Training in tools and modelling to develop risk ranking and quantitative microbial risk assessment for cheese along the Polish and Spanish food Authors: Olga Sęczkowska, Ana Baenas Soto, Pablo S Fernández Escámez, Alfredo Palop Gómez, Alberto Garre Published: 2025 ABSTRACT AbstractTo ensure that ready‐to‐eat products are safe to eat, it is important to develop new tools to prevent and control bacterial contamination. Risk assessment has been gaining popularity in recent years and provides tools with which we can develop mathematical models for robust food safety management tools for use by health authorities, consumer protection agencies and industry stakeholders This training programme was developed under the EU‐FORA programme and had the goal of training a fellow in modern quantitative microbiological risk assessment (QMRA) methods through a hands‐on approach within the Unit of Microbiology and Risk Assessment at Universidad Politecnica de Cartagena (UPCT). The programme involved working on a case study focused on the microbiological safety of artisanal cheese, covering every step of a QMRA, including hazard identification, exposure assessment, hazard characterisation and risk characterisation. The case study was related to listeriosis, which constitutes a significant public health concern. Ready‐to‐eat foods, such as artisanal cheeses, provide an ideal environment for growth Listeria monocytogenes. The fellow developed skills in microbiological techniques such as preparation of thermal and growth experiments, development of a protocol for preparing artisanal cheeses at laboratory scale, as well as data analysis using the R programming language through the bioinactivation, biogrowth and biorisk packages. Experimental findings demonstrated that thermal treatment of milk at 60°C for 9 min causes more than 6 log‐reductions of Listeria below the detection limit, with similar kinetics in cow's and goat's milk. Conversely, storing the finished cheese at elevated temperatures (e.g. above 25°C) was shown to promote Listeria proliferation. This information was combined into kinetic models based on predictive microbiology, and applied in an academic exercise to evaluate the exposure of the consumer to the hazard and to estimate the illness per serving.Keywords: artisanal cheese, artistenal cheese, predictive microbiology, QMRA model INTRODUCTION The growing demand from consumers for healthy and safe food products has led to increased interest in developing and implementing effective quality and safety management systems for ready‐to‐eat products (RTE) (Koutsoumanis et al., 2021). Dairy‐derived foods represent a common preference among consumers, with cheese being one of the most popular ones in terms of production volume within European Union countries (Possas et al., 2022). As RTE products lack a cooking step by the consumer, it is important to adopt a broader perspective on the microbiological risks and infection‐related challenges associated with such products. Listeria monocytogenes is a pathogenic bacterium that causes listeriosis, one of the most serious diseases affecting consumers. It is one of the foodborne diseases with the highest mortality rate (EFSA BIOHAZ Panel, 2018; Garre et al., 2019), mainly in immunocompromised people, pregnant women, newborns and the elderly (Martín et al., 2022). Although listeriosis was the fifth most commonly reported zoonosis in humans in the European Union (EU), it caused the highest number of hospitalisations and deaths (EFSA and ECDC, 2023; Wang et al., 2025). In 2022, there were 2738 confirmed cases of invasive listeriosis in the EU, resulting in 1330 hospitalisations and 286 deaths (EFSA and ECDC, 2023). The bacterium shows high tolerance to low temperatures and has the ability to grow in various environments (Bonanno et al., 2024; Martín et al., 2022). Fresh cheeses provide an optimal environment for listeriosis to develop. Even when stored in refrigerated conditions, they provide the bacteria with the ideal environment for growth, given that Listeria can grow at low temperatures and pH, provided the water activity remains high (Falardeau et al., 2021; Martín et al., 2022). Listeria spp. may be affected by natural competition in ready‐to‐eat products, in what is often called the ‘Jameson effect’. Particularly in cheese, there is competition between lactic acid bacteria (LAB) and L. monocytogenes. In practice, the dominant microorganism limits the growth of others, reducing the concentration of L. monocytogenes (Serra‐Castelló et al., 2022). This is due to LAB producing antimicrobial compounds, such as lactic acid, hydrogen peroxide and bacteriocins, which can inactivate or inhibit the growth of pathogens (Cadavez et al., 2019; Serra‐Castelló et al., 2022). Predictive microbiology is a tool that enables the prediction of the behaviour of microorganisms under specific conditions that may occur at various stages of the food production chain – from preparation and processing to distribution and storage. Temperature plays a key role in shaping the growth dynamics of microorganisms, and this parameter is an integral part of predictive modelling (Thomas et al., 2019). In the context of risk analysis, the focus of the food safety management system is on assessing the health risks associated with food consumption, and on selecting and implementing effective strategies to reduce these risks (Koutsoumanis et al., 2021). The work programme focused on training the fellow on methodologies related to quantitative microbiological risk assessment (QMRA). The fellow joined the research group at the host institution, gaining practical experience in experimental methods (inactivation and growth experiments), statistical analysis, mathematical modelling and stochastic simulations. The programme covered all stages of risk assessment, including hazard identification, exposure assessment, hazard characterisation and risk characterisation. DATA AND METHODOLOGIES Methodologies Strain preparation L. monocytogenes EGD‐e was selected, as this is a common reference strain for L. monocytogenes. The strains were individually activated by transferring one colony of the stored culture to 5 mL of TSB and incubated at 37°C during 24 h before either growth or inactivation experiments. Thermal treatments Thermal treatments were carried out using a Mastia thermoresistometer (Conesa et al., 2009), developed by the host group. The Mastia thermoresistometer enables the resistance of microorganisms to high temperatures to be measured under both isothermal and non‐isothermal conditions (Conesa et al., 2009). The Mastia thermoresistometer is designed to maintain a constant temperature within the analysed sample throughout the entire process. This is made possible by the design of the vessel, which is equipped with a stirrer. The sample is poured into the vessel and the stirrer maintains the set temperature by stirring the sample throughout the experiment (Conesa et al., 2009). Isothermal experiments were applied at 57°C, 60°C and 62°C for 45, 9 and 4.5 min, respectively. Experiments were done in tryptic soy broth (TSB, Scharlau) as laboratory medium, as well as goat's milk (from a local store). Both media were poured into the vessel of the thermoreistometer at a volume of 400 mL, heated to the set temperature and inoculated 0.2 mL of L. monocytogenes EDG‐e, resulting in a concentration of ~6 log CFU/mL. Samples were collected at set timepoints, serially diluted and spread on TSA plates. The plates were then stored for 24 h at 37°C before counting. The experiments were conducted on three independent replicates in both media for all three temperature profiles. Growth experiments Development of a laboratory‐scale cheese model The first stage involved the development of a laboratory protocol for a model cheese from unpasteurised goat milk. The raw milk was provided by a local supplier (Cartagena, Spain) and microbial rennet from a Polish supplier (Browin, Poland). The cheeses were produced on a laboratory scale in accordance with a protocol developed by the fellow holder and the UPCT group. Milk used to produce cheese was tested for pH and water activity, then salted, enriched with calcium chloride, warmed and combined with rennet to form curd, which was cut, drained in moulds and left to set. The resulting cheese was portioned and prepared for storage. The final cheese product was portioned into 10 g pieces and transferred to Falcon tubes. To artificially contaminate the test samples, each Falcon tube was inoculated by placing 10 μL‐aliquots of strain L. monocytogenes culture (3 log cfu/g) in the central perforation of each test sample using a micropipette. Samples were grouped into control and test samples and stored at 25°C for 48 h. Each test tube represented a specific point in time. A temperature of 25 degrees was chosen as the worst‐case scenario for consumers. Microbiological analysis The growth of L. monocytogenes (in test samples) and LAB (both in test and control samples) was studied at 25°C. Samples were taken after 0, 2, 4, 6, 14, 16, 18, 20, 22, 24 and 48 h. Test samples were placed into sterile bags (LABOLAN S.L., Navarra, Spain) with the addition of 90 mL of peptone water 1.0% (dilutor system, IUL Instruments®, Barcelona, Spain) and homogenised for 1.5 min using a Stomacher. The homogenised sample was serially diluted in peptone water before plating. The concentration of L. monocytogenes was quantified on Chromogenic Listeria Agar, Ottaviani and Agosti (ALOA, Condalab®, Spain), whereas the concentration of LAB was measured in MRS. Plates were incubated at 37°C (48 h) for L. monocytogenes and 30°C (72 h) for LAB. Colonies were counted and results were expressed as log cfu/g. Physicochemical analyses The pH and aw of the control samples were measured throughout storage. The measurement for pH was made by immersing the electrode in the sample 10 g of sample was mixed with 10 mL of distilled water at 40°C, according with PN‐73/A‐86232 ‘Milk and dairy products. Cheese. Test methods’. After determining electrode readings, the result was read with an accuracy of 0.01. Three measures of each physicochemical parameter were made for each sample. Water activity (aw) was measured at 25 ± 0.2°C using an AquaLab 4TE device (MeterGroup, USA) after allowing samples to reach equilibrium in the sealed chamber. Predictive microbiology models Microbial inactivation The data obtained from isothermal treatments was analysed using the Bigelow and Geerards model. As presented in Equation (1), the Bigelow model assumes that the logarithm of the microbial concentration (N) decreases linearly over time during the treatment, relative to the initial concentration (N 0 ). In this approach, the rate of inactivation is described by the D‐value (D). In the context of secondary models, the Bigelow model describes a log‐linear dependence between the D‐value and the treatment temperature (T). As illustrated in Equation (2), the microorganism's sensitivity to temperature variation is quantified by the z‐value (z). To enhance the identifiability of the model parameters, a reference temperature (T ref) is introduced, with D ref representing the D‐value at T ref. Geeraerd's model (Geeraerd et al., 2000) is a logarithmic‐linear model of microorganism inactivation, extended by an adaptation phase (shoulder) and an asymptotic value (tail). As the data obtained did not show a clear tail, it was described using the algebraic solution for the Geeraerd model without tail (Equation 3). In this equation, the shoulder length is described by parameter Sl. This model introduces a log‐linear secondary model for k, similar to the one defined within the Bigelow model (Equation 4). This introduces kref as the value of k at Tref. The models were fitted using the online version of bioinactivation (Garre et al., 2025) available at https://foodlab‐upct.shinyapps.io/bioinactivation4/. Microbial growth The data obtained from growth experiments was analysed using the Baranyi model. This model takes into account the adaptation phase (lag phase), the logarithmic growth phase and the stationary phase. Under isothermal conditions, the Baranyi model is defined by Equation (5) The model is parameterised by the maximum specific growth rate (μmax), the lag phase duration (λ) and the maximum concentration in the stationary phase (Nmax), as well as the initial concentration (N0). These parameters was estimated from the experimental data using the online version of the biogrowth package (Garre et al., 2023) available online at https://foodlab‐upct.shinyapps.io/biogrowth4/. QMRA The final stage of the project was to build a QMRA model as an academic exercise, combining the predictive microbiology models developed with literature data. The exercise included four different scenarios: Cheeses produced from pasteurised milk including bacterial competition (Baseline) Cheeses produced from pasteurised milk including bacterial competition (Baseline) Cheeses produced from pasteurised milk without bacterial competition (P + C − ) Cheeses produced from pasteurised milk without bacterial competition (P + C − ) Cheeses prepared from unpasteurised milk including bacterial competition (P − C + ) Cheeses prepared from unpasteurised milk including bacterial competition (P − C + ) Cheeses prepared from unpasteurised milk without bacterial competition (P − C − ) Cheeses prepared from unpasteurised milk without bacterial competition (P − C − ) The QMRA was developed as a Modular Process Risk Model (Nauta, 2008) Besides the predictive models for growth and inactivation, the model requires probability distributions for the logistic parameters of the food supply chain (pasteurisation time and temperature; storage time and temperature), the initial concentration (prevalence and concentration of L. monocytogenes in raw milk) and the consumer behaviour (serving size). These data were taken from the scientific literature and described using Normal, Uniform, Empirical and Triangular distributions. For the initial concentration, we used data according to Dalzini et al. (2016), ranging from < 1 to > 1000 cfu−1. Due to the lack information of storage time in 25°C data for artisanal cheese in literature, we assumed the best‐case scenario was 3 h and the worst‐case scenario was 48 h. We assumed the serving size to be 100 g. The QMRA model was evaluated using Monte Carlo simulations (Pouillot & Guillier, 2020) to perform forward uncertainty propagation. All calculations were performed in R (version 4.2.3) using the biorisk package, which is available at https://github.com/albgarre/biorisk. One million Monte Carlo simulations were calculated for each scenario. Figure 1 illustrates the elements of the QMRA model built for the baseline scenario (green). The P−C+ and P−C− do not include a pasteurisation step. Therefore, the concentration of L. monocytogenes in raw milk reported by Dalzini et al. (2016) is used as direct input for the growth element (red outline in Figure 1). Illustration of the elements of the QMRA model developed as an academic exercise within the training programme. Regarding the competition, our laboratory results showed that the presence of LAB only influenced the maximum concentration of L. monocytogenes in stationary phase (probably due to Jameson effect). Therefore, scenarios without competition (P+C− and P−C−) assigned logNmax a constant value of 9 log CFU/g (based on expert opinion), whereas scenarios with competition (P+C+ and P−C+) assigned the parameter values estimated from experimental data. RESULTS The first part of the training programme generated the experimental data required to build growth and inactivation models for L. monocytogenes in cheese and milk, respectively. Table 1. summarise the parameters of the Geeraerd and Baranyi model fitted to these observations. TABLE 1Parameters of the Geeraerd and Baranyi models obtained for the inactivation kinetics of Listeria monocytogenes in goat's milk and for its growth in goat's cheese.ResponseParameterUnitEstimate ± SD errorInactivation logDref min0.30 ± 0.11 z °C5.69 ± 1.73Growth logN0 log10 CFU/g3.10 ± 0.19μ (mu)0.30 ± 0.12λ (lambda)h3.51 ± 1.69 logNmax log10 CFU/g5.71 ± 0.10 Parameters of the Geeraerd and Baranyi models obtained for the inactivation kinetics of Listeria monocytogenes in goat's milk and for its growth in goat's cheese. The models summarised in Table 2 was the basis for the QMRA model. The QMRA model was used to estimate the exposure of the consumer to L. monocytogenes in fresh cheeses, as well as the associated probability of illness during storage at 25°C. For scenarios Baseline and C−P+, a large number of simulations (75%) had a microbial dose of 0 cells. However, only 0.005% of the simulations in scenarios C+P− and C−P− would have a dose of 0 cells. This shows the importance of the pasteurisation treatment for the risk in this academic scenario, as it would ensure the absence of the microorganism in the serving. TABLE 2Number of cases per million servings estimated from the academic QMRA model built for the risk of listeriosis in cheese.Type of scenarioMedian90th percentile99th percentile Baseline 0059 C − P + 0080,634 C + P − 256184 C − P − 6192,979131,801Baseline: pasteurised milk including bacterial competition.C−P+: pasteurised milk without bacterial competition.C+P−: unpasteurised milk including bacterial competition.C−P−: unpasteurised milk without bacterial competition. Number of cases per million servings estimated from the academic QMRA model built for the risk of listeriosis in cheese. Baseline: pasteurised milk including bacterial competition. C−P+: pasteurised milk without bacterial competition. C+P−: unpasteurised milk including bacterial competition. C−P−: unpasteurised milk without bacterial competition. Table 2 summarises the results of the QMRA model. The baseline scenario has an expected number (median of 1,000,000 MC simulations) of listeriosis cases per million servings of zero. On the other hand, the worst‐case scenario has an expected number of 61 cases per million. This demonstrates the relevance of competition and/or the pasteurisation treatment on the risk of listeriosis in cheese for this academic exercise. The scenario without pasteurisation (C + P − ) had a higher expected number of cases per million servings (25 cases) than the one without competition (C − P + ; 0 cases). This implies that the pasteurisation treatment is more relevant than the competition for the expected number of cases. The output in Table 2 presents uncertainty by including the 90th and 99th percentile besides the median number of cases. The comparison between the Baseline and Worst‐case scenario show similar results as before, with the baseline having a 99th percentile of 59 cases and the worst case 131,801 cases. However, the interpretation of the results for the two intermediate scenarios is not so evident. Scenario C+P− has a higher number of expected cases (25 cases) than C−P+ (0 cases), but a lower number in the 99th percentile (84 vs. 80,634 cases). This can be explained based on the microbial concentration at exposure, illustrated in Figure 2. The pasteurisation treatment has a big impact on the average microbial concentration, reducing the overall risk. However, the competition defines a maximum threshold for the microbial concentration. As a result, scenario C+P− has overall higher exposure, resulting in higher cases. However, competition prevents exposure to extremely high microbial concentrations. On the other hand, the C−P+ scenario does not define that threshold. Therefore, extreme events are more likely, despite a lower overall risk. Microbiological concentration at the consumption stage for all scenarios considered for cases where the dose was higher than zero. Baseline – baseline model + competition (75% simulations with 0 dose); C−P+ − baseline model no competition (75% simulations with 0 dose); C+P− – raw milk + competition (0.005% simulations with 0 dose); C−P− – raw milk no competition (0.005% simulations with 0 dose). CONCLUSION The EU‐FORA programme gave the fellow the chance to learn novel methodologies for microbiological risk assessment. The programme provided theoretical and hands‐on training on microbiological risk assessment through the participation in various training courses and activities at the Polytechnic University of Cartagena. Building on this experience, the fellow conducted microbial growth and inactivation experiments, analysing the data to build predictive microbiology models that were the basis for an academic QMRA model implemented in the R programming language. For this, the fellow also became familiar with relevant databases for food safety. The EU‐FORA programme also equipped the scholarship holder with valuable skills in project management, planning and conducting experiments. ADDITIONAL ACTIVITIES The fellow joined a research group involved in various microbiological risk assessment projects, gaining comprehensive experience in QMRA, from data collection to developing food safety recommendations. The fellow participated in various seminars and practical classes on food safety. She will also be visiting the Spanish Agency for Food Safety and Nutrition (AESAN) in Madrid, and the National Centre for Food laboratory, in Majadahonda on 18–19 February. The presentation covered AESAN's activities in the areas of risk assessment, management and communication. The fellow also shared information about her work and activities within the EU‐FORA programme. The grantee participated in the IX Day of Scientific Coopertaion with Spanish organisations collaborating with EFSA on April 24 in Madrid, where she presented her experience in the EU‐FORA programme. In the second half of the year, the fellow will present the results of his work at an International Conference. ABBREVIATIONS Spanish Agency for Food Safety and Nutrition Chromogenic Listeria Agar, Ottaviani and Agosti water activity pasteurised milk + bacterial competition Panel on Biological Hazards raw milk + bacterial competition colony forming units raw milk ‐ bacterial competition pasteurised milk ‐ bacterial competition lactic acid bacteria de Man, Rogosa and Sharpe agar quantitative microbiological risk assessment ready to eat food refrigerated ready to eat food tryptic soy agar tryptic soy broth Universidad Politecnica de Cartagena COPYRIGHT FOR NON‐EFSA CONTENT EFSA may include images or other content for which it does not hold copyright. In such cases, EFSA indicates the copyright holder and users should seek permission to reproduce the content from the original source. Sęczkowska, O. , Soto, A. B. , Fernández Escámez, P. S. , Gómez, A. P. , & Garre, A. (2025). Training in tools and modelling to develop risk ranking and quantitative microbial risk assessment for cheese along the Polish and Spanish food. EFSA Journal, 23(S1), e231114. 10.2903/j.efsa.2025.e231114 Approved: 29 September 2025 The declarations of interest of all scientific experts active in EFSA's work are available at https://open.efsa.europa.eu/experts. REFERENCES