved in manufacturing, preparing, or producing the specific food in question. They are set for products at defined stages of the production process and do not apply to foods that have already been placed on the market.

Table 1 summarises the microbiological criteria applicable to minced meat and meat preparations under assimilated Regulation (EC) No 2073/2005 Annex I. For high-throughput establishments, FBOs shall take samples for microbiological analysis at least once a week; for E. coli and ACC, the frequency may be reduced to fortnightly testing if satisfactory results are obtained for 6 consecutive weeks. A similar approach may be taken for testing for Salmonella spp. if satisfactory results are obtained for 30 consecutive weeks or where an effective national or regional control programme is in place and demonstrates low prevalence (EU, 2005; FSA, 2005). In the UK, establishments producing an average of less than 2 metric tonnes per week of combined minced meat and meat preparations intended to be eaten cooked are currently not required to take any samples (FSA, 2024a).

Raw fresh meat has typically acidic to neutral pH of 5.3-5.8, and water activity (a_w) of 0.56-0.73 (USDA, 2025). Salmonella spp. are highly adaptable bacteria capable of growing across a broad environmental range. They multiply between 8-45 C (optimum 37 C), tolerate pH 4.0-9.5 (optimum 6.5-7.0), and can survive in low water activity (<a_w 0.94) environments (Chlebicz et al., 2018). Therefore, if Salmonella spp. are present in meat, its physicochemical properties would support the pathogen’s survival.

This risk assessment used the results from the FSA’s sampling survey FS900516: ‘A survey of Salmonella spp., E. coli and aerobic colony count (ACC) in minced meat and meat preparations from cutting plants in England, Wales and NI’, carried out between 7^th October to 15^th November 2024. The survey was designed to address a data evidence gap to inform a strategic risk assessment on process hygiene criteria. Samples of minced meat and meat preparations were collected over a 6-week period, from both high-throughput and low-throughput meat cutting plants. The plants were based in England, Wales and NI and were visited by unannounced FSA inspectors. Each plant was visited once a week, and 5 samples were collected per visit as described in the legislation (Gov.UK, 2005a), either of different animal species, if more than one was processed, or of different batches if only one species was processed. Samples were collected at random, prior to packaging, then sent to the laboratory within 24 h from sampling and tested within 24 h of receipt. The survey was carried out to establish new baseline levels for these microorganisms within the food products in scope, to inform an FSA assessment of the relative risk to consumers from products from cutting plants processing two metric tonnes of meat a week or less, which are currently exempted from sampling.

Data provided by FSS from establishments approved by FSS was also analysed. This dataset included results of sampling by FBOs in Scotland which are either non-exempt or which might be exempt from testing (those under two-tonne production per week) but choose to sample.

The number of establishments for which data were available in both datasets can be found in Table 2. The number of total approved establishments is also included in the table for context. The number of samples taken per establishment size and animal species can be found in Table 3.

The dataset contained limited or no sampling results for venison and game, and therefore only animal species such as beef, pork, lamb and chicken (chicken) have been analysed and discussed in this risk assessment. Similarly, data were not sufficient to compare mixed or other (buffalo and venison) products. Unknown products could not be classified by species and were not included in the analysis either.

Data were available for England, Wales, Northern Ireland and Scotland. The data were combined for analysis. The variables listed in Table 4 were considered for suitability for the analysis.

The dataset contained a total of 3,285 datapoints (number of test results for all samples, including for ACC, E. coli and Salmonella spp.). After filtering out poorly defined species such as “Mixed”, the datapoints available dropped to 3,015. The results_unit variable also contained some blanks which were filtered out to avoid ambiguity. Therefore, the total number of datapoints used in the analysis was 2,717.

Not all data and metadata were available for all of the datapoints. Censored data (observations where the exact value is only partially known, either because it falls above, below, or within a certain range, such as when a value is below the limit of quantification (LOQ) or limit of detection (LOD) of the test used) were treated using substitution based on the analyte in question. This is explained in more detail in the respective analyte sections.

The objective was to compare microbiological results (ACC, E. coli, Salmonella spp.) across different cutting plant sizes (≥2 tonnes/week vs <2 tonnes/week).

The following explanatory variables were considered:

Data obtained from FSA’s sampling survey FS900516 and from Scottish FBOs were not balanced in the same way (different data distribution across species, establishment sizes, etc.) when descriptive statistics were examined before considering data source as a variable in the models, and this was taken into account when the data was analysed.

Testing of minced meat and meat preparation samples for Salmonella spp. detection, enumeration, and serotyping was carried out in accordance with ISO 6579-1 using a horizontal method (a standardised procedure applicable across different food matrices rather than tailored to a specific product type). Following confirmation of Salmonella spp. isolates, serotyping was performed by determining the presence or absence of specific O-, H- and Vi-antigens.

For Salmonella spp., an initial mixed-effects logistic regression model was explored to assess associations with establishment throughput while accounting for batch-level clustering. The logistic regression was chosen because Salmonella spp. is in the form of binary data (“Detected” and “Not Detected”). Batch number was included as a random effect to account for non-independent repeated samples within a batch. However, due to the very low number of positive detections, the model provided limited scope for reliable inference. Consequently, a simpler batch-level pass/fail approach was used. Each batch was classified as Fail if it contained at least one positive result and Pass if it contained no positive results for Salmonella spp.

A species (beef, pork, lamb and chicken) stratified descriptive analysis was conducted to compare batch failures for Salmonella spp. between high-throughput and low-throughput cutting plants. Additionally, a Fisher’s exact test was conducted for each species to determine if there is enough evidence of a difference between high- and low-throughput cutting plants. Odds ratios (ORs) and confidence intervals (Cis) were also calculated to give extra context for the magnitude and uncertainty of this difference.

Due to the low Salmonella spp. detection rates and small effective batching sizes, statistical power was limited.

Enumeration of ACC was based on colony count at 30 °C by the pour plate technique, according to ISO 4833-1:2013.

For ACC, a linear mixed-effects model was fitted to log₁₀-transformed results, with data source, throughput, species, throughput and species interaction included as fixed effects. Batch number was included as a random effect to account for non-independent repeated samples within a batch. The linear model was chosen because ACC data is continuous. Further species-specific analysis was performed using estimated marginal means (EMMs) to determine which species showed significant difference between high- and low-throughput establishments. EMMs were calculated from the fitted model to compare mean ACC levels between high- and low-throughput establishments.

Descriptive checks suggested that FS900516 survey data (FSA) and Scottish FBOs data (FSS) were not distributed in the same way across species and throughput. Therefore, source (FSA or FSS) was added as an extra explanatory variable. Where censored ACC values were reported, upper-censored values were replaced with their stated limits (e.g. >3,000,000 is given as 3,000,000) and lower-censored values were replaced with half their stated limits (e.g. <10 is given as 5). This approach was considered acceptable for ACC because the proportion of censored data was low (9.4%), and any assumptions made with these conversions expected to have minimal impact on the results. A sensitivity analysis was additionally performed excluding all censored observations to assess the robustness of the conclusions.

The E. coli enumeration method was based on a horizontal method for the enumeration of beta-glucuronidase-positive Escherichia coli, according to ISO 16649-2:2001 (colony count technique at 44 °C using 5-bromo-4-chloro-3-indolyl beta-D-glucuronide).

For E. coli, applying the same analytical approach as for ACC was not considered appropriate across all species because the number of results below the LOQ differed substantially between chicken and non-chicken species. Non-chicken (beef, pork and lamb) animal species had very high levels of censoring, with approximately 85-90% of non-chicken results below the LOQ, predominantly as <10 cfu/g. In contrast, chicken data had little to no censoring, as samples were commonly contaminated. Therefore, E. coli was analysed separately for chicken and non- chicken species.

For non-chicken, results were converted to binary outcomes based on whether a result was censored (reported as <10 cfu/g) or uncensored (reported as numerical values). Some results were reported as less than 100 cfu/g (2.9%), rather than less than a 10 cfu/g, so these were excluded from this binary classification to avoid ambiguity with respect to the less than a 10 threshold.

A mixed-effects logistic regression model was fitted with throughput, species and their interaction included as explanatory variables. Batch number was included as a random effect to account for repeated samples within batches. Source was not included as a fixed effect as the proportion of censored data for non-chicken E. coli was very high, leading to less information in the non-censored part of the data. Adding another fixed effect would have reduced model stability/interpretability. Secondary analysis was performed to check the effects of adding source as a fixed effect. Further species-specific analysis was performed using EMMs from the fitted model. Within each species, comparisons between high- and low-throughput establishments were evaluated, consistent with the approach used for ACC.

For chicken, where censoring was minimal (1/56 samples), a linear mixed-effects model was fitted with log₁₀-transformed results with throughput as a fixed effect and batch as a random effect. Source was not included as a fixed effect for the chicken analysis as the data from the FSS source had no chicken present.

It is important to note that because the censoring differed greatly between non-chicken and chicken, the E. coli analysis addressed related but distinct questions depending on species. The non-chicken analysis compared the likelihood of results being below the LOQ (<10) with measurable values, whereas the chicken analysis compared differences in measured E. coli levels. Although both assess potential differences between low-throughput and high-throughput establishments, they do so in different ways and should not be interpreted as directly comparable tests.

The data was analysed with the R programming language, version 4.5.1. The tidyverse, emmeans, logistf, lme4, lmerTest, and Matrix suite of packages were used (D. Bates et al., 2015, 2026; Heinze et al., 2025; Kuznetsova et al., 2017; Lenth et al., 2026).

Data was visualised with the R programming language, version 4.5.1, with tidyr, dplyr, stringr, ggplot2, broom, scales, and patchwork suite of packages (Pedersen, 2025; Wickham, 2016; Wickham et al., 2019, 2025b, 2025a).

For the purposes of this risk assessment, it was assumed that the handling of meat products after they leave the cutting plant is similar regardless of the size of the plant in which they were produced. The assessment therefore assumes a reasonably foreseeable and most likely consumer exposure scenario, with Salmonella spp. contamination at the point of dispatch as a starting point, as Salmonella spp. is the only foodborne pathogen included in this assessment.

This exposure scenario considers consumer purchasing habits, the potential for temperature abuse during transport, storage and handling at a domestic setting, the probability of cross-contamination during meal preparation, typical consumption patterns, population immunity to Salmonella spp. with expected clinical outcomes, and national data on reported Salmonella spp. occurrence in animals. As there is no evidence to suggest that these post-processing factors vary systematically between meat originating from different processing plant sizes, consumer risk is assumed to be directly related to the level of contamination present at the point the product leaves the cutting plant.

In the following sections it is explained why post-processing handling was assumed to be a constant for the purposes of this risk assessment.

Typical symptoms of Salmonella spp. infection can vary from asymptomatic carriage to symptoms that include abdominal pain and diarrhoea, sometimes accompanied by vomiting, nausea, fever, chills, muscle or joint pain, cramps, and reduced appetite (Chlebicz et al., 2018). After symptoms resolve, carriage may persist for up to 4 weeks in adults and 7 weeks in children. However, some people including young children less than 5 years of age, adults aged 65 years and older, or those with weakened immune systems, are at higher risk of developing more severe illness or longer-term complications(UKHSA, 2025b).

The incubation period ranges from 6-72 hours, depending on dose and host factors, and illness typically resolves within 7 days (Chlebicz et al., 2018; Eikmeier et al., 2018). The infectious dose can vary widely, usually falling between 10⁶-10⁸ cells, but in highly susceptible individuals, as few as 10 cells may trigger disease (Chlebicz et al., 2018). The presence of Salmonella spp. in food (especially high in high fat content) is particularly hazardous to children and other high-risk populations, as these foods can protect the pathogen. during passage through the stomach. Even low doses (perhaps as few as 10 to 100 cells) pose a significant risk to health (Podolak et al., 2010).

The aerobic colony count (ACC), also known as the total viable count, is an indicator of the overall microbiological quality of a product rather than safety, and cannot generally be used alone to assess the presence or absence of foodborne hazards (UKHSA, 2024). Elevated ACC values are unlikely to directly cause disease in consumers; however, they provide contextual information on the hygienic conditions under which meat has been processed, handled or stored.

In the context of hazard characterisation, ACC results can signal underlying issues such as inadequate hygiene practices, poor temperature control, prolonged storage or reduced shelf life. For this reason, ACC testing is often done alongside pathogen testing in labs, and it is used to support consistent interpretation of overall microbiological findings (UKHSA, 2024). However, ACC levels exceeding the guideline thresholds cannot, on their own, justify product recalls or legal action; instead, they should be used to highlight possible weaknesses in food processing, preparation, storage, or shelf life (UKHSA, 2024). Therefore, within this assessment, ACC results are used to contextualise Salmonella spp. results, which is the only pathogen in scope, i.e., ACC results were used to provide general context on overall hygiene and microbial levels when interpreting Salmonella spp. findings.

Generic E. coli is used in food microbiology as an indicator of faecal contamination, rather than a direct measure of consumer health risk. While their detection alone does not necessarily indicate that a food product is unsafe, it usually signals a breakdown in food safety controls and an increased likelihood that faecal pathogens may be present in the product. Detection of E. coli therefore raises concerns about potential contamination (UKHSA, 2024).

Evidence supports the association between the detection of E. coli with detection of foodborne pathogens. For example, McLauchlin et al. (2018) reported that for all samples investigated in their research (curry, betel and banana leaves), where Salmonella spp. was detected, unsatisfactory or borderline levels of E. coli were also found, with significant association between the presence of Salmonella spp. and increasing levels of E. coli (P<0.001; OR=1.72, 95% CI 1.31–2.25) (McLauchlin et al., 2018).

Therefore, within this assessment, generic E. coli results are used to contextualise Salmonella spp. results, which is the only pathogen in scope.

Salmonella spp. positive samples were more frequently reported in low-throughput compared to high-throughput establishments when results were pooled across all animal species, with failure rates of 4.5% and 1% respectively, although this result was not statistically significant (Fisher’s exact test, p=0.068). Additionally, the estimated odds of Salmonella spp. batch failure were approximately 4.8 times higher in low-throughput establishments than in high-throughput ones (OR=4.82). However, due to the low failure rates, the corresponding 95 % CI (0.68 to 54.16) was very wide and included 1, which indicates uncertainty and no statistically significant difference between throughput groups.

No statistically significant difference was found for indicator E. coli between establishments of different sizes as defined in this assessment. Finally, while there was no overall significant difference between low- and high-throughput establishments on ACC levels, species-specific differences were detected: higher ACC levels were recorded in beef and pork products from high-throughput establishments, but lower in lamb products (not statistically different). No statistically significant difference was observed for chicken products.

When considered together, the ACC and indicator E. coli results suggest that differences observed in overall microbial loads (ACC) are not mirrored by corresponding differences in faecal contamination indicators. This pattern is consistent with the interpretation that elevated ACC levels, where observed, were more likely driven by general hygiene, process or storage control failures rather than failures in faecal contamination controls. The absence of throughput-related differences in indicator E. coli supports conclusions such as that core hygiene barriers relevant to pathogen introduction, particularly those linked to gut-derived contamination, were broadly similar across establishments of different sizes. Any species-specific differences in ACC levels between establishments of different sizes are likely to reflect differences in the processing steps and risk mitigation measured implemented by those establishments.

Sampling itself does not directly reduce public health risk. Any public health risk is minimised only when unsatisfactory process hygiene criteria (PHC) results lead to the implementation of appropriate corrective actions by food business operators. Such actions may include the evaluation of monitoring data over time to identify adverse patterns, followed by process improvements intended to reduce future contamination levels. In establishments exempt from testing, no equivalent trigger is present, although interventions may still be implemented for other operational reasons, for example following process modifications, identification of hygiene or equipment issues, internal reviews, or outcomes of official inspections. Implementation of PHC requirements, as described in the legislation may also be tightened or revised in response to outbreaks, rising disease incidence, and public health concern.

Consumption patterns vary by both animal species and age group, with consumers of all ages consistently reporting more frequent beef minced consumption compared to products from other animal species, reflecting its relatively high consumption across much of the population. In contrast, poultry minced consumption is very low compared with other minced meats and therefore contributes minimally to overall exposure. Whilst consumption of undercooked minced meat products does occur, it is uncommon and it largely confined to certain adult sub-groups.

Factors such as cross-contamination during food preparation and temperature abuse during transport storage or handling can increase microbial growth, but there is no evidence to suggest that these consumer level behaviours differ systematically according to the size of the cutting plant of origin. As a result, these factors were treated as a constant in this comparative risk assessment.

Importantly, adequate cooking (core temperature to 70 °C for at least two minutes or equivalent) is sufficient to eliminate microbial cells present in food, pathogenic or otherwise.

Livestock surveillance data indicate heterogeneous trends in Salmonella spp. occurrence across animal species and despite general decline in prevalence in recent years, human salmonellosis cases show an increasing trend. These trends suggest that mitigations successfully implemented at the primary production level do not necessarily translate into a reduction in the burden of disease. As FSA and FSS incident data indicate, meat and meat products remain prominent in food safety incidents and outbreak investigations, with such products frequently implicated, reflecting their susceptibility to contamination and the opportunities for bacterial growth and cross-contamination along the supply chain.

On the basis of the available data considered in this risk assessment, there was insufficient statistical evidence to demonstrate a difference in the per-portion likelihood of microbiological exposure of UK consumers to Salmonella spp. from minced meat and meat preparations intended to be eaten cooked, originating from UK cutting plants producing less than two metric tonnes per week, in comparison to those producing more than two metric tonnes. However, the low number of positive Salmonella spp. samples and failing batches resulted in substantial uncertainty and limited statistical power, meaning that a true difference in risk between throughput groups cannot be excluded. The uncertainty associated with this risk characterisation is High.

Statistically significant species-specific differences were detected for ACC contamination, an indicator of hygiene or temperature process failures. However, those did not translate into a consistent overall throughput size effect, i.e., while contamination levels differed by animal species, there was no clear pattern showing that either size plants consistently had higher or lower contamination. These results are more likely to reflect species-specific processing steps and variability in operational practices, rather than a consistent effect of establishment size. E. coli levels were broadly comparable between throughput groups and species, indicating that controls of faecal contamination appear similar across establishment sizes.

It should be noted that this is an assessment of the relative risk between high- and low-throughput cutting plants and does not represent an assessment of the risk to UK consumers from minced meat and meat preparations as a product category or from cutting plants as a sector.

The uncertainty is primarily driven by the limited availability of data and the low Salmonella spp. detections, which also resulted in statistical uncertainties as highlighted in previous sections and in Section 6. The uncertainty also lies around the sampling differences between data sources as well as number of establishments included in the datasets.

The severity of detriment of salmonellosis is considered to be Low for the general population and Medium for vulnerable groups (as described in 4.1.1). Due to the range of serovars detected in the survey underlying this risk assessment, which are associated with different clinical outcomes, the uncertainty on the severity of detriment is Medium.

For the definitions of risk, uncertainty and severity characterisations, please see the Annex.