Risk of infection of dairy cattle in the EU with highly pathogenic avian influenza virus affecting dairy cows in the United States of America (H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13)

Abstract

This opinion assesses the risk of avian influenza H5N1 B3.13 genotype virus infection in EU dairy cattle. Introduction of the virus into EU dairy cattle, poultry or wild birds via trade or migratory birds from the US is assessed as highly unlikely. The potential impact in case of introduction is estimated as high for most Member States. Measures assessed as most effective to prevent introduction of the virus are avoiding importation of cattle and poultry from infected countries, and cleaning and disinfection of milking equipment. Measures assessed as most effective to prevent spread of the virus in the EU are milking hygiene, banning movement of cattle in infected areas, avoiding the exchange of workers, vehicles and equipment, and implementing biosecurity measures before entering farms. Regarding adaptation of current highly pathogenic avian influenza (HPAI) surveillance, a proportional, risk‐based preparedness strategy is recommended, including (i) awareness raising to strengthen passive and syndromic surveillance, (ii) targeted investigations of suspicions/outbreaks after confirmed exposure of cattle to HPAI, (iii) using sensitive diagnostics on multiple sample types and (iv) regional active surveillance (bulk milk reverse transcription quantitative polymerase chain reaction (RT‐qPCR)) following first detections in cattle. In case the virus is identified in wild birds or poultry, surveillance of dairy farms in the affected area should be considered. The contamination of bulk milk is considered very likely, if EU dairy herds become infected, as cows may not show clear clinical signs and may shed the virus before changes in milk become apparent. If EU bulk milk or colostrum are contaminated with the virus, food‐borne exposure of consumers to viable virus would be highest for raw drinking milk, raw colostrum and raw milk cream. To date, however, no confirmed cases of food‐borne human infection with H5N1 B3.13 genotype virus have been reported.

SUMMARY

This assessment focuses on the risk of infection of dairy cattle in the EU with H5N1 virus, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13. Evidence from outbreaks in the US shows that cattle can become infected with this virus via milk and milking equipment, and at least three separate introductions in dairy cattle of different genotypes of clade 2.3.4.4b H5N1 highly pathogenic avian influenza (HPAI) viruses have occurred. Despite the continuous circulation of this clade in Europe since the 2020/2021 epidemic, no spillover to cattle has been detected in the EU. Nevertheless, preparedness is warranted given the potential economic, animal health and welfare, and zoonotic implications and public health risks.

ToR 2a: The potential impact of an infection of dairy cows in the EU with the H5N1 B3.13 genotype virus has been assessed with the ‘Living One Health Risk Assessment risk assessment model’ (L'ORA). The introduction of H5N1 B3.13 genotype virus from the US into EU dairy cattle or poultry via trade is assessed as highly unlikely (annual probability < 10⁻⁵). Introduction via wild bird migration is also considered highly unlikely (annual probability < 10⁻⁵), except for western Ireland and western France regions (annual probability up to 10⁻²–10⁻¹), especially in autumn. These estimates carry uncertainty due to limited data on transatlantic wild bird movements and infection prevalence in US wild bird populations. The model estimates the occurrence of outbreaks of H5N1 B3.13 genotype virus in EU poultry, dairy cattle and wild birds as highly unlikely (annual probability from 10⁻¹⁴ to 10⁻⁴). If outbreaks were to occur in the EU despite these very low probabilities, the largest epidemic would result from introductions by migratory wild birds, with estimates of ~250–923 k wild birds affected, compared to ~240–713 poultry farms and ~26–81 dairy farms. Without taking control measures in dairy cattle, outbreaks in EU cattle could expand substantially (~161–17,000 farms). In case of outbreaks, the overall impact, considering economic, societal and environmental dimensions, is predicted to be high for most Member States. The implementation of control measures in cattle would markedly reduce the number of infected dairy farms, but not the overall economic, societal and environmental impact.

ToR 2b: The measures that could prevent introduction of the H5N1 B3.13 genotype virus in EU dairy cattle and poultry as well as possible risk mitigating measures to prevent its spread in the EU has been assessed regarding their effectiveness and feasibility through an expert opinion elicitation process involving four EFSA experts.

Regarding measures to prevent introduction into EU dairy cattle via wild birds, cleaning and disinfection of milking equipment is assessed as the most effective of the assessed measures, as it directly prevents intramammary infection. It is also considered the most feasible, since such practices are already routine on most farms. Preventing access of wild birds and mammals to the milking parlour and equipment is assessed as the next most effective measure, reducing the likelihood of contamination of equipment and subsequent intramammary infection. Its feasibility is considered moderate, as structural modifications may be needed to fully implement it. Vaccination of cattle against HPAI is considered effective in principle, as it could reduce the susceptibility of dairy cattle. However, this measure is associated with greater uncertainty than the previous two, and it remains unclear whether infections can be prevented. Its feasibility is considered low, as there are currently no licensed vaccines and implementation would require considerable logistical and financial effort. Management of feed and water within the premises has been ranked lower in effectiveness, since the oral route of infection seems to play a limited role in virus introduction. Its feasibility is considered moderate, as changes in management practices would be generally achievable, but require on‐farm modifications. Making the farm surroundings unattractive for wild birds, reporting dead wild birds in the vicinity of farms for removal and testing and keeping dairy cattle indoors in high‐risk areas and during high‐risk periods are considered the least effective of the assessed measures regarding their ability to prevent introductions. Feasibility varied from extremely low (due to important structural farm changes, ecological habitats changes) to high (provided responsibilities for removal and testing of dead wild birds are clarified).

Regarding measures to prevent introduction into EU dairy cattle via trade, avoiding importation of cattle from infected countries is assessed as the most effective of the assessed measures, as dairy cattle appear to be the main reservoir in the US. It is also considered highly feasible, because such restrictions are already established under EU legislation for other diseases. Avoiding importation of potentially contaminated animal products or germinal products from infected countries is ranked the next most effective measure, as, with the exception of germinal products, these commodities are less likely to come into contact with EU dairy cattle than live animals. Its feasibility is considered high, as such restrictions can be implemented under current EU legislation. Vaccination of cattle against HPAI in farms receiving trade products is considered less effective than the two previous measures due to uncertainty about whether full protection could be achieved. Its feasibility is considered low, as there are currently no licensed vaccines and implementation would require substantial logistical and financial effort. Pre‐movement testing of outgoing cattle and quarantine of newly imported cattle from infected countries is considered the least effective of the assessed measures, as diagnostic tests can miss infected animals and limited clinical signs in cattle could allow infected animals to pass undetected, although testing during quarantine may mitigate this risk. Feasibility varied from moderate (quarantine, due to the infrastructure required for dedicated facilities) to high (pre‐movement testing, which is readily implementable).

Regarding measures to prevent introduction into EU poultry flocks via trade, avoiding importation of poultry from infected countries is assessed as the most effective of the assessed measures, as infected live poultry poses a high risk for virus introduction. It is also considered highly feasible, since such restrictions can be readily implemented under EU legislation. Avoiding importation of potentially contaminated poultry products or germinal products (e.g. turkey semen or embryonated chicken eggs) is ranked the next most effective measure, as these commodities are considered less likely to introduce the virus compared to live animals. Its feasibility is considered high, since such restrictions can be readily implemented under EU legislation. Pre‐movement testing of outgoing animals and quarantine of newly imported poultry from infected countries are considered the least effective of the assessed measures. Pre‐movement testing may not have 100% sensitivity, and while quarantine is generally effective, ducks may require additional testing. Feasibility is judged high for both measures, as they are standard practices for disease control in poultry.

Regarding measures to prevent within‐farm spread in infected dairy cattle farms in the EU, milking hygiene is assessed as the most effective of the assessed measures, as it reduces the likelihood of infection through the intramammary route. Its feasibility is considered moderate, as it requires consistent and thorough hygiene practices during milking, which can be demanding in large dairy farms. Vaccination of cattle against HPAI is assessed as less effective than the previous measure due to uncertainty about whether full protection could be achieved. Its feasibility is considered low, as there are currently no licensed vaccines, and implementation would require considerable logistical and financial effort. Feeding calves only with heat treated milk or milk replacers, isolation of sick animals, proper disposal of carcasses and animal products (raw milk) and culling of infected animals are considered the least effective of the assessed measures. The first only reduces the probability of infection in calves; the success of the second and the fourth largely depends on early detection, which is difficult in the presence of sub‐clinically infected animals; for the third, live animals are considered as more important for within‐farm virus spread than dead infected animals or animal products. Feasibility is considered low to moderate, due to logistical, financial and societal constraints.

Regarding measures to prevent spread from infected dairy farms to other dairy farms in the EU, banning the movement of cattle in infected areas is assessed as the most effective of the assessed measures, as it directly addresses the main mode of transmission identified during the US outbreak. Its feasibility is considered high, since cattle movement bans are already routinely implemented in the control of other livestock diseases. Pre‐movement testing of cattle from infected areas, quarantine of new cattle entering farms, early detection and isolation/culling of infected cattle farms, vaccination of cattle against HPAI and avoiding the exchange of workers, vehicles and equipment are considered less effective than movement bans. Diagnostic tests may not have 100% sensitivity, and subclinical infections may go unnoticed, delays between sampling and testing may limit their impact, and there is evidence of possible human‐mediated transmission to cows, albeit to a lesser extent than cattle movements. Feasibility varies from extremely low (culling), low (quarantine, as it requires dedicated facilities and infrastructure, vaccination due to reasons previously listed), moderate (early detection and isolation, which would require an established surveillance system, workers/equipment exchanges, which are often essential for farm operations) to high (pre‐movement testing, which is readily implementable). Indoor confinement or limiting outdoor access of cattle in infected areas, removing, reporting and testing dead wild birds in the vicinity of cattle farms, managing feed and water within the premises, avoiding non‐essential visits, applying biosecurity measures before entering farms, restricting vehicle movements from infected areas, proper disposal of carcasses and animal products, banning the movement of untreated milk and prohibiting slurry or manure spreading from infected farms are considered the least effective of the assessed measures. Feasibility varies from low (some of these measures would require considerable infrastructure changes and logistics), to moderate (indoor confinement may conflict with animal welfare and production standards, measures about managing feed/water/manure depend on seasonal conditions) to high (avoiding non‐essential visit is readily applicable, as well as removal, reporting and testing dead wild birds, provided responsibilities are clearly defined).

Regarding measures to prevent spread from infected poultry farms to dairy cattle farms in the EU, vaccination of cattle or poultry against HPAI, avoiding the exchange of workers, vehicles and equipment, and implementing biosecurity measures before entering the farm are assessed as the most effective of the assessed measures, as they would reduce both the susceptibility of cattle and the shedding of the virus by infected poultry and limit transmission between the two. The feasibility of vaccination of poultry is considered moderate due to logistical and financial requirements, whereas vaccination of cattle is assessed to have low feasibility for the reasons explained in previous sections. Avoiding exchanges of workers, vehicles and equipment and implementing biosecurity measures are considered low to moderately feasible, for the reasons previously explained. Restricting the movement of cattle from infected areas, indoor confinement or limiting outdoor access of cattle during high‐risk periods, pre‐movement testing of outgoing cattle and quarantine of newly introduced cattle, as well as removing, reporting and testing dead wild birds in the vicinity of cattle farms, avoiding non‐essential visits, managing feed and water within the premises, restricting vehicles movements from infected areas and proper disposal of carcasses and animal products are assessed as less effective than the previous measures. The feasibility varies from low to high: low for quarantine of new animals; moderate for indoor confinement (which may conflict with animal welfare and production standards), measures related to managing feed/water/manure (which depend on seasonal conditions) and restrictions on vehicle movements and carcass or product disposal (which would require specific logistics and adaptation to milk collection infrastructure); and high for movement restrictions, pre‐movement testing, wild bird removal and avoidance of non‐essential visits.

Regarding measures to prevent spread from infected dairy farms to poultry flocks in the EU, movement restriction of dairy cattle from infected areas, avoiding exchange of workers, vehicles and equipment between poultry and infected cattle farms, and vaccination of poultry against HPAI are assessed as the most effective measures, as they would directly reduce the probability of virus transmission from infected dairy farms to poultry by limiting contact with infected cattle, contaminated fomites or infected wildlife and by decreasing the susceptibility of poultry. The feasibility varies from moderate (avoiding exchanges of workers, vehicles and equipment, vaccination of poultry) to high (restricting cattle movements) for the reasons explained in previous sections. Early detection combined with culling or isolation of infected cattle farms, vaccination of cattle against HPAI and banning the movement of milk unless treated to inactivate the virus prior to movement are assessed as less effective than the previous measures. These would reduce transmission but not fully prevent virus spread due to delays in detection, partial vaccine protection and the limited contribution of contaminated milk to between‐farm spread. The feasibility varies from low to moderate due to the logistical, financial and acceptability constraints already discussed in previous sections. Indoor confinement or limiting outdoor access of poultry during high‐risk periods, pre‐movement testing of outgoing cattle from infected areas, avoiding non‐essential visits to infected cattle farms, implementing biosecurity measures before entering cattle farms, movement restriction of vehicles from infected areas, and proper disposal of carcasses and animal products are assessed as less effective than the previous measures, as they would mainly reduce indirect transmission and have limited impact on the overall risk of virus introduction into poultry. The feasibility varied from low (biosecurity measures before entering farms and cattle vaccination) to moderate (movement restrictions, proper disposal of carcasses and animal products, indoor confinement) for the reasons explained previously. Removal, reporting and testing of dead wild birds in the vicinity of cattle farms and prohibiting slurry or manure spreading from infected cattle farms are assessed as the least effective of the assessed measures, since wild bird removal would only have a minor impact on spread from dairy to poultry farms, and infected cattle are believed to shed limited amounts of virus in their faeces. The feasibility is considered moderate to high for the reasons explained in previous sections.

ToR 2c: The possible adaptations of the current EU surveillance for HPAI that would allow detection of an introduction into the EU dairy cows were identified by the EFSA experts using a qualitative approach for the synthesis of scientific evidence, descriptions of current Union surveillance programmes and information on MS's surveillance activities. Passive surveillance in dairy farms remains the main option of surveillance, but its sensitivity is expected to be low, as clinical signs of HPAI‐affected cows are variable and often non‐specific and subclinical infections likely represent the majority of cases. Recognition of clinical signs might be hampered by confusion with mastitis caused by other aetiological agents, leading to delayed recognition. Raising awareness among farmers, veterinarians and laboratories is therefore critical. Syndromic surveillance including monitoring of production parameters may provide early warning, though sensitivity is limited if only a small proportion of cattle are clinically affected. Active surveillance through testing of bulk milk (BM) by RT‐qPCR holds the greatest potential for timely detection. The existing EU infrastructure for bulk milk monitoring could be adapted, and a regional, risk‐based implementation is considered to be most efficient. Surveillance should be accompanied by enhanced sequencing of viruses detected in dairy cattle, wild birds and poultry for genotype identification. Serological methods for antibody detection against avian influenza viruses (AIVs) in bovine serum samples add value for retrospective and herd‐level investigations.

It is therefore recommended to develop a proportional and risk‐based preparedness strategy for the detection of an introduction of H5N1 B3.13 genotype virus and other genotypes into the dairy cattle population. This should include awareness raising to strengthen passive and syndromic surveillance; targeted investigations of suspicions/outbreaks after confirmed exposure of cattle to HPAI virus, using sensitive diagnostics on multiple sample types; and regional active surveillance (bulk milk RT‐qPCR) following first detections in cattle, as well as follow‐up monitoring with frequent bulk milk RT‐qPCR testing, later complemented by serology to confirm the absence of diseases in a given area. In case the H5N1 B3.13 genotype virus is identified in wild birds or poultry, surveillance of dairy farms in the affected area should be considered.

ToR 2d: To assess the likelihood of bulk milk being contaminated if EU lactating dairy cows are infected with the HPAI H5N1 B3.13 genotype virus, an expert knowledge elicitation was used. The group considers contamination of bulk milk, given infection in a dairy herd, very likely (90%, 95% certainty interval: 80%–100%). Reasons provided for the judgements include the consideration that cows may shed the virus before any changes in the milk become apparent, and that a large proportion of lactating cows would shed virus in their milk without showing clinical signs.

ToR 2e: To estimate the levels of viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products produced in the EU at the end of processing, modelling was used. The highest estimated levels would be found in raw drinking milk, raw milk cream and raw colostrum, with an estimated median concentration in the final product of 4.4 log₁₀ EID₅₀/mL or /g (1.0–7.2 log EID₅₀/mL or /g 95% variability interval in different servings). In production settings in which mixing of bulk tank milk is common, the levels would be lower depending on the proportion of milk from infected herds in the mixture. As the H5N1 B3.13 genotype virus is heat sensitive, levels of viable virus in pasteurised milk and colostrum are significantly reduced compared to raw foodstuffs. Based on a thermal inactivation model, LTLT pasteurised (63°C – 30 min) products and HTST pasteurised cream (82°C – 15 s) and derived products are expected to be free of viable H5N1 B3.13 genotype virus. HTST pasteurised milk and colostrum (72°C – 15 s) may still contain viable H5N1 B3.13 genotype virus, although at low levels (median concentration of −0.5 log₁₀ EID₅₀/mL), and time/temperature profiles applied in industrial processes (including heating, temperature holding and cooling) may provide higher reduction of viable H5N1 B3.13 genotype virus. Thermisation of milk at 57°C for 30 min is also very effective, and products derived from this process are expected to be free of viable H5N1 B3.13 genotype virus, while the shorter thermisation process (65°C – 15 s) yields limited reduction (0.9 median log₁₀ reduction) but can be used as an additional hurdle in dairy processing. The H5N1 B3.13 genotype virus is not completely inactivated through acidification/fermentation of raw milk to the pH values (pH 4.6) expected in fermented milk or fresh cheeses derived from raw milk (3.5 median log₁₀ reduction) or to the pH values (pH 5.0) expected in fermented cream from raw milk (1.9 median log₁₀ reduction). Apart from acidification, soft and semi‐soft raw milk cheeses also undergo ripening for up to 6 weeks and 3 months, respectively. This leads to a further reduction of viable H5N1 B3.13 genotype virus (1.0 and 1.9 median log₁₀ reduction for soft cheese and semi‐soft cheese, respectively), but infectious virus could still be present in the final product. Cheeses with more prolonged ripening periods (e.g. semi‐hard, hard, extra‐hard cheeses) are expected to contain lower titres of viable virus compared to soft and semi‐soft cheeses. Products derived from pasteurised milk or colostrum (such as yoghurt, pasteurised fermented milk, pasteurised milk cheeses) are expected to contain lower viable H5N1 B3.13 genotype virus levels compared to pasteurised milk or colostrum due to the combined inactivation of thermal treatment, acidification, ripening and other processes.

To estimate the mean probability of infection and illness per serving or, alternatively, the probability of exposure to different levels of viable H5N1 B3.13 genotype virus for consumers in the EU via the food‐borne route, it was assessed whether the available evidence provides a dose–response (DR) model that is applicable to humans through the oral route. It was concluded that, to date, no confirmed cases of human infection with H5N1 B3.13 genotype virus through the food‐borne route have been reported. Based on the comparison of available DR models (a human inhalation model, a model based on the inhalation model modified to approximate an ingestion DR relationship and an animal oral ingestion model), it was considered not appropriate to select one DR model and therefore exposure levels were evaluated against two critical thresholds (10 ⁷ EID ₅₀ and 10 ¹⁰ EID ₅₀ ) that were used as a reference for classification and comparison across products. The first threshold (10⁷ EID₅₀) was selected considering that, based on the human inhalation model and considering the impact of gastric digestion on H5N1 viability, the probability of human infection through the oral route is expected to be much lower than 1% at doses up to 7 log₁₀ EID₅₀. This threshold is supported by experimental data in non‐human primates (macaques) that displayed seroconversion but no clinical signs at an oral dose of 7.48 log EID₅₀. The second threshold (10¹⁰ EID₅₀) was instead selected considering that the inhalation model modified to approximate an ingestion DR relationship provides a probability of human infection through the oral route of 1% at doses of 10 log₁₀ EID₅₀. For each of the selected foodstuffs, the probability was calculated for the doses ingested through consumption to be larger than selected threshold values. In a hypothetical scenario that bulk milk or bulk colostrum is found to be contaminated with H5N1 B3.13 genotype virus in the EU, the exposure to viable virus, expressed as the median number of servings per thousand of the foodstuffs, consumed as they are and considering the selected processing parameters, providing a dose exceeding 10⁷ EID₅₀, as estimated by the model, was high (median ≥ 100) for raw drinking milk, raw colostrum, raw milk cream and milk thermised at 65°C for 15 s with the latter, however, not typically consumed as such. It was intermediate (10 ≤ median < 100) for raw fermented cream and low (1 ≤ median < 10) for raw fermented milk, raw milk fresh cheese, raw milk semi‐soft cheese and HTST pasteurised (72°C – 15 s) drinking milk and colostrum. It was estimated as zero per 1000 servings for raw milk soft cheese, pasteurised cream (82°C – 15 s), milk thermised at 57°C for 30 min and LTLT pasteurised (63°C – 30 min) drinking milk or colostrum. For freeze‐dried colostrum it was not determinable due to the lack of evidence on the effect of this process on H5N1 B3.13 genotype virus viability. It was remarked that the exposure to viable H5N1 B3.13 genotype virus in products derived from thermised (65°C for 15 s) or pasteurised milk (e.g. fermented milk, fermented cream, different types of cheese) is expected to be lower than in the equivalent products derived from raw milk. The products classified under the high exposure group using the threshold of 10⁷ EID₅₀ exceed, at lower frequency (up to 10 median servings per 1000 servings) also the critical threshold of 10¹⁰ EID₅₀, based on the model estimates. While the model integrates some uncertainties probabilistically, other uncertainty sources have been identified and their impact on the conclusions described.

To evaluate the available measures along the food chain (from initially contaminated bulk milk or bulk colostrum up to consumer) and their effectiveness to reduce the levels of viable H5N1 B3.13 genotype virus, the outcome of the thermal inactivation model and the evidence derived from the literature review was used. It is concluded that thermal treatment is highly effective at reducing the levels of infectious H5N1 B3.13 genotype virus in milk and milk products and colostrum and colostrum‐based products: thermisation of milk for 30 min at 57°C and LTLT pasteurisation (63°C – 30 min) of milk or colostrum, and HTST pasteurisation of cream (82°C – 15 s) provides an estimated reduction of infectious virus by > 10 log₁₀; HTST pasteurisation of milk or colostrum (72°C – 15 s) is expected to reduce viral titres by 4.9 [1.7–20.2] log₁₀, which is expected to be further reduced in the industrial processes with the heating and cooling down steps; while other combinations of time/temperature can achieve significant reductions and have been listed. Moreover, various non‐thermal treatments, including high‐pressure processing, pulsed electric field technology, high‐pressure homogenisation, ultraviolet‐C irradiation, etc., have the potential to reduce the levels of viruses in milk or colostrum. These technologies are at a variable degree of implementation by the food/dairy industry.

1. INTRODUCTION

1.1. Background and Terms of Reference as provided by the requestor

As from late March 2024, the United States of America (US) are facing unprecedented events of detection of highly pathogenic avian influenza (HPAI) virus in bovine animals. US Department of Agriculture (USDA), Food and Drug Administration (FDA), Centres for Disease Control and Prevention (CDC), and state veterinary and public health officials have been investigating the occurrence of infection of dairy cows with HPAI H5N1 virus considering animal health, food safety and public health angles. The virus has spread between herds by animal movements specially lactating cows, but most reported spreading is within herds. Milk has been demonstrated to be contaminated with the virus, and the main intra‐herd transmission pathway is considered to be the aerosols with milk droplets in milking parlours. Although the cows do not show clinical signs of respiratory disease, they suffer with mastitis and there is a change in the quality of milk. In addition, this virus, which is also occurring in wild birds and poultry (as well as two cases recently detected in pigs), has demonstrated certain zoonotic ability: as of 6 November 2024, 25 mild human cases have been reported in individuals exposed to dairy cows that were infected or presumed to be infected, and one unexplained event without evident contact with animals. However, there has been to date no evidence of human‐to‐human transmission.

The US veterinary authorities, chiefly USDA‐APHIS services, have taken several measures (some regulatory, but mainly advisory and financial) to limit the spread of the virus, still the virus has spread to 15 States so far and is still spreading within few of them. The US federal authorities provided for numerous recommendations on the animal health and public health sides. They are also requiring compulsory pasteurisation of milk of affected or at‐risk herds as a risk mitigating measure to prevent the spread of the virus from animals to humans.

The virus type affecting dairy cows in the US (H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13) has not been reported so far in the EU or elsewhere than USA. The EURL, EFSA and ECDC are closely monitoring this situation. Even if the EU is not currently affected by this virus genotype, it is prudent to consider various potential options for risk assessment of the situation with a view of exploring possible action in animals, notably in dairy cows, ahead of a possible future introduction.

It is relevant to ask support from EFSA and EURL, to analyse the situation in the US and get scientific advice assessing animal health and food safety risks linked with this HPAI strain. The scientific advice should address in particular its likelihood of entry into Europe through trade and migratory wild birds, and if likely, the estimated expected timespan for entry via migratory wild birds, its potential impact on the EU and possible risk mitigation measures.

Terms of Reference

In the light of the above:

1. in accordance with Article 31 of Regulation (EC) No 178/2002, the Commission requests EFSA to provide scientific and technical assistance on the risk posed by the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13, currently circulating in the US; the following aspects are of particular relevance for the scientific report:

   1. provide a summary of the virological information currently available on the virus, of the outbreaks in dairy cattle in the US, and the measures that have been recommended or applied by the US authorities;
   2. describe the potential pathways for entry of the virus into the European continent via trade and via migratory birds, and timelines associated with the potential entry via migratory birds;
2. in accordance with Article 29 of Regulation (EC) No 178/2002, the Commission requests EFSA to provide a scientific opinion on risk analysis of the infection of dairy cows in the EU with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13, currently circulating in the US; the following aspects are of particular relevance for the scientific opinion:

   1. assess the potential impact of the infection of dairy cows in the EU;
   2. describe possible measures to prevent the introduction of the virus into dairy cows and poultry in the EU as well as possible risk mitigating measures to prevent its spread in the EU;
   3. describe possible options for adaptations of the current EU surveillance for HPAI that would allow detection of an introduction into the EU dairy cows populations posing a significant animal or public health threat;
   4. in a hypothetical scenario that lactating dairy cows are found to be infected with this virus, assess, taking into account the application or not of risk mitigating measures described above, the likelihood of bulk milk to be contaminated;
   5. in a hypothetical scenario that bulk milk is found to be contaminated with this virus in the EU, (i) assess the levels of viable virus at the point of consumption in raw milk, colostrum, dairy products and colostrum‐based products produced and consumed in the EU, (ii) estimate the risk for humans due to the exposure to those assessed levels of viable virus via the foodborne route, and (iii) critically review the available measures to mitigate the estimated risk.

Consider and describe the uncertainty related to any of the above.

1.2. Interpretation of the Terms of Reference

As per request of the European Commission, the scope of the assessments in this opinion is limited to the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 (further referred to as ‘H5N1 genotype B3.13 virus’) that has caused an outbreak in US dairy cattle. However, it should be noted that experimental evidence indicates that H5N1 B3.13 genotype virus is unlikely to behave differently from H5N1 clade 2.3.4.4b strains already circulating in Europe, and the absence of evidence of cattle infections despite widespread and repeated exposure to HPAI in wild birds and poultry in Europe indicates a very narrow bottleneck for spillover to dairy cattle under European conditions

ToR 1a. and 1b. have been addressed in a scientific report (EFSA, 2025).

ToR 2a–2d. relate to animal health questions and are addressed by the AHAW Panel, while ToR 2e. relates to food safety questions which are addressed by the BIOHAZ Panel.

The EC confirmed that the phrase ‘posing a significant animal or public health threat’ in ToR 2c. can be disregarded.

As regards ToR 2d., bulk milk is considered as the pooled milk collected from multiple cows of a dairy farm, stored in a bulk milk tank on the dairy farm before transport outside the farm. In this document, the term ‘farm’ is used to refer to establishments in which cattle or poultry are kept.

For ToR 2e., it has been clarified that the foods to be assessed in ToR2e (i) are raw milk, raw colostrum, dairy and colostrum‐based products (referred to as ‘relevant foodstuffs’ throughout this document) derived from dairy cows in the EU. Other lactating animal species are not considered. Raw milk is defined in Regulation No (EC) 853/2004 ¹ as ‘milk produced by the secretion of the mammary gland of farmed animals that has not been heated to more than 40°C or undergone any treatment that has an equivalent effect’. Dairy products are defined as ‘processed products resulting from the processing of raw milk or from the further processing of such processed products’. Colostrum is defined as ‘the fluid secreted by the mammary glands of milk‐producing animals up to 3–5 days post parturition that is rich in antibodies and minerals and precedes the production of raw milk’. Colostrum‐based products are defined as ‘processed products resulting from the processing of colostrum or from the further processing of such processed products’. The hypothetical scenario of contamination of bulk milk was considered for milk and dairy products while for colostrum and colostrum‐based products the hypothetical scenario was that the bulk colostrum is contaminated. The processing steps that were considered as relevant to include in the assessment are those having or presumably having an impact on the levels of viable H5N1 B3.13 genotype virus. The production scale was not considered per se, but artisanal production processes will not be excluded. For the scope of the assessment, a viable virus is defined as an infectious virus, i.e. a virus retaining infectivity towards its hosts or in in vitro models used to assess infectivity. It was considered that good hygiene practices (GHP) and good manufacturing practices (GMP) during food production are implemented and that the foods are consumed ‘as they are’ at the end of processing as worst‐case; thus, storage at retail or at the consumer level or cooking during food preparation were not covered. The risk for humans in ToR 2e. (ii) was translated into the probability of infection/illness or the probability of exposure to different levels of viable virus per serving for a consumer in the EU population. In case of absence of dose–response (DR) data in humans, animal model data (either mammals or just non‐human primates) were considered to assess the risk for humans due to the (oral/food‐borne) exposure. Apart from the ingestion of the food, also accidental aspiration during food or liquid consumption was considered potentially leading to a respiratory infection. For ToR 2e. (iii), measures along the food chain (starting from bulk milk/colostrum) to reduce the estimated levels from ToR 2e. (i) were listed, considering those either already in place or under experimental testing for possible implementation, and included, when relevant and feasible, a qualitative evaluation of their effectiveness (effect on levels). Economic or environmental impacts or user acceptance were not considered. The specific requirements for heat treatment of raw milk, colostrum and dairy or colostrum‐based products are reported in Regulation (EC) No 853/2004. Food business operators must ensure that the treatment satisfies the requirements laid down in Regulation (EC) No 852/2004 ² :

• Pasteurisation is achieved by a treatment involving: (i) a high temperature for a short time (HTST) (at least 72°C for 15 s); (ii) a low temperature for a long time (LTLT) (at least 63°C for 30 min); or (iii) any other combination of time–temperature conditions to obtain an equivalent effect, such that the products show, where applicable, a negative reaction to an alkaline phosphatase (ALP) test immediately after such treatment.

• Ultra‐high temperature (UHT) treatment is achieved by a treatment: (i) involving a continuous flow of heat at a high temperature for a short time (not less than 135°C in combination with a suitable holding time) such that there are no viable microorganisms or spores capable of growing in the treated product when kept in an aseptic closed container at ambient temperature and (ii) sufficient to ensure that the products remain microbiologically stable after incubating for 15 days at 30°C in closed containers or for 7 days at 55°C in closed containers or after any other method demonstrating that the appropriate heat treatment has been applied.

The mandate ToRs were translated into the following assessment questions (AQs) and subquestions (SQs):

ToR 2a.:

Assessment question 1:

What would be the potential impact of the infection of dairy cows in the EU with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13?

1. What would be the extent of spread?

2. What would be the economic impact caused by trade restrictions, control measure implementation and disease in animals?

3. What would be the societal impact caused by animal health impact, animal welfare impact, impact of animal disease on human society and human disease burden?

4. What would be the environmental impact?

ToR 2b.:

Assessment question 1:

Which measures could prevent the introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 into dairy cows and poultry in the EU?

1. Which measures could prevent the introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 through legal trade and through wild birds into dairy cows in the EU?

2. Which measures could prevent the introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 through legal trade into poultry in the EU?

Assessment question 2:

Which measures could prevent the spread of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 in dairy cows and poultry in the EU?

1. Which measures could prevent the spread of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 between EU dairy cows in the same establishment (within‐farm spread)?

2. Which measures could prevent the spread of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 between EU dairy cow establishments (between‐farm spread)?

3. Which measures could prevent the spread of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 between EU poultry establishments (between‐farm spread)?

This document does not include any assessment of measures to prevent introduction of the virus into poultry by wild birds nor measures to prevent the spread between poultry farms by wild birds, as these have already been assessed previously and are reflected in current legislation regarding control of HPAI in poultry flocks. Data from the US provide no indications that the HPAI B3.13 virus behaves differently in poultry flocks from the clade 2.3.4.4b variants that have circulated in EU in the past few years.

ToR 2c.:

Assessment question 1:

How would the current EU surveillance for HPAI need to be adapted to allow detection of an introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 into the EU dairy cows populations posing a significant animal or public health threat?

1. What are the current EU surveillance activities for HPAI?

2. Which EU surveillance activities focus on dairy cows?

3. How can the introduction of the H5N1 B3.13 virus into EU dairy cows be detected?

ToR 2d.:

Assessment question 1:

What is the likelihood of bulk milk being contaminated if EU lactating dairy cows are infected with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13?

1. What is the likelihood that an infected cow sheds the virus in the milk?

2. What is the likelihood that an infected dairy cow shedding the virus with the milk shows clinical signs that will enable the animal caretaker to notice that the cow is infected (e.g. mastitis, reduced milk production, reduced feed intake, changed behaviour)?

ToR 2e.:

Assessment question 1: What are the estimated levels of viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products produced in the EU at the end of processing, assuming contamination of bulk milk or bulk colostrum with this virus in the EU?

• SQ 1.1: In addition to raw milk and colostrum, which dairy products and colostrum‐based products consumed in the EU, are the relevant foodstuffs for the assessment?

• SQ 1.2: What are the relevant processing steps and associated parameters for the foodstuffs in SQ 1.1?

• SQ 1.3: What are the estimated levels of viable H5N1 B3.13 genotype virus in bulk milk and bulk colostrum?

• SQ 1.4: What are the estimated levels of viable H5N1 B3.13 genotype virus at the end of processing in the identified foodstuffs based on SQ 1.1 and SQ 1.2?

Assessment question 2: What is the mean probability of infection and illness per serving or, alternatively, what is the probability of exposure to different levels of viable H5N1 B3.13 genotype virus for consumers in the EU via the food‐borne route, given that the serving originates from milk or colostrum from a contaminated bulk tank in the EU?

• SQ 2.1: What is the probability of exposure to different levels of viable H5N1 B3.13 genotype virus for consumers in the EU through the consumption of servings of the identified foodstuffs (from SQ 1.4), assuming that the servings originate from milk or colostrum from a contaminated bulk tank in the EU?

• SQ 2.2: Does the available evidence provide a dose–response model that is applicable to humans, for the assessment of the probability of infection and the probability of illness given infection, given a known exposure to viable H5N1 B3.13 genotype virus through the oral route?

• SQ 2.3: If a dose–response model is available, what is the mean probability of infection and, if it can be assessed, illness per serving for consumers in the EU through the consumption of the identified foodstuffs considering the estimated levels of viable H5N1 B3.13 genotype virus (from SQ 1.4) in servings originating from milk or colostrum from a contaminated bulk tank in the EU?

Assessment question 3: What are the available measures along the food chain (from initially contaminated bulk milk or bulk colostrum up to consumer), and, if feasible, their effectiveness, to reduce the levels of viable H5N1 B3.13 genotype virus?

1.3. Additional information

In July 2025, four risk assessments addressed the risk associated with the occurrence of highly pathogenic avian influenza virus (HPAIv) in milk, colostrum or products thereof.

The Food and Agriculture Organization (FAO) ³ concluded in its preliminary rapid risk assessment of food‐borne avian influenza A (H5N1) virus that the risk of people acquiring avian influenza A (H5N1) from food is negligible based on the absence of any evidence of food‐borne transmission of influenza viruses to humans over the several decades of surveillance of influenza illnesses and the unlikely chance of exposure to the virus through food or drink. Avian influenza A (H5N1) virus is rendered non‐infectious by pasteurisation and adequate cooking temperatures (70°C). To further safeguard against all food‐borne illnesses, consumers are advised to drink only milk that has been pasteurised and to cook foods of animal origin prior to consumption assuring that no viable virus is left.

US Food and Drug Administration (US FDA) conducted a rapid assessment of the predicted risk to US consumers of cow's milk with a ‘bottom‐up’ and ‘top‐down’ approach. The bottom‐up approach consisted of a quantitative risk assessment model that integrated data on virus levels in milk (based on Spackman, Jones, et al., 2024), milk consumption and dose–response (using a modified version of the FSIS‐FDA‐APHIS model for influenza inhalation) (see Section 3.6.2). The top‐down epidemiological analysis linked current novel flu illness detection to the consumption of raw and pasteurised milk. The risk assessment model identified pasteurisation as a critical control step for H5N1 in milk and highlighted the need for (i) the targeted sampling of bulk raw milk in affected states pre‐pasteurisation, (ii) raw milk herd surveillance and sampling and (iii) a better understanding of ingestion as a route of H5N1 infections for humans (Chen et al., 2025).

UK Food Standards Agency (UK FSA) assessed the hypothetical risk to UK consumers from H5N1 B3.13 genotype virus in UK dairy cattle, milk, dairy products, colostrum and colostrum‐based products. The assessment sets out the risk to UK consumers in a hypothetical scenario where they are exposed to the virus by consuming milk and dairy products from cattle herds, if the virus began circulating in UK cattle. These products include pasteurised and unpasteurised cow's drinking milk, milk and dairy products, as well as colostrum and colostrum‐based products. The available evidence suggests that the virus is susceptible to heat treatment, and normal pasteurisation methods have been shown to reduce live infectious virus in milk by 4.44 log₁₀ to more than 6 log₁₀. This assessment determined that the probability of individuals being exposed to an infectious dose of H5N1 virus from pasteurised milk, pasteurised colostrum or dairy products made from pasteurised milk was negligible, on a per portion basis; this probability of exposure was given a medium level of uncertainty. Being exposed to an infectious dose of H5N1 virus via raw cow's drinking milk was instead attributed a medium probability, on a per portion basis. Considering the data gaps and applying a precautionary approach, also the probability of individuals being exposed to an infectious dose per portion of dairy products made with unpasteurised milk was determined to be medium; a high level of uncertainty was given to this probability of exposure (Adams et al., 2025).

The assessment of the Ministry of Agriculture, Fisheries, Food Security and Nature of the Netherlands also considered this hypothetical scenario using a descriptive exposure assessment. They concluded that the virus remains infectious in refrigerated raw milk, and that only the pasteurisation process (or stronger heating processes) has so far been proven effective in making raw, HPAIv‐contaminated milk safe for human and animal consumption. Human cases of disease related to dairy cattle in the US are linked to exposure to the animals and not to drinking raw milk (products), and there is currently no evidence of infection with HPAIv through food to humans. The probability of infection by HPAIv via the consumption of contaminated dairy products is currently considered low. However, transmission through food could increase due to genetic changes in the virus (NVWA, 2025).

2. DATA AND METHODOLOGIES

2.1. ToR 2a. Assessment question 1: Potential impact of the infection of dairy cows in the EU with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13

To address this ToR, an existing risk assessment model, the ‘Living One Health Risk Assessment tool (L'ORA) was used (De Vos et al., 2025). A detailed description of the data and methodology used can be found in Annex A.

Briefly, the tool consists of four modules. The first module describes the spatio‐temporal distribution of cases in US dairy cattle and poultry using data from the World Animal Health Information System (WAHIS), the US Department of Agriculture (USDA) and the Global Initiative on Sharing All Influenza Data (GISAID).

The second module estimates the rate of virus introduction into the EU. It combines the probability of infection of each functional group (dairy cattle, poultry, wild birds) via each possible pathway with the number of livestock trade and migratory bird movements into each NUTS2 (Nomenclature of Territorial Units for Statistics; level 2 corresponds to the provincial or state‐like level) region and time period.

The third module estimates the extent of spread and number of outbreaks (the occurrence of an outbreak would be size 1 or more), country and time period. It combines epidemiological parameters (like basic reproduction number, length of high‐risk period, time between infection generation) with population data.

The fourth module estimates the potential impact of infection in terms of (a) economic impact caused by direct and indirect agricultural costs, trade restriction costs, human disease costs and costs of side effects, ⁴ (b) societal impact caused by human disease burden, impact of control measures on animal welfare, impact of disease on animal health and welfare and human societal burden and (c) environmental impact caused by spread to wildlife (here: wild birds), spread to threatened species, wildlife case fatality and size of the exposed area. These estimates combine the results from the third module e.g. number of infected epidemiological units and defined economical, societal and environmental parameters related to control measures taken.

The results include an assessment of the uncertainty of the estimates that are related to the epidemic size and affected geographical areas.

2.1.1. Data needs and sources of data

The model was adapted to the assessment of the potential impact of the infection of EU dairy cows with the H5N1 genotype B3.13 virus. Overall, data were drawn from different sources: virus characteristics (from the relevant sections of the scientific report (EFSA, 2025)), outbreaks in the US (from WAHIS, USDA and GISAID sources), pathways for entry through trade and migratory birds (from EU TRACES and eBird sources), population data (from Eurostat, eBird, FEDIAF European PetFood). Where information was lacking in peer‐reviewed and grey literature (e.g. epidemiological parameters for HPAI transmission in cattle), assumptions were defined through discussions with the working group experts and by extrapolation from parameters available for poultry and other species. The species and commodities that were considered in the assessment by the tool are shown in Annex A.

2.1.2. Evidence synthesis

The model outcomes were synthesised into an estimate of risk of introduction for the NUTS2 regions of EU member states, and the estimated extent of spread for the different NUTS2 regions of the EU member states as a result of introductions that occur in domestic and wildlife amplifying species (wild birds). Further, the estimated impact of disease for the different member states in the EU is provided, given that it is introduced in the NUTS2 region of interest, either via domestic animals or wild birds.

2.2. ToR 2b. Assessment question 1: Measures that could prevent the introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 into dairy cows in the EU

To address this ToR, a combined evidence assessment and uncertainty analysis using a structured expert opinion elicitation process was applied to answer the questions on which measures could prevent the introduction of the H5N1 B3.13 genotype virus into dairy cows and into poultry in the EU.

2.2.1. Data needs and sources of data

The data needed to answer the questions concern the virus, the outbreak in the US and the measures taken to control it as summarised in the scientific report (EFSA, 2025). In addition, previous EFSA opinions (, 2025; EFSA AHAW Panel, 2024) were reviewed for further measures that could be used to prevent the introduction of the virus into EU dairy cows or poultry. The measures listed in these documents were collated to create the list of measures for the expert opinion elicitation process (see Appendix B).

2.2.2. Evidence dossier

The list of measures that could be used to prevent the introduction of the virus into EU poultry or dairy cows was reviewed and, if necessary, was completed by the working group members. This list and the results of the scientific report (EFSA, 2025) constituted the evidence dossier for the structured expert opinion process.

2.2.3. Expert opinion elicitation questions

In the assessment, two different scenarios were considered for a situation where the virus has been introduced for the first time to the EU. These included the introduction into dairy cattle (scenario 1, with sub‐scenario 1A covering the introduction into a dairy farm in the EU via wild birds and sub‐scenario 1B covering the introduction into a dairy farm in the EU via trade) and the introduction into poultry farms in the EU via trade (scenario 2).

Scenario 1 → Introduction into dairy cattle: The application of the measures should lead to a significant reduction in the probability of introduction of H5N1 B3.13 genotype virus into a dairy farm via exposure to wild birds or traded live animals and/or their products, assuming this was the first introduction of the virus into the EU. Two sub‐scenarios were considered depending on the route of introduction. The experts were requested to evaluate each measure based on its perceived effectiveness to achieve the following objectives:

• Sub‐scenario 1A – Introduction into dairy cattle farms by wild birds: minimise the number of newly H5N1 B3.13 genotype virus‐infected dairy farms that would occur, if the measure was implemented, assuming that virus introduction occurs via migratory birds.

• Sub‐scenario 1B – Introduction into dairy cattle farms by trade: minimise the number of newly H5N1 B3.13 genotype virus‐infected dairy farms that would occur, if the measure was implemented, assuming that virus introduction occurs via trade.

Scenario 2 → Introduction into poultry: The application of the measures should lead to a significant reduction in the probability of introduction of H5N1 B3.13 genotype virus into a poultry farm via exposure to traded live animals and/or their products, assuming this was the first introduction of the virus into the EU. The experts were requested to evaluate each measure based on its perceived effectiveness to achieve the following objectives:

• Scenario 2 – Introduction into poultry farms by trade: minimise the number of newly H5N1 B3.13 genotype virus‐infected poultry farms that would occur, if the measure was implemented, assuming that virus introduction occurs via trade.

For each measure listed for prevention of the spread of the virus in the EU, the EFSA experts were asked to rate its effectiveness under the different scenarios. The list of measures assessed for the different scenarios is shown in Appendix B. In addition to assessing the different measures separately for each scenario, the experts provided the rationale explaining their judgements.

In a second step, WG members were asked to qualitatively assess the feasibility of each measure according to the agreed scale shown in Table 1 and to provide the rationale for their judgement.

TABLE 1.

Ratings of the feasibility of measures (e.g. changing or abandoning current practices, implementing new practices and/or measures) to prevent the introduction of the virus into dairy cows and poultry in the EU.

Feasibility	Explanation
Extremely low	The measure has several major technical, economic or social difficulties, making its implementation in practice almost impossible.
Low	The measure can be implemented with substantial technical, economic or social difficulties.
Moderate	The measure can be implemented in practice with moderate technical, economic or social difficulties.
High	The measure is already in use and can be easily implemented in practice with limited technical, economic or social difficulties.
Unknown	Knowledge about the feasibility of the measure is (currently) lacking.

2.2.4. Execution of the expert opinion elicitation process

The experts involved in the opinion elicitation consisted of the four working group experts. The expert opinion process was carried out in three steps.

In the first step, the working group experts reviewed the evidence dossier and individually rated the effectiveness of each measure regarding the different scenarios/sub‐scenarios. To this end, they were asked to allocate points to each measure based on their perceived effectiveness to achieve the objectives below, with more points indicating a higher perceived effectiveness of the measure for the given scenario. The total number of points that could be assigned in each scenario/sub‐scenario was equal to the number of measures to be assessed for that scenario multiplied by 10, with no restriction of the number of points that could be allocated per measure. Following the effectiveness assessment, experts assessed the feasibility of the listed prevention measures for each introduction sub‐scenario for the two target populations. They noted the reasoning for their answers regarding effectiveness and feasibility for each measure. The list of measures assessed for the different scenarios is shown in Appendix A.

For the second step, individual judgements were anonymised, summarised graphically (for effectiveness) or as tables (for feasibility), and collated in a report including all judgements for each measure‐scenario combination along with the reasoning provided by the experts. The report was shared with the experts to review their judgements in the light of the points raised by others.

In the third step of the process, a group meeting was conducted during which the individual judgements were graphically displayed. The experts discussed the reasoning for their individual assessments. Individual judgements were used to group the measures in terms of their effectiveness based on the similarity of the total number of points received, with no limit on the number of measures per rank. The measures considered more effective were allocated to the first group (rank 1), followed by the next lower assessed measures (rank 2), etc. It should be noted that this ranking indicates the relative effectiveness of the measures depending on the scenario, and that the difference between the ranks is not necessarily the same across the scenarios. For the feasibility assessment, a consensus was reached on the most appropriate category Table 1 for each measure (see Table 1). The rationale for each group judgement was documented.

2.2.5. Synthesis

The results of the evidence assessment and uncertainty analysis were summarised in a tabular format for each assessment question, showing the outcome for each measure‐scenario combination regarding effectiveness and feasibility. Based on these results, a concluding section was drafted, followed by recommendations.

2.3. ToR 2b. Assessment question 2: Measures that could prevent the spread of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 in dairy cows and poultry in the EU

To address this ToR, a combined evidence assessment and uncertainty analysis using a structured expert opinion elicitation process was used to answer the questions on which measures could prevent the spread of the H5N1 B3.13 genotype virus in dairy cows and in poultry in the EU, following the process described in Sections 2.2.1, 2.2.2, 2.2.4 and 2.2.5.

The different scenarios for spread considered in this assessment included the within‐farm spread of the H5N1 B3.13 virus in dairy farms (scenario 3), secondary spread to dairy cattle farms (scenario 4, with sub‐scenario 4a covering infections due to contact with a previously infected dairy farm and sub‐scenario 4b covering infections due to contact with a previously infected poultry farm) and spread from dairy to poultry farms (scenario 5).

Scenario 3 → Spread within dairy cattle farms: The experts were requested to evaluate each measure based on its perceived effectiveness to achieve the following objectives:

• Scenario 3 – Spread within dairy farms: minimise the number H5N1 B3.13 genotype virus‐infected dairy cows that would occur if the measure was implemented.

Scenario 4 → Secondary spread to dairy cattle farms: This scenario assumes H5N1 B3.13 genotype virus has already been detected in the EU livestock population (either in cattle or poultry). Two sub‐scenarios were considered depending on the route of spread. The experts were requested to evaluate each measure based on its perceived effectiveness to achieve the following objectives:

• Sub‐scenario 4a – Spread from dairy to dairy farms: minimise the number of secondary H5N1 B3.13 genotype virus‐infected dairy farms that would occur if the measure was implemented, assuming that virus spread occurs through contact with another infected dairy farm in the EU

  Sub‐scenario 4b – Spread from poultry to dairy farms: minimise the number of H5N1 B3.13 genotype virus‐infected dairy farms that would occur if the measure was implemented, assuming that virus spread occurs through contact with an infected poultry farm in the EU

Scenario 5 → Spread from dairy to poultry farms: The experts were requested to evaluate each measure based on its perceived effectiveness to achieve the following objectives:

• Scenario 5 – Spread from dairy to poultry farms: minimise the number of newly H5N1 B3.13 genotype virus‐infected poultry flocks that would occur if the measure was implemented, assuming that virus spread occurs through contact with an infected dairy farm in the EU.

For each measure listed for prevention of the spread of the virus in the EU, the EFSA experts were asked to rate its effectiveness under the different scenarios, as explained in Section 2.2.3. The list of measures assessed for the different scenarios is shown in Appendix B. In addition to assessing the different measures separately for each scenario, the experts provided the rationale explaining their judgements.

In a second step, WG members were asked to qualitatively assess the feasibility of each measure according to the agreed scale shown in Table 1 and to provide the rationale for their judgement.

2.4. ToR 2c. Assessment question 1: Adaptations of the current EU surveillance for HPAI needed to allow detection of an introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 into the EU dairy cows populations

To address this ToR, evidence from scientific literature and study reports were used to answer the question of which surveillance activities could detect an introduction of the H5N1 B3.13 genotype virus into EU dairy cattle.

2.4.1. Data needs and sources of data

The current EU surveillance activities for HPAI and additional EU surveillance activities focussing on dairy cows for other purposes were collated from legal documents and EFSA opinions. This list was reviewed and, where needed, completed by the WG experts. In addition, EFSA asked the members of its Network on Animal Health to provide a brief description (e.g. sampling approach, test used) of any HPAI surveillance done in dairy cattle in their country.

The EURL for Avian Influenza and Newcastle Disease provided information on the diagnostic options for the H5N1 B3.13 genotype virus in milk and other specimens, including available information on their sensitivity and specificity.

The data on the virus, the outbreak in the US and pathways for entry through trade and migratory birds described in the relevant sections of EFSA's scientific report were used (EFSA, 2025).

2.4.2. Evidence appraisal

The evidence identified was appraised through discussion by the EFSA experts.

2.4.3. Evidence synthesis

The working group synthesised the evidence using a qualitative approach. Utilising the temporal and spatial results of the assessment of the migratory pathways for H5N1 B3.13 genotype virus introduction, the information on ongoing surveillance of HPAI and on the diagnostic options, the experts described possible surveillance approaches that would allow the detection of the introduction of the HPAI B3.13 genotype virus into the EU dairy population. Sources of uncertainty were qualitatively explored and their effect on the results was discussed.

2.5. ToR 2d. Assessment question 1: Likelihood of bulk milk to be contaminated if EU lactating dairy cows are infected with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13

To address this ToR, a combined evidence assessment and uncertainty analysis using a structured expert opinion elicitation process has been used.

2.5.1. Data needs and sources of data

The experts based their assessment on data regarding the virus, the outbreak in the US and pathways for entry through trade and migratory birds described in EFSA's scientific report (EFSA, 2025). Specifically, information on virus shedding in the milk of infected dairy cattle, the within‐herd prevalence of H5N1 B3.13 genotype virus infection and clinical signs observed in infected US dairy farms, and the duration of the incubation period, were highlighted as relevant for the assessment.

2.5.2. Expert opinion elicitation question

The experts agreed two conditions had to be fulfilled for bulk tank milk contamination to occur in an infected farm: (1) one or more infected cows must be shedding the virus and (2) the H5N1 B3.13 genotype virus infection must remain undetected, allowing milk of the infected cows to be added to the bulk tank. Factor (1) determines whether any milk in the farm can be contaminated and depends on the proportion of cattle shedding the virus which is related to the prevalence of infection. Factor (2) determines whether the milk of infected cows reaches the bulk tank. This depends on the probability that a lactating H5N1 B3.13 genotype virus‐shedding dairy cow remains undetected, which in turn depends on the proportion of infected cows that show clinical signs (mastitis, reduced milk production, reduced feed intake, changed behaviour) and the probability of including H5N1 B3.13 genotype virus in the differential diagnosis in case of observing such clinical signs. It is assumed that once a cow is observed to be clinically affected, its milk will no longer be added to the bulk tank. It is also assumed that once H5N1 B3.13 genotype virus infection is detected at the farm (i.e. there is at least one cow with clinical signs in which H5N1 B3.13 genotype virus is detected), animals will be closely monitored and control measures implemented. From that point onward, the probability of further contamination of bulk tank milk is considered to be zero, although milk produced before detection could already have been contaminated.

Based on the above considerations, the experts were asked to answer the question ‘If 100 dairy farms ⁵ become infected with H5N1 B3.13 genotype virus in the EU in the next year, what is the proportion of farms in which bulk milk will become contaminated before the H5N1 B3.13 genotype virus infection is detected in the farm ⁶ ?’. Experts had to provide the median value as well as the 2.5% lower limit and the 97.5% upper limit for their judgement. The uncertainty of the experts' judgements was captured based on the width of the interval chosen, as proposed in EFSA's Guidance on Uncertainty Analysis in Scientific Assessments. ⁷

2.5.3. Execution of the expert opinion process

The expert opinion process was carried out in three steps as described in Section 2.2.4. The experts involved in the elicitation consisted of four EFSA experts.

2.5.4. Synthesis

The results of the evidence assessment and uncertainty analysis were summarised graphically, and a concluding section was drafted, followed by recommendations.

2.6. ToR 2e. Probability of infection/illness for a consumer in the EU population due to the exposure to viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products through the food‐borne route and measures to mitigate the risk

The approach to answer ToR 2e was defined in advance and can be found in the protocol (Annex B). Additional information is provided here below. The literature searches were conducted in April, June and on 15 September 2025. The publication status of the preprints was last checked on 21 October 2025.

Data on infectious virus particles were collected from various literature sources according to the adopted titration method (i.e. embryonated chicken eggs or cell cultures) and were expressed as Egg Infectious Dose 50% (EID₅₀, number of virus particles required to infect 50% of embryonated chicken eggs), Tissue Culture Infectious Dose 50% (TCID₅₀, amount of virus required to infect 50% of cell culture wells) or Plaque Forming Units (PFU, amount of virus capable of lysing host cells and form a plaque). EID₅₀ was used in the assessment, and for some data a conversion was performed as described in Appendix B. The uncertainty associated to the different sensitivity of the titration systems was mentioned in the uncertainty analysis (Appendix G).

For SQ 1.3, apart from relying on observed levels of viable H5N1 B3.13 genotype virus in bulk milk and bulk colostrum in the US, a modelling approach ― based on Koebel et al. (2024) (referred to as bulk tank model) ― was used to explore the impact of farm specific parameters (e.g. proportion of animals contributing contaminated milk to the bulk tank, herd size) on the levels of viable H5N1 B3.13 genotype virus in bulk milk (and bulk colostrum). The bulk tank was considered a tank in which the milk from one herd is collected. Mixing milk from different herds or farms in larger bulk tanks or in milk silos at the processing level, which would result in dilution of milk contaminated with the virus, was not considered. The high‐risk products, including raw milk and unpasteurised dairy products considered in the assessment, will primarily be produced and sold locally, in small‐scale production settings. Literature was screened to obtain data on the shedding of H5N1 B3.13 genotype virus in milk from individual infected cows. Samples from clinically diagnosed animals were not considered eligible as diagnosed animals would be excluded from milk production (Appendix C). Additionally, literature was screened to gather data on other parameters, such as the average herd size in the EU, the daily amount of milk produced per cow, and the production loss of infected animals. Appendix D contains the description of the exposure assessment model used in this opinion, including the modelling approach used to estimate the concentration of viable H5N1 B3.13 genotype virus in bulk milk. The full exposure assessment model is available at https://zenodo.org/uploads/17457398.

For SQ 1.4, to assess thermal inactivation, the dataset was built using the D‐values (i.e. decimal reduction time or the time required at a given temperature to kill 90% of the exposed microorganisms) extracted by Hessling (2022). Three rectifications to published values were made ⁸ and exclusion criteria were applied to ensure that the final dataset reflected only conditions and matrices relevant and comparable to those expected in milk, colostrum and related products. In detail, only studies reporting D‐values within the relevant temperature range (48°C < T < 90°C) were retained. Data obtained in dried egg products and on surfaces were excluded, as these matrices are associated with low‐water activity (a_w), which is known to have a protective effect on virus survival (Morris et al., 2021). Additionally, data from manure were excluded due to the potential confounding effect of elevated pH levels (often >8), which may increase virus sensitivity to heat.

The classical Bigelow model was applied to the refined dataset in which the logarithm of the decimal reduction time (${\log }_{10}\left(D\right)$) is assumed to vary linearly with temperature. The model is defined by two parameters: the reference temperature ($T_{\mathrm{ref}}$, set at 70°C), at which ${\log }_{10}\left(D\right)={\log }_{10}\left(D_{\mathrm{ref}}\right)$, and the z‐value ($z_T$), which represents the temperature increase required to reduce $D$ by one log₁₀ unit (i.e. a 10‐fold reduction). The model is expressed as follows:

$${\log }_{10}\left(D\left(T\right)\right)={\log }_{10}\left(D_{\mathrm{ref}}\right)-\frac{T-T_{\mathrm{ref}}}{z}$$

The model was fitted to the selected dataset using non‐linear least squares in R through the ‘nls’ function with appropriate bounds and the ‘port’ algorithm to constrain the parameter space. The fit was visually inspected, confirming a satisfactory agreement between observed and predicted values. To quantify uncertainty around the parameter estimates, a non‐parametric bootstrap procedure was performed using the ‘rsample’ and ‘broom’ packages. A total of 1000 bootstrap resamples were generated, and the model was re‐fitted on each resample. The resulting distributions of the parameters were summarised using empirical quantiles (2.5th, 50th and 97.5th percentiles) to provide robust confidence intervals (CI). Bootstrap estimates were separated into two components — one for ${\log }_{10}\left(D_{\mathrm{ref}}\right)$ and one for $z_T$. Pairs of values were used to assess uncertainty of D‐values according to temperature conditions. This secondary Bigelow‐type model was further evaluated using experimental data reporting HPAI H5N1 inactivation in milk products based on the literature search. This validation focused on studies providing either point estimates or right‐censored log₁₀ reductions, without explicit estimation of D‐values. Two temperature conditions were considered: 63°C and 72°C (see Appendix E). The developed thermal inactivation model was used to derive the predicted log₁₀ reductions achieved for each selected food category according to the thermal treatment process applied. The full model and data used are available at https://zenodo.org/uploads/17457287.

The impact of each of the other processing steps, such as acidification/coagulation, draining and ripening (depending on the foodstuff) was evaluated using a single value or a distribution of the estimated log₁₀ reduction, based on the available evidence derived from the literature search. For acidification/fermentation, it was assumed that the coagulation effect is linked to the final pH and that duration has no impact. Additionally, for cheese production, in the draining step viral particles present in milk can partition between the curd and whey based on their interactions with milk components, especially proteins in the curd matrix. To reflect this mechanism, a virus distribution model was drafted, using the partition coefficient to curd ($P_C$), defined as follows with $C_{\text{curd}}$ and $C_{\text{whey}}$ as the virus concentrations in the curd and whey, respectively:

$$P_C=C_{\text{curd}}/C_{\text{whey}}$$

The partition coefficient $P_C$ reflects the virus's affinity for curd relative to whey. When $P_C=1$, the virus distributes equally between both phases. A value $P_C>1$ indicates preferential association with the curd. Conversely, $P_C<1$ suggests that the virus remains mostly in the whey.

Assuming mass conservation, the initial virus load in milk is distributed between curd and whey as follows, considering $Y$ = fraction of milk retained in the curd and $\left(1-Y\right)$ = fraction going to whey:

$$C_{\text{total}}=Y\times C_{\text{curd}}+\left(1-Y\right)\times C_{\text{whey}}$$

Substituting $C_{\text{whey}}=C_{\text{curd}}/P_C$ into the equation and solving for $C_{\text{curd}}$:

$$C_{\text{curd}}=P_C\times C_{\text{total}}/\left(Y+\left(1-Y\right)/P_C\right)$$

The overall log₁₀ reduction and concentration of viable virus in one unit (volume or weight, according to the product) of the final foodstuff was calculated assuming a mean initial concentration of infectious H5N1 B3.13 genotype virus in bulk milk and considering the log₁₀ reductions achieved by the different processing steps.

For SQ 2.1, as additional information to the outcome of the levels in the foodstuffs, estimates for the serving sizes (in g or mL) of the selected food products based on the mean daily intakes using the most appropriate FoodEx2 category (i.e. the standardised food classification and description system developed by EFSA) were extracted from the most recent survey per Member State (MS) for adults in the EFSA food consumption database. ⁹ Data were available from 27 countries. ¹⁰ The survey starting date ranged from 2003 to 2021. The overall mean of the mean for each country on consuming days only was calculated. For colostrum, as no data were available, data were retrieved from other sources, such as literature and webpages.

Available dose–response (DR) models were retrieved through the literature search for SQ 2.2. Information on experimental infections with H5Nx strains through the oral route (gavage, intragastric, per os) in mammalian animal models was also extracted from the literature. Only studies in animals with no previous immunity were considered. Responses were summarised considering clinical infection. The probability of infection through the oral route in animal models was estimated according to the species using a beta‐Poisson model and a DR model was developed for ferrets. Appendix F contains the information retrieved from the literature and the DR models. The DR model developed for ferrets can be found in https://zenodo.org/uploads/17457024.

For SQ 2.3, for each of the selected foodstuffs, the probability was calculated for the doses ingested through consumption to be larger than threshold values selected through expert judgement (and expressed in EID₅₀).

As recommended by the EFSA guidance and related principles and methods on uncertainty analysis in scientific assessments (EFSA Scientific Committee, 2018a, 2018b), an uncertainty analysis was undertaken as described in the protocol (Annex B), the sources of the main uncertainties related to ToR 2e were identified by the experts in the WG and their impact on the conclusions described (Appendix G). The impact of the uncertainty on the answers to SQ2.3 was derived through the uncertainty estimates obtained from the model only.

3. ASSESSMENT

When considering the outcome of the assessments done for this opinion, it is important to note that experimental evidence indicates that H5N1 B3.13 genotype virus is unlikely to behave differently from H5N1 clade 2.3.4.4b strains already circulating in Europe (see e.g. Halwe et al., 2025), and the absence of evidence of cattle infections despite widespread and repeated exposure to HPAI viruses in wild birds and poultry in Europe indicates that spillover to dairy cattle under European conditions is subject to a very narrow transmission bottleneck.

3.1. Assessment of the potential impact of the infection of dairy cows in the EU with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 (ToR 2a.)

Below is a summary of the outcome of the impact assessment with the L'ORA tool. More details can be found in Annex A.

The results of the L'ORA model indicate that an introduction of HPAI B3.13 into EU poultry or dairy cattle by trade from the US is highly unlikely (annual rate of less than 10⁻⁵, score = 0) for all NUTS2 regions within the EU. This low likelihood is attributable to: (1) very limited trade between the US and the EU in commodities such as live animals, fertilised eggs and germinal products and (2) low probabilities that susceptible livestock will be exposed to imported products of animal origin, such as unpasteurized dairy products. Bretagne (France) shows a slightly higher estimated rate of introduction into poultry compared to other NUTS2 regions due to some imports of live poultry and fertilised eggs from the US, although its overall introduction score remained in the same category (score = 0). The rates may be underestimated for certain EU member states, because postcode data were missing or invalid for many consignments of fertilised eggs and germ plasm from the US. Recently, H5N1 B3.13 RNA was detected in semen collected from one of three bulls tested in a large US dairy operation. Further attempts to isolate live virus or to generate a complete genome sequence failed. Moreover, absence of a clear seroconversion in the bull that tested positive do not allow a clear interpretation of this finding (i.e. accidental contamination at the time of collection or viral elimination by this route) (Lim et al., 2025).

The probability of introduction of HPAI B3.13 into EU wild birds by migratory wild birds from the US also remains highly unlikely (annual rates of less than 10⁻⁵, score = 0) for most NUTS2 regions within the EU (Figure 1). This low likelihood is attributable to the limited migration events from North America to the EU, and the absence of direct migration routes from the US. However, L'ORA allowed for the possibility that H5N1 B3.13 genotype virus could be present in the North American region due to birds in the US migrating north into Canada and Greenland and subsequently mixing with those wild birds migrating to Northwest Europe. Bird ringing data from the UK supports this possibility, as some birds caught in the UK had also been caught in Northern Canada and Greenland, both part of the North America region. As a result, some NUTS2 regions show higher rates (up to 10⁻²–10⁻¹, score = 4), such as the west of Ireland (IE04 – Northern & Western Ireland; IE05 – Southern Region) and west of France (FRH0 – Bretagne; FRI3 – Poitou‐Charentes) and thus are the regions with the highest risk of introduction of H5N1 B3.13 genotype virus. Estimated monthly rates of introduction into regions are highest in autumn period (September–October). These higher rates are linked to predicted migratory movement of wild birds, especially brent geese and barnacle geese, which migrate in large numbers from Northern Canada and Greenland to Northwest Europe. While no direct migratory route from the US was confirmed, L'ORA allowed for the possibility that H5N1 B3.13 genotype virus could be present in the North American region due to birds in the US migrating north that some birds caught in the UK had also been caught in Northern Canada and Greenland, both part of the North America region. However, overall, the estimates of introduction rates are subject to considerable uncertainty. The number of wild birds migrating between US and Europe is highly uncertain and likely overestimated, while limited data on the actual prevalence of H5N1 B3.13 genotype virus in North American wild birds adds further uncertainty.

FIGURE 1.

Score for the annual rate of introduction (across all pathways) into wild birds in the EU (NUTS2 level) for the period April 2024 – March 2025. (Risk scores: 0 = < 10⁻⁵; 1 = 10⁻⁵–10⁻⁴; 2 = 10⁻⁴–10⁻³; 3 = 10⁻³–10⁻²; 4 = 10⁻²–10⁻¹).

The results of L'ORA indicate that the H5N1 B3.13 genotype virus outbreak occurrence in EU poultry, dairy cattle and wild birds is highly unlikely for all NUTS2 regions within the EU (interquartile probability estimates mostly ranging from 10⁻¹⁴ to 10⁻⁴) (Figure 2) (Annex A, module 2. Rate of incursion). This low likelihood reflects the very low estimated rate of introduction described above, as the probability of outbreak occurrence in the L'ORA model is driven by the expected rate of introduction and the basic reproduction numbers (R0) within and between species.

FIGURE 2.

Estimated probability of outbreak occurrence into poultry, dairy cattle and wild birds in the EU (NUTS2 level) for the period April 2024 – March 2025.

The total number of outbreaks in the L'ORA model is determined by the sum of the number of outbreaks during the high‐risk period and those occurring once control measures are applied. These numbers are calculated based on a list of transmission parameters (in particular R0, incubation period and generation time), which take different values during these two periods (meaning for ex. that R0 is reduced to reflect control measures implementation, without specifying particular interventions). Outbreak size is thus reflecting the potential spread after an assumed introduction has occurred.

If an outbreak was to occur, the L'ORA predictions showed that the expected epidemic size in EU poultry (Interquartile range (IQR) 240–713), dairy cattle (IQR 26–81) and wild birds (IQR 250k‐923k) 1 year after introduction would be most substantial if HPAI B3.13 were introduced by migratory wild birds (Table 2). The reason is that no effective measures can be implemented to reduce virus spread among wild birds, resulting in continued exposure. The numbers of outbreaks were generally larger when the first outbreak occurred in winter (January) compared to summer (July) (Annex A, module 3. Extent of spread.) If control measures were applied in poultry only Table 2, the expected numbers of outbreaks in EU cattle would be substantially larger (IQR 161‐17k). The modelled numbers rely on several input parameters (detailed in Annex A), particularly the basic reproduction number (R0), which is assumed to decrease when control measures are applied in a given population. However, because control measures are considered more difficult to implement or less effective in wild birds, their R0 is assumed to remain unchanged, resulting in larger predicted numbers of outbreaks with introductions from wild birds compared with introductions from dairy cattle or poultry, for which R0 is assumed to be reduced.

TABLE 2.

Estimated numbers of outbreaks ¹¹ in each population summarised over all EU Member States during the high‐risk period and once control measures are applied in all populations or only in poultry populations (median and IQR).

	Estimated number of outbreaks (median and IQR)
	Dairy cattle		Poultry		Cats		Spillover to humans		Wild birds
Source population for introduction	Control measures applied in all populations	Control measures applied in poultry only	Control measures applied in all populations	Control measures applied in poultry only	Control measures applied in all populations	Control measures applied in poultry only	Control measures applied in all populations	Control measures applied in poultry only	Control measures applied in all populations	Control measures applied in poultry only
Dairy cattle	8 (8–8)	3531 (669–17,218)	4 (4–4)	19 (7–37)	8 (8–8)	5275 (244–25,146)	9 (9–9)	5296 (244–25,672)	5297 (5240‐5317)	21,772 (8371–31,436)
Poultry	4 (3–4)	556 (185–618)	43 (41–44)	43 (33–45)	4 (3–4)	576 (246–624)	5 (4–5)	576 (246–624)	49,576 (45,573‐51,304)	50,625 (12,962–51,598)
Wild birds	51 (26–81)	2082 (161–3586)	470 (240–713)	396 (154–802)	52 (26–82)	2490 (242–3857)	69 (35–107)	2494 (242–3859)	531,331 (244,333‐923,062)	597,883 (158,356–1,028,946)

Note that the outbreak occurrence probability is estimated to be very low, meaning that these estimates represent only a hypothetical scenario in the unlikely event that an outbreak occurs. While the IQR reflects the variation across calculations, influenced by the month of virus introduction, the NUTS2 region of introduction and the country, there is additional uncertainty on the estimates obtained related to lack of knowledge about key parameters used in the analysis such as wild bird population sizes, parameter values (R0) and model assumptions (e.g. homogeneous mixing, fully susceptible population) and how these may change over time. As a result, the estimates should not be interpreted as absolute estimates, rather, the table is intended to illustrate the relative differences in estimated numbers of outbreaks depending on which populations are targeted by control measures. For example, the number of human infections is likely overestimated, as the model includes the entire human population rather than only those with occupational or environmental exposure to dairy cattle, poultry (such as farmers) or wild birds. Using census or occupational data, where available, would yield more realistic data, although accounting for exposure to wild birds remains challenging. Consequently, the modelled numbers of human infections are not in line with the numbers observed during the outbreaks in US dairy farms. Similarly, the estimated wild bird cases might also differ from observations due to differences in case definition between countries, differences in surveillance efforts and not all birds dying from infection. Finally, assuming that the H5N1 B3.13 genotype virus behaves similarly to clade 2.3.4.4b viruses currently circulating in Europe, the estimated number of outbreaks in cattle appears unexpectedly high. Overall, this implies that the model outputs should be interpreted with caution and not regarded as exact predictions, but rather as comparative indicators to evaluate different scenarios (e.g. with and without control measures applied to dairy cattle) within the model's context.

The overall impact in the L'ORA model is determined by the estimated epidemic size and a list of impact parameters. Each type of impact is further influenced by control measures, which in the L'ORA model include both compulsory measures applied in poultry and additional measures for cattle, such as movement restrictions (economic impact), carcasses disposal (economical and environmental impact), bulk milk surveillance (economic impact), indoor confinement (societal impact), creating unattractive environment for wild birds (environmental impact) and culling of the first (up to five) infected cattle farms (economic impact). Again, overall impact is independent of the outbreak occurrence probability.

If an outbreak was to occur, and the control measures defined in the L'ORA model were to be applied in both poultry and cattle, the predictions showed, that the overall impact (economic, societal and environmental) of H5N1 B3.13 genotype virus outbreaks in the EU was generally high (score 3) for most countries (Figure 3). The highest scores were in poultry and wildlife in France and Spain (score = 4). This is due to a combination of high costs due to larger outbreaks involving more animals, a higher environmental impact due to the presence of endangered wildlife species in these regions and for Spain a larger impact of indoor confinement on the welfare of cattle in winter. The overall impact in dairy cattle was relatively lower than in poultry, in particular due to the economic component, primarily caused by the large difference in the number of culled farms (which was only the first five infected farms in cattle).

FIGURE 3.

Overall impact scores obtained for three populations of introduction (poultry, dairy cattle and wild birds) when HPAI B3.13 is introduced in winter (January). Left panel with control measures applied in poultry and dairy cattle populations, right panel with control measures in poultry only. The bars show the variation in scores between NUTS2 regions of introduction per country, meaning that the relative frequency corresponds to the percentage of NUTS2 regions within each country assigned to each impact score category. Scores categories are a combination of the economic, environmental and societal scores which represent a mix of monetary, ecological and welfare dimensions, the resulting score has no physical unit.

If the control measures defined in the L'ORA model were applied in poultry only, the infection was expected to spread more widely among cattle and also in poultry farms, due to spread from infected cattle farms. However, the higher costs of larger outbreaks in poultry would be largely offset by the cost savings from not implementing control measures in dairy cattle, resulting in similar overall direct economic cost compared to the scenario where measures are applied in both sectors. On the other hand, the increase in cattle outbreaks would likely exacerbate other impacts, such as production losses and societal impacts related to reduced animal welfare due to increased disease burden in cattle.

Focusing on the two MS showing the highest risk of outbreak occurrence (Figure 3), the economic impact could not be reliably compared between Ireland and France, as several key factors remained uncertain, such as the extent of potential culling of infected dairy cows and the scale of testing (Annex A, module 4. Impact of disease).

Concluding remarks

• –The introduction of the H5N1 B3.13 genotype virus from the US into EU dairy cattle or poultry via trade is highly unlikely. Slightly higher introduction rates into poultry are estimated for Bretagne (France), due to some imports of live poultry and fertilised eggs but still remain in the ‘highly unlikely’ category.
• –The introduction of H5N1 B3.13 genotype virus from the US into EU wild birds is also highly unlikely, except for western Ireland and western France, with an increased risk in autumn period (September–October) linked to the migratory movements of wild birds. However, high uncertainties remain concerning the wild bird populations migrating between the two continents and the prevalence levels in these.
• –The occurrence of H5N1 B3.13 genotype virus outbreaks in EU poultry, dairy cattle and wild birds is highly unlikely for all NUTS2 regions within the EU.
• –If an outbreak was to occur, the predicted number of outbreaks would only be substantial if H5N1 B3.13 genotype virus is introduced by migratory wild birds. In this scenario, the largest epidemic is predicted in wild birds (IQR 250–923 k birds), compared to poultry (IQR 240–713 farms) or dairy farms (IQR 26–81 farms). Predicted numbers of outbreaks are generally larger when the first outbreak occurred in winter (January) compared to summer (July). If an outbreak was to occur and control measures were only applied in poultry (e.g. not in dairy cattle), the predicted numbers of outbreaks in EU cattle would be substantially larger (IQR 161–17K farms), mostly driven by cattle‐to‐cattle transmission.
• –If an outbreak was to occur, the overall impact (economic, societal and environmental) of H5N1 B3.13 genotype virus outbreaks in the EU would be generally high for most countries. The highest scores would be in poultry and wildlife in France and Spain, due to a combination of high costs due to more outbreaks involving more animals, a higher environmental impact due to the presence of vulnerable wildlife species in these regions and for Spain a larger impact of indoor confinement on the welfare of cattle in winter. If control measures were applied in poultry only, the overall impact would be quite similar as the increased impact for some of the categories considered balance out with the decreased impact of others.
• –Between France and Ireland, the only two EU countries with introduction rates higher than in the rest of the EU, societal impacts would be similar.
• –Assuming that the H5N1 B3.13 genotype virus behaves similarly to clade 2.3.4.4b viruses currently circulating in Europe, the number of outbreaks in cattle predicted by the L'ORA model appears unexpectedly high. This implies that the model outputs should be interpreted with caution and not regarded as exact predictions, but rather as comparative indicators to evaluate different scenarios (e.g. with and without control measures applied to dairy cattle) within the model's context.

3.2. Measures that could prevent the introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 into dairy cows and poultry in the EU (ToR 2b.1)

3.2.1. Scenario 1a: Measures to prevent the introduction into EU dairy farms via wild birds

The ranking of the assessed measures regarding their effectiveness to prevent the introduction of the H5N1 genotype B3.13 virus into a dairy farm in the EU via wild birds and the results of the assessment of their feasibility are shown in Table 3.

TABLE 3.

Scenario 1a: Assessment of the effectiveness and feasibility of measures to prevent the introduction into EU dairy farms via wild birds.

Measure assessed	Effectiveness rank	Feasibility category
Cleaning and disinfection of milking equipment to remove or inactivate any virus contamination before use	Rank 1	High
Prevent access of wild birds/mammals to the milking parlour and equipment	Rank 2	Moderate
Vaccination of cattle against HPAI	Rank 3	Low
Management of feed and water for cattle within the premises to avoid contamination from wild birds	Rank 4	Moderate
Make the surroundings of the farm unattractive for wild birds (e.g. no pools of water, no trees, concrete surfaces on farm)	Rank 5	Extremely low
Remove, report and test dead wild birds in the vicinity of cattle farms		High
Keep dairy cattle indoors in high‐risk areas and during high‐risk periods		Moderate

Effectiveness

Rank 1

All currently available evidence indicates that the intramammary route is essential both for the successful infection of dairy cattle with H5N1 genotype B3.13 virus and its sustained transmission. Thorough cleaning and disinfection of milking equipment that removes or inactivates any virus contamination before use is considered the most effective measure since it should prevent infections through the intramammary route.

Rank 2

Preventing access of wild birds/mammals to the milking parlour and its equipment is considered effective to prevent its contamination, reducing the probability of infections through the intramammary route.

Rank 3

Though vaccination of cattle against HPAI was considered in principle an effective measure to reduce the susceptibility of dairy cattle to infection, there is uncertainty about whether a vaccine could result in full protection of all vaccinated animals against infection.

Rank 4

Management of feed and water for cattle within the premises to reduce contamination from wild birds is considered less effective, as the oral route of infection seems to play a limited role in the introduction of the virus, based on currently available information.

Rank 5

Making the surroundings of the farm unattractive for wild birds (e.g. no pools of water, no trees, concrete surfaces on farm), reporting dead wild birds in the vicinity of dairy farms for removal and testing by the authorities and keeping dairy cattle indoors in high‐risk areas and during high‐risk periods might limit the wild bird density around the farm and reduce the exposure of dairy cattle to infected wild birds, but not to the degree that introduction could effectively be avoided. Therefore, the effectiveness of these measures is ranked lower.

Feasibility

High

Cleaning and disinfection of milking equipment to remove or inactivate any virus contamination before use is considered highly feasible, as such practices are already in place on most farms (e.g. as part of mastitis prevention) and might only require a change of the disinfectant used. Reporting dead wild birds in the vicinity of dairy farms for removal and testing by the authorities is also considered highly feasible, provided that the responsibility for the removal is clarified.

Moderate

Experts consider that good management of feed and water for cattle within the premises to avoid contamination from wild birds as well as keeping dairy cattle indoors in high‐risk areas and during high‐risk periods have a moderate feasibility, as without structural changes of the farm premises these measures would be difficult to implement. In addition, the latter measure may lead to social and economic conflicts, especially for certain cattle breeds that need to be kept outdoors to comply with the production standards (such as organic production, etc.).

Low

Vaccination of cattle against HPAI is considered to have low feasibility, as there are currently no licensed vaccines, and their development and implementation require considerable efforts in terms of funding and logistics.

Extremely low

Making the surroundings of the farm unattractive for wild birds (e.g. no pools of water, changing tree coverage, concrete surfaces on farm) is considered to have an extremely low feasibility due to the considerable investments and time that would be needed to modify the farm's infrastructure and surroundings and the negative impact this measure may have on the surrounding environment and habitats.

3.2.2. Scenario 1b: Measures to prevent the introduction into EU dairy farm via trade

The ranking of the assessed measures regarding their effectiveness in preventing the introduction of the H5N1 genotype B3.13 virus into a dairy farm in the EU via trade and the results of the assessment of their feasibility are shown in Table 4.

TABLE 4.

Scenario 1b: Assessment of the effectiveness and feasibility of measures to prevent the introduction into EU dairy farms via trade.

Measure assessed	Effectiveness rank	Feasibility assessment
Avoid importation of cattle from infected countries (in accordance with EU legislation)	Rank 1	High
Avoid importation of potentially contaminated animal products/germinal products from infected countries (in accordance with EU legislation)	Rank 2	High
Vaccination of cattle against HPAI	Rank 3	Low
Pre‐movement test of outgoing animals from infected countries (in accordance with EU legislation)	Rank 4	High
Quarantine of new animals from infected countries (in accordance with EU legislation)		Moderate

Effectiveness

Rank 1

Avoiding importation of cattle from infected countries (in accordance with EU legislation) is considered the most effective measure to reduce the probability of virus introduction, as, based on the current epidemiological situation, dairy cattle appear to be the main reservoir.

Rank 2

Avoiding importation of potentially contaminated animal products/germinal products from infected countries (in accordance with EU legislation) is ranked second, as most of these commodities are considered to be less likely to come into contact with EU dairy cattle than live animals, with the exception of germinal products, for which experimental evidence shows that in‐utero infection of mice with H5N1 B3.13 genotype virus is possible (Seibert et al., 2025).

Rank 3

Vaccination of cattle against HPAI in the farm receiving the trade products is considered, in principle, an effective measure to reduce the susceptibility; however, there is uncertainty about whether a vaccine could provide full protection to all vaccinated animals.

Rank 4

Pre‐movement testing of outgoing animals from infected countries (in accordance with EU legislation) is considered less effective, as diagnostic tests may not have 100% sensitivity and animals could become infected in the period between testing and movement. Quarantine ¹² of new animals from infected countries (in accordance with EU legislation) is also considered less effective, due to the limited clinical signs observed in cattle infected with H5N1 B3.13 genotype virus in the US. This limitation could be mitigated by testing during the quarantine.

Feasibility

High

The feasibility of avoiding importation of cattle or potentially contaminated animal products/germinal products from infected countries, as well as pre‐movement testing of outgoing animals from infected countries, is considered to be high, as these are common disease control measures that can be readily implemented.

Moderate

The experts judge the feasibility of quarantine of new animals imported from affected areas to be moderate because of the special infrastructure requirements for cattle quarantine facilities.

Low

The feasibility of vaccinating cattle is judged to be low due to the reasons provided in Section 3.2.1.

3.2.3. Scenario 2: Measures to prevent the introduction into EU poultry flocks via trade

The ranking of measures assessed regarding their effectiveness to prevent the introduction of the H5N1 genotype B3.13 virus into a poultry flock in the EU via trade and the results of the assessment of their feasibility are shown in Table 5.

TABLE 5.

Scenario 2: Assessment of the effectiveness and feasibility of measures to prevent the introduction into the EU poultry flocks via trade.

Measure assessed	Effectiveness ranking	Feasibility assessment
Avoid importation of poultry from infected countries (in accordance with EU legislation)	Rank 1	High
Avoid importation of potentially contaminated animal products/germinal products from infected countries (in accordance with EU legislation)	Rank 2	High
Pre‐movement test of outgoing animals from infected countries (in accordance with EU legislation)	Rank 3	High
Quarantine of new animals from infected countries (in accordance with EU legislation)		High

Effectiveness

Rank 1

Avoiding importation of poultry from infected countries (in accordance with EU legislation) is considered the most effective, as infected live poultry poses a high risk for virus introduction.

Rank 2

Avoiding importation of potentially contaminated animal products/germinal products (e.g. turkey semen or embryonated chicken eggs) from infected countries (in accordance with EU legislation) is ranked second as these commodities were considered to play a smaller role in the virus introduction.

Rank 3

Pre‐movement testing of outgoing animals from infected countries (in accordance with EU legislation) is considered less effective for the reasons mentioned in Section 3.2.2. Quarantine of new animals from infected countries (in accordance with EU legislation) is equally considered less effective in preventing introduction, given that poultry usually shows acute clinical signs and therefore would not pass the clinical inspection, except for ducks which show less clinical signs.

Feasibility

High

All assessed measures to prevent the introduction into a poultry flock via trade are considered highly feasible. Avoiding importation of poultry or potentially contaminated poultry products/germinal products from infected countries, as well pre‐movement test of outgoing animals from infected countries, are common disease control measures that can readily be implemented. Quarantining new poultry imported from affected areas is a common measure taken for other diseases that would usually involve the entire flock.

3.3. Measures that could prevent the spread of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 in dairy cows and poultry in the EU (ToR 2b.2)

3.3.1. Scenario 3: Measures to prevent within‐farm spread in infected dairy cattle farms in the EU

The ranking of the measures assessed regarding their effectiveness to prevent within‐farm spread of the H5N1 genotype B3.13 virus in infected dairy cattle farms in the EU and the results of the assessment of their feasibility are shown in Table 6.

TABLE 6.

Scenario 3: Assessment of the effectiveness and feasibility of measures to prevent within‐farm spread in infected dairy cattle farms in the EU.

Measure assessed	Effectiveness ranking	Feasibility assessment
Milking hygiene (biocidal teat treatment, avoid vacuum fluctuations)	Rank 1	Moderate
Vaccination of cattle against HPAI	Rank 2	Low
Feeding only milk replacement to calves	Rank 3	Moderate
Isolation of any sick animal		Moderate
Proper disposal of carcasses and animal products (raw milk) in a dedicated area outside the farm zone		Moderate
Culling of infected animals		Low

Effectiveness

Rank 1

Milking hygiene (application of biocidal teat treatment (such as cleaning of udder and teats, pre‐ and post‐milking dip), avoiding vacuum fluctuations) is considered the most effective measure, as it has also been shown to significantly reduce the transmission of mastitis pathogens.

Rank 2

Vaccination of cattle against HPAI is considered less effective, as it could reduce the shedding of the virus, but not completely prevent it.

Rank 3

Feeding calves only with milk replacers is ranked less effective, as it would only reduce the probability of infection of calves. A lower effectiveness is also assigned to isolation and culling of sick animals, as it largely depends on early detection of the infected ones, which, in the absence of testing, is difficult to achieve due to the presence of sub‐clinically infected animals. The experts consider the effectiveness of proper disposal of carcasses and animal products (raw milk) in a dedicated area outside the farm zone equally low, as direct contact between live animals is considered to be more important for within‐farm spread of the virus than contaminated products.

Feasibility

High

The feasibility of applying milking hygiene (avoiding vacuum fluctuations, application of biocidal teat treatment) is considered to be moderately feasible, as it would have to be applied after milking each cow, which requires a considerable effort in practice, especially in large dairy farms.

Moderate

Feeding calves only with milk replacers is also considered moderately feasible, depending on funding being made available to farmers. Isolation of any sick animals and proper disposal of carcasses and animal products (raw milk) in a dedicated disposal area (usually outside the main animal housing zones, where rendering companies collect such material) are also considered to have only moderate feasibility, as in many farms, such areas would need to be improved or established.

Low

The feasibility of culling infected animals is ranked low due to its demands in terms of logistics and costs. The experts also consider that the acceptability of this measure by society and farmers would be limited. The feasibility of vaccinating cattle is judged to be low (see Section 3.2.1).

3.3.2. Scenario 4a: Measures to prevent spread from infected dairy farms to other dairy farms in the EU

The ranking of the assessed measures regarding their effectiveness to prevent between‐farm spread of the H5N1 genotype B3.13 virus in dairy cattle in the EU and the results of the assessment of their feasibility are shown in Table 7.

TABLE 7.

Scenario 4a: Assessment of the effectiveness and feasibility of measures to prevent between‐farm spread in dairy cattle in the EU due to transmission from infected dairy cattle farms.

Measure assessed	Effectiveness ranking	Feasibility assessment
Movement ban of cattle in infected areas	Rank 1	High
Pre‐movement test of outgoing cattle from infected areas	Rank 2	High
Quarantine of new cattle entering the farm		Low
Early detection and culling of infected cattle farms		Extremely low
Vaccination of cattle against HPAI		Low
Avoid exchange of workers, vehicles and equipment		Moderate
Early detection and isolation of positive farms		Moderate
Indoor confinement/limit of outdoor access in infected areas for cattle	Rank 3	Moderate
Remove, report and test dead wild birds in the vicinity of cattle farms		High
Management of feed and water in cattle farms to avoid contamination from wild birds/mammals		Moderate
Avoid non‐essential visits		High
Biosecurity measures before entering the farm (e.g. cleaning and disinfecting essential vehicles, equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors)		Low
Movement restriction of vehicles from infected areas		Moderate
Proper disposal of carcasses and animal products (waste raw milk)		Moderate
Movement ban of untreated milk (treatment to inactivate the virus not applied) – Treatment of milk prior to movement to inactivate the virus		Moderate
No slurry/manure spreading from infected farms		Moderate

Effectiveness

Rank 1

The highest effectiveness is assigned to the ban of movement of cattle in infected areas, as it would address the most important mode of transmission identified in the US outbreak.

Rank 2

Early detection of infected farms and culling or isolation of infected cattle farms (by restricting cattle movements, avoiding non‐essential visits, correct biosecurity, movement restrictions of vehicles and movement ban of untreated milk, and no slurry/manure spreading) are considered slightly less effective than the rank 1 measure for preventing between‐farm spread due to the delay between sampling and testing. Pre‐movement testing of outgoing cattle from infected areas and quarantine of new cattle entering the farm are also considered less effective for the reasons explained in Section 3.2.2. Avoiding exchange of workers, vehicles and equipment is judged to be less effective, especially for workers involved in the milking process, as there is evidence of possible infection of workers in dairy farms, which they might transmit to cows (Uyeki et al., 2024). Vaccination of cattle is ranked lower than cattle movement bans for the reason explained in Section 3.2.1.

Rank 3

Indoor confinement or limiting outdoor access in infected areas for cattle was considered to be less effective than rank 2 measures, as it would not stop indirect transmission through, e.g. wild birds. Removing, reporting and testing dead wild birds in the vicinity of cattle farms and the management of feed and water in cattle farms to reduce contamination from wild animals, e.g. wild birds, rodents, etc. are considered to have a limited influence on virus transmission between dairy farms. Avoiding non‐essential visits and use of biosecurity measures before entering the farm (e.g. cleaning and disinfecting vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for all workers/visitors to change clothing/boots, and clean and disinfect hands) are considered to be effective mainly if the visitors and equipment had been in contact with contaminated milk. Movement restriction of vehicles from infected areas is also considered to have a limited effectiveness because the role of fomites in the transmission of the virus in the US outbreak is unclear. Proper disposal of carcasses and animal products (waste raw milk) and banning the movement of milk, unless treated to inactivate the virus prior to movement, are considered to be only effective to reduce indirect, but not direct transmission. Prohibiting slurry or manure spreading from infected farms is considered to have a lower effectiveness, as there seems to be limited shedding of the virus in faeces.

Feasibility

High

Movement bans of cattle in infected areas, avoiding non‐essential visits and pre‐movement testing of outgoing cattle from infected areas are considered to have a high feasibility as these measures are already applied in the control of outbreaks of other diseases. Removing, reporting and testing dead wild birds in the vicinity of cattle farms is also considered to be highly feasible, provided that the responsibility for the removal is clarified.

Moderate

Avoiding the exchange of workers, vehicles and equipment is considered to have a moderate feasibility, as in some settings and farm types, such exchanges are important for the farm operations. Early detection and isolation of positive farms is considered to be moderately feasible and would depend on the establishment of a surveillance system, e.g. based on bulk milk tank testing. Also, the management of feed and water in cattle farms to reduce contamination from wild birds/mammals and the proper disposal of carcasses and animal products (waste raw milk) are considered to have moderate feasibility, as these measures would require some structural changes on many dairy farms. The feasibility of not spreading slurry/manure from infected farms depends on the season and is therefore judged to be moderate. Banning the movement of milk, unless treated to inactivate the virus, is considered moderately feasible as it requires suitable equipment. The restriction of movements of vehicles from infected areas is judged to be moderately feasible, as milk collection would have to continue, requiring special logistics.

Low

Quarantine of new cattle entering the farm and vaccination of cattle against HPAI are considered to have a low feasibility due to the reasons explained in Sections 3.2.1 and 3.2.2, respectively. Biosecurity measures before entering the farm are judged to have a low feasibility as considerable infrastructure changes would be needed for most dairy farms for these measures.

Extremely low

Early detection and culling of infected cattle farms is considered to have an extremely low feasibility due to the involved logistics and the limited acceptability of this measure by farmers and society.

3.3.3. Scenario 4b: Measures to prevent spread from infected poultry farms to dairy cattle farms in the EU

The ranking of the assessed measures regarding their effectiveness in preventing spread of the H5N1 B3.13 genotype virus from poultry to dairy cattle in the EU and the results of the assessment of their feasibility are shown in Table 8.

TABLE 8.

Scenario 4b: Assessment of the effectiveness and feasibility of measures to prevent between‐farm spread in dairy cattle in the EU due to transmission from infected poultry flocks.

Measure assessed	Effectiveness ranking	Feasibility assessment
Vaccination of cattle against HPAI	Rank 1	Low
Avoid exchange of workers, vehicles and equipment		Moderate
Biosecurity measures before entering the farm (e.g. cleaning and disinfecting vehicles, equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors)		Low
Vaccination of poultry against HPAI		Moderate
Movement restriction of animals (cattle) from infected areas	Rank 2	High
Indoor confinement of cattle/limit of outdoor access during high‐risk periods/areas		Moderate
Remove, report and test dead wild birds in the vicinity of cattle farms		High
Pre‐movement test of outgoing animals (cattle) from infected areas		High
Quarantine of new animals (cattle)		Low
Management of feed and water to avoid contamination from wild birds/mammals		Moderate
Avoid non‐essential visits		High
Movement restriction of vehicles from infected areas		Moderate
Proper disposal of carcasses and animal products (waste raw milk)		Moderate

Effectiveness

Rank 1

Vaccination of cattle and of poultry against HPAI is judged to be effective, as it would reduce the susceptibility of cattle and reduce the shedding of virus by infected poultry, respectively. Avoiding exchange of workers, vehicles and equipment and implementing biosecurity measures before entering the dairy farm (e.g. cleaning and disinfecting vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors) are also considered effective as they would reduce indirect transmission.

Rank 2

Restricting the movement of cattle from infected areas, indoor confinement of cattle/limit of outdoor access during high‐risk periods/areas, pre‐movement testing of outgoing animals (cattle) from infected areas and quarantine of new animals (cattle) are considered less effective because these measures would not reduce the probability of indirect transmission from poultry to cattle (e.g. through exchange of workers or by wild birds). Removing, reporting and testing dead wild birds in the vicinity of cattle farms is considered to have lower effectiveness as it would only limit indirect transmission. The same applies to avoiding non‐essential visits, and management of feed and water to reduce contamination from wild birds/mammals. Also, movement restriction of vehicles from infected areas and proper disposal of carcasses and animal products (waste raw milk) are considered to be less effective as these measures would have little impact on the transmission risk posed by infected poultry.

High

Movement restrictions of animals (cattle) from infected areas, removing, reporting and testing dead wild birds in the vicinity of cattle farms, and avoidance of non‐essential visits, as well as pre‐movement testing of outgoing animals (cattle) from infected areas are considered to have a high feasibility due to the reasons explained in Sections 3.2.2 and 3.2.3, respectively.

Feasibility

Moderate

Avoiding exchange of workers, vehicles and equipment, indoor confinement of cattle/limit of outdoor access during high‐risk periods/ in high‐risk areas, management of feed and water to reduce contamination from wild birds/mammals, movement restriction of vehicles from infected areas and proper disposal of carcasses and animal products (waste raw milk) are considered to be moderately feasible for the reasons explained in Sections 3.2.1 and 3.3.2. Vaccination of poultry against HPAI is also considered to have a moderate feasibility because of the logistical and financial efforts needed.

Low

Vaccination of cattle against HPAI, implementing biosecurity measures before entering the farm (e.g. cleaning and disinfecting of vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors) and quarantine of new animals (cattle) are judged to have a low feasibility for the reasons explained in Sections 3.2.1, 3.3.2 and 3.2.2.

3.3.4. Scenario 5: Measures to prevent spread from infected dairy farms to poultry flocks in the EU

The ranking of the assessed measures regarding their effectiveness to prevent between‐farm spread of the H5N1 B3.13 genotype virus in poultry flocks in the EU due to transmission from infected dairy farms and the results of the assessment of their feasibility are shown in Table 9.

TABLE 9.

Scenario 5: Assessment of the effectiveness and feasibility of measures to prevent between‐farm spread in poultry flocks in the EU due to transmission from infected dairy farms.

Measure assessed	Effectiveness ranking	Feasibility assessment
Movement restriction of cattle from infected areas	Rank 1	High
Avoid exchange of workers, vehicles and equipment in infected cattle farms		Moderate
Vaccination of poultry against HPAI		Moderate
Early detection and stamping out of infected cattle herds	Rank 2	Extremely low
Vaccination of cattle against HPAI		Low
Movement ban of untreated milk (treatment to inactivate virus not applied) – Treatment of milk prior to movement to inactivate the virus		Moderate
Early detection and isolation of positive farms		Moderate
Indoor confinement/limit of outdoor access during high‐risk periods/areas of poultry	Rank 3	Moderate
Pre‐movement test of outgoing cattle from infected areas		High
Avoid non‐essential visits to infected cattle farms		High
Biosecurity measures before entering cattle farms (e.g. cleaning and disinfecting vehicles, equipment in dedicated area outside the farm zone, use of anteroom for essential workers/visitors)		Low
Movement restriction of vehicles from infected areas		Moderate
Proper disposal of carcasses and animal products (waste raw milk)		Moderate
Remove, report and test dead wild birds in the vicinity of cattle farms	Rank 4	High
No slurry/manure spreading from infected cattle farms		Moderate

Effectiveness

Rank 1

Movement restriction of dairy cattle from infected areas is ranked as most effective because it would limit the number of infected dairy farms and therefore also reduce the probability of spread to poultry flocks. Avoiding exchange of workers, vehicles and equipment between infected dairy farms and poultry is considered equally effective, as this would limit the exposure of poultry to contaminated fomites from dairy farms. Similarly, vaccination of poultry against HPAI is judged to be an effective measure as it would reduce the susceptibility of poultry to infection.

Rank 2

Early detection combined with culling of cattle of infected farms or isolation of positive farms is considered to be less effective in preventing poultry outbreaks due to the delay between sampling and testing. Vaccination of cattle is also ranked less effective for the reasons explained in Section 3.2.1. Banning the movement of milk unless treated to inactivate the virus prior to movement is considered less effective as well, as it would only reduce indirect transmission.

Rank 3

Indoor confinement/limit of outdoor access of poultry during high‐risk periods/areas is considered to have a lower effectiveness, as it would only reduce indirect transmission of the virus in commercial free‐range operations where no direct contact with dairy cows is expected. Similarly, pre‐movement testing of outgoing cattle from infected areas is judged to have a lower effectiveness due to the possibility of missing infected cattle tested during the incubation period. Avoiding non‐essential visits to infected cattle farms, applying biosecurity measures before entering poultry farms (e.g. cleaning and disinfecting vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors), movement restriction of vehicles from infected areas as well as proper disposal of carcasses and animal products (waste raw milk) would only reduce indirect virus exposure and are therefore considered to have a lower effectiveness.

Rank 4

The lowest effectiveness is assigned to the removal, reporting and testing of dead wild birds in the vicinity of cattle farms and to avoiding the spreading of slurry/manure from infected cattle farms. This is due to a limited effect of the first on indirect transmission from dairy to poultry farms and the limited amount of virus that is shed by infected cattle in their faeces.

Feasibility

High

Restricting the movement of cattle from infected areas, pre‐movement testing of outgoing cattle from infected areas, avoiding non‐essential visits to infected cattle farms and removing, reporting and testing dead wild birds in the vicinity of cattle farms are considered to be highly feasible measures for the reasons given in Sections 3.2.3 and 3.3.2.

Moderate

Avoiding exchange of workers, vehicles and equipment in infected cattle farms, vaccination of poultry against HPAI, banning the movement of milk unless treated prior to movement to inactivate the virus, early detection and isolation of positive farms, restriction of movement of vehicles from infected areas, proper disposal of carcasses and animal products (waste raw milk) and avoiding the spreading of slurry/manure from infected cattle farms are considered moderately feasible measures, for the reasons shown in Sections 3.3.1, 3.3.2 and 3.3.3. Indoor confinement/limit of outdoor access during high‐risk periods/areas of poultry is also judged moderately feasible due to not being acceptable by farmers and society at large and also conflicting with existing outdoor requirements of specific production systems.

Low

Vaccination of cattle against HPAI and implementing biosecurity measures before entering cattle farms (e.g. cleaning and disinfecting of essential vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors) are considered to have low feasibility due to the reasons explained in Sections 3.2.1 and 3.3.2.

Extremely low

Lastly, early detection and culling of infected cattle farms is judged to have an extremely low feasibility as explained in Section 3.3.2, unless applied to a very limited number of farms (Table 10).

TABLE 10.

Overview of the assessment of the effectiveness and feasibility of measures to prevent introduction and spread of H5N1 B3.13 genotype virus in EU dairy cattle and poultry (1–5 = effectiveness ranking, 1 = highest rank, 5 = lowest rank; green = high feasibility, yellow = moderate feasibility, orange = low feasibility, red = extremely low feasibility).

Measures	Feasibility	Introduction into EU dairy herds via wild birds	Introduction into EU dairy herds via trade	Introduction into the EU poultry flocks via trade	Within‐herd spread in infected dairy cattle herds in the EU	Between‐herd spread in dairy cattle, dairy‐dairy	Between‐herd spread in dairy cattle, dairy‐poultry	Between‐herd spread in poultry flocks
Cleaning and disinfection of milking equipment to remove or inactivate any virus contamination before use	High	1
Prevent access of wild birds/mammals to the milking parlour and equipment	Moderate	2
Vaccination of cattle against HPAI	Low	3	3		2	2	1	2
Management of feed and water for cattle within the premises to reduce contamination from wild birds	Moderate	4				3	2
Make the surroundings of the farm unattractive for wild birds (e.g. no pools of water, no trees, concrete surfaces on farm)	Extremely low	5
Report dead wild birds in the vicinity of dairy farms for removal and testing by the authorities	High	5
Keep dairy cattle indoors in high‐risk areas and during high‐risk periods	Moderate	5
Avoid importation of cattle/poultry from infected countries (in accordance with EU legislation)	High		1	1
Avoid importation of potentially contaminated animal products/germinal products from infected countries (in accordance with EU legislation)	High		2	2
Pre‐movement testing of outgoing animals from infected countries (in accordance with EU legislation)	High		4	3		2	2	3
Quarantine on new animals from infected countries (in accordance with EU legislation)	Moderate		4
Milking hygiene (biocidal teat treatment, avoid vacuum fluctuations)	Moderate				1
Feeding of only milk replacer to calves	Moderate				3
Isolation of any sick animal	Moderate				3
Proper disposal of carcasses and animal products (raw milk) in a dedicated area outside the farm zone	Moderate				3	3	2	3
Culling of infected animals	Low				3
Movement ban of cattle in infected areas	High					1	2	1
Quarantine of new cattle entering the farm	Low					2	2
Early detection and culling of infected cattle herds	Extremely low					2		2
Avoid exchange of workers, vehicles and equipment	Moderate					2	1	1
Early detection and isolation of positive farms	Moderate					2		2
Indoor confinement/limit of outdoor access in infected areas for cattle	Moderate					3	2	3
Remove, report and test dead wild birds in the vicinity of cattle farms	High					3	2	4
Avoid non‐essential visits	High					3	2	3
Biosecurity measures before entering the farm (e.g. cleaning and disinfecting of essential vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors)	Low					3	1	3
Movement restriction of vehicles from infected areas	Moderate					3	2	3
Banning the movement of milk, unless treated to inactivate the virus prior to movement	Moderate					3		2
No slurry/manure spreading from infected farms	Moderate					3		4
Vaccination of poultry against HPAI	Moderate						1	1

3.4. Adaptations of the current EU surveillance for HPAI needed to allow detection of an introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 into the EU dairy cow populations (ToR 2c.)

Effective surveillance is essential for the early detection and control of HPAI infections in dairy cattle, particularly in light of the recent H5N1 outbreaks in the US. EFSA has proposed criteria for monitoring AI in dairy cattle, emphasising the importance of initiating surveillance in dead or clinically affected animals, especially on farms with epidemiological links to infected poultry, wild birds or mammals. EFSA further recommended to extend AI surveillance to dairy cattle showing (i) unexplained clinical signs (i.e. without common and clear aetiology) like mastitis, decreased lactation, reduced feed intake and respiratory issues and/or (ii) unusual or mass mortality events, particularly in AI high‐risk areas or during periods of active virus circulation. In addition, apparently healthy cattle should be closely monitored or selectively tested when exposed to infected poultry, for example in mixed‐species farms, as subclinical infections may occur in cattle and contribute to onward spread (EFSA AHAW Panel, 2025).

Complementary guidance from FAO recommends passive surveillance for H5N1 alongside routine and opportunistic sampling to assess cattle health status (El Masry et al., 2024). In addition, FAO advises targeted or risk‐based surveillance for cattle exposed to poultry or wild birds in at‐risk regions, to ensure timely detections and investigation of potential outbreaks.

This section outlines potential surveillance approaches for HPAI in dairy cattle, available diagnostic options and essential requirements for the detection of the H5N1 B3.13 genotype virus in the EU. It is important to note that HPAI surveillance in dairy cattle does not differentiate between H5N1 B3.13 genotype virus and other HPAI viruses.

3.4.1. Passive surveillance

Passive surveillance refers to the detection of the disease through the observation of clinical signs or unexpected events that prompt further investigation. This approach relies on farmers and veterinarians to recognise such abnormalities and initiate diagnostic testing. In dairy cattle, the most typical manifestation of infections with H5N1 B3.13 genotype virus is a change in milk appearance, which appears like colostrum. In the EU, these signs would most likely lead farmers/veterinarians to suspect mastitis, the most common production‐related disease in dairy farms. H5N1 B3.13 genotype virus‐infected cows often test positive in the California Mastitis Test (CMT) because of the increased number of leukocytes in the milk (Halwe et al., 2025). This would likely prompt antimicrobial treatment targeting common mastitis pathogens (e.g. Staphylococcus aureus or Streptococcus spp.). However, these treatments are ineffective against viral infections, and depending on milking hygiene practices, the virus may continue to spread within the infected dairy farm. When the initial treatment fails, farmers/veterinarians typically submit milk samples for microbiological culture which usually return negative results in case of viral infections. Such a negative result, combined with continued spread of the disease, may then prompt consideration of less common causes of mastitis, including HPAI virus. As a consequence of the above, the time from the introduction of a H5N1 B3.13 genotype virus into the first farm to the identification of the H5N1 B3.13 genotype virus infection as the cause may span several weeks. This delay could even be longer, given that a substantial proportion of infected dairy cows in USA did not show overt clinical signs (Peña‐Mosca et al., 2025).

In the US, neurological signs have been observed in cats that were fed milk from infected cows. Consequently, observations of neurological signs or unexplained deaths in farm cats could also be considered part of passive surveillance and may warrant testing of milk for the presence of HPAIv. It should be noted, however, that this observation is based on US experience, and the Panel does not recommend the deliberate use of cats as sentinels for HPAI virus detection in dairy cattle. Similarly, the presence of dead wild birds or detection of HPAI virus in poultry, wild birds or mammals on dairy farm premises should raise suspicion of HPAI virus infection in cattle and prompt earlier diagnostic investigation (EFSA AHAW Panel, 2025). Testing dairy cattle in such situations is considered warranted given the accumulating evidence of the susceptibility of ruminants to HPAI exposure. In addition to the outbreaks in dairy cattle reported in US dairy cattle, new‐born goats in the US have been found dead and positive for HPAI after being raised in close proximity to infected poultry on a mixed‐species farm. In the UK, milk of a sheep kept together with infected poultry tested positive for HPAI H5N1 virus of the 2.3.4.4b clade (DI genotype) in early 2025 (Mahase, 2025). Similarly, antibodies against H5 were recently detected in sheep in Norway, 11 months after an outbreak of HPAI H5N1 virus in nearby seabirds, indicating prior exposure of these ruminants to the virus (Hol Fosse et al., 2025).

In unvaccinated poultry, the effectiveness of the current passive surveillance strategies for detecting outbreaks caused by HPAI would also detect infection with the H5N1 B3.13 genotype virus, so no adaptations would be required for these surveillance activities. However, it is worth noting that virus sequencing, required to identify the H5N1 B3.13 genotype specifically, is not routinely performed for all detected viruses in the EU or in real time. Consequently, the fact that the outbreaks are caused by H5N1 B3.13 genotype virus in poultry or wild birds may go undetected or be identified with a delay, and the associated transmission risk to cattle may be underestimated.

Raising awareness among farmers and veterinarians about the possibility of HPAI as a cause of altered milk quality and decreased milk production, in particular in high‐risk areas and periods (e.g. after detection of H5N1 B3.13 genotype virus or other HPAI virus strains in wild birds, poultry or mammals in Greenland or in Europe), may help shorten the time between virus introduction and detection by passive surveillance. More generally, awareness that the virus can spread to cattle and that any known or suspected exposure to infected poultry, wild birds or mammals in or around the farm should prompt increased vigilance and early investigation for potential H5N1 B3.13 genotype virus infections is essential. Other issues, such as the farmers' consent to participate in the surveillance, may also need to be considered. In addition, people with professional exposure to potentially infected animals need to be informed about their infection risk and how to reduce exposure to infected animals and their environment (ECDC, EFSA, 2025).

3.4.2. Syndromic surveillance

Syndromic surveillance refers to the monitoring of routine animal health or production data, such as mortality or milk parameters such as somatic cell counts (SCC) to detect unexpected changes that may indicate emerging health issues, including infection with HPAI viruses. Syndromic surveillance is designed to pick up deviations in routine data early. Infection with H5N1 B3.13 genotype virus can lead to reduced milk production and an increase in SCC. The latter is a key milk quality parameter, and within the EU, SCC in bulk milk must not exceed 400,000 cells/mL. Therefore, an unexplained drop in milk production or a rise in SCC could serve as a trigger to investigate a possible HPAI infection. In Denmark, when SCC values exceed the threshold value, the data are reported to the Danish Veterinary and Food Administration (DVFA), which monitors farm‐level prevalence of high SCCs. It is expected that H5N1 B3.13 genotype virus spreading in a dairy farm may result in epidemic mastitis with markedly increased SCC levels. The Netherlands also has syndromic surveillance in place for udder health, focussing on SCC, antimicrobial use and milk yield drop (Santman‐Berends, van den Heuvel, et al., 2021). However, as in passive surveillance (Section 3.4.1), it is likely that initial investigations would still first focus on common mastitis pathogens. Moreover, the proportion of cows affected by HPAI‐related production loss in the US is typically only 10%–20% of a farm (Peña‐Mosca et al., 2025), and scientific literature does not yet provide clarity on SCC levels in cows with subclinical HPAI virus infection. As a result, it may take several weeks after introduction of H5N1 B3.13 genotype virus before changes in milk production and SCC level become noticeable at the farm level.

Another option could be to monitor mortality. However, since mortality due to H5N1 B3.13 genotype virus in cattle is generally low (< 2%; USDA APHIS, 2024), syndromic surveillance based on increased mortality is not considered a sensitive indicator for H5N1 B3.13 genotype virus introduction. Nevertheless, a rise in the number of cows culled due to severe mastitis could serve as a trigger for further investigations.

As with passive surveillance, raising awareness among farmers and veterinarians of HPAI viruses as a potential underlying cause of changes in milk production and SCC, especially in high‐risk areas and during high‐risk periods (e.g. after detection of H5N1 B3.13 genotype virus in wild birds, poultry or mammals nearby), could help support early detection by syndromic surveillance.

3.4.3. Active surveillance

Active surveillance refers to the systematic collection of samples from animals regardless of clinical signs to detect the presence of infection or exposure to pathogens. Currently, there is no structured active surveillance for H5N1 B3.13 virus in dairy cattle in the EU. However, several MS have active surveillance programmes in place for other infectious diseases in cattle, which could be adapted or extended to include H5N1 B3.13 virus detection if needed. These programmes typically utilise blood, serum or milk samples, with testing primarily aimed at detecting antibodies rather than a pathogen.

In the US, bulk milk testing is widely used to detect HPAI‐infected farms by RT‐qPCR and plays an important role in disease control (Stachler et al., 2025). In the EU, bulk milk testing is a well‐established tool for the control or elimination of various infectious diseases in cattle, such as Brucellosis, bovine viral diarrhoea (BVD), infectious bovine rhinotracheitis (IBR), Leptospirosis, Neospora caninum infection, Bluetongue, Salmonella Dublin infection, Liver fluke infection and Q‐fever (see e.g. Santman‐Berends, Mars, et al., 2021). Nevertheless, surveillance strategies and programme participation vary between MS, depending on the national priorities and disease classifications under the Animal Health Law. In addition, many of these programmes rely on ELISA‐based serological assays, which require basic laboratory equipment and low biosecurity environments, while relatively few laboratories are equipped to perform molecular testing for the detection of HPAI viruses.

As a result, there is currently no established EU‐wide system for channelling bulk milk samples into laboratories capable of testing for HPAI viruses. Nonetheless, existing infrastructures could be leveraged for HPAI virus surveillance. For efficiency, implementation could follow a risk‐based approach, for example initiating targeted testing of existing sample streams in high‐risk areas, following a H5N1 B3.13 genotype virus detection in wild birds, poultry or mammals within the EU or in migratory source regions or after having detected an HPAI virus in a dairy farm. It should be noted that certain farm types, such as smallholders without bulk milk tanks or farms processing milk on‐site, would fall outside such a system. Still, given that the milk from bulk milk tanks is routinely collected from most dairy farms and often sampled for milk quality monitoring, it provides a practical and accessible matrix for early detection of H5N1 B3.13 genotype virus. The frequency of testing required for effective early detection will depend on the virus transmission dynamics and determining the optimal frequency would require dedicated modelling approaches to evaluate and compare different surveillance time intervals. DeWitt et al. (2024) simulated that for a 500‐head dairy farm in the US, daily testing would be effective for early detection, whereas weekly testing would be insufficient. Modelling based on molecular testing of bulk milk from HPAI outbreaks in California dairy farms estimated the probability of detecting HPAI infection at different times relative to the onset of clinical signs. The model suggested average detection probabilities of 0.16 when testing 14–7 days before onset, 0.69 when testing 7–0 days before onset and 0.97 in the week following onset. These estimates are derived from the US production context, which is characterised by very large dairy farms (i.e. often many thousands of heads), and may not be directly transferable to the European production systems and epidemiological scenario.

Wastewater surveillance is also used in the US as a form of active surveillance for HPAI virus, with weekly samples collected from over 400 sites nationwide (CDC, Wastewater Data for Avian Influenza A (H5), online) and it allows identifying the virus subtypes. ¹³ While this method is efficient for monitoring regional infection dynamics in an already affected country, its effectiveness for early detection in dairy cattle within a previously disease‐free region may be limited, particularly because many dairy farms may not be connected to the municipal sewage systems. Nevertheless, in the European context, wastewater surveillance could still serve as a valuable and efficient tool to detect HPAI viruses if targeted at dairy processing plants, where milk from multiple farms is collected and processed.

Upon EFSA's request, several members of EFSA's Animal Health Network provided information on current or planned active surveillance efforts. Several European countries have already undertaken active surveillance to detect HPAI viruses in their cattle populations, using either archived samples or newly collected samples. In Germany, 1500 bulk milk samples were tested in 2024 using RT‐qPCR. Additionally, serological surveys were conducted in 2024 and 2025, each involving 1000 cattle in Mecklenburg‐Western Pomerania. Serology involved initial screening for NP antibodies. No evidence of H5N1 B3.13 genotype virus or other HPAI viruses was found in German cattle. In the Netherlands, retrospective analysis of 2190 archived serum samples from 367 farms was conducted using an influenza A blocking ELISA. ELISA‐positive samples were subjected to confirmatory testing and subtyped using a Luminex H5/H7 assay. These tests also showed no indications of HPAI virus presence (Fabri et al., 2025). Norway tested approximately 2000 serum samples, primarily from beef cattle and a smaller number from dairy cows. Half of the samples dated from 2019 (prior to HPAI detection in Norway), and the other half from risk areas within 25 km of confirmed outbreaks in poultry or wild birds. Samples were screened using an ELISA for influenza A virus antibodies (with 1%–2% positive samples), followed by H5‐specific ELISA and HI tests. All follow‐up tests were negative. Ireland is currently planning serological surveillance in dairy cattle, using blood samples collected as part of existing programmes for diseases such as brucellosis. Switzerland is expected to implement active surveillance starting in 2026, particularly if H5N1 is detected in cattle elsewhere in Europe. The proposed strategy includes biannual bulk milk sampling from all dairy farms, with serological testing and regional pooling (e.g. 42 farms per pool). The approach may be adjusted based on the evolving epidemiological situation, focussing on high‐risk farms where relevant.

In conclusion, surveillance for H5N1 B3.13 genotype virus in dairy cattle requires a multifaceted approach adapted to the epidemiological, clinical and logistical challenges involved. Passive surveillance remains an essential first line of detection but is constrained by the common occurrence of mastitis on dairy farms, likely leading to delayed recognition. Syndromic surveillance can provide early warning through existing monitoring systems, though its sensitivity may be limited when only a small proportion of infected animals are clinically affected. Active surveillance, especially through bulk milk sampling, holds the greatest potential for timely detection, but will require substantial investment in infrastructure and testing capacity. A more feasible approach may be to integrate HPAI virus testing into existing bulk milk testing programmes, particularly in high‐risk areas and during high‐risk periods. Rapid sequencing of viruses detected in poultry and wild birds in such risk areas would also help timely detection of a H5N1 B3.13 genotype virus introduction. Across all surveillance strategies, raising awareness among farmers, veterinarians and laboratories is crucial to enhance early detection. Ultimately, a coordinated, risk‐based approach at both national and EU levels, using existing infrastructure, holds the greatest promise to improve preparedness.

3.4.4. Diagnostic testing methods for detection of HPAI in bovine samples

The atypical pathogenesis of H5N1 B3.13 genotype virus in dairy cattle necessitates re‐evaluation of sampling strategies and methods applied to ensure high sensitivity and reliable results. Unlike poultry, where cloacal/tracheal/oropharyngeal swabs are considered the standard for diagnosis, dairy cattle require matrices aligned with viral shedding patterns. Milk is the preferential matrix for detection of H5N1 B3.13 genotype virus due to the high viral loads and non‐invasive sampling, but consideration should be given to sample collection, transportation methods and RNA extraction methods to ensure reliability of tests. Serum is secondary for active outbreaks but critical for retrospective studies into transmission history and epidemiological investigations; however, at the time being, there are only limited validated options for serological screening in dairy cattle. Environmental samples also represent an opportunity, as they could be valuable for tracing indirect transmission in milking systems.

A summary of advantages and disadvantages for the different sample matrices is given in Table 11

TABLE 11.

Sample Matrices for HPAI Genotype B3.13 Detection in Dairy Cattle.

Sample matrix	Description and use	Advantages	Limitations	References
Milk (Bulk Tank or Silos)	Pooled milk from farm‐level collection; reflects aggregate viral shedding.	–Non‐invasive –Cost‐effective	–RNA degradation without stabilisation –PCR inhibitors (fats, proteins) –Dilution effect	APHIS (2024), APHA (2024), NMPF (2024), Halwe et al. (2025), Snoeck et al. (2024)
Milk (Individual Cow)	Milk collected directly from lactating cows; can be used for confirmatory testing during outbreaks.	–High viral load (10⁶–10⁸ RNA copies/mL) –Early detection (pre‐seroconversion)	–Labour‐intensive for large farms –Intermittent shedding patterns	APHIS (2024), Shittu et al. (2025), Caserta et al. (2024), Burrough et al. (2024), Giménez‐Lirola et al. (2024)
Nasopharyngeal Swabs	Swabs collected from the upper respiratory tract; can be used to confirm infection in non‐lactating animals to detect active infection.	–Feasible to collect on farm –Highest viral load after milk in infected cows	–Lower sensitivity compared to milk (40%–60% concordance) –Variable viral load	APHIS (2024), Baker et al. (2024), Halwe et al. (2025), Oguzie et al. (2024), Shittu et al. (2025)
Blood	Limited value for detection of HPAIv; blood‐derived serum can be used for serological assays (e.g. ELISA, HI) to detect previous exposure.	–Identifies past infections –Allows establishing seroprevalence (cumulative infections)	–Limited utility for active infections –Cross‐reactivity with other type A influenza viruses depending on the tests used –Labour‐intensive for large farms	FLI (2024), Giménez‐Lirola et al. (2024), Abousenna et al. (2025), Lang et al. (2025)
Post‐Mortem Tissues	Mammary tissue, lung or lymph nodes collected during necropsy for confirmation and to evaluate tissue distribution	–Confirmatory diagnosis –Can be collected at slaughterhouse in case of suspect based on lesions	–Limited use to investigate farm prevalence	Baker et al. (2024), Halwe et al. (2025), Facciuolo et al. (2025), Shi et al. (2025)
Environmental Samples	Manure, milking equipment or water troughs, wastewater, air; assesses viral persistence in the environment.	–Indirect surveillance –Identifies fomite transmission risks –Regional surveillance	–Low viral load –Risk of cross‐contamination	Singh et al. (2024), Falender et al. (2025), Campbell et al. (2025), Sutton et al. (2025)

3.4.4.1. Viral detection

Molecular methods

Detection of viral RNA by RT‐qPCR offers a rapid and sensitive approach for testing both individual milk samples and bulk or silos milk. The choice of diagnostic method depends on the sample matrix and the specific testing objective – whether detecting viral presence or antibodies indicating prior exposure. Real‐time reverse transcription PCR (RT‐qPCR) remains the gold standard for identifying viral RNA. In the US, validated protocols are available for application to milk, nasal swabs and tissue samples (APHIS, 2024). However, the detection of the virus in complex biological fluids like milk poses greater challenges than in simpler matrices such as swab supernatants, requiring methodological optimisation. Raw milk is a biologically complex and enzymatically active matrix. Among the primary concerns is the rapid degradation of RNA following sample collection. Raw milk naturally contains a variety of enzymes, including lipases and proteases, which actively break down biological molecules. Lipases contribute to this degradation by disrupting the lipid membranes of enveloped viruses, thereby destabilising the viral structure and exposing RNA to further attack. Proteases compound the problem by digesting viral capsid proteins, accelerating RNA release and making it highly vulnerable to ribonucleases (RNases) present in the milk (Koczera et al., 2016).

In addition to enzymatic degradation, bacterial contamination in raw milk introduces further RNase activity during storage (refrigeration might not be particularly useful in case of contamination by psychrotrophic species like Pseudomonas spp.). The combined effects of enzymatic and microbial degradation can significantly impair RNA integrity (Apura et al., 2021).

To mitigate these effects, prompt and appropriate sample handling is essential. The USDA recommends that milk samples be refrigerated to 4°C within 30 min of collection to inhibit enzymatic activity. In field and laboratory settings alike, the addition of RNA stabilisation buffers can help protect RNA. Beyond degradation, the intrinsic chemical composition of milk presents additional barriers to effective molecular diagnostics. Milk is rich in fats, proteins and minerals, all of which can inhibit reverse transcription and polymerase chain reaction (RT‐qPCR) assays. Milk fats, for instance, tend to form micelles that can entrap viral particles, preventing efficient lysis and RNA extraction. Calcium ions, abundant in milk, have been shown to bind DNA polymerases and impair their enzymatic function during amplification. Similarly, casein proteins tend to precipitate during nucleic acid extraction, often dragging RNA with them and reducing the overall yield.

These inhibitory effects manifest as decreased RT‐qPCR efficiency and increased risk of false negatives. Nonetheless, practical experience, particularly from the United States, indicates that viral RNA can be readily detected in retail pasteurised milk samples, suggesting that, for qualitative (positive/negative) outcomes, the impact of inhibitors is limited; however, careful sampling remains essential to maximise the likelihood of successful virus isolation and subsequent characterisation.

To overcome these challenges, a number of optimised extraction strategies have been developed and have been recently compared by Snoeck et al. (2024). In brief, pre‐dilution of milk samples before RNA extraction has been suggested in some studies (Blais‐Savoie et al., 2025) and official guidelines (FDA, 2024; IZSVe, 2024) to reduce the impact of inhibitory components and, if dilution media is also a stabilising agent, can help to preserve RNA integrity.

Considering the genetic characteristics of H5N1 genotype B3.13, the recommended methods by the European Reference Laboratory for avian influenza and Newcastle disease (EURL AI/ND) remain valid for the detection of genome targets for identification of avian influenza, subtyping and molecular pathotyping (M, H5‐HA and N1‐NA genes).

Virus Isolation

Viral isolation by inoculation in specific pathogen free (SPF) embryonated chicken eggs or onto suitable cell lines is the only method that can confirm virus infectivity. However, it is time‐consuming, requires BSL‐3 facilities and must be performed by trained staff. Its use should be reserved for confirmatory testing, viral phenotype characterisation or to demonstrate the efficacy of inactivation methods.

Antigenic detection

For the detection of the antigen, lateral flow devices have been used in the US during field investigations for the presumptive diagnosis of H5N1 B3.13 genotype virus directly on milk samples. Although their sensitivity is lower than PCR and they do not provide additional information on the viral subtype, these devices may be valuable for initial screening or in situations where there is no access to fully equipped laboratories.

3.4.4.2. Serological assays

Serological assays are essential for identifying previous exposure to H5N1 B3.13 genotype virus in dairy cattle, supporting both surveillance and epidemiological investigations. Unlike direct viral detection techniques, these assays identify host antibodies formed in response to infection, complementing molecular methods and providing data on the spread and prevalence of infection in farms. When integrated into broader surveillance frameworks, they can enhance our ability to map the extent of viral circulation and support timely control measures. Notably, serological assays may offer increased sensitivity compared to molecular diagnostics in certain contexts, as they capture the cumulative exposure history of the farm rather than a snapshot of active infection, thus improving the likelihood of detecting past or transient infections in dairy cattle farms.

Enzyme‐Linked Immunosorbent Assay (ELISA)

Among serological techniques, ELISAs are particularly well‐suited for large‐scale testing due to their high throughput capacity, relatively low cost and ease of automation. This makes them ideal as a first‐line screening method, especially when large numbers of samples need to be processed efficiently in the context of outbreak investigations or structured surveillance programmes.

Commercially available ELISA kits target either conserved viral proteins, such as nucleoprotein (NP) or subtype‐specific antigens like hemagglutinin (HA), including H5. NP‐based ELISAs are valuable for detecting antibodies against a broad range of influenza A viruses, while H5‐specific ELISAs offer the specificity needed to confirm exposure to H5 subtypes, such as H5N1 B3.13 genotype virus. One significant advantage of ELISA in this context is the availability of competitive or blocking formats, which do not require species‐specific secondary antibodies. This feature is particularly important for non‐avian species such as cattle (or ruminants in general), where traditional indirect assays may be limited by the availability or performance of anti‐bovine conjugates. Competitive ELISAs allow for accurate antibody detection regardless of species, making them especially suitable for multi‐species surveillance settings.

A recent study from the Netherlands (Fabri et al., 2025) retrospectively screened 2190 archived bovine serum samples (collected from 367 farms between 2022 and 2024) using an influenza A blocking NP‐ELISA, followed by confirmation with a Luminex assay. According to the manufacturer's instruction, this ELISA has been validated for use in cattle. ¹⁴ The ELISA initially flagged four positive samples (0.2%), which were all ultimately negative on the subtype‐specific confirmatory test, suggesting a specificity of 99.8% in cattle.

Recent experience in the United States further underscores the potential of serological surveillance in dairy cattle, particularly through the use of milk as a diagnostic matrix. During the 2024 H5N1 B3.13 genotype virus outbreaks in U.S. dairy farms, serological testing played a crucial role in understanding farm‐level exposure and infection dynamics (Peña‐Mosca et al., 2025). Preliminary evaluations demonstrated that milk could be used to detect influenza A antibodies using ELISA platforms, though the assay's performance in this matrix requires careful optimisation. As highlighted by Giménez‐Lirola et al. (2024), treatments might be required (e.g. rennet‐coagulation to separate milk serum) prior to testing, to mitigate matrix effects that can interfere with assay performance. This highlights the importance of adapting protocols specifically for milk and underlines the need for validation in this matrix before widespread implementation.

Although real‐life evidence for the use of milk for HPAI serology is limited (Abousenna et al., 2025), the use of milk as a serological diagnostic matrix is particularly attractive in dairy farms, as it allows for non‐invasive sampling and repeated monitoring over time. Bulk milk testing, in particular, can serve as a cost‐effective tool for farm‐level surveillance, enabling early detection of exposure and guiding follow‐up investigations where needed.

Hemagglutination Inhibition (HI) Test

The Hemagglutination Inhibition (HI) test remains a reference standard serological assay for the detection of subtype‐specific antibodies against influenza A viruses, including H5N1 B3.13 genotype virus. Unlike ELISA, which can detect antibodies to conserved or broad‐spectrum viral proteins, HI assays specifically measure antibodies that inhibit the binding of hemagglutinin (HA) to red blood cells, thereby providing functional evidence of immune response to a given subtype.

HI assays have traditionally been used in avian species, but recent work has expanded their application to non‐avian hosts, including dairy cattle. HI tests require species‐specific protocol optimisation, particularly for sample pre‐treatment and standardisation of red blood cells when expanding their application beyond avian species. The HI assay's strength lies in its antigenic specificity, making it well‐suited to differentiate between influenza subtypes and even clades within the same subtype. However, this specificity can also pose a limitation, as antigenic drift may reduce cross‐reactivity and compromise sensitivity if test antigens are not closely matched to circulating strains. It is worth noting that HI assays have also been adapted for Influenza D virus (IDV) serology in cattle (Gaudino et al., 2021), although the biology of IDV is distinct from avian influenza viruses. This example illustrates both the feasibility of adapting HI protocols to ruminant hosts and the need to address important limitations, including non‐specific serum inhibitors, virus‐specific differences in RBC agglutination and the lack of established validation cut‐offs for HPAI in cattle when considering HI assays for serological surveys.

Application of HI to bovine milk remains largely unexplored, though theoretically feasible. Given the assay's sensitivity to inhibitors commonly present in biological fluids, direct use of whole milk without treatment is unlikely to yield reliable results. Any future adaptation for milk testing would require rigorous validation and potentially complex sample preparation, such as the removal of fat, casein and other matrix components that may interfere with hemagglutination (Superti et al., 2019). As such, HI testing currently remains confined to serum in the context of bovine surveillance, serving as a valuable confirmatory assay where subtype‐specific resolution is needed.

While HI is more labour‐intensive and technically demanding than ELISA, it can be performed under low biosecurity measures as it can employ inactivated antigens in contrast with virus neutralisation, which requires live viruses. Therefore, HI represents a valuable confirmatory assay proven thorough standardisation and accurate selection of antigens.

Virus neutralisation test

The virus neutralisation (VN) test is considered the most specific serological assay for detecting antibodies against HPAI H5N1, as it directly measures the ability of host antibodies to prevent viral infection in cell culture systems. Unlike ELISA and HI, which rely on antigen–antibody binding or inhibition of hemagglutination, VN assesses whether antibodies present in the serum can neutralise live virus, making it the reference method for evaluating protective immunity.

VN assays have been successfully applied in various mammalian species, including experimental and field studies in cattle. In recent investigations, neutralising antibodies were detected in dairy cattle naturally and experimentally infected with H5N1 B3.13 genotype virus. VN sensitivity in other mammalian species has been shown to be superior to HI, making VN a useful tool to confirm results obtained by ELISA, particularly in the case of weakly positive ELISA results. Moreover, neutralisation assays, together with HI testing, have proven instrumental in characterising the immune response to different H5N1 clades and tracking antigenic drift, particularly in studies involving emerging 2.3.4.4b viruses.

Despite its advantages, VN testing requires high biosafety levels (typically BSL‐3) due to the use of infectious HPAI viruses. This limits its application to specialised laboratories and make routine use impractical in large‐scale surveillance programmes. To overcome this, pseudotyped virus neutralisation assays (PVNAs) have been developed, which can be handled under BSL‐2 conditions using viral vectors bearing the HA protein of interest (Temperton et al., 2007; Tsai et al., 2009). These assays use replication‐defective viral vectors (often lentiviral or vesicular stomatitis virus‐based backbones) that express the hemagglutinin (HA) protein of interest. They measure actual neutralising capacity of serum samples and their application for cattle sera has been evaluated (Miyakawa et al., 2024). However, these assays are not widely available, and differences with traditional VN assays have been highlighted. As for HI tests, the importance of challenge virus (or HA origin for PVNAs) for testing is critical to obtain reliable results.

At present, VN is mainly used as a confirmatory tool or for research purposes, particularly in vaccine or immunogenicity evaluations and detailed serological profiling. Its application to milk samples remains unvalidated and given the complex matrix and potential for cytotoxic effects in cell culture, serum remains the preferential type of sample for this assay.

Concluding remarks

Given the current evidence, the probability of introduction of H5N1 B3.13 genotype virus is very low, with wild bird migration representing the most likely pathway of introduction. Moreover, experimental evidence indicates that H5N1 B3.13 genotype virus is unlikely to behave differently from H5N1 clade 2.3.4.4b strains already circulating in Europe, and the absence of evidence of cattle infections despite widespread and repeated exposure to HPAI in wild birds and poultry in Europe indicates a very narrow bottleneck for spillover to dairy cattle under European conditions. Consequently, the initiation of a broad, EU‐wide active surveillance for H5N1 B3.13 genotype virus in cattle is neither efficient nor proportionate. At the same time, the intensity and extent of the surveillance of wild birds varies across countries, and whole‐genome sequencing of HPAI viruses is not done for all viruses detected and not done in real‐time, which means introductions of H5N1 B3.13 genotype virus by wild birds may be missed or detected only after some time.

Against this background it is recommended to focus preparedness for HPAI viruses in dairy cattle on a graduated, risk‐based approach that combines awareness, outbreak investigation and targeted testing:

1. Awareness raising to improve passive/syndromic surveillance

Farmers, veterinarians and laboratories must be aware that HPAI should be considered in the differential diagnosis of atypical mastitis (colostrum‐like milk, negative bacterial cultures, non‐response to antimicrobial treatment), unexplained milk yield drops, elevated somatic cell counts and neurological signs or unexplained death in cats fed raw milk. Syndromic surveillance systems tracking milk quality indicators (SCC, yield, antimicrobial use) should also take HPAI into account. Awareness raising is particularly relevant in areas where HPAI is circulating or where the risk of H5N1 B3.13 genotype virus introduction is highest.

• 2 Targeted outbreak investigation after exposure to HPAI

When dairy cattle are exposed to HPAI (through infected poultry, wild birds or contaminated environments), a structured outbreak investigation should be conducted. The sampling design involves testing at least 10–15 animals per epidemiological unit, prioritising lactating cows (EFSA AHAW Panel, 2025; Table 5).

• ○Individual cow milk samples: from clinically affected cows, preferably, completed with asymptomatic ones. Primary screening should rely on highly sensitive methods, such asRT‐qPCR for influenza A M gene and RT‐qPCR targeting H5‐HA and N1‐NA genes. Given the high viral load in milk in clinically ill cows, the use of on‐site testing could also be considered as a preliminary triage tool.
• ○Milk from bulk milk tanks (BTM): tested by RT‐qPCR for influenza A M gene and the H5‐HA and N1‐NA genes.
• ○Nasopharyngeal swabs: Recommended for non‐lactating animals or when milk sampling is unavailable.
• ○Serum samples: Useful for retrospective investigations and to assess farm‐level exposure, using influenza A blocking ELISA followed by H5‐specific ELISA or HI test for confirmation of positive samples.
• ○Environmental samples (in order of priority: milking equipment, air samples, wastewater): Optional, supportive evidence of viral contamination.
• ○Post‐mortem tissue samples (mammary gland, lung, lymph nodes): Recommended for confirmatory diagnosis in dead animals.

All PCR‐positive samples suitable for sequencing should be submitted for whole‐genome sequencing to confirm genotype.

• 3 Regional active surveillance after first detection

Once an infection with HPAI viruses has been confirmed in cattle, active surveillance is recommended in the region surrounding the affected farm to establish the extend of spread. The most efficient approach is PCR testing of bulk milk across all dairy farms in an area defined by the veterinary authority. Where feasible, testing of milk at dairy processing plants could complement on‐farm sampling. If H5N1 B3.13 genotype virus is detected in wild birds or poultry in a region where dairy farms are present, similar regional bulk milk testing should be considered. Although the likelihood of outbreaks of H5N1 B3.13 genotype virus in dairy farms originating from poultry or wild birds under European circumstances is considered very low, a detection of this specific genotype would likely raise concern given its widespread circulation among US dairy farms. Another argument for initiating surveillance following such findings is that uncertainty remains regarding the specific transmission dynamics of the H5N1 B3.13 genotype virus in field populations comprised of various susceptible species, even though experimental studies indicate behaviour similar to that of other viruses of that clade. Field observations have shown that different HPAI virus genotypes may behave differently under natural conditions, as illustrated by the observed differences between the BB and DI genotypes in poultry and wild birds, despite their comparable behaviour in experimental infections in chickens.

• 4 Follow‐up and containment

After infected farms are identified in the region and control measures put in place, the surveillance zone can be defined and continued bulk milk testing can be used as a cost‐effective monitoring tool. Testing frequency should initially be high (e.g. bi‐weekly) and preferably tailored using modelling of expected spread within and between farms. Serology (ELISA, HI) can be deployed later (even on BTM, to reduce the impact on Veterinary Services and widen the assessment) to assess the extent of exposure once virus circulation is under control.

Overall, the most effective and proportionate strategy for the EU to detect H5N1 B3.13 genotype virus and other HPAI variants is to strengthen vigilance through passive and syndromic surveillance, to conduct structured outbreak investigations with the most sensitive diagnostic tools (RT‐qPCR) and to escalate to targeted regional active surveillance when epidemiological signals warrant it.

3.5. Likelihood of bulk milk to be contaminated if EU lactating dairy cows are infected with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13 (ToR 2d.)

A combined evidence assessment and uncertainty analysis using a structured expert opinion process was used to estimate the likelihood of bulk milk being contaminated, if lactating EU dairy cows are infected with the H5N1 B3.13 genotype virus.

3.5.1. Evidence considered

Experimental infections using the intramammary or oronasal routes suggest a short incubation period (< 5 days): Following intramammary inoculation of H5N1 B3.13 genotype virus in lactating cows, signs of mastitis (positive California Mastitis Test and changes in milk colour and consistency) were reported 2 days after inoculation (i.e. 1 day after the virus was first detected in milk) and until 14 days after inoculation (Baker et al., 2024). Another experimental study reported that the first clinical signs (impaired general condition, abnormal posture, lethargy) were observed as early as 1 day after intramammary inoculation of lactating cattle, while the earliest clinical signs (nasal secretion and coughing) were observed 2 days after oronasal inoculation of Holstein calves (Halwe et al., 2025)

Based on observational farm‐based studies from the US, clinical signs have been first observed 10–20 days after entry of the virus: According to the USDA, the incubation period of the H5N1 B3.13 genotype virus in dairy cattle is 12–21 days (USDA‐APHIS, 2024). A recent study describing an outbreak in one dairy farm reported that the first clinical signs appeared 13 days after introducing seemingly healthy lactating cattle from an affected farm (Peña‐Mosca et al., 2025).

According to USDA's national epidemiologic brief, the morbidity in the affected dairy cattle farms was less than 20% on average and the mortality and culling was up to 2% on average (USDA‐APHIS, 2024). However, industry reported up to 15% mortality in infected cattle (Los Angeles Times, online).

According to the USDA, ‘> 85% of farms reported abnormal lactation and decreased feed consumption, > 80% of farms reported thickened or clotted milk’, so up to 15% of infected farms may not observe clinical signs.

Shedding of H5N1 B3.13 genotype virus can occur in clinically and non‐clinically affected dairy cattle, as well as during the incubation period (Nguyen et al., 2024). Spackman, Jones, et al. (2024) detected a mean virus titre of 3.5 log₁₀ 50% egg infectious doses per mL, or about 3000 virus particles per mL in bulk tank milk from naturally infected farms.

Regarding experimental evidence, two studies assessed shedding in cattle challenged with HPAIv strains, with similar results in terms of time from inoculation to shedding. The first study by Baker and collaborators reported the detection of H5N1 B3.13 genotype virus in milk samples from the inoculated quarters already 1 day post‐inoculation of two dairy cows via the intramammary route, while only low Ct values were observed at certain days in non‐inoculated quarters (these positive results were attributed to cross‐contamination) (Baker et al., 2024). In the second study by Halwe and collaborators, the onset of shedding in six cattle experimentally inoculated by intramammary route also peaked 3 days after infection. Infectious virus was isolated from milk samples 1–8 days post‐experimental infection (Halwe et al., 2025).

Although the prevalence of infected animals in affected farms can be high, only a small proportion of animals may show clinical signs. In a study by Peña‐Mosca et al. (2025), farm staff identified clinical influenza in 777 out of 3876 cows (20%) based on reduced milk production and clinical signs such as inappetence, apathy and decreased rumination time recorded using the AfiCollar® system. However, serum testing ¹⁵ 105 days after the first reported case revealed a seroprevalence of 89% among 637 cows (including both lactating and dry cows that were at the farm during the outbreak). Notably, 76% of the seropositive cows had not shown clinical signs.

Further, the study by Spackman, Jones, et al. (2024) reported detection of H5N1 B3.13 genotype virus by PCR in up to 57.5% milk samples (24.8% positive for virus isolation) from affected and non‐affected dairy farms. However, samples were blinded, and thus it is not possible to know the proportion of positive samples from farms not known to be affected. Some occasional reports of farms detected by passive sampling (e.g. a case in Nevada) were mentioned, and it is thus plausible that infected farms in the US can remain unnoticed long enough to have contaminated bulk milk.

3.5.2. Individual judgement

Following the first round generating individual expert judgements (Figure 4) and the related reasoning, these were discussed in a group meeting. Reasons provided for the judgements include the consideration that a large proportion of lactating cows would shed virus with their milk without showing clinical signs. Further, due to the fact that the cows are milked two to three times per day and that diagnostic test results might only be available 1–2 days after sample collection, experts reasoned that bulk milk will get contaminated before H5N1 B3.13 genotype virus is detected in the farm.

FIGURE 4.

Individual judgements of the probability of bulk milk to be contaminated if EU lactating dairy cows are infected with the H5N1, B3.13 genotype virus.

3.5.3. Outcome of the group judgement

In their group judgement, the experts agreed on several assumptions. First, the scenario that experts consider in their judgement should focus on the very first detections in the EU, i.e. there would not be a high awareness, and thus several days would pass before farmers or veterinarians consider HPAI virus among the differential diagnosis for any clinical signs observed in affected farms.

Second, a potential lack of sensitivity of tests applied to detect bulk milk contamination would not be considered in the assessment (i.e. the parameter of interest is whether bulk milk is contaminated, regardless of whether contamination load is sufficiently high to be detected, not taking into consideration a possible dilution effect).

Further, it was agreed that, as long as bulk milk is contaminated at some point after the infection is present in the farm, the result should be considered positive (e.g. regardless of whether infection may affect one single cow and then shortly fade out, if the cow is milked and the milk reaches the bulk milk tank, it would be considered contaminated, irrespective of the detection threshold).

Lastly, the only way bulk milk would not be contaminated if a dairy farm is infected with the H5N1 B3.13 genotype virus would be if the infection occurred in a very limited number of animals, all of which would show overt clinical signs, and the farmer or veterinarian would stop milking these animals to the bulk milk tank and swiftly submit samples for testing. Experts should consider this scenario unlikely.

The consensus judgement for the proportion of farms out of 100 that become infected with H5N1 B3.13 genotype virus in the EU in the next year, in which bulk milk would become contaminated before the infection is detected at the farm, given the current situation, is 90% (95% certainty interval 80%–100%). Therefore, in the current situation, it is considered very likely (80%–100%) that bulk milk from infected dairy farms will become contaminated before infection with H5N1 B3.13 genotype virus is detected at the farm level.

3.6. Probability of infection/illness for a consumer in the EU population due to the exposure to viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products through the food‐borne route and measures to mitigate the risk (ToR 2e.)

3.6.1. Levels of viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products

Dairy products and colostrum‐based products consumed in the EU with their processing steps and processing conditions

• Description of dairy products produced and consumed in the EU

Industrial and artisanal dairy products in the EU are very diverse. Figure 5 presents a schematic overview, for the aim of this risk assessment, of the diverse range of products that can be derived from milk through various processing pathways. The diagram shows the technological processes, starting with clarification and, optionally, separation, which divides milk into two primary streams: skim milk and cream. This separation creates distinct processing branches that lead to different product categories based on their technological treatments.

FIGURE 5.

Schematic overview of the dairy production chain from raw milk to final products. The flowchart illustrates the main processing pathways and highlights products selected for risk assessment (ovals with red outline). Rectangles: Processing steps in the dairy production chain with dashed rectangles representing optional processing steps; ovals: The products.

The processing diagram reveals several technological approaches distinguished by their heat treatment intensity: products processed from raw milk or raw cream (no heat treatment), thermised milk products (mild heat treatment) and pasteurised products (see Section 1.2 for further details on these treatments). Additionally, products requiring more intensive thermal processing, which include UHT treatments and sterilisation processes, are depicted. Lastly, specific products such as extended shelf‐life (ESL) milk can also be produced through non‐thermal processes such as microfiltration, usually combined with a final thermal pasteurisation treatment to meet regulatory requirements (EFSA, 2021).

Dairy processing encompasses various technological operations that influence the properties and safety of the resulting products. Acidification and fermentation play crucial roles in product development and microbial stability. Each cheese‐making technology involves a specific succession of unit operations with distinct characteristics, leading to the large diversity of cheeses with different characteristics. However, cheeses share several key stages in their production process. In the early stages of production, milk coagulation and curd syneresis by the action of proteolytic enzymes (i.e. chymosin) and lactic acid bacteria (i.e. starter cultures) result in whey loss and a drop in pH. The salting stage creates unfavourable growth conditions for bacteria by lowering the water activity (a_w). During ripening, biochemical processes contribute in some cheese categories to a pH increase, and moisture loss of the cheese is observed. Based on the characteristics of these major production steps and on the final moisture content, a cheese classification can be proposed ¹⁶ (Almena‐Aliste & Mietton, 2014; Papademas & Bintsis, 2017): extra‐hard cheeses (very low moisture 30%–32%), with aging for up to 30 months (e.g. Sbrinz, Parmigiano Reggiano, Grana Padano); hard cheeses (low moisture 33%–43%), with aging of usually 6–18 months (e.g. Gruyère, Pecorino Romano, Cheddar); semi‐hard cheeses (moderate moisture 44%–55%), with aging of 4–8 months (i.e. Appenzeller, Comte, Castelmagno); semi‐soft cheeses (high moisture > 55%), with aging of usually 1–3 months (e.g. Reblochon, Taleggio, Saint‐Nectaire); soft cheeses (high moisture > 55%), with aging of usually 2–6 weeks (e.g. Brie, Camembert); fresh cheeses (very high moisture > 60%), usually with no aging (e.g. Fromage blanc, fresh goat cheese (Chevre), Cottage cheese, Petit Swiss). Similar categorisations based on pH and a_w have been also proposed (Trmčić et al., 2017).

• Description of colostrum‐based products produced and consumed in the EU

An overview of the range of products that can be derived from raw colostrum is presented in Figure 6. The processing steps include pasteurisation, microfiltration, freeze‐drying or low‐temperature spray drying. At the farm, colostrum may be mechanically collected in tanks of variable sizes and, to preserve quality, it is usually directly frozen after collection. The quantities not consumed raw at farm level are then delivered frozen to the processing unit where, following defrosting under controlled conditions and eventually pooling, it might be directly pasteurised or first microfiltrated to separate fat (bacterial load is reduced as well), protein (casein) and then the serum part of the colostrum may be heat processed (LTLT: 63°C for 30 min for batch process, and HTST: 72°C for 15 s for continuous process). Colostrum will then undergo a drying process, usually freeze‐drying, although low‐temperature spray drying is also used. Drum drying is also reported (Adams et al., 2025) as a method for producing milk powders, but is considered the least favourable option especially for colostrum drying, as it may cause excessive caramelisation. Powdered colostrum or capsules are used as a dietary supplement or as ingredients in other products (Kaplan et al., 2021).

FIGURE 6.

Schematic overview of the processing of different colostrum‐based products. The flowchart illustrates the main processing pathways and highlights the product selected for risk assessment (ovals with red outline). Rectangles: Processing steps in the dairy production chain; ovals: The products.

Process pathways omitting any heat treatment or microfiltration of colostrum before further processing (i.e. freeze‐drying) have been reported mainly on commercial websites of colostrum processors, while an EU patent ¹⁷ has been granted in which colostrum is microfiltrated and then freeze‐dried, reaching a maximum temperature of 35°C–40°C during processing.

• Selection of milk and colostrum products for risk assessment

For the assessment, selection was based on identifying those products presenting the highest potential risk, characterised by the absence or minimal application of processing steps intended to inactivate viruses. The selection criteria prioritised products without or with mild thermal processing, as heat treatment represents an effective intervention for viral inactivation in dairy processing. Products undergoing UHT sterilisation were not specifically included in the assessment, as these intense thermal treatments (135–150°C, 1–10 s) would effectively eliminate viral contamination. The selected products can be found in Table 12.

TABLE 12.

Key processing parameters across different dairy product categories selected for risk assessment.

Categories of foodstuff	Heat treatment	Acidification/coagulation	Draining	Yield from milk to product a	Ripening
Raw (drinking) milk	No	No	No	100 L	No
Thermised milk	57°C – 30 min 65°C – 15 s	No	No	100 L	No
Pasteurised (drinking) milk	63°C – 30 min (LTLT) 72°C – 15 s (HTST)	No	No	100 L	No
Raw fermented milk	No	pH 4.6 1 day	No	100 L	No
Raw milk fresh cheese	No	pH 4.6 1 day	Yes	10–15 kg	No
Raw milk soft cheese	No	pH 4.6 5 days b	Yes	13–15 kg	4 weeks (2–6 weeks)
Raw milk semi‐soft cheese	No	pH 5.2 14 days b	Yes	15 kg (11–20 kg)	2 months (1–3 months)
Raw milk cream	No	No	No	3.5–4 L	No
Raw fermented cream	No	pH 5.0 3 days	No	3.5–4 L	No
Pasteurised cream	82°C – 15 s (HTST)	No	No	3.5–4 L	No
Raw colostrum	No	No	No	100 L	No
Pasteurised colostrum	63°C – 30 min (LTLT) 72°C – 15 s (HTST)	No	No	100 L	No
Freeze‐dried colostrum powder	No	No	No	11.5 kg (8–15 kg)	No

Abbreviations: HTST, high temperature for a short time; LTLT, low temperature for a long time.

^a Expressed per 100 L of starting milk or colostrum.

^b The pH drops after coagulation for the reported duration and then increases towards pH 7 during the ripening period.

From the raw milk pathway, several examples of products were selected due to the absence of heat treatment: raw (drinking) milk for direct consumption, raw fermented milk, raw milk fresh cheese, raw milk soft cheese and raw milk semi‐soft cheese produced from raw milk through coagulation, draining and ripening processes. Hard, semi‐hard and extra‐hard cheeses made from raw milk typically undergo heating of the curd and extended ripening periods (minimum 4 months), providing additional hurdles for viral survival and thus representing lower risk compared to soft and semi‐soft raw milk cheese types.

Thermised milk is not consumed as such, and thermisation is usually performed to extend the shelf‐life of milk prior to further processing or in the manufacture of some cheeses (IDF bulletin, 2022). The thermised milk pathway was hence included to evaluate the impact of mild heat treatment on viral inactivation. Products from this category, including thermised milk cheeses, represent an intermediate risk level between raw and fully pasteurised products.

From the pasteurised milk branch, pasteurised drinking milk was selected as the representative product for this processing category. Products derived from pasteurised milk (such as pasteurised milk cheeses, yoghurt, milk powder and dairy products with additives) undergo additional processing steps beyond the initial pasteurisation treatment, which are expected to further add to the viral reduction achieved through pasteurisation.

From the cream processing branch, besides raw cream for direct consumption, raw fermented cream was selected as it undergoes only fermentation without thermal treatment. From the pasteurised cream branch, pasteurised cream was selected as the representative product, as other cream‐based products such as pasteurised butter undergo additional processing steps (churning) beyond the initial pasteurisation treatment, which would further reduce viral survival and thus present lower risk levels.

Three colostrum products were selected to represent different processing intensities: raw colostrum without any treatment, pasteurised colostrum undergoing standard heat treatment and freeze‐dried colostrum powder processed through freezing and dehydration to achieve a reduced a_w and extended shelf‐life.

• Description of the main processing steps for the selected products

Table 12 presents the key processing parameters that define the main technological characteristics of each selected product category. While the most critical steps affecting product safety and stability are listed, additional processing operations may be applied to the products, including clarification, usually, separation at the initial stages and homogenisation for certain products to achieve desired textural properties. Heating of curd during cheesemaking was not considered a thermal treatment. Furthermore, cold storage of milk is a critical stage between milk collection and processing, during which raw milk is typically kept at refrigeration temperatures (4°C or below) for varying time periods.

Processing parameters are described for the distinct technological profiles across the selected products. Heat treatment conditions vary significantly, from no thermal processing in raw products to standard pasteurisation regimes for milk (63°C for 30 min or 72°C for 15 s). ¹⁸ Pasteurised cream usually receives the most intensive treatment (82°C for 15 s). ¹⁹ Thermisation conditions applied to extend the shelf‐life of milk prior to further processing are in the range of 57–68°C for 20–5 s and usually 63–65°C for 15–20 s, while thermisation used for cheesemaking in some Alpine countries varies in the range between 57°C for 30 min and 65°C for 15 s. ²⁰

Acidification processes create different pH environments across fermented products. It should be noted that pH conditions are inherently dynamic throughout the manufacturing process, with continuous changes occurring during fermentation, draining and ripening stages. However, for the purpose of this risk assessment, the process has been simplified by presenting characteristic pH values at key stages. Furthermore, given the high variability of local and regional artisanal products, for the purpose of this assessment, some typical conditions were chosen (Almena‐Aliste & Mietton, 2014; Boutrou et al., 2006; Gobbetti et al., 1997) and it is acknowledged that some products may display different processing parameters. Raw fermented milk achieves the lowest pH (4.6) in 1 day, while cheeses display progressive acidification over extended timeframes. Soft cheeses usually reach pH 4.6 over 5 days, while semi‐soft cheeses achieve a higher pH of 5.2 over 14 days. Raw cream products show intermediate pH values (5.0–5.5) depending on the fermentation process applied.

The draining step, exclusive to cheese manufacturing, involves physical separation of whey from the curd matrix, concentrating solids and reducing moisture content. Ripening periods provide additional processing time under controlled environmental conditions (the pH during ripening to be higher than 5.5), ranging from 2 to 6 weeks for soft cheeses to 1–3 months for semi‐soft varieties (Almena‐Aliste & Mietton, 2014).

Standard pasteurisation temperatures (63°C for 30 min or 72°C for 15 s) were considered for colostrum (Costa et al., 2023; Kaplan et al., 2021).

Levels of viable H5N1 B3.13 genotype virus in bulk milk and bulk colostrum (SQ 1.3)

Infectious virus of H5N1 B3.13 genotype virus has been detected in raw milk samples from bulk tanks (capacity to hold milk from 600 to 700 cows) and in a single retail sample of raw milk. ²¹ In their sampling of raw milk from bulk tanks in four US states, US FDA found 57.5% of 275 samples positive for viral RNA, with 24.8% of those containing infectious virus. Infectious virus titres ranged from 1.3 to 6.3 log₁₀ EID₅₀ per mL with a median titre of 3.5 log₁₀ EID₅₀/mL (Spackman, Anderson, et al., 2024). Metadata on the 39 bulk milk samples in which infectious virus was found are not available as the samples were double‐blinded (personal info from Erica Spackman, 23 July 2025).

Data on titres of infectious virus HPAI H5N1 clade 2.3.4.4b in milk from individual cows (Caserta et al., 2024) and bulk tank samples (Spackman, Anderson, et al., 2024), together with conditions of sampling and testing methods, are summarised in Table C.1 (Appendix C). As herd sizes and bulk tank volumes in the US are, in general, larger than those in the EU, and this may impact the viral concentrations (Eales et al., 2025), a modelling approach was used to estimate the viral concentrations in bulk milk in the EU (see Appendix D). The expected concentrations were simulated as a mixing process (Nauta, 2005), where the variability of viral concentrations in the bulk milk is a function of the number of animals providing milk to the bulk tank, the proportion of animals contributing contaminated milk to the bulk tank and the viral concentrations in the milk of these animals. The number of contributing animals was assumed to vary following a binomial distribution, characterised by the number of lactating animals in a herd and the proportion of lactating animals in a herd that are shedding contaminated milk. The assumed distribution of virus concentrations in milk at the cow level was a normal distribution with mean 5.71 log₁₀ TCID₅₀/mL and SD 1.47. This was based on the dataset from Caserta et al. (2024) on concentrations in the milk of individual infected cows, which was considered most informative considering the number of available samples and their sampling from different farms affected by the introduction of H5N1 B3.13 genotype virus. The maximum concentration in the milk from individual cows was assumed to be 9 log₁₀ TCID₅₀ per mL considering the highest concentration of 8.8 log₁₀ TCID₅₀ per mL in the milk at cow level found by Caserta et al. (2024) and the viral shedding reported in other studies on naturally (Guan et al., 2024) and experimentally (Halwe et al., 2025) infected cows (see Appendix E). Together with data from Koebel et al. (2024) that detail the milk yield per cow and the reduction therein for infected cows, this provides the variability in concentrations between bulk tanks as a function of the proportion of lactating animals in a herd that are shedding contaminated milk and the herd size. The potential impact of temporal variation in the number of animals shedding contaminated milk was not considered. As, in the model, this proportion of lactating animals that are shedding contaminated milk is interpreted as the probability per cow in an infected herd to shed the virus in the bulk milk, a proportion of bulk tanks would not be contaminated. These were omitted from the results, as according to the ToRs, only contaminated bulk tanks were to be considered. For details on the model, see Appendix D.

To explore the model performance, six scenarios were evaluated considering a low (p_herd = 0.01); median (p_herd = 0.05) and high proportion of shedding cows (p_herd = 0.1), as well as a low (n_low: H_all = 65, the size‐biased EU mean herd size) and high (n_high: H_all = 500) herd size. The maximum concentration in milk of individual cows (and thus in bulk tanks) was considered to reach 10 log₁₀ EID₅₀/mL, equivalent to the maximum concentration of 9 log₁₀ TCID₅₀/mL. As illustrated in Figure 7, a lower proportion of shedding cows implies a lower mean concentration in the bulk tank; and when combined with a low herd size, the variability is considerable. The distributions of the variability in concentrations between bulk tanks obtained by model simulations, expressed into log₁₀ EID₅₀/mL using the regression described in Appendix B, are compared with the experimental data from bulk tanks in the US (Spackman, Anderson, et al., 2024). Both the raw data and a normal distribution fitted through the log₁₀ transformed data are shown in Figure 7. The scenario for the EU was with a proportion of shedding cows of 0.01, and a herd size of 65 (see Appendix D for details). It is acknowledged that the proportion of animals shedding contaminated milk is a parameter affected by significant uncertainty and that the value may be higher, in relation to the strategies to detect infection in the farm (i.e. the interval from infection to detection will be related to the time required until HPAIV tests are applied).

FIGURE 7.

Variability in concentrations of viable H5N1 B3.13 genotype virus in bulk milk. The first six boxplots represent the outcome of model simulations of bulk tanks in the EU, assuming various scenarios of proportion of lactating animals contributing contaminated milk to the bulk tank (p) and various herd sizes (H), while the last two boxplots represent the experimental data from bulk tanks in the US and its fit to a normal distribution (Spackman, Anderson, et al., 2024). The box plots show the 95% percentile, 75%, median (50% percentile), 25% and 5% percentile. The first scenario (p = 0.01, H = 65) is used as the baseline scenario in this risk assessment.

In this scenario (the light blue bar in Figure 7), the mean concentration in contaminated bulk milk is 4.26 log₁₀ EID₅₀/mL, SD = 1.64, 2.5 and 97.5th percentiles are 0.90 and 7.27 log₁₀ EID₅₀/mL. The difference with the US bulk milk data collected by Spackman, Anderson, et al. (2024) is rather small. However, the highest concentration reported in that study is 6.3 log₁₀ EID₅₀/mL of viable virus, whereas a fitted distribution through the data includes much higher values.

As no data were found on the levels of viable H5N1 B3.13 genotype virus in bulk colostrum, the same titres as in bulk milk were considered.

Levels of viable H5N1 B3.13 genotype virus in the selected foodstuffs at the end of processing (SQ 1.4)

Surveys have been conducted by the US FDA on the presence of H5N1 B3.13 genotype virus in milk and dairy products at retail level in the US. Viral RNA was found in 20% of pasteurised retail milk products (N = 297) sampled in April 2024, and virus concentrations, based on quantitative PCR results of 60 samples, were up to 5.4 log₁₀ EID₅₀/mL, with a mean and median of 3.0 and 2.9 log₁₀ EID₅₀/mL, respectively (Spackman, Jones, et al., 2024). Furthermore, viral RNA was detected in 17.4% of the tested retail dairy products, including a variety of cheeses, butter, ice cream and fluid milk (i.e. whole, 2%, 1%, skim milk and heavy cream) (N = 167) sampled in June–July 2024 (Suarez et al., 2024). None of the samples from these two studies showed presence of viable virus in the samples. Evidence of infectious virus being present in commercially sold raw milk was found following an outbreak involving seven household cats that had consumed raw milk from batches that had tested positive for avian influenza and that had been therefore subjected to recall by the California Department of Public Health. Fever and neurologic signs developed, and five cats died; H5N1 presence was confirmed in four animals. ²²

The effect of the different processing steps presented in Table 12 is described below to evaluate the levels of viable H5N1 B3.13 genotype virus in the listed foodstuffs.

• Assessment of the effect of milk storage on virus viability

According to Martin et al. (2024), it is likely that H5N1 would retain infectivity in raw milk until pasteurisation, as in the dairy industry this is usually performed soon after collection to preserve the quality. Multiple studies consistently demonstrate that H5N1 virus remains rather stable in raw milk under refrigeration conditions (4°C) for extended periods.

Bovine and avian H5N1 viruses (i.e. HPAI A/cattle/Texas/063224‐24‐1/2024 (A/ca/Tex/24) and LPAI A/duck/Hokkaido/Vac‐1/2004 (A/du/Hok/04)) viruses showed no significant loss of infectivity when incubated for 4 weeks in heat‐inactivated bovine milk at 4°C (Lenz‐Ajuh et al., 2025). Most influenza viruses remained infectious for over 7 days at 4°C in pasteurised whole milk, and in some cases for considerably longer. For example, the H5N1 strain A/chicken/Scotland/054477/2021 (AIV09/AB genotype) declined about 2 log₁₀ over 12 days, and a reassortant strain with the internal genes of A/Puerto Rico/8/1934 (H1N1) and the HA, NA gene of A/dairy cow/Texas/24‐008749001‐original/2024 (H5N1), named PR8‐Cattle, showed a reduction of approximately 2.5 log₁₀ over 15 days at 4°C (Schafers et al., 2025). Further, a virus half‐life of 2.1 (95% CrI 1.5–3.4) days at 4°C was found in irradiated raw milk spiked with HPAI H5N1 clade 2.3.4.4b. The time needed for a 10 log₁₀ reduction in virus titre was estimated at 69 days (95% CrI 51–112) at 4°C (Kaiser et al., 2024). HPAI A(H5N1) virus‐positive raw cow's milk declined only 2 log₁₀ when stored at 4°C for over 5 weeks (Guan et al., 2024). The stability of influenza virus strains in pasteurised milk, unpasteurised colostrum and unpasteurised mature milk was tested for up to 4 days. At 4°C, approximately a 1 log₁₀ decline was observed after 4 days for the H5N1 strains included in the study (i.e. TX/24, wild‐type strain A/Texas/37/2024 (H5N1); ty/IN/22, wild‐type highly pathogenic strain A/turkey/Indiana/3707–003/2022 (H5N1)) in the different types of samples (Caceres et al., 2024).

As the virus maintains appreciable infectious titres with minimal or no reduction in viability over 1 week at 4°C, for modelling purposes, it was considered appropriate to assign a 0 log₁₀ reduction value for the refrigeration storage step of raw milk, as milk is usually processed within 18 h for the production of raw milk cheese (Arias‐Roth et al., 2022; TetraPack, 2025) and, in raw drinking milk at refrigeration conditions, unacceptable bacterial growth is reached within 3–7 days (Li et al., 2023; Muir et al., 1978).

• Assessment of the effect of thermal treatment on virus viability

Based on the bootstrap procedure (n = 1000 resamples) using the refined dataset from Hessling (2022) (see Figure E.1A), the median estimate of ${\log }_{10}\left(D_{\mathrm{ref}}\right)$ was −1.05, with a 95% CI ranging from −1.29 to −0.80. This corresponds to a D‐value of approximately 0.088 minutes (5.2 s) at the reference temperature of 70°C. The estimated z‐value ($z_T$) had a median value of 9.70°C, with a 95% CI from 8.15°C to 11.92°C. For comparison, Nooruzzaman, Covaleda, et al. (2025) estimated $z_T$ as 9.93°C when carrying out tests using raw milk spiked with HPAI H5N1.

Figure E.1B shows the predicted ${\log }_{10}\left(D\right)$ values as a function of temperature based on the fitted Bigelow model. The confidence band around the ${\log }_{10}\left(D\right)$–temperature curve, as well as the set of bootstrap‐derived $D_{\mathrm{ref}}$ and z‐values, reflect both variability and uncertainty, since the underlying data were compiled from heterogeneous studies. Although it is generally difficult to disentangle the contributions of strain, laboratory or matrix effects from pure uncertainty, several studies report clear strain‐dependent differences under controlled conditions. Based on this evidence, these distributions are interpreted hereafter as representing biological variability. The validation of the developed secondary Bigelow‐type model using studies reporting HPAI H5N1 inactivation in milk products providing either point estimates or right‐censored log₁₀ reductions, without explicit estimation of D‐values, is presented in Figure E.2. Two temperature conditions were considered: 63°C and 72°C. At 63°C, the model predicts a median log₁₀ reduction of approximately 6 in less than 5 min, while at 72°C, it predicts a median log₁₀ reduction of approximately 5 within 15 s of treatment. These predictions are consistent with the available experimental evidence.

The validation of the secondary Bigelow model, comparing model predictions with experimental D‐values measured in milk matrices, is shown in Figure 8. The position of seven D‐values obtained from three independent studies (Kaiser et al., 2024; Lenz‐Ajuh et al., 2025; Nooruzzaman, Covaleda, et al., 2025) is presented in relation to the model's median prediction line, CI and prediction bands. Six of the seven observed values fall very close to the median prediction line, indicating a good agreement between model and empirical data. The remaining point, measured at 60°C, shows slightly higher sensitivity (i.e. a lower D‐value) but still falls within the 95% prediction band.

FIGURE 8.

Experimental D‐values of H5N1 in milk and predicted thermal inactivation based on the Bigelow model developed for various influenza viruses and media.

Taken together, the consistency between model predictions and the censored inactivation data (Figure E.2), as well as with the directly measured D‐values in milk (Figure 8), supports the validity of the fitted Bigelow model. This model can thus be considered a suitable representation of H5N1 thermal inactivation dynamics in dairy matrices under the tested conditions.

Table 13 provides the predicted log₁₀ reductions achieved for each selected food category according to the ‘notional’ thermal treatment (without considering the effect of heat‐up and cool‐down times on virus inactivation) applied. These reductions represent the minimum expected inactivation for the process; more realistic reduction values using simulated profiles based on an available industrial time/temperature profile are also displayed for illustrative purposes.

TABLE 13.

Predictions of log₁₀ reductions of various thermal treatment processes ($R_{\text{thermal}}$) of milk or colostrum to produce the selected foodstuff.

Categories of foodstuff	Temperature treatment	Log10 reduction (quantile 50% [2.5%–97.5%])
		Using the ‘notional’ heat treatments	Using a simulated industrial profile a
Thermised milk	57°C – 30 min	NT57: 15.5 [12.2–19.1]	– b
	65°C – 15 s	NT65: 0.9 [0.5–1.9]	IP65: 1.3 [0.8–2.5]
Pasteurised (drinking) milk	63°C – 30 min (LTLT)	NT63: 66.2 [43.8–119.4]	– b
	72°C – 15 s (HTST)	NT72: 4.9 [1.7–20.2]	IP72: 7.0 [2.8–26.5]
Pasteurised cream	82°C – 15 s (HTST)	NT82: 55.2 [9.8–641.8]	– b
Pasteurised colostrum	63°C – 30 min (LTLT)	NT63: 66.2 [43.8–119.4]	– b
	72°C – 15 s (HTST)	NT72: 4.9 [1.7–20.2]	IP72: 7.0 [2.8–26.5]

Abbreviations: HTST, high temperature for a short time; IP, industrial profile, the adjacent number indicates the temperature; LTLT, low temperature for a long time; NT, notional treatment.

^a The simulated profile is based on an industrial profile (Strahm & Eberhard, 2010); it has the heating and cooling down steps with a slope of 1.3°C/s.

^b Not determined as the predicted log₁₀ reduction using the notional heat treatment is substantial (> 10 log₁₀).

The FDA study on the effectiveness of pasteurisation for the inactivation of A/turkey/Indiana/22‐003707‐003/2022 (clade 2.3.4.4b, designated H5N1 (TK/IN/22)), which used a pilot‐scale continuous flow system to closely simulate commercial pasteurisation systems, found no infectious virus after the ramped heating step in which the milk is heated before entering the holding tube (to receive a treatment at 72°C for 15 s). From this observation yielding a ~6 log₁₀ reduction, it was extrapolated that reductions of > 12 log₁₀ can be expected from industry‐standard pasteurisation practices (Spackman, Anderson, et al., 2024).

• Assessment of the effect of acidification on virus viability

For addressing the impact of acidification, it was assumed that the coagulation effect is linked to the final pH and that duration has no impact. Three pH values were considered: pH 4.6, 5.0 and 5.2. Table 14 provides the estimated log₁₀ reductions achieved for each selected food category according to the acidification process applied.

TABLE 14.

Estimated or considered log₁₀ reductions ($R_{\text{acidification}}$) using various acidification/coagulation conditions of milk to produce the selected foodstuffs.

Categories of foodstuff	Acidification/coagulation	Log10 reduction	Reference
Raw fermented milk	pH 4.6–1 day	3.5	Lenz‐Ajuh et al. (2025), Tamime and Robinson (2007)
Raw milk fresh cheese	pH 4.6–1 day
Raw milk soft cheese	pH 4.6–5 days
Raw milk semi‐soft cheese	pH 5.2–14 days	Pert (0.25,1.9,2.5)	Lenz‐Ajuh et al. (2025), Nooruzzaman, de Oliveira, et al. (2025)
Raw fermented cream	pH 5.0–3 days

The studies described in Lenz‐Ajuh et al. (2025), Nooruzzaman, de Oliveira, et al. (2025); Moreno et al. (2025), Lang et al. (2024), Crossley et al. (2025) and Blondin‐Brosseau et al. (2025) were screened for their evidence to derive the log₁₀ reduction estimates for the effect of acidification on virus viability. Three studies were not withheld; Crossley et al. (2025) used lower pH values (pH of ∼4.0–4.4) for investigating the effectiveness of acidification of nonsaleable raw whole milk to inactivate H5N1 while Moreno et al. (2025) evaluated the survival and inactivation of AIV strains during the production and ripening of Grana‐type hard cheeses made from raw cow's milk. Blondin‐Brosseau et al. (2025) evaluated the survival of H1N1 in some fresh type cheeses made in the laboratory with raw milk, as well as inoculated into commercially produced raw milk cheeses, but figures for the sole effect of acidification cannot be deduced.

Lenz‐Ajuh et al. (2025) incubated strain A/ca/Tex/24 (H5N1) for 30 min at 21°C with buffers adjusted to defined pH values. The infectious titre gradually decreased when the pH was stepwise lowered from pH 6.0 to 4.0, with no infectivity left at pH 4.0. At pH 4.6, a 3.5 log₁₀ reduction was achieved. When the same strain was added to diluted yoghurt (pH drop to 4.2) and incubated for 72 h at 4°C, a 6 log₁₀ reduction was observed. This pH value, however, is below our lowest pH value of 4.6 considered. In the experiment by Lang et al. (2024), homemade yoghurt was prepared by adding commercial yoghurt to AIV spiked milk (using strains A/swan/Germany/R65/2006) (clade 2.2.2, designated H5N1‐06) and A/chicken/Germany/AI04286/022 (clade 2.3.4.4b, designated H5N1‐22), giving a pH drop to 5.8. Following incubation at 42°C for 8 h, a 3–3.5 log₁₀ reduction was found. Unfortunately, the final pH was not reported, but based on literature, it is expected to be in the range pH 4.0–4.6 (Tamime & Robinson, 2007). Based on these data, a 3.5 log₁₀ reduction was deduced for milk acidification to pH 4.6.

Three sources of evidence from the paper by Lenz‐Ajuh et al. (2025) were retained for pH 5.0 and 5.2. In the same experiment using buffers, a 1.9 log₁₀ reduction and 1.6 log₁₀ reduction were achieved when incubating strain A/ca/Tex/24 (H5N1) for 30 min at 21°C in a buffer at pH 5.2 and pH 5.0, respectively. When strain A/bo/Tex/24‐HAmb (H5N1) was added to a buffer at pH 5.0 for 72 h at 4°C, a 2.5 log reduction was found. Incubation of strain A/ca/Tex/24 (H5N1) for 72 h at 4°C with semi‐hard cheese (pH 5.05–5.15) only gave a 0.25 log₁₀ reduction compared to heat‐inactivated milk (pH 6.5). In the study by Nooruzzaman, de Oliveira, et al. (2025), HPAI H5N1 TX2/24 virus (H5N1 B3.13 genotype virus) was spiked in milk followed by direct acidification using 50% lactic acid. After adjustment to pH 5.0, a 2.4 log₁₀ reduction was achieved, while a 2.8 log₁₀ reduction was achieved after incubation at 34°C for 1 h. However, cheese may not contain this amount of lactic acid; therefore, these values were considered an overestimation. Based on these data, a 1.9 log₁₀ reduction, with a min‐max range 0.25–2.5, was estimated for pH 5.0 and pH 5.2 (Pert (0.25, 1.9, 2.5)).

• Assessment of the effect of draining on viable virus occurrence

In the draining steps of cheese production, viral particles occurring in milk can partition variably between the curd and whey due to different interactions with milk components.

To address the impact of draining in different cheese categories, information on the effect of draining on the occurrence of avian influenza and other viruses in curd and whey was screened from Cliver (1973), Blackwell (1976), Lang et al. (2024), Nooruzzaman, de Oliveira, et al. (2025), Moreno et al. (2025), ANSES (2025) and Blondin‐Brosseau et al. (2025). Four studies were not retained: Cliver (1973) and Moreno et al. (2025) evaluated inactivation of AIV during production of two hard/extra‐hard cheeses (cheddar and a Grana‐type cheese) but figures for the partition on virus in curd and whey could not be extracted due to values either below or above the analytical range; in Lang et al. (2024) no viable H5N1 virus was detected in homemade cheese, yoghurt or whey derived from these productions, probably reflecting the intense treatment applied in the study and data obtained by molecular tests (PCR) were not considered exclusively determined by virus partition in the different phases; lastly, the partition factor applied in ANSES (2025) for tick‐borne encephalitis virus (TBEV) (Flaviviridae family) was not considered, as not derived from experimental data.

The study by Blackwell (1976) on cheddar cheese produced using milk from cows infected with foot‐and‐mouth disease virus (Picornaviridae family) showed that the virus titres in the experimental replicates were 3.9–4.6 and 4.3–4.7 log₁₀ PFU/mL in curd and whey respectively at the time of the addition of the coagulant. Subsequently, at the separation post‐heating, titres were 3.2–4.3 log₁₀ PFU/mL in curd and 4.3–4.6 log₁₀ PFU/mL in whey and lowered to 2.5 log₁₀ PFU/mL in curd and 2.0–1.0 log₁₀ PFU/mL in whey at the separation post‐salting.

Nooruzzaman, de Oliveira, et al. (2025), reported the effects on HPAI H5N1 TX2/24 virus (H5N1 B3.13 genotype virus) of cheesemaking at different pH values (5.0, 5.8 and 6.6) and tested viable virus after acidification both in the curd and the end of process, and in the whey collected after coagulation (whey 1), salting (whey 2) and draining (whey 3). While infectious virus was absent in all phases at pH 5.0, at pH 5.8 the titres of infectious virus were, on average, approximately 4.9, 5.0, 4.0 and 3.7 log₁₀ EID₅₀/mL in whey 1, whey 2, whey 3 and curd, respectively, and were 5.9, 5.3, 5.3 and 5.3 log₁₀ EID₅₀/mL at pH 6.6. Therefore, particularly considering data less affected by the combined effect of acidification, minor differences in the concentrations of infectious viruses in the curd and whey phases were detected.

The study by Blondin‐Brosseau et al. (2025) reported the effect of homemade cheesemaking of a Swiss‐style cream cheese and a feta‐style cheese on the viability of H1N1 virus, used as a surrogate for H5N1, in three independent experiments. Infectious H1N1 was detected in 2.5 g of the cream cheese curd in all three experiments and in 1 mL the cream cheese whey from one experiment. On the other hand, infectious H1N1 was detected in 2.5 g of the feta curd from two experiments and no infectious H1N1 was detected in 1 mL of the whey from feta. As the amount of sample tested differed and a significant loss of virus (3–4 log PFU) occurred during sample extraction, it was considered that concentrations of infectious virus in whey and curd did not differ significantly.

Based on the information available, it was concluded that there is not enough evidence to consider a preferential affinity of HPAI to curd or whey. Therefore, no affinity was considered in the assessment ($P_C=1$).

• Assessment of the effect of ripening on virus viability

The impact of ripening on virus stability is considered for raw milk soft cheese for 4 weeks (2–6 weeks) and for raw milk semi‐soft cheese for 2 months (1–3 months).

Nooruzzaman, de Oliveira, et al. (2025) evaluated the stability of HPAIv H5N1 in raw milk cheeses using a mini cheese model prepared with HPAI‐spiked raw milk under varying pH levels (pH 6.6, 5.8 and 5.0). Cheeses were made by direct acidification (lactic acid) of raw milk spiked with HPAI H5N1 TX2/24 virus (H5N1 B3.13 genotype virus). The D‐values of H5N1 virus in the pH 6.6 and 5.8 groups were estimated by the authors to be 29.2 days and 48.3 days, respectively, based on 120 days of ripening at 4°C. In commercial raw milk cheese inadvertently produced with naturally contaminated raw milk (mean viral titres of 4.21 ± 0.48 log₁₀ EID₅₀/g of infectious virus and mean pH of 5.37 ± 0.06 at the beginning of investigation, on day 24 of aging), the virus remained relatively stable until day 120 of aging at 4°C (mean viral titres of 3.6 ± 0.89 log₁₀ EID₅₀/g of HPAI H5N1 and mean pH 5.34 ± 0.2). These data on commercial cheese accidentally produced from naturally contaminated milk confirm sustained stability of HPAIv H5N1 in raw milk cheese, with limited viral inactivation between day 24 and 120 of ripening, possibly displaying higher D‐values under non‐experimental conditions.

Blondin‐Brosseau et al. (2025) examined survival over 8‐week ripening at 4°C of various commercial unpasteurised‐milk cheeses (soft cheese not aged, mixed rind semi‐soft brie‐like cheese, cheddar and a washed rind firm aged cheese) that were inoculated with H1N1 virus. The D‐values for H1N1 on firm cheeses were greater than the soft and semi‐soft cheeses; 20.5 ± 3.4 days and 17.3 ± 3.2 days for cheddar (moisture: 39% ‐ pH 5.4) and washed rind firm cheese (moisture: 35% ‐ pH 5.4), respectively versus 16.1 ± 3.8 days and 1.8 ± 0.1 days for semi‐soft cheese (moisture: 49% ‐ pH 5.5) and soft cheese (no ageing, moisture: 56%, pH 4.9), respectively.

The described ripening experiments were carried out under refrigerated conditions, and higher reductions are to be expected when cheeses are ripened at higher temperatures (typical ripening temperatures of cheeses are 10°C–18°C, Tetra Pak, 2025).

Based on above, the D‐value is expected to range from 16.1 to 48.3 days. The log₁₀ reductions for raw milk soft cheese and raw milk semi‐soft cheese weeks were estimated as 1.0 [0.4–2.3] and 1.9 [0.8–4.4], respectively (Table 15), based on the combination of the D‐values distribution and ripening duration distribution (both uniform distributions).

TABLE 15.

Estimated log₁₀ reductions ($R_{\text{ripening}}$) using various ripening conditions and variability therein.

Categories of foodstuff	Ripening/storage	Log10 reduction (quantile 50% [2.5%–97.5%])	Reference
Raw milk soft cheese	4 weeks (2–6 weeks)	1.0 [0.4–2.3]	Nooruzzaman, de Oliveira, et al. (2025), Blondin‐Brosseau (2025)
Raw milk semi‐soft cheese	2 months (1–3 months)	1.9 [0.8–4.4]

• Assessment of the effect of freeze‐drying on virus viability

No information on the effect of freeze‐drying on HPAIv or any other virus in either colostrum or milk were retrieved in the literature. One paper addressed the effect of spray‐drying on bovine leukaemia virus (Retroviridae family) in colostrum (Lomónaco et al., 2020), however, given the differences of spray‐drying and freeze‐drying in terms of process and applied temperatures, this paper was not considered. Information on the effects of freeze‐drying on virus viability was available for applications other than colostrum processing, such as long‐term storage and preservation of viral stocks and preparation of plasma derivatives. As the first processes are specifically optimised to improve preservation of virus viability, data associated with these applications were not considered representative for commercial freeze‐drying of bovine colostrum. As far as plasma‐derived products, Unger et al. (2009) reported the effects of five different lyophilisation procedures of hemoderivatives (with their respective time–temperature profiles) and their effect on hepatitis A virus (HAV, Picornaviridae family), pseudorabies virus (PRV, Orthoherpesviridae family) and bovine viral diarrhoea virus (BVDV, Flaviviridae family). Results showed that, depending on the applied process, mean reduction factors ranged from 4.0 to 4.8 for HAV, from 1.9 to 3.6 for PRV and from 1.7 to 2.5 for BVDV, and that different reduction factors occurred depending on the specific hemoderivative.

Given the lack of data specific for freeze‐drying of HPAIv in colostrum or milk and the indications on variability of effects depending on virus, matrix and process, it was not possible to conclude on the effects of freeze‐drying on the level of infectious H5N1 B3.13 genotype virus in colostrum.

• Assessment of the overall effect of the different processes on virus viability

The estimated log₁₀ reductions of viable H5N1 B3.13 genotype virus during processing for the selected foodstuffs, and the estimated concentration of viable virus in one unit (volume or weight, according to the product) of the final foodstuff can be found in Table 16. The estimates assumed the distribution of the initial concentration of infectious H5N1 B3.13 genotype virus in bulk milk and bulk colostrum (see Section 3.6.2) and considered the impact of the key processing steps defined in Table 12 on the log₁₀ changes during processing.

TABLE 16.

Estimated log₁₀ reductions of viable H5N1 B3.13 genotype virus in the different processes to produce the selected foodstuffs and the estimated levels/concentrations in the final products. The variability between concentrations in servings of the foodstuff categories is shown, using most likely values for log₁₀ reductions.

Categories of foodstuff	Calculation of log10 reductions during processing (Rtotal)	Median log10 reduction in production process (2.5, 97.5%‐ile)	Median concentration in final product (log EID50/mL or g) a (2.5, 97.5%‐ile)
Raw (drinking) milk	NA	0 (0, 0)	4.4 (1.0, 7.2)/mL
Thermised milk b	= R thermal (Table 13)	NT57: 15.5 (12.2, 19.1) NT65: 0.9 (0.5, 1.9) IP65: 1.3 (0.8, 2.5)	NT57: ‐Inf (‐Inf, ‐Inf)/mL NT65: 3.5 (0.1, 6.3)/mL IP65: 3.1 (−0.3, 6.0)/mL
Pasteurised (drinking) milk	= R thermal (Table 13)	NT63: 66.2 (43.8, 119.4) NT72: 4.9 (1.7, 20.2) IP72: 7.0 (2.8, 26.5)	NT63: ‐Inf (‐Inf, ‐Inf)/mL NT72: −0.5 (‐Inf, 2.3)/mL IP72: ‐Inf (‐Inf, 0.3)/mL
Raw fermented milk	= R acidification (Table 14)	3.5 (3.5, 3.5)	0.9 (‐Inf, 3.7)/mL
Raw milk fresh cheese	= R acidification (Table 14)	3.5 (3.5, 3.5)	0.9 (‐Inf, 3.7)/g
Raw milk soft cheese	= R acidification + R ripening (Tables 14 and 15)	4.5 (3.9, 5.8)	−0.1 (‐Inf, 2.8)/g
Raw milk semi‐soft cheese	= R acidification + R ripening (Tables 14, 15)	3.7 (2.2, 6.2)	0.7 (‐Inf, 3.6)/g
Raw milk cream	NA	0 (0, 0)	4.4 (1.0, 7.2)/g
Raw fermented cream	= R acidification (Table 14)	1.8 (0.9, 2.4)	2.6 (−0.8, 5.5)/g
Pasteurised cream	= R thermal (Table 13)	NT82: 55.2 (9.8, 641.8)	‐Inf (‐Inf, ‐Inf)/g
Raw colostrum	NA	0 (0, 0)	4.4 (1.0, 7.2)/mL
Pasteurised colostrum	= R thermal (Table 13)	NT63: 66.2 (43.8, 119.4) NT72: 4.9 (1.7, 20.2) IP72: 7.0 (2.8, 26.5)	NT63: ‐Inf (‐Inf, ‐Inf)/mL NT72: −0.51 (‐Inf, 2.34)/mL IP72: ‐Inf (‐Inf, 0.28)/mL

Abbreviations: IP, industrial profile, the adjacent number indicates the temperature; NA, not applicable; NT, notional treatment; NT57, 57°C for 30 min; NT63, 63°C for 30 min; NT65, 65°C for 15 s; NT72, 72°C for 15 s; NT82, 82°C for 15 s.

^a ‐Inf = log(0) indicates 0 EID₅₀/mL or g.

^b Thermised milk is not consumed as such. A drinking portion size is assumed in the assessment for comparison purposes.

3.6.2. Probability of exposure via the food‐borne route and, if feasible, probability of infection/illness per serving

Probability of infection and illness given infection, upon human exposure to the viable H5N1 B3.13 genotype virus through the oral route (SQ 2.2)

From April 2024 to June 25th 2025, 70 human cases and 7 probable cases of infection with H5 influenza A virus have been reported in the US. According to the analysis by Rolfes et al. (2025) and considering both sequence‐confirmed and inferred cases, 52 of the 70 cases (74%) were associated to the H5N1 B3.13 genotype virus, while the remaining were represented by the D1.1 (17 cases) and the D1.3 (1 case) genotypes. For 73 of the 77 ascertained or probable cases, exposure was traced to contact with infected dairy herds, activities associated with poultry farms and culling operations or other kinds of animal exposure (including contacts with backyard flocks, wild birds or other mammals). Only for three cases and one probable case, the source of exposure could not be identified (CDC Bird Flu surveillance ²³ ). Food‐borne infection has not been reported as a cause for any of the cases. Therefore, to date, food‐borne transmission ― and particularly transmission through milk, colostrum, dairy products or colostrum‐based products ― of H5N1 B3.13 genotype virus has not been reported. The disease in individuals infected with H5 influenza A virus through contact with sick or infected animals has often been mild, being characterised by conjunctivitis and flu‐like symptoms (e.g. mild fever, headache, cough, rhinorrhea ²⁴ ).

It is currently not possible to determine if, in case of a food‐borne infection, symptomatology may differ from that observed for respiratory infection, if specific subpopulations (e.g. elderly, pregnant women) may display different susceptibility, or if infection in breastfeeding women may lead to virus occurrence in human milk, as reported for other viruses (Francese et al., 2023). Furthermore, it is unknown how antibody response developed through previous exposure to human influenza viruses (e.g. H1N1) may affect a DR relationship via the oral route, though it is acknowledged that antibodies to the conserved stalk domain of the hemagglutinin may provide partial protection towards the infection with H5N1 viruses (Garretson et al., 2025).

• Infection through accidental inhalation during ingestion

Accidental inhalation of foods, liquids or other materials into the airways during ingestion (also described as ‘silent aspiration’) is relatively rare in healthy adults but is a phenomenon often described in individuals with dysphagia, and in patients with prolonged intubation or with neurological disorders; also, specific age groups (elderly, infants) are more prone to silent aspiration (Butler et al., 2018). Silent aspiration has been associated with pneumonia, and while pulmonary infections by food‐borne pathogens such as Listeria monocytogenes have been reported, no clear evidence of their association with silent aspiration is present (Morgand et al., 2018). Nonetheless, given the presence of influenza receptors in the respiratory tract (α2,3 sialic acid receptors in ciliated cells in bronchioles and alveoli and α2,6 in the ciliated and non‐ciliated cells in respiratory tract), infection with avian influenza, including H5N1 B3.13 genotype virus, through silent aspiration cannot be ruled out and a risk assessment scenario for this transmission has been proposed in Chen et al. (2025). For the current assessment, only infection through food ingestion will be explicitly considered. It is however recognised that, likewise food ingestion, infection through silent aspiration may pose a risk to consumers, particularly for liquid foods and in the presence of predisposing conditions.

• Infection through ingestion

Several studies have reported the detection and even isolation of various influenza A viruses from human stool samples (Minodier et al., 2015), suggesting possible intestinal involvement in infection. The presence of both α2,6‐linked sialic acid receptors, particularly in the ileal epithelium and of α2,3‐linked sialic acid receptors, which are preferentially recognised by avian influenza viruses, in the colon epithelium (Kuchipudi et al., 2021) further supports this possibility. In addition to sialic acids, several other cellular receptors and co‐factors – not confined to the respiratory tract and also expressed in extra‐respiratory tissues such as the gastrointestinal tract – may facilitate infection by both human and avian influenza viruses (Karakus et al., 2020, 2024).

Various animal species – including mice, ferrets, hamsters, guinea pigs and non‐human primates – have been used to model human infection with AIVs, including modelling of oral infection; laboratory studies evaluating the risk of infection in these species following oral or gastric inoculation with influenza A viruses are summarised in Bullock Trent et al. (2025). Mice are widely used due to the advantages they offer (cost, ease of handling, etc.), but they are not naturally susceptible to many wild‐type AIVs (virus adaptation is needed) and they typically do not exhibit hallmark features of human influenza disease (fever, sneezing or respiratory droplet transmission). Ferrets, by contrast, are considered the most representative model for human influenza. They are naturally susceptible to both human and avian influenza viruses without prior adaptation and develop symptoms closely resembling those seen in humans. Ferrets also support efficient replication of AIVs in the upper respiratory tract and are capable of transmitting the virus via respiratory droplets, making them ideal for studying transmission dynamics. However, it is important to note that ferrets do not fully replicate human immune complexity or the nature of human social interactions relevant to transmission.

Data from experimental infection through the oral route in animal models and previously developed DR models used for human oral exposure are provided in Appendix F.

Three published DR models were available: (i) the FSIS‐FDA‐APHIS model (FSIS, FDA and APHIA, 2010), relying on an exponential function and based on data on intranasal exposure of human volunteers to influenza viruses (data from Beare & Webster, 1991); (ii) the FDA model (Chen et al., 2025), using the FSIS‐FDA‐APHIS model for inhalation, modified to approximate an ingestion DR relationship through a 3 log₁₀ reduction of the r‐value (virulence parameter) of the model, to account for reduction of viable virus during gastric passage; and (iii) the Koebel model (Koebel et al., 2024), also using an exponential function, estimated using data from a ferret ingestion model (data from Lipatov et al., 2009). While the DR models based on inhalation in humans had the undeniable advantage of eliminating uncertainties related to extrapolating from animal models to humans, extrapolation from intranasal exposure to oral exposure introduces uncertainty to the assessment, given the different mechanisms of disease development of the two exposure routes. Conversely, DR models based on oral/ingestion in animal models are subjected to the significant uncertainty of extrapolation to humans but have the advantage of providing data on H5Nx strains (including H5N1 B3.13 genotype virus) through a comparable exposure route.

The data retrieved from experimental infection through the oral route in animal models (eight studies, including ferrets, mice, macaques, hamsters and cats) were assessed considering evidence of infection. Based on data patterns, mice displayed a significantly higher susceptibility to infection compared to the other species, including non‐human primates. It was therefore agreed not to consider this species for DR model development. The single study on macaque exposed to the virus via the oral route to 10⁷ TCID₅₀ of HPAIv H5N1 clade 2.3.4.4b (A/bovine/OH/B24OSU‐342/2024) (Rosenke et al., 2025) showed limited infection and seroconversion, with the animals remaining asymptomatic throughout the study. Conversely, intranasal or intratracheal inoculation of macaques caused systemic infection resulting in mild and severe respiratory disease, respectively. Although the macaque study provided important proof‐of‐principle evidence that food‐borne infection with HPAIv H5N1 can occur (as shown by seroconversion), orogastric inoculation led only to localised, low‐level virus replication with minimal pathology and no clinical disease, indicating low susceptibility to systemic infection via this route.

For the other animal models, ferrets were those with the largest amount of data and were therefore selected for model development using a beta‐Poisson model. The DR relationships for H5N1 oral infection in ferrets showed an estimated median ID₅₀ of 1.2 × 10⁸ EID₅₀ (95% CI: 4.0 × 10⁷ to 5.3 × 10⁸).

Comparison of the developed ferret ingestion model with the FDA model (inhalation model modified to approximate oral ingestion; ID₅₀ = 5.1 × 10¹¹ EID₅₀, 95% CI: 5.1 × 10¹⁰ to 2.8 × 10¹²) showed major differences (Figure 9), the FDA model providing an estimated ID₅₀ approximately 4000 higher than the ferret model. Conversely, it was observed that the ID₅₀ of the developed ferret ingestion model closely resembled the value of FSIS‐FDA‐APHIS model for inhalation in humans, providing a slightly lower ID₅₀ (1.2 × 10⁸ compared to 5.1 × 10⁸ EID₅₀). This was regarded as a sign of overestimation of the response by this animal ingestion DR model compared to human ingestion, as the HPAI ID for oral exposure in humans is expected to be higher than for respiratory exposure. Oral ingestion, indeed, exposes H5N1 to inactivation during gastric digestion through a pH‐induced conformational change of the hemagglutinin proteins (David Shannon et al., 2023), which reduces the amount of virus able to initiate infection. The impact of digestion on some relevant respiratory viruses, such as coronaviruses, for example, is affected by several factors, including the variability of gastric pH associated to fed/fasting‐state, the protective effect of ingested food (Harlow et al., 2022) – which may vary according to food type and composition – and the gastric emptying time (Esseili Malak, 2023). Furthermore, it is recognised that the use of proton pump inhibitors and conditions associated with hypochloridria may affect susceptibility to infections through the ingestion route (Charpiat et al., 2020; Martinsen et al., 2005). Given the number of factors affecting the impact of gastric digestion on viable H5N1 and the scarcity of specific data on the effect of gastrointestinal fluids on the H5N1 B3.13 genotype virus, and considering the variability of the food products selected for the assessment (with different buffering capacity, protective effect and gastric emptying times), it was agreed that the selection of a reduction value associated with gastric digestion – as per the FDA model developed for raw and pasteurised milk – would be affected by a significant uncertainty.

FIGURE 9.

Comparison of three dose–response models describing the probability of clinical infection with H5N1 through inhalation and oral exposure in humans (grey and blue line, respectively) according to FSIS, FDA and APHIS (2010) and Chen et al. (2025), and through oral exposure in ferrets (red line; this opinion). The solid red line indicates the median predicted probability of infection, and the shaded ribbon represents the 95% confidence band.

Overall, considering these outcomes and the limitations and uncertainties of all models, it was agreed that it was not appropriate to select a DR model. In their risk assessment, Browne et al. (2025) concluded that the orally infectious dose for humans is unlikely to be much lower than 10⁷ TCID₅₀ (or 3 × 10⁷ EID₅₀ using our conversion factor).

In the absence of a DR model, exposure levels have been evaluated against critical thresholds derived from the available evidence through expert judgement. Two thresholds were selected, considering a 1% probability of the median estimate for the FSIS‐FDA‐APHIS human inhalation model (i.e. 6.9 log₁₀ EID₅₀ rounded to 7 log₁₀ EID₅₀) and the FDA inhalation model modified to approximate oral ingestion (9.9 log₁₀ EID₅₀ rounded to 10 log₁₀ EID₅₀). The first one was selected taking into account that the human inhalation model, combined with the considerations on the impact of gastric digestion on H5N1 viability, supports that the probability of human infection through the oral route is much lower than 1% at doses up to 7 log₁₀ EID₅₀, and that in the study on macaques inoculated with H5N1 B3.13 genotype virus via the oral route (10⁷ TCID₅₀, equivalent to 7.48 log EID₅₀) seroconversion occurred with the animals remaining asymptomatic. The second threshold was instead selected considering that the FSIS‐FDA‐APHIS model for inhalation, modified to approximate an ingestion DR relationship, with the assumption of a 3 log₁₀ reduction during gastric passage, provides a probability of human infection through the oral route of 1% at doses of 10 log₁₀ EID₅₀. These levels were used as a reference for classification and comparison across products.

Probability of exposure to different levels of viable H5N1 B3.13 genotype virus through the consumption of the selected foodstuffs (SQ 2.1)

Consumer exposure to viable H5N1 B3.13 genotype virus in the selected food products was calculated using the concentrations present in the final products just after manufacturing (Table 16), while accounting for standardised serving sizes. As serving sizes, the mean daily intake of adults as given in the EFSA food consumption database was used (Table 17), except for colostrum. For colostrum, the serving size was based on Playford et al. (2001), considering suggested amounts in trial studies on consumption of bovine colostrum to reduce intestine permeability in individuals with chronic assumption of non‐steroidal anti‐inflammatory drugs. Fixed values without variability or uncertainty were used in the model.

TABLE 17.

Serving size for adults derived from the EFSA food consumption database.

Categories of foodstuff	FoodEx 2 level/reference	Serving size (g)
Raw (drinking) milk	L5_Cow milk	201
Thermised milk	L5_Cow milk	201 a
Pasteurised (drinking) milk	L5_Cow milk	201
Raw fermented milk	L3_Fermented milk products	192
Raw milk fresh cheese	L3_Fresh uncured cheese	72
Raw milk soft cheese	L4_Soft ‐ ripened cheese	51
Raw milk semi‐soft cheese	L5_Firm/semi‐hard cheese (Gouda and Edam type)	41
Raw milk cream	L4_Cream, plain	33
Raw fermented cream	L4_Sour cream, plain	40
Pasteurised cream	L4_Cream, plain	33
Raw colostrum	NA	125 b
Pasteurised colostrum	NA	125 b
Freeze‐dried colostrum powder	L4_Milk powder	16 or 7.4 (weighted)

Abbreviation: NA, not available.

^a Thermised milk is not consumed as such. A drinking portion size is assumed in the assessment for comparison purposes.

^b Playford et al. (2001).

For all categories of foodstuffs, exposure is expressed as the frequency of servings (per thousand) that contain a dose per serving larger than each of the three selected critical thresholds (EID₅₀ = 0; log₁₀ EID₅₀ = 7 and log₁₀ EID₅₀ = 10), assuming that the foods are consumed as they are and the product derives from milk or colostrum from a contaminated bulk tank (Table 18). The uncertainty in this frequency, as indicated in Figure 10 and Table 18, is the uncertainty associated with the data on viral loads shed by individual cows, uncertainty in the translation from TCID₅₀ to EID₅₀ and the uncertainty in the sampling of servings from the bulk milk.

TABLE 18.

Number of servings per thousand exceeding a certain dose (i.e. 0 EID₅₀, 10⁷ EID₅₀ or 10¹⁰ EID₅₀) of viable H5N1 B3.13 genotype virus for consumers in the EU through the consumption of servings of various foodstuffs (consumed as such), assuming that the servings originate from milk or colostrum from a contaminated bulk tank in the EU (see Section 3.6.1). Numbers given represent median, (5%–95% uncertainty interval).

Categories of foodstuff	Heat treatment a	Exposure threshold
		0 EID50	107 EID50	1010 EID50
Raw (drinking) milk	NA	999, (997–1000)	340, (253–434)	10, (4–17)
Thermised milk b	NT57	0, (0–0)	0, (0–0)	0, (0–0)
	NT65	995, (989–1000)	163, (104–226)	1, (0–3)
	IP65	992, (983–998)	113, (66–163)	1, (0–1)
Pasteurised (drinking) milk	NT63	0, (0–0)	0, (0–0)	0, (0–0)
	NT72	622, (584–659)	8, (4–13)	0, (0–0)
	IP72	433, (393–473)	1, (0–2)	0, (0–0)
Raw fermented milk	NA	939, (900–972)	3, (1–6)	0, (0–0)
Raw milk fresh cheese	NA	911, (862–953)	1, (0–2)	0, (0–0)
Raw milk soft cheese	NA	761, (686–832)	0, (0–0)	0, (0–0)
Raw milk semi‐soft cheese	NA	827, (770–881)	2, (1–4)	0, (0–0)
Raw milk cream	NA	997, (992–1000)	188, (122–260)	2, (1–3)
Raw fermented cream	NA	978, (958–992)	32, (16–50)	0, (0–0)
Pasteurised cream	NT82	7, (6–9)	0, (0–0)	0, (0–0)
Raw colostrum	NA	999, (996–1000)	296, (213–383)	6, (2–11)
Pasteurised colostrum	NT63	0, (0–0)	0, (0–0)	0, (0–0)
	NT72	604, (565–644)	6, (3–10)	0, (0–0)
	IP72	414, (374–454)	1, (0–2)	0, (0–0)

^a NA, not applicable; NT, notional treatment; IP, industrial profile, the adjacent number indicates the temperature. NT57, 57°C for 30 min, NT63: 63°C for 30 min; NT65, 65°C for 15 s; NT72, 72°C for 15 s; NT82, 82°C for 15 s.

^b Thermised milk is not consumed as such. A drinking portion size is assumed in the assessment for comparison purposes.

FIGURE 10.

Probability of exposure to different levels of viable H5N1 B3.13 genotype virus for consumers in the EU through the consumption of raw milk servings, assuming that the servings originate from milk from a contaminated bulk tank in the EU. The black line is the median, bands indicate the 50% and 90% uncertainty interval.

As an example, for the consumption of raw drinking milk, it is assumed that a serving of 201 mL is directly taken from a contaminated bulk tank. Taking the variability between concentrations in milk from different bulk tanks into account, this leads to a distribution representing the variability in ingested doses between servings. This distribution is shown in Figure 10. As a median estimate, the results show for example that 340 of 1000 servings of raw drinking milk contain a dose of more than 7 log₁₀ EID₅₀, with a 90% uncertainty interval of 253–434 per 1000 servings.

Results show that exposure is highest for raw drinking milk, raw colostrum and raw milk cream, followed by thermised milk (65°C – 15 s), all exceeding both the 10⁷ EID₅₀ threshold (> 100 servings per thousand) and the 10¹⁰ EID₅₀ threshold (between 1 and 10 servings per thousand) (Table 18). Of these products, it is considered that thermised milk is not consumed as such as thermisation is typically performed to extend the shelf‐life of milk prior to further processing or in the manufacture of some cheeses.

Considering the threshold value of 10⁷ EID₅₀, exposure is lower for raw fermented cream (32 servings per thousand) and further lower (between 1 and 10 servings per thousand) for raw fermented milk, fresh and semi‐soft cheese made from raw milk, and drinking milk and colostrum subjected to HTST pasteurisation (72°C – 15 s). Industrial time/temperature profile (i.e. IP72) instead of the notional time/temperature for thermisation or pasteurisation reduces the exposure estimates and the infection dose for oral exposure is expected to be higher than for respiratory exposure considering the connotation of the 10⁷ EID₅₀ threshold (1% probability of infection according to the human inhalation model, see Section 3.6.2).

Pasteurisation of milk or colostrum using a LTST treatment at 63°C for 30 min and thermisation of milk at 57°C for 30 min inactivates all HPAI H5N1 present, thus not leading to any exposure to viable virus.

The exposure to viable virus in products derived from thermised or pasteurised milk (e.g. fermented milk, fermented cream, different types of cheese) is expected to be lower than in the equivalent products derived from raw milk.

3.6.3. Measures along the food chain to reduce the levels of viable H5N1 B3.13 genotype virus

The impact of thermal treatments of milk, such as pasteurisation and thermisation, to reduce the levels of H5N1 B3.13 genotype virus can be found in Table 13. Based on the notional heat treatment for thermisation (65°C – 15 s) or HTST pasteurisation (72°C – 15 s) for example, the predicted reduction of infectious virus is 0.9 [0.5–1.9] and 4.9 [1.7–20.2] log₁₀, respectively. Other combinations of time/temperature can achieve significant reductions as well. Table 19 and Figure 11 present the estimated time (in minutes) required to achieve different levels of log₁₀ reduction (4, 5, 6 and 7 log₁₀) at temperatures ranging from 50°C to 85°C. Two values are reported: the median estimate (typical scenario) and the 97.5th percentile (conservative scenario, accounting for biological variability).

TABLE 19.

Predicted time ^a (in minutes) to achieve a given log₁₀ reduction of viable H5N1 B3.13 genotype virus in milk or colostrum based on the Bigelow model and bootstrap parameter distributions.

Temperature (°C)	D‐value median (97.5th)	Predicted time (min)
		4 log10		5 log10		6 log10		7 log10
		Median	97.5th	Median	97.5th	Median	97.5th	Median	97.5th
50	10.6 (22.6)	42	90	53	110	64	140	74	160
55	3.2 (4.3)	13	17	16	22	19	26	22	30
60	0.93 (1.2)	3.7	4.8	4.7	6.0	5.6	7.2	6.5	8.4
65	0.28 (0.48)	1.1	1.9	1.4	2.4	1.7	2.9	2.0	3.3
70	0.08 (0.20)	0.33	0.81	0.42	1.0	0.50	1.2	0.58	1.4
75	0.02 (0.08)	0.10	0.35	0.12	0.43	0.15	0.52	0.17	0.60
80	0.007 (0.03)	0.03	0.15	0.04	0.18	0.04	0.22	0.05	0.25
85	0.002 (0.015)	< 0.01	0.06	< 0.01	0.08	0.01	0.09	0.02	0.11

^a Times are derived from the fitted Bigelow secondary model using a bootstrap distribution of D₇₀ and z‐values. To enhance readability, results are rounded according to their magnitude. When the predicted value is extremely small, a threshold notation is used (e.g. ‘< 0.01’). The median and 97.5th percentile times are shown.

FIGURE 11.

Predicted time (in minutes) for the median estimate (typical scenario) and the 97.5th percentile (conservative scenario, accounting for biological variability) to achieve log₁₀ reductions of viable H5N1 B3.13 genotype virus in milk or colostrum based on the Bigelow model and bootstrap parameter distributions. The dotted lines present the 4, 5, 6 and 7 log₁₀ reductions.

Novel thermal processes (i.e. ohmic and induction heating) will inactivate the virus by raising the temperature of the milk in a more rapid/efficient and/or uniform way. The principle of inactivation stays the same as for the standard thermal treatments. For these novel thermal processes there are no specific data on virus inactivation in milk.

Non‐thermal technologies have been largely driven by consumer demand for minimally processed food products with the flavour and nutritive properties of fresh foods. Deeth & Lewis (2017) describe the following non‐thermal technologies alone or in conjunction with additional thermal processing for producing milk and dairy products: microfiltration, high‐pressure processing (HPP), pulsed electric field technology (PEF), high‐pressure homogenisation (HPH), bactofugation, Ultraviolet‐C irradiation (UV‐C), Gamma (γ) irradiation, carbon dioxide and (high‐pressure) carbon dioxide (HPCD). Microfiltration is of limited or no efficacy for virus removal as the pore size of the membranes (considering ceramic membranes with a pore size of 1.4 μm) is a limiting factor in the interaction with microorganisms of small size (Eugster & Jakob, 2019). There is lack of evidence on the efficacy of other processes like bactofugation, γ‐irradiation and HPCD for virus inactivation in milk/colostrum. The advantages, limitations and stage of commercial application of these technologies have been reviewed by Deeth & Lewis (2017). The dairy applications of HPP, PEF technology and HPH have been reviewed by Deeth et al. (2013), while Neoκleous et al. (2022) also reviewed the effect on safety and quality characteristics of non‐thermal processing technologies for dairy products.

Table 20 provides an overview of the potential of various non‐thermal technologies to reduce the levels of viruses in milk or colostrum. It needs to be considered that the processes are at a variable degree of implementation by the food industry (and especially the dairy industry), ranging from pilot scale to full industrial use and that most of the available data concern human and animal viruses other than HPAIv. It should also be noted that, in several instances, experimental data involving human viral pathogens were gathered on breast milk (e.g. for milk donation) and not on bovine milk.

TABLE 20.

Potential of various non‐thermal technologies to reduce the levels of viable viruses in milk or colostrum.

Technology	Inactivation mechanism	Type of virus and typical inactivation conditions	Notes	Reference
High‐Pressure Processing (HPP)	High pressures disrupt noncovalent bonds such as ionic and hydrophobic bonds (i.e. biomolecules proteins and nucleic acids)	Cytomegalovirus (CMV) in human milk reduced by > 0.9 log10 using 350 MPa for 8 min at room temperature Hepatitis A virus (HAV) in human milk reduced by > 4 log10 using 350 MPa for 8 min at room temperature	HPP is effective against enveloped and non‐enveloped viruses	Pitino et al. (2022)
		Human Coronavirus (HCoV) 229E in human milk reduced by 1.4 log10 using 4 cycles (5 min) of 350 MPa at 38°C and non‐significantly reduced using a single cycle (5 min) of 600 MPa at 20°C. Hepatitis E Virus (HEV) in human milk non‐significantly reduced using 4 cycles (5 min) of 350 MPa at 38°C and reduced by 1.6 log10 using a single cycle (5 min) of 600 MPa at 20°C.	HPP cannot ensure viral removal in donor milk when compared to ‘holder pasteurisation’ using the standard protocol of 62.5°C for 30 min based on two models for enveloped and non‐enveloped viruses	Bouquet et al. (2023)
		Bovine herpes virus (BHV) in bovine colostrum reduced by 4.5 log10 using 300 MPa for 30 min at room temperature Feline calicivirus (FCV) in bovine colostrum reduced by 5 log10 using 300 MPa for 30 min at room temperature	HPP of bovine colostrum maintains an acceptable IgG level while decreasing bacterial and viral counts	Foster et al. (2016)
High‐Pressure Homogenisation (HPH)	High pressure combined with cavitation, shear stress and turbulence, impingement and temperature, which may increase due to friction	Murine norovirus (MNV‐1) in skim milk reduced by ~1.3 log10 using 300 MPa for < 2 s FCV‐F9 in skim milk reduced by ≥ 4 log10 using 300 MPa and by ~1.3 log10 using 250 MPa for < 2 s Homogenisation block temperature: 2°C; recorded exposure temperatures of 70°C and 75°C for < 2 s at 250 and 300 MPa, respectively	HPH appears promising for commercial use to inactivate norovirus surrogates in fluid foods at pressures ≥ 300 MPa	Horm et al. (2012)
Ultraviolet C (UV‐C) Irradiation	DNA/RNA damage by UV‐C light at 254 nm	The UV‐C resistant phage MS2 (ssRNA virus) in skim milk (106 PFU/mL) was reduced by 5 log10 using 150 mJ/cm2 The viral surrogate phage T1UV (dsRNA virus) in skim milk (106 PFU/mL) was reduced by 5 log10 using ~30 mJ/cm2		Gunter‐Ward et al. (2018)
		CMV in human milk reduced using optimal (1 cm) and suboptimal (5 cm) UV‐C irradiation protocols at 254 nm wavelength; 64 mJ/cm2 UV‐C was shown to eliminate replicative virus	Mainly surface or thin liquid treatment	Lloyd et al. (2016)
		Escherichia coli phage T4 in human milk (104 PFU/mL) was completely inactivated using an average UV‐C dose of 1879.2 J/m2 during the 18‐min treatment time Staphylococcus aureus phage BYJ20 in human milk (104 PFU/mL) was completely inactivated using an average UV‐C dose of 1879.2 J/m2 during the 18‐min treatment time		Stinson et al. (2023)
		Bacteriophage MS2 (104–105 PFU/mL) in whole milk reduced by 0.7 log10 using far UV‐C light at a dose of 300 mJ/cm2		Memic et al. (2025)
Pulsed Electric Fields (PEF)	Disrupts viral membranes and proteins via electric pulses	Bacterial count in human milk was reduced by ~4 log10 using 6000 pulses at 15 kV and 20 Hz or 50 Hz frequency	Possibly effective against viruses	Zhang et al. (2023)

Abbreviations: BHV, Bovine herpes virus; CMV, cytomegalovirus; dsRNA virus, double stranded RNA virus; FCV, Feline calicivirus; HAV, Hepatitis A virus; HCoV, Human Coronavirus; HEV, Hepatitis E Virus; HPH, high‐pressure homogenisation; HPP, high‐pressure processing; IgG, immunoglobulin; MNV, Murine norovirus; PEF, pulsed electric fields; ssRNA virus, single stranded RNA virus; UVC, ultraviolet C.

HPP is a non‐thermal treatment in which foods are subjected to isostatic pressures. Pressure is transmitted rapidly and uniformly in an isostatic manner so that all parts of the food are subjected to the same pressure simultaneously. An EFSA scientific opinion (EFSA BIOHAZ Panel, 2022) found that, within the industrial context, pressures of between 400 and 600 MPa are most often applied for microbial inactivation in foods, with common holding times ranging from 1.5 to 6 min. The water used as pressure‐transmitting fluid is often pre‐chilled at 4–8°C. Although the use of HPP in milk was one of the first applications, up to now, the industrial relevance of HPP in the dairy sector is low and there are still very few applications for dairy products, although patents for milk and dairy products exist. On the other hand, HPP is successfully applied to high‐quality dairy products such as cheese to inactivate pathogens. HPP application on cheese can increase the shelf‐life and safety of the product. Target pressure–holding time combinations that would achieve defined log₁₀ reductions (i.e. 5, 6, 7 and 8 log₁₀ reductions) for various pathogens were identified in EFSA BIOHAZ Panel (2022). However, for the relevant virus identified in the context of that opinion, TBEV, minimum HPP requirements could not be set as no data were available. HPP efficacy towards viruses in milk was investigated by Pitino et al. (2022), who demonstrated the reduction of cytomegalovirus (CMV) and HAV in human milk, by Bouquet et al. (2023), which reported the reduction of human coronavirus and Hepatitis E virus in human donor milk, and by Foster et al. (2016), who showed the reduction of bovine herpes virus (BHV) and feline calicivirus (FCV) in bovine colostrum. More information can be found in the review by Osaili et al. (2025) on the ability of HPP to control pathogenic and spoilage microorganisms in liquid foods, including milk.

HPH is based on the same principles as the standard homogenisation process used in the dairy industry for reducing the size of fat globules, but it works at higher pressures (100–400 MPa). Not only pressure, but the combined action of other physical effects (such as cavitation, shear force, turbulence, impingement and temperature, which may increase due to friction) causes inactivation of vegetative microorganisms (Patrignani & Lanciotti, 2016). Horm et al. (2012) demonstrated the effect of HPH to inactivate two human norovirus surrogates, murine norovirus (MNV‐1) and FCV‐F9, in skim milk.

Ultraviolet light in the germicidal range (UV‐C) from 200 to 280 nm, is being investigated as an alternative to thermal treatment for inactivating pathogens and improving shelf‐life and safety of skim milk. UV‐C light is well proven as a non‐thermal method of disinfecting drinking, waste and recreational water. Recent advances in science and engineering have clearly demonstrated that UV‐C technology holds considerable promise as an alternative to traditional thermal processes such as pasteurisation for food preservation (Gunter‐Ward et al., 2018). Tamarack's TruActive® process using advance UV light technology has recently received the approval from FDA that this method meets the official efficacy criteria for pasteurisation for the production of powdered dairy ingredients such as whey protein concentrate, milk protein concentrate and immune‐supporting compounds like lactoferrin. ²⁵ Lloyd et al. (2016) described the inactivation of cytomegalovirus, a virus commonly excreted into breast milk, using UV‐C irradiation, while Gunter‐Ward et al. (2018) inactivated model viruses (phages) in skim milk by using a bench‐top Collimated Beam system. Stinson et al. (2023), inoculated two different phages in donor milk and illustrated that UV‐C treatment efficiently destroys both thermotolerant and thermosensitive phages.

PEF treatments consist of subjecting a product placed between two electrodes, usually immersed in an aqueous solution, to high‐intensity electric fields (between 0.5 and 30 kV/cm) by applying intermittent pulses of short duration (microseconds to milliseconds) without increasing the product's temperature (Martínez et al., 2023). PEF was used by Zhang et al. (2023) to reduce bacterial counts in donor milk. Although they focused on microbial inactivation, the possibility that PEF could also be active against viruses was clearly noted.

Other non‐thermal processes such as cold plasma are prompted as potential candidates for inactivating viruses in foodstuffs but have not yet been tested in milk (Harikrishna et al., 2023). Likewise, infra‐red (IR) technology has been described by Danesi et al. (2024) as a promising solution for reducing microbial loads in raw milk but has not yet been tested on food‐borne viruses. Further, Guimarães et al. (2019) reviewed the application of high‐intensity ultrasound treatment to dairy products and milk, focusing on microbial inactivation with no reference to viruses.

3.7. Uncertainty analysis

Several sources of uncertainty affecting the outcome of the scientific assessment were identified:

ToR 2a.: Assessment of the potential impact of the infection of dairy cows in the EU with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13

Several sources of uncertainty affecting the risk assessment performed through the L'ORA tool are described in Annex A. Briefly, these are:

• –Uncertainties in estimating disease occurrence due to the use of incomplete data and voluntary reporting and to the lack of specific data on transmissibility of the H5N1 B3.13 genotype virus
• –Uncertainties in estimating the rate of incursion derived from (a) the assumption that H5N1 B3.13 genotype virus will behave similar to other HPAI strains in poultry and will behave similarly in all bird species considered (poultry or wild regardless of age except for fertilised eggs); (b) the assumption that cattle and poultry have separate production systems; (c) limitations in TRACES data regarding production type of imported cattle and NUTS2 region involved; (d) lack of data on the presence of B3.13 in cattle and cattle products; (e) lack of data regarding contact of wild birds and contaminated fomites and of products of animal origin and livestock/wildlife; (f) limitations of data used on wild bird distribution and on wild bird movement patterns;
• –Uncertainties in estimating the extent of spread due to limitations on the data regarding size of wild bird populations and on the transmission dynamics in affected populations, and due to the assumptions made in the models used (homogeneous mixing, lack of pre‐existing immunity).
• –Uncertainties in estimating the impact of disease related to lack of details on the specific measures that would be implemented (and to what extent) in case of disease introduction, to the uncertainties affecting the estimates of the extent of spread mentioned above, to the use of limited local data for estimating the cost and to the lack of data on the public response to the introduction of H5N1 B3.13 genotype virus.

The level of uncertainty in the risk assessment derived from the sources described above was assessed as ranging from low to moderate on different aspects of estimating disease occurrence, rate of incursion and impact of disease and moderate to high for aspects related to estimating the spread of disease (Table 25 in Annex A).

ToR 2b.: Measures that could prevent the introduction and spread of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 in dairy cows and poultry in the EU

• –The assessment of the efficacy of the measures is based on the assumption that they were correctly implemented, which may be more challenging in certain situations (e.g. the level to which contamination of feed and water from wild birds can be reduced will depend on the facilities; the efficacy of proper disposal of contaminated materials from a farm in a dedicated area will depend on the specifics of the area).
• –Effectiveness of measures aiming at early detection of infected animals/farms is dependent on the incubation and latency periods in cattle, aspects for which still limited field data is available.
• –The assessment of the effectiveness of measures based on altering transmission patterns (by limiting effective contacts or allowing the disease to spread to quickly reduce the proportion of susceptible hosts in a farm) is hampered by the limited data on transmission dynamics under field conditions in cattle.
• –The assessment of the effectiveness of vaccination in cattle is based on the limited data currently available from experimental trials, which indicates that it has potential to become a useful tool for disease control, but this is still preliminary information.
• –Feasibility of the measures is assessed considering an average dairy cattle farm in the EU, but there are large variations in production systems that could have a major impact on how feasible implementing certain measures is (e.g. indoor confinement of dairy cattle in certain organic productions, etc.).

The effect of these sources of uncertainty on the assessment was not quantified.

ToR 2c.: Adaptations of the current EU surveillance for HPAI needed to allow detection of an introduction of the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 into the EU dairy cow populations

The assessment of the usefulness of possible adaptations of current EU surveillance for HPAI for detection of an introduction of H5N1 B3.13 genotype virus in the EU dairy cow populations is limited by the lack of data on the performance of passive, syndromic and, to a lower extent, active surveillance strategies for HPAI detection in cattle. Relative roles of intramammary versus other infection routes, within‐herd prevalence over time and potential mechanical vectors are still being clarified (EFSA, 2025). Moreover, it has been difficult to reproduce field observations in experimental infections. These uncertainties directly affect sampling frames and farm revisit intervals. The effect of these sources of uncertainty on the assessment was not quantified.

ToR 2d.: Likelihood of bulk milk to be contaminated if EU lactating dairy cows are infected with the HPAI virus H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13

The data available to inform the elicitation of values for the parameter of interest is affected by several sources of uncertainty:

• The only data informing of the time elapsed between infection and shedding come from experimental studies performed with very low number of animals (n = 2–3) that were infected in controlled challenges and thus potentially exposed to doses (and even routes) not fully representative of what may be observed in the field.

• Estimates for the duration of the incubation period from observational studies are scarce and reliant on the date of the introduction of likely infected animals as the primary source of infection in the farm. Furthermore, there is no information on how certain management factors (farm size, biosecurity, etc.) may affect disease dynamics (e.g. speed of transmission).

• While experimentally infected animals showed clinical signs 1–2 days post‐infection, sample sizes were very small and field data indicate that the chance of having subclinical/non‐clinically affected animals in the farm is non‐negligible (and the proportion of subclinical animals in a positive farm can be > 50% of those infected), but there is limited data on this.

• Time from infection to detection will be related not only to the incubation period (time for clinical signs to appear) but also to the time passing until HPAI tests are applied (e.g. after other likely causes and common treatments have been considered); there is however no information on the length of this period.

ToR 2e.: Probability of infection/illness for a consumer in the EU population due to the exposure to viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products through the food‐borne route and measures to mitigate the risk.

The sources of the main uncertainties related to ToR 2e can be found in Appendix G.

4. CONCLUSIONS

While the request of the EC focusses on the H5N1 B3.13 genotype virus, it should be noted that experimental evidence indicates this genotype virus is unlikely to behave differently from H5N1 clade 2.3.4.4b strains already circulating in Europe. Further, the absence of evidence of cattle infections despite widespread and repeated exposure to HPAIv in wild birds and poultry in Europe indicates a very narrow bottleneck for spillover to dairy cattle under European conditions.

Conclusions about the potential impact of the infection of dairy cows in the EU with H5N1 B3.13 genotype virus (ToR 2a.)

The introduction of HPAI B3.13 from the US into EU dairy cattle or poultry via trade is considered highly unlikely (annual probability < 10⁻⁵). Introduction via wild bird migration is also highly unlikely (annual probability < 10⁻⁵), except for western Ireland and western France regions (annual probability up to 10⁻²–10⁻¹), especially in autumn. However, the latter estimates are subject to considerable uncertainty due to limited data on transatlantic wild bird movements and HPAI prevalence in US wild bird populations.

The occurrence of HPAI B3.13 outbreaks in EU wild birds, poultry or dairy cattle through introductions from US is also considered highly unlikely (annual probability from 10⁻¹⁴ to 10⁻⁴).

If outbreaks were to occur despite these very low probabilities, estimates suggest that ~244–923 k wild birds could be affected, along with ~240–713 poultry farms and ~26–81 dairy farms. In the absence of control measures, outbreaks in EU dairy cattle could expand substantially (~161‐17 k farms).

In case of such numbers of outbreaks, the overall impact, considering economic, societal and environmental dimensions, would be high for most Member States, especially in regions with extensive outdoor production systems and with vulnerable bird wildlife, such as regions in France and Spain. If no control measures were implemented in cattle, the overall impact would remain similar because the decreased impact for some of the categories (e.g. costs of control measures in dairy cattle) is balanced out with the increased impact of others (e.g. costs of culling infected poultry farms).

Conclusions on measures that could prevent the introduction and spread of HPAI B3.13 genotype virus in EU dairy cattle and poultry (ToR 2b.)

The experts conclude that cleaning and disinfection of milking equipment to remove or inactivate any virus contamination before use is the most effective measure to prevent an introduction into EU dairy herds via wild birds, while they consider avoiding the importation of cattle/poultry from infected countries (in accordance with EU legislation) the most effective measure to prevent an introduction into EU dairy herds or poultry flocks via trade.

It is concluded that milking hygiene (such as avoiding vacuum fluctuations and applying biocidal teat treatment before and after milking) is the most effective measure to prevent within‐herd spread in infected EU dairy cattle herds. Banning the movement of cattle in infected areas is considered to be the most effective measure to prevent spread from infected dairy farms to other dairy farms in the EU, while vaccination of poultry against HPAI, applying biosecurity measures before entering the farm (e.g. cleaning and disinfecting of vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors), avoiding exchange of workers, vehicles and equipment and vaccination of cattle against HPAI are considered to be the most effective measures to prevent spread from infected poultry farms to dairy cattle farms in the EU. Finally, the experts consider that vaccination of poultry against HPAI, banning the movement of cattle in infected areas and avoiding exchange of workers, vehicles and equipment are the most effective measures to prevent spread from infected dairy farms to poultry flocks in the EU.

The experts conclude that cleaning and disinfection of milking equipment to remove or inactivate any virus contamination before use, reporting dead wild birds in the vicinity of dairy farms for removal and testing by the authorities, avoiding importation of cattle/poultry and potentially contaminated animal products/germinal products from infected countries (in accordance with EU legislation), pre‐movement testing of outgoing animals from infected countries (in accordance with EU legislation), banning the movement of cattle in infected areas, removing, reporting and testing dead wild birds in the vicinity of cattle farms as well as avoiding non‐essential visits are measures with a high feasibility, i.e. these measures are already in use and can be implemented in practice with limited technical, economic or social difficulties.

It should be noted that the effectiveness and feasibility of the measures may vary depending on production systems. The assessments conducted by the EFSA experts identify similar introduction and spread pathways and a comparable set of risk‐mitigation measures that have recently been published as an OFFLU guidance (Carnegie et al., 2025).

Conclusion on adaptations of the current EU surveillance for HPAI needed to allow detection of an introduction of H5N1 B3.13 genotype virus into EU dairy cows populations (ToR 2c.)

Passive surveillance remains the essential first line of detection of HPAI viruses in dairy cattle, but detection is likely to be delayed due to clinical presentation being easily confused with mastitis caused by different aetiological agents. Raising awareness among farmers and veterinarians of HPAI viruses as a potential infectious agent of dairy cattle and causative agent of clinical and subclinical mastitis is critical to shorten time‐to‐detection.

Syndromic surveillance based on monitoring milk production, somatic cell counts (SCC) and antimicrobial use could provide early warning signals for HPAI infections in dairy cattle, although its sensitivity may be limited when only a small proportion of cattle are clinically affected.

Active surveillance based on molecular testing of bulk milk samples by RT‐qPCR has the greatest potential for timely detection of infected farms. Given the low probability of virus introduction into European dairy cattle, a regional, risk‐based application of active surveillance would be more efficient than broad EU‐wide active surveillance for HPAI virus infections in cattle. The existing EU infrastructure for bulk milk testing could be adapted but this requires scaling and coordination.

Milk is the most suitable sample matrix due to high viral loads excreted by infected animals. In non‐lactating animals, collection of nasopharyngeal swabs represents a suitable alternative. During post‐mortem investigations, the collection of the mammary tissue is recommended. RT‐qPCR targeting the M‐gene of AIVs is the preferred assay for screening and detection, followed by confirmatory subtyping and pathotyping assays, as rare infections by other influenza type A viruses have been reported in cattle. Whole‐genome sequencing is required to confirm the genotype.

Retrospective and herd‐level investigations can be performed by serological methods using commercial competition‐ELISA options on serum samples or by HI and VN assays.

Rapid sequencing of viruses identified in wild birds and poultry outbreaks is essential to timely detect introductions of H5N1 B3.13 genotype virus in avian populations in Europe.

Conclusions on the likelihood of bulk milk to be contaminated if EU lactating dairy cows are infected with the H5N1 B3.13 genotype virus (ToR 2d.)

If EU lactating cows are infected with H5N1 B3.13 genotype virus, it is very likely (80%–100%) that bulk milk from infected dairy farms would become contaminated before infection is detected. This is because most infected lactating cows are expected to shed virus in milk without or prior to showing clinical signs. Given frequent milking and diagnostic delays of 1–2 days, bulk milk contamination would almost certainly occur prior to detection of an infection with H5N1 B3.13 genotype virus at farm level.

Conclusions on the probability of infection/illness for a consumer in the EU population due to the exposure to viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products through the food‐borne route and measures to mitigate the risk (ToR 2e.)

AQ 1: What are the estimated levels of viable H5N1 B3.13 genotype virus in raw milk, colostrum, dairy products and colostrum‐based products produced in the EU at the end of processing, assuming contamination of bulk milk or bulk colostrum with this virus in the EU?

If bulk milk or colostrum is found to be contaminated with H5N1 B3.13 genotype virus in the EU, among various foodstuffs, the highest estimated levels of viable H5N1 B3.13 genotype virus will be found in raw drinking milk, raw milk cream and raw colostrum. The median level is estimated as 4.4 log₁₀ EID₅₀/mL or /g, with a 1.0–7.2 log EID₅₀/mL or /g 95% variability interval in different servings.

In production settings in which mixing of bulk milk is common, the levels of H5N1 B3.13 genotype virus would be lower depending on the proportion of milk from infected herds in the mixture.

As the H5N1 B3.13 genotype virus is heat sensitive, levels of viable H5N1 B3.13 genotype virus in pasteurised milk and colostrum are significantly reduced compared to raw foodstuffs. This is especially the case for LTLT pasteurisation (63°C – 30 min), whose products are free of viable H5N1 B3.13 genotype virus. HTST processes (72°C – 15 s) are less effective (4.9 median log₁₀ reduction) and HTST pasteurised milk and colostrum may still contain viable H5N1 B3.13 genotype virus, although at low levels (median concentration of −0.5 log₁₀ EID₅₀/mL), and time/temperature profiles applied in industrial processes (including heating, temperature holding and cooling) may provide higher reduction (of the amount) of viable H5N1 B3.13 genotype virus compared to the notional HTST treatment. The more intense HTST pasteurisation (82°C – 15 s) of cream renders pasteurised cream and derived products (e.g. pasteurised butter) free of viable H5N1 B3.13 genotype virus.

Thermisation of milk (57°C – 30 min) is also very effective, making products derived from this process free of viable H5N1 B3.13 genotype virus. The shorter thermisation process (65°C – 15 s) yields limited reduction (0.9 median log₁₀ reduction) but can be used as an additional hurdle in dairy processing, as typically thermisation is performed to extend the shelf‐life of milk prior to further processing or in the manufacture of some cheeses and thermised milk is not consumed as such.

The H5N1 B3.13 genotype virus is not completely inactivated through acidification of raw milk to the pH‐values (pH 4.6) expected in fermented milk or fresh cheeses derived from raw milk (3.5 median log₁₀ reduction). Acidification at higher pH values (pH 5.0), as in fermented cream from raw milk, is less effective at H5N1 B3.13 genotype virus inactivation (1.9 median log₁₀ reduction).

Apart from acidification, soft and semi‐soft raw milk cheeses also undergo ripening for up to 6 weeks and 3 months, respectively. This leads to a further reduction of viable H5N1 B3.13 genotype virus (1.0 and 1.9 median log₁₀ reduction for soft cheese and semi‐soft cheese, respectively), but infectious virus could still be present in the final product. Cheeses with more prolonged ripening periods (e.g. semi‐hard, hard, extra‐hard cheeses) are expected to contain lower titres of viable virus compared to soft and semi‐soft cheeses.

Products derived from pasteurised milk or colostrum (such as yoghurt, pasteurised fermented milk, pasteurised milk cheeses) are expected to contain lower viable H5N1 B3.13 genotype virus levels compared to pasteurised milk or colostrum due to the combined inactivation of thermal treatment, acidification, ripening and other processes.

As only key processing steps across different dairy product categories are considered, the levels of viable H5N1 B3.13 genotype virus will be lower for specific foodstuffs in which additional production steps (e.g. brining/dry salting of cheeses or heating of curd in hard cheese varieties) will yield additional reduction.

AQ 2: What is the mean probability of infection and illness per serving or, alternatively, what is the probability of exposure to different levels of viable H5N1 B3.13 genotype virus for consumers in the EU via the food‐borne route, given that the serving originates from milk or colostrum from a contaminated bulk tank in the EU?

To date, no confirmed cases of human infection with H5N1 B3.13 genotype virus through the food‐borne route have been reported.

Based on the comparison of available dose–response models (a human inhalation model based on experimental data on influenza viruses in human volunteers, a model based on the inhalation model modified to approximate an ingestion DR relationship, and an animal oral ingestion model developed in this assessment), it was considered not appropriate to select one DR model and exposure levels have been evaluated against two critical thresholds (10⁷ EID₅₀ and 10¹⁰ EID₅₀) that were used as a reference for classification and comparison across products.

The first threshold (10⁷ EID₅₀) was selected considering that, based on the human inhalation model and taking into account the impact of gastric digestion on H5N1 viability, the probability of human infection through the oral route is expected to be much lower than 1% at doses up to 7 log₁₀ EID₅₀. This threshold is supported by experimental data in non‐human primates (macaques) that displayed seroconversion but no clinical signs at an oral dose of 7.48 log EID₅₀. The second threshold (10¹⁰ EID₅₀) was instead selected considering that the inhalation model modified to approximate an ingestion DR relationship provides a probability of human infection through the oral route of 1% at doses of 10 log₁₀ EID₅₀.

If bulk milk or colostrum is found to be contaminated with H5N1 B3.13 genotype virus in the EU, the exposure to viable H5N1 B3.13 genotype virus expressed as the median number of servings per thousand of the foodstuffs, consumed as they are and considering the selected processing parameters, ²⁶ providing a dose exceeding 10⁷ EID₅₀, as estimated by the model, is ²⁷ :

• High (median ≥ 100) for raw drinking milk, raw colostrum, raw milk cream and milk thermised at 65°C for 15 s with the latter, however, not typically consumed as such;

• Intermediate (10 ≤ median < 100) for raw fermented cream;

• Low (1 ≤ median < 10) for raw fermented milk, raw milk fresh cheese, raw milk semi‐soft cheese and HTST pasteurised (72°C – 15 s) drinking milk and colostrum;

• Zero per 1000 servings for raw milk soft cheese, pasteurised cream (82°C – 15 s), milk thermised at 57°C for 30 min and LTLT pasteurised (63°C ‐ 30 min) drinking milk or colostrum;

• Not determinable for freeze‐dried colostrum due to the lack of evidence on the effect of this process on H5N1 B3.13 genotype virus viability;

• The exposure to viable H5N1 B3.13 genotype virus in products derived from thermised (65°C for 15 s) or pasteurised milk (e.g. fermented milk, fermented cream, different types of cheese) is expected to be lower than in the equivalent products derived from raw milk.

The products classified under the high exposure group using the threshold of 10⁷ EID₅₀ exceed at lower frequency (up to 10 median servings per 1000 servings) also the critical threshold of 10¹⁰ EID₅₀, based on the model estimates.

While the model integrates some uncertainties probabilistically, other uncertainty sources have been identified and their impact on the conclusions described. ²⁸

AQ 3: What are the available measures along the food chain (from initially contaminated bulk milk or bulk colostrum up to consumer), and, if feasible, their effectiveness, to reduce the levels of viable H5N1 B3.13 genotype virus?

Thermal treatment is highly effective at reducing the levels of infectious H5N1 B3.13 genotype virus in milk and milk products and colostrum and colostrum‐based products:

• Thermisation of milk for 30 min at 57°C and LTLT pasteurisation (63°C – 30 min) of milk or colostrum, and HTST pasteurisation of cream (82°C – 15 s) provide an estimated reduction of infectious virus by > 10 log₁₀;

• HTST pasteurisation of milk or colostrum (72°C – 15 s) is expected to reduce viral titres by 4.9 [1.7–20.2] log₁₀, which is expected to be further increased in the industrial processes with the heating and cooling down steps;

• Other combinations of time/temperature can achieve significant reductions and have been listed in the opinion. ²⁹

Various non‐thermal treatments, including high‐pressure processing, pulsed electric field technology, high‐pressure homogenisation, ultraviolet‐C irradiation, etc., have the potential to reduce the levels of viruses in milk or colostrum. These technologies are at a variable degree of implementation by the food/dairy industry. Furthermore, available data concern human and animal viruses other than H5N1 B3.13 genotype virus and, in several instances, data involving human viral pathogens were reported on breast milk (e.g. for milk donation) and not on bovine milk.

5. RECOMMENDATIONS

Potential impact of the infection of dairy cows in the EU with H5N1 B3.13 genotype virus (ToR 2a.)

Given that the probability of introduction of H5N1 B3.13 genotype virus into EU dairy cattle and poultry is assessed as highly unlikely across all NUTS2 regions, but that the potential impact of outbreaks could be high, it is recommended to strengthen vigilance through passive and syndromic surveillance and consider risk‐mitigation measures to prevent virus spread from wild birds to cattle, particularly in regions with higher estimated introduction rates and especially during autumn migration and winter.

In case of outbreaks, to reduce the potential impact in the EU it is recommended to apply control measures in both poultry and dairy cows, instead of in poultry only.

Possible measures to prevent the introduction of the virus into dairy cows and poultry in the EU as well as possible risk mitigating measures to prevent its spread in the EU (ToR 2b.)

To reduce the likelihood of introduction into EU dairy cattle via wild birds, it is recommended to prioritise cleaning and disinfection of milking equipment to remove or inactivate any virus contamination before use.

To reduce the likelihood of introduction into EU dairy cattle via trade, it is recommended to prioritise avoiding importation of cattle from infected countries.

To reduce the likelihood of introduction into EU poultry flocks via trade, it is recommended to prioritise avoiding importation of poultry from infected countries.

To reduce the likelihood of within‐farm spread in infected dairy cattle farms in the EU, it is recommended to prioritise milking hygiene (application of biocidal teat treatment (such as cleaning of udder and teats, pre‐ and post‐milking dip), avoiding vacuum fluctuations).

To reduce the likelihood of spread from infected dairy farms to other dairy farms in the EU, it is recommended to prioritise banning the movement of cattle in infected areas.

To reduce the likelihood of spread from infected poultry farms to dairy cattle farms in the EU, it is recommended to prioritise avoiding the exchange of workers, vehicles and equipment, and implementing biosecurity measures before entering the farm (e.g. cleaning and disinfecting of essential vehicles and equipment in a dedicated area outside the farm zone, use of an anteroom for essential workers/visitors). Vaccination of poultry, and of cattle if vaccines become available for this species, against HPAI could be considered as an additional protective measure, based on a risk assessment considering the local conditions.

To reduce the likelihood of spread from infected dairy farms to poultry flocks in the EU, it is recommended to prioritise applying movement restrictions of dairy cattle from infected areas, avoiding exchange of workers, vehicles and equipment between infected dairy farms and poultry farms, and vaccination of poultry against HPAI.

While the feasibility of increasing general biosecurity measures in European dairy farms was considered low in the current system due to outdoor access and structural and behavioural constraints linked to farmers/farm workers, campaigns to increase biosecurity on dairy farms are recommended.

Possible options for adaptations of the current EU surveillance for HPAI that would allow detection of an introduction into the EU dairy cows population (ToR 2c.)

Awareness raising

It is recommended to train and inform farmers, veterinarians and laboratories about HPAI as a differential diagnosis in cases of atypical mastitis (e.g. colostrum‐like milk, negative bacterial cultures, non‐response to antimicrobials), unexplained milk yield drops and elevated SCC. In farm cats residing on dairy premises and presenting with neurological signs, HPAI should be included among differential diagnoses, particularly where access to raw (unpasteurised) milk is possible. This HPAI awareness should be integrated in existing udder health and milk quality monitoring programmes.

Targeted Outbreak Investigations

When exposure to HPAIv is suspected (via infected poultry, wild birds or contaminated environments), or when clinical signs and laboratory results warrant further investigation, a structured outbreak investigation should be performed (see EFSA AHAW Panel, 2025). Sampling strategies should prioritise lactating cows, with testing of individual milk from clinically affected animals and of bulk milk tank samples. If clinically affected cows do not contribute to the bulk milk, then individual samples should be collected from all sick animals. In non‐lactating animals (e.g. dry cows, calves, bulls) nasopharyngeal swab collection is recommended. RT‐qPCR targeting M‐gene of AIvs is recommended as the primary test, complemented by sequencing for genotype confirmation. If serum samples collected for outbreak investigation or to complement molecular testing show positive results, these should be confirmed by a subtype‐specific assay (e.g. HI), as cattle can occasionally be infected with other Influenza Type A viruses.

Regional Active Surveillance

Once an infection with HPAI viruses has been confirmed in cattle, it is recommended to carry out active surveillance in the region surrounding the affected farm, to establish the extent of spread. Where feasible, testing of milk at dairy processing plants is recommended to complement on‐farm sampling. If H5N1 B3.13 genotype virus is detected in wild birds or poultry in a region where dairy farms are present, similar regional bulk milk testing should be considered.

Follow‐up Monitoring

It is recommended to maintain bulk milk testing by PCR in affected regions, initially at a high frequency (e.g. bi‐weekly), later at a frequency adapted based on epidemiological modelling. Serology should be used to assess herd‐level exposure or to uncover the extent of virus spread once virus circulation is under control.

Likelihood of bulk milk contamination if EU lactating cows are infected with the H5N1 B3.13 genotype virus (ToR 2d.)

To minimise the risk of bulk milk contamination, awareness should be raised among farmers, veterinarians and dairies about HPAI H5N1 in dairy cattle. In case of suspicions, sample submission should be expedited, and diagnostic turnaround times should be shortened followed by rapid subtyping/confirmation.

Food safety risk and mitigation measures (ToR 2e.)

The harmonisation of approaches (plaque/focus forming assays versus endpoint dilution assays, mathematical formulas applied to endpoint assays, etc.) for titration of infectious viruses in studies related to food safety should be promoted.

Studies addressing H5N1 B3.13 genotype virus survival in milk, colostrum and derived products following thermal treatment, acidification, freeze‐drying, high‐pressure processing and other processing technologies should be promoted.

In case of introduction of H5N1 B3.13 genotype virus or other HPAI viruses into EU dairy farms, systematic sampling and quantitative testing of bulk milk for infectious HPAIv should be considered.

ABBREVIATIONS

AIv

Avian influenza virus

ALP

alkaline phosphatase

AQ

assessment question

a_w

water activity

BHV

Bovine herpes virus

BVDV

bovine viral diarrhoea virus

CI

confidence interval

CMV

cytomegalovirus

CrI

credibility interval

DR

dose–response

dsRNA virus

Double Stranded RNA virus

D‐value

decimal reduction time or the time required at a given temperature to kill 90% of the exposed microorganisms

EID₅₀

egg infectious dose 50%

ESL

extended shelf‐life

EURL AI/ND

European Reference Laboratory for avian influenza and Newcastle disease

FAO

Food and Agriculture Organization

FCV

Feline calicivirus

HA

hemagglutination

HAV

hepatitis A virus

HCoV

Human Coronavirus

HEV

hepatitis E virus

HI

hemagglutination inhibition

HPAIv

highly pathogenic avian influenza virus

HPCD

high‐pressure carbon dioxide

HPH

high‐pressure homogenisation

HPP

high‐pressure processing

HTST

high temperature for a short time

IgG

immunoglobulin

IR

infra‐red

LOD

limit of detection

LTLT

low temperature for a long time

MDCK

Madin‐Darby canine kidney

MNV

Murine norovirus

MS

Member State

$P_C$

partition coefficient reflecting the virus's affinity for curd relative to whey

PEF

pulsed electric fields

PFU

Plaque Forming Units

PRV

pseudorabies virus

Ro

the basic reproductive number that indicates how contagious an infectious agent is

RT‐qPCR

reverse transcription quantitative PCR

SD

standard deviation

SQ

sub‐question

ssRNA virus

single stranded RNA virus

TBEV

tick‐borne encephalitis virus

TCID₅₀

tissue culture infectious dose 50%

UHT

ultra‐high temperature

UK FSA

UK Food Standards Agency

US FDA

U.S. Food and Drug Administration

UVC

Ultraviolet C irradiation

z‐value

the temperature increase required to reduce D by one log₁₀ unit (i.e. a 10‐fold reduction)

γ – irradiation

Gamma – irradiation

REQUESTOR

European Commission

QUESTION NUMBER

EFSA‐Q‐2024‐00712

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.

PANEL MEMBERS

Julio Alvarez, Anette Ella Boklund, Sabine Dippel, Fernanda Dórea, Jordi Figuerola, Mette S. Herskin, Virginie Michel, Miguel Ángel Miranda Chueca, Eleonora Nannoni, Søren Saxmose Nielsen, Romolo Nonno, Anja B. Riber, Jan Arend Stegeman, Karl Ståhl, Hans‐Hermann Thulke, Frank Tuyttens, and Christoph Winckler.

GENERIC MAP DISCLAIMER

The designations employed and the presentation of material on any maps included in this scientific output do not imply the expression of any opinion whatsoever on the part of the European Food Safety Authority concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries.

SPECIFIC MAP DISCLAIMER

Any designation of Kosovo is without prejudice to positions on status and is in line with United Nations Security Council Resolution 1244 and the International Court of Justice Opinion on the Kosovo Declaration of Independence.

Supporting information

Annex A: Incursion risk of highly pathogenic avian influenza H5N1 Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 for the European Union: A risk assessment based on the L'ORA tool

Annex B: Protocol to answer ToR 2e

APPENDIX A. Measures assessed regarding their effectiveness and feasibility regarding prevention of introduction and spread of the H5N1 B3.13 genotype virus

Scenario	Measures assessed
Scenario 1a	Cleaning and disinfection of milking equipment to remove or inactivate any virus contamination before use
Scenario 1a	Keep dairy cattle indoors in high‐risk areas and during high‐risk periods
Scenario 1a	Make the surroundings of the farm unattractive for wild birds (e.g. no pools of water, concrete surfaces on farm)
Scenario 1a	Management of feed and water for cattle within the premises to reduce contamination from wild birds
Scenario 1a	Prevent access of wild birds/mammals to the milking parlour and equipment
Scenario 1a	Remove, report and test dead wild birds in the vicinity of cattle farms
Scenario 1a	Vaccination of cattle against HPAI
Scenario 1b	Avoid importation of cattle from infected countries (in accordance with EU legislation)
Scenario 1b	Avoid importation of potentially contaminated animal products/germinal products from infected countries (in accordance with EU legislation)
Scenario 1b	Pre‐movement test of outgoing animals from infected countries (in accordance with EU legislation)
Scenario 1b	Quarantine for new animals from infected countries (in accordance with EU legislation)
Scenario 1b	Vaccination of cattle against HPAI
Scenario 2	Avoid importation of potentially contaminated animal products/germinal products from infected countries (in accordance with EU legislation)
Scenario 2	Avoid importation of poultry from infected countries (in accordance with EU legislation)
Scenario 2	Pre‐movement test of outgoing animals from infected countries (in accordance with EU legislation)
Scenario 2	Quarantine for new animals from infected countries (in accordance with EU legislation)
Scenario 3	Culling of infected animals
Scenario 3	Feeding of only milk replacement to calves
Scenario 3	Isolation of any sick animal
Scenario 3	Milking hygiene (biocidal teat treatment, avoid vacuum fluctuations)
Scenario 3	Proper disposal of carcasses and animal products (raw milk) in a dedicated area outside the farm zone
Scenario 3	Vaccination of cattle against HPAI
Scenario 4a	Avoid exchange of workers, vehicles and equipment
Scenario 4a	Avoid non‐essential visits
Scenario 4a	Biosecurity measures before entering the farm (e.g. cleaning and disinfecting of essential vehicles, equipment in dedicated area outside the farm zone, use of anteroom for essential workers/visitors)
Scenario 4a	Early detection and isolation of positive farms
Scenario 4a	Early detection and culling of infected cattle farms
Scenario 4a	Indoor confinement/limit of outdoor access in infected areas for cattle
Scenario 4a	Management of feed and water in cattle farms to reduce contamination from wild birds/mammals
Scenario 4a	Movement ban of cattle in infected areas
Scenario 4a	Movement ban of untreated milk (treatment to inactivate virus not applied)
Scenario 4a	Movement restriction of vehicles from infected areas
Scenario 4a	No slurry/manure spreading from infected farms
Scenario 4a	Pre‐movement test of outgoing cattle from infected areas
Scenario 4a	Proper disposal of carcasses and animal products (waste raw milk)
Scenario 4a	Quarantine for new cattle entering the farm
Scenario 4a	Remove, report and test dead wild birds in the vicinity of cattle farms
Scenario 4a	Vaccination of cattle against HPAI
Scenario 4b	Avoid exchange of workers, vehicles and equipment
Scenario 4b	Avoid non‐essential visits
Scenario 4b	Biosecurity measures before entering the farm (e.g. cleaning and disinfecting of essential vehicles, equipment in dedicated area outside the farm zone, use of anteroom for essential workers/visitors)
Scenario 4b	Indoor confinement of cattle/limit of outdoor access during high‐risk periods/areas
Scenario 4b	Management of feed and water to reduce contamination from wild birds/mammals
Scenario 4b	Movement restriction of animals (cattle) from infected areas
Scenario 4b	Movement restriction of vehicles from infected areas
Scenario 4b	Pre‐movement test of outgoing animals (cattle) from infected areas
Scenario 4b	Proper disposal of carcasses and animal products (waste raw milk)
Scenario 4b	Quarantine for new animals (cattle)
Scenario 4b	Remove, report and test dead wild birds in the vicinity of cattle farms
Scenario 4b	Vaccination of cattle against HPAI
Scenario 4b	Vaccination of poultry against HPAI
Scenario 5	Avoid exchange of workers, vehicles and equipment in infected cattle farms
Scenario 5	Avoid non‐essential visits to infected cattle farms
Scenario 5	Biosecurity measures before entering cattle farms (e.g. cleaning and disinfecting of essential vehicles, equipment in dedicated area outside the farm zone, use of anteroom for essential workers/visitors)
Scenario 5	Early detection and isolation of positive farms
Scenario 5	Early detection and culling of infected cattle farms
Scenario 5	Indoor confinement/limit of outdoor access during high‐risk periods/areas of poultry
Scenario 5	Movement ban of untreated milk (treatment to inactivate virus not applied)
Scenario 5	Movement restriction of cattle from infected areas
Scenario 5	Movement restriction of vehicles from infected areas
Scenario 5	No slurry/manure spreading from infected cattle farms
Scenario 5	Pre‐movement test of outgoing cattle from infected areas
Scenario 5	Proper disposal of carcasses and animal products (waste raw milk)
Scenario 5	Remove, report and test dead wild birds in the vicinity of cattle farms
Scenario 5	Vaccination of cattle against HPAI
Scenario 5	Vaccination of poultry against HPAI

APPENDIX B. Correlation of virus titration units

To correlate TCID₅₀ to EID₅₀ titration values, data from three papers that performed both tests using different H5N1 isolates were considered (Table B.1). The dataset displayed good correlation between the two titration techniques on H5N1 isolates (correlation coefficient r ² = 0.836), supporting the possible conversion of the data.

TABLE B.1.

TCID₅₀ and EID₅₀ titration values derived from studies performing both tests using different H5N1 isolates.

Reference	Strain	Additional info	TCID50 (log10 unit)	EID50 (log10 unit)
Dovas et al. (2010)	A/mute swan/Greece/436/06	HPAIv Virus in water; titration in CEFs	3.81	5.0
Bordes et al. (2024)	H5N1–2020	H5N1 clade 2.3.4.4b, EU lineage Viral stock; titration in MDCK	4.82	6.0
	H5N1–2021 PB2‐627K	H5N1 clade 2.3.4.4b, EU lineage Viral stock; titration in MDCK	4.85	6.0
	H5N1–2021 PB2‐627E	H5N1 clade 2.3.4.4b, EU lineage Viral stock; titration in MDCK	5.61	6.0
Nooruzzaman, Covaleda, et al. (2025)	HPAI H5N1 TX2/24	H5N1 clade 2.3.4.4b, genotype B3.13 Virus in milk; titration in Cal.1 cells	4.72	6.11
			4.80	5.30
			5.22	6.11
			5.14	4.97
			4.97	4.92
			4.72	4.72
			3.88	3.28
			3.38	2.67
			2.30	2.67
			4.55	4.56
			4.88	4.50
			4.55	4.56
			3.30	3.50
			4.05	3.50
			3.72	3.50

Abbreviations: CEFs, chicken embryo fibroblasts; MDCK, Madin–Darby Canine Kidney cells.

For the conversion of TCID₅₀ into EID₅₀ data, only the data from Nooruzzaman, Covaleda, et al. (2025) was used as providing the closest testing conditions to the assessment (H5N1 B3.13 genotype virus, milk matrix) (see Figure B.1). The following equation was derived (r² = 0.748):

$${\log }_{10}\left({\mathrm{EID}}_{50}\right)=1.159\times {\log }_{10}\left({\text{TCID}}_{50}\right)-0.636.$$

FIGURE B.1.

Regression line used to derive log EID₅₀/mL values from log TCID₅₀/mL, based on data from Nooruzzaman, Covaleda, et al. (2025) with 95% prediction interval.

To correlate plaque forming unit (PFU) to the other units, the following formula was considered, according to the ATCC Virology Culture Guide ³⁰ :

$$\mathrm{PFU}=0.7\times {\text{TCID}}_{50}$$

APPENDIX C. Titres of infectious H5N1 clade 2.3.4.4b virus in milk samples from infected cows

TABLE C.1.

Titres of infectious H5N1 clade 2.3.4.4b virus in individual and bulk milk samples from naturally or experimentally infected cows.

Product	Virus clade/genotype	Country Year	No of tested samples	Infectivity testing method	Estimated titres	Remarks	Ref.
Bulk milk	H5N1 2.3.4.4b	USA (multiple states) a 18.04.2024 to 27.04.2024	157 (39 positive)	Embryonating chicken eggs (3 replicates per dilution; inoculum through the chorioallantoic sac route; detection through HA assay). LOD: not reported.	Median titre in positive samples: 3.5 log10 EID50/mL Range titre in positive samples: 1.3–6.3 log10 EID50/mL	Preliminary screening by RT‐qPCR (158/275 samples); infectivity assay performed on PCR‐positive samples	Spackman, Jones, et al. (2024)
Milk from naturally infected cows	H5N1 2.3.4.4b	Texas, Ohio (USA) b 11.02.2024 to 19.03.2024	69 (42 positive)	Cal‐1 cells TCID50/mL; immunofluorescence (anti‐NP Ab), titre calculated by the Spearman and Karber's method. LOD: 101.05 TCID50/mL	Range titre in positive samples: 4.05 to 8.8 log10 TCID50/mL	Preliminary screening by RT‐qPCR (129/192 samples); infectivity assay performed on a selection of PCR‐positive samples	Caserta et al. (2024)
Milk from naturally infected cows	H5N1 2.3.4.4b	New Mexico (USA) March 2024	4 (TCID50 method)	MDCK cells; TCID50 titre calculated by the Reed and Muench method. LOD: 1.5 log10 TCID50/mL and 10 PFU/mL	KS#6 (5 titrations): 5.33; 5.33; 4.67; 4.67; 5 log10 TCID50/mL KS#3 (5 titrations): 6; 6; 5.33; 5.33; 5.67 log10 TCID50/mL NM#93 (2 titrations): 7.13; 7.33 log10 TCID50/mL NM#115 (2 titrations): 5.4; 5.33 log10 TCID50/mL		Guan et al. (2024)
			8 (PFU method)		#90: 5 × 102#93: 6 × 107 PFU/mL #96: <10 PFU/mL #115: 2 × 105 PFU/mL #99, #105, #108, #119: 0 PFU/mL
Milk from experimentally infected cows c	H5N1 clade 2.3.4.4b, B3.13	–	9	MDCK‐II cells TCID50/mL; titre calculated by the Specific INfection (SIN) method. LOD: not reported.	#87: 2.75 × 106 (d1); 1.62 × 109 (d2); 4.68 × 108 (d3) TCID50/mL #92: 3.55 × 105 (d1); 1.45 × 109 (d2); 5.13 × 106 (d3) TCID50/mL #47: 6.31 × 104 (d1); 4.17 × 109 (d2); 1.23 × 108 (d3) TCID50/mL	Milk positive from 1 dpi and peak at 3 dpi. RNA detectable 20 dpi. Antibodies from 9 dpi.	Halwe et al. (2025)
	euDG H5N1 clade 2.3.4.4b				#66: 1.66 × 103 (d1); 6.92 × 106 (d2); 2.24 × 107 (d3) TCID50/mL #88: 2.29 × 103 (d1); 4.68 × 108 (d2); 6.17 × 108 (d3) TCID50/mL #72: 5.75 × 103 (d1); 4.68 × 107 (d2); 1.78 × 106 (d3) TCID50/mL

Abbreviations: dpi, days post infection; EID₅₀, 50% Egg Infectious Dose; HA, hemagglutination; LOD, limit of detection; MDCK, Madin–Darby canine kidney; PFU, plaque forming units; RT‐qPCR, reverse transcription quantitative PCR; SIN, specific infection; TCID₅₀, 50% Tissue Culture Infectious Dose.

^a Samples from farms known to be affected and from farms not known to be affected in regions known to be affected.

^b Samples from three affected farms. Summary of the clinical investigation on the farms: (i) TX1 (4000 cows, 20% of clinically affected cows during the outbreak, sampling date: 09–13.03.24); (ii) TX2 (35,000 cows, 3.42% of clinically affected cows during the outbreak, sampling date: 11.2.24–13.3.24); (iii) OH1 (3350 cows, 23.4% of clinically affected cows during the outbreak, sampling date: 21–29.3.24). Clinically affected cows displayed the following symptoms: decreased feed intake, decreased rumination time, mild respiratory signs, lethargy, milk with abnormal yellowish colostrum‐like colour, thick and sometimes curdled consistency. In addition, drop in milk production ranging from 20% to 100%. Clinical disease length: 5–14 days.

^c Intramammary inoculation of 3 lactating cows for each strain. Inoculum: 0.5 mL per teat of 10^5.9 TCID₅₀/mL (US strain) or 10^6.1 (euDG). Impaired general conditions, postural abnormalities and lethargy as early as 1 dpi. Fever and reduced feed intake, reduced milk production, change in milk colour/consistency from 2 dpi.

APPENDIX D. Exposure assessment model

D.1. CONCENTRATION OF VIABLE H5N1 B3.13 GENOTYPE VIRUS IN BULK MILK TANKS

As in Koebel et al. (2024), a modelling approach was used to estimate the concentration of viable H5N1 B3.13 genotype virus in bulk milk. This approach allows to estimate the concentrations in typical EU bulk tanks and to explore the impact of farm specific parameters (e.g. proportion of animals in the herd contributing contaminated milk to the bulk tank, herd size) on the levels of viable H5N1 B3.13 genotype virus in bulk milk (and bulk colostrum). The bulk tank was considered to be a tank in which the milk from one herd is collected. Table D.1 presents the parameters used in the bulk tank model and the approach taken.

TABLE D.1.

Parameters used in the bulk tank model to explore the impact of farm specific parameters on the levels of viable H5N1 B3.13 genotype virus in bulk milk.

Parameter	Symbol	Values/equation	Variability	Reference
Proportion of animals in the herd contributing contaminated milk to the bulk tank	pherd	User defined Baseline model: 0.01		Caserta et al. (2024), Koebel et al. (2024)
Herd size	Hall	User defined Baseline model: 65		Eurostat data
Number of infected cows in herd	Hinf	Binomial(Hall, pherd), Zero‐truncated	Per herd or bulk tank
Healthy cow milk yield (L/day)	YH,i	Pert(30.3, 34.1, 37.9)	Cows i = 1..Hall	USDA‐NASS (2024), Koebel et al. (2024)
Relative yield infected cow	ry,i	Pert(0.5,0.6,1)	Cows i = 1..Hinf	Durst (2024), Koebel et al. (2024)
Infected cow milk yield (L/day)	YI,i	YH,i × ry,i	Cows i = 1..Hinf
Viral titre infected cows	VTi		Cows i = 1..Hinf	Caserta et al. (2024), Nooruzzaman, Covaleda, et al. (2025)
(log10 TCID50) per mL		Truncated Normal(5.71, 1.47, max = 9)
(log10 EID50) per mL		Truncated Normal(5.98, 1.70, max = 10)
Viral load in milk (TCID50) per cow	VLCi	YI,i × 1000 × 10^VTi	Cows i = 1..Hinf
Milk volume in bulk tank (L)	Vbulk	Σ i = 1..Hinf YH,i + Σ i = Hinf + 1..H YI,i	Per herd or bulk tank
Viral load in bulk tank (log10 TCID50) per mL	VLB	log10(Σ i = 1..Hinf VLCi/Vbulk) – 3	Per herd or bulk tank

For the purposes of the model, the proportion of animals in the herd contributing H5N1 B3.13 genotype virus‐contaminated milk to the bulk tank (within‐herd prevalence, p_herd), is interpreted as the probability per cow in the herd to be infected at the moment of lactation when contributing milk to the bulk tank and modelled as the p‐value in a Binomial distribution. To assure that only contaminated bulk tanks are analysed (as requested in ToR2e), the zero‐truncated version of this Binomial distribution is used. Note that the mean of this zero‐truncated distribution is larger than the mean of the Binomial distribution that it is derived from.

For this value it was considered that not all cows are infected simultaneously, and that clinically ill cows are segregated for other milk‐producing animals (i.e. do not contribute their milk to the tank), while infected cows contribute their milk to the bulk tank while they are asymptomatic. According to data from a study conducted in 2500‐cow farm by the National Milk Producers Federation, ³¹ in the period between first detection of H5N1 virus in bulk milk and the detection of the herd as positive (approximately 20 days) a slight increase of the reported daily illnesses from the baseline daily average of 4.18 to 4.47. This would lead, in the aforementioned period, to an estimated cumulative number of cows showing clinical signs of illness due to H5N1 ‐ and shedding virus in milk – in the range of 5 to 37, depending on the proportion of daily new cases attributed to H5N1 (i.e. a fixed proportion over the period or a progressively increasing proportion up to 100%, as in an expanding disease transmission). This would give, in the period before reporting of the outbreak, an average within‐herd prevalence of H5N1 clinically ill cows of around 0.001–0.005 (with a maximum, at the time of detection of the outbreak, of 0.002–0.015). However, considering that according to literature only 20% of infected cows show clinical signs (Peña‐Mosca et al., 2025), this would imply a more realistic within‐herd average infection prevalence of approximately 0.006–0.023 (with a maximum, at the time of detection of the outbreak, of 0.01–0.074). Based on discussion in the WG, considering the evidence, a within‐herd prevalence of 0.01 was assumed as a baseline scenario.

Herd size estimates in Europe can be obtained from Eurostat. It provides statistics for 2013 for the number dairy farms, the number of dairy cows per country and some categorical data on herd sizes. When selecting a representative herd size for the EU, the relevant mean would be the mean herd size for a random portion of milk (or the mean herd size for a random cow), which is not the same as the mean herd size, as larger herds contribute more milk to the total milk production. This is known as accounting for a size‐biased distribution, and the appropriate (size.biased) mean is:

$$E\left[X^{*}\right]=E\left[X^2\right]/E\left[X\right]$$

With $X$ = herd size, $E\left[X\right]$ is the sum of the herd sizes and $E\left[X^2\right]$ is the sum of the square of the herd sizes.

To calculate this, the variance in the herd sizes per EU MS is required, which is not given in the Eurostat data. We can however use the equation using the data form Eurostat shown in Figure D.1, which gives $E\left[X^{*}\right]$ = 83. Assuming that about 80% of the cows in a herd is lactating, this implies a mean H_all = 65.

FIGURE D.1.

Number of dairy cows per farm in the EU according to the Eurostat data 2013. Blue bars show the percentage of farms with the indicated herd sizes; orange bars show the percentage of cows of the total number in the EU in these farm categories.

The Eurostat data show a considerable variation in herd sizes between EU MS (Figures D.2 and D.3).

FIGURE D.2.

Mean farm sizes in EU MS and size‐biased means, based on Eurostat data 2013.

FIGURE D.3.

Herd sizes in EU MS as the relative numbers of farms and dairy cows in different EU MS.

The viral concentrations in milk from individual infected cows shedding the virus in the milk, is obtained from Caserta et al. (2024). These authors report concentrations from pooled samples and individual animals. Results for these single cow samples (n = 39), are shown in Figure D.4. To account for the small dataset, a normal distribution was fitted through the observed concentrations (log₁₀ TCID₅₀/mL) to obtain distributions for the variability in concentrations in milk shed by individual infected animals. For this fit it was assumed that the data reported as 4.05 log₁₀ TCID₅₀/mL (9 of 39 data points) actually lie in the range between the reported Limit of Detection (LOD; 1.05) and 4.05 log₁₀ TCID₅₀/mL. Additionally, the distribution was truncated at 9 log₁₀ TCID₅₀ per mL, which the WG experts considered to be a plausible biological maximum level for naturally infected cows. This threshold was selected taking into account: (i) the highest viral shedding reported both in Caserta et al. (2024) and in other studies on naturally (Guan et al., 2024) and infected cows (8.8 log₁₀ TCID₅₀/mL and 7.3 log₁₀ TCID₅₀/mL, respectively); (ii) data on experimentally infected cows (9.2–9.6 log₁₀ TCID₅₀/mL at the peak of excretion, Halwe et al., 2025), considered as an extreme for shedding; (iii) the conclusions of the UK FSA report (2025) on the maximum concentration of infectious H5N1 B3.13 genotype virus (8.5 log₁₀ infectious particles/mL) that, as a worst‐case scenario, can be expected based on the lowest PCR values obtained in milk testing in the study by Burrough et al. (2024). The fitted distribution has a mean of 5.71 log₁₀ TCID₅₀/mL and standard deviation (SD) of 1.47 log₁₀ TCID₅₀/mL. Transformed to EID₅₀ per mL, using the regression line obtained in Appendix D, this is similar as a distribution with mean 5.98, SD 1.70 log₁₀ EID₅₀/mL, truncated at 10 log₁₀ EID₅₀/mL.

FIGURE D.4.

Concentrations of H5N1 2.3.4.4b virus in the milk from individual animals (n = 39) as derived from Caserta et al. (2024).

The viral load in the bulk tanks is transformed from TCID₅₀ to EID₅₀ by using the regression line and prediction interval described in Appendix B. Using this approach, the variability of concentrations of viral load between typical bulk tanks in the EU is estimated to have a mean 4.26 and standard deviation 1.64 log₁₀ EID₅₀ per mL (p = 0.01, H_all = 65), or mean 5.29 and SD 1.36 log₁₀ EID₅₀ per mL (p = 0.05, H_all = 65) as shown in Figure D.5.

FIGURE D.5.

Variability between concentrations in bulk tanks obtained from the model when p = 0.01 and H = 65 (left) and 0.05 and H = 65 (right).

To assess the sensitivity of the bulk tank model assumptions, the impact of the assumption of prevalence (p_herd, the probability per cow to be infected and contribute milk to the bulk tank = 0.01) and the number of cows contributing milk (herd size, H_all = 65) is explored further.

By using a Binomial model, we assume that the number of infected cows contributing milk to the bulk tank, H_inf, varies. For the scenario used, the variation in H_inf is illustrated in Figure D.6. It shows that with H_all = 65, only one infected cow is contributing milk in 71% of the bulk tanks. H_inf values increase with increasing herd size and increasing within‐herd prevalence p_herd.

FIGURE D.6.

Variation in the number of cows shedding contaminated milk in a contaminated bulk tank in scenarios with a within‐herd prevalence of 0.01 or 0.05 and herd sizes 65 or 500.

Note that the actual prevalence in the herd may be larger than p_herd because the binomial distribution implies a probability of outcome 0, no infected cows. Uncontaminated bulk tanks are discounted by using a zero‐truncated binomial. The effect of this phenomenon is especially large when H_all and p_herd are small. Some examples are given in Table D.2.

TABLE D.2.

The mean number of cows contributing contaminated milk to the bulk tank and the herd prevalence in contaminated bulk tanks when using a zero‐truncated binomial distribution.

H_all	pherd	Mean number of infected cows per herd	Herd prevalence for contaminated bulk tanks
65	0.01	1.36	0.021
65	0.05	3.38	0.052
65	0.1	6.51	0.100
500	0.01	5.03	0.010
500	0.05	25	0.05
500	0.1	50	0.1

The amount of viable virus shed into the bulk tank increases with H_inf. This increase can be described by summing samples from a lognormal distribution, which is mathematically complex (and analytically impossible). With Monte Carlo simulation, however, it is possible to obtain the expected median estimate of the concentration in bulk tanks of different size as a function of the prevalence, or alternatively the number of infected cows contributing milk. Assuming no difference in yield, this looks as illustrated in Figure D.7.

FIGURE D.7.

Expected median estimate of the concentration in bulk tanks of different size as a function of the number of infected cows contributing milk, or alternatively of the proportion of animals contributing contaminated milk (p_herd).

It shows that with an increasing number of infected cows, the concentration in the bulk tank increases and with a larger herd size, dilution gives a decrease in concentration.

D.2. EXPOSURE ASSESSMENT

The exposure assessment uses the bulk tank concentrations as starting point. Milk from the bulk tank is processed into one of the dairy products. Mixing with milk from other bulk tanks is ignored. The processing results in inactivation, as described in Section 3.6.1, which reduces the concentration. Variation in the inactivation is interpreted as variability and combined with the variability in bulk tank concentrations.

Next, one serving of dairy product is sampled, using this is a Poisson process with as mean value the concentration (in EID₅₀ per mL) multiplied by the serving size (see Table 17). The obtained EID50 per serving is the ingested dose, i.e. the outcome of the exposure assessment.

The model considers variability and uncertainty separately. The uncertainty in the mean and standard deviation of the variability in viral shedding between individual cows is obtained by bootstrapping (R package fitdistrplus). The prediction interval of regression line is used to describe the uncertainty in the transformation of TCID₅₀ into EID₅₀.

D.3. SENSITIVITY ANALYSIS

The sensitivity of the exposure assessment results for the assumptions on the bulk tank concentration distribution is explored by scenario analysis. First, for raw drinking milk, the frequency of exposure to doses above 0, 10⁷ and 10¹⁰ EID₅₀ is modelled for the scenarios shown in Figure 3 (Section 3.6.1), see Table D.3. Next, the analysis is done for the critical level 10⁷ EID₅₀ for all foodstuffs for a scenario where p_herd = 0.05 instead of 0.01 (Table D.4).

TABLE D.3.

Exposure assessment in raw milk for different scenarios, ranges indicate uncertainty obtained from modelling the uncertainty in the data on viral shedding of individual cows by Caserta et al. (2024), uncertainty in the translation from TCID₅₀ to EID₅₀, and uncertainty in the partitioning from sampling one serving per bulk tank.

Raw (drinking) milk scenario	Exposure threshold
	0 EID50	107 EID50	1010 EID50
Pherd = 0.01, Hall = 65	999, (997–1000)	339, (253–434)	9, (4–17)
Pherd = 0.05, Hall = 65	1000, (999–1000)	612, (499–709)	25, (11–43)
Pherd = 0.1, Hall = 65	1000, (1000–1000)	828, (728–897)	51, (23–87)
Pherd = 0.01, Hall = 500	1000, (999–1000)	446, (312–571)	2, (1–5)
Pherd = 0.05, Hall = 500	1000, (1000–1000)	922, (825–965)	19, (7–37)
Pherd = 0.1, Hall = 500	1000, (1000–1000)	979, (945–990)	49, (17–98)
Data Spackman	1000, (1000–1000)	150, (150–150)	0, (0–0)
Fit Spackman	1000, (1000–1000)	197, (120–280)	1, (0–6)

The outcome of the sensitivity analysis shows that exposure increases with increasing p_herd. For exposure above 10⁷ EID₅₀ it also increases with herd size, but for 10¹⁰ EID₅₀ it does not, probably due to decreased variability. When the Spackman, Jones, et al. (2024) data are used instead of the modelling approach, exposure estimates are lower. Still, in all scenarios raw drinking milk exposure should be characterised as ‘High’.

The increased exposure with increased p_herd is observed for all foodstuffs. The relative differences between the foodstuffs remain similar, and the characterisation of the exposure of the different foodstuffs also remains highly similar.

TABLE D.4.

Number of servings per thousand exceeding the dose 10⁷ EID₅₀ of infectious H5N1 virus, assuming a scenario where p_herd = 0.05 instead of 0.01. Numbers given represent median, (5%–95% uncertainty interval).

Categories of foodstuff	T profile	Exposure threshold: 107 EID50
Raw (drinking) milk		6611, (498–708)
Thermised milk a	NT57	0, (0–0)
	NT65	3343, (235–452)
	IP 65	251, (158–344)
Pasteurised (drinking) milk	NP63	0, (0–0)
	NP72	21, (10–32)
	IP72	3, (1–6)
Raw fermented milk		8, (3–15)
Raw milk fresh cheese		3, (1–6)
Raw milk soft cheese		0, (0–1)
Raw milk semi‐soft cheese		2, (1–4)
Raw milk cream		392, (275–504)
Raw fermented cream		76, (40–121)
Pasteurised cream		0, (0–0)
Raw colostrum		556, (436–658)
Pasteurised colostrum	NP63	0, (0–0)
	NP72	15, (7–25)
	IP72	2, (1–4)

Abbreviations: IP, industrial profile, the adjacent number indicates the temperature; NT, notional treatment.

^a Thermised milk is not consumed as such. A drinking portion size is assumed in the assessment for comparison purposes.

APPENDIX E. Secondary Bigelow‐type model and its validation

The refined dataset from Hessling (2022) is presented in Figure E.1A. Figure E.1B shows the predicted log₁₀(D) values as a function of temperature based on the fitted Bigelow model, along with the 95% CI (grey/blue area) reflecting parameter uncertainty and the 95% prediction band (pink/red area) accounting for residual variability around the model.

FIGURE E.1.

Model fitting for influenza virus thermal reduction time (D) in log₁₀ as a function of temperature. (A) Final dataset used for model fitting, after exclusion of extreme temperatures and unsuitable matrices (dried eggs, surfaces, manure) and (B) model fit with 95% confidence and prediction bands derived from bootstrap estimates.

The validation of the developed secondary Bigelow‐type model using experimental data reporting HPAI H5N1 inactivation in milk products is presented in Figure E.2. This validation focused on studies providing either point estimates or right‐censored log₁₀ reductions, without explicit estimation of D‐values. At 63°C, the model predicts a strong reduction of viral infectivity over time, with a median log₁₀ reduction of approximately 6 in less than 5 min. These predictions are consistent with the available experimental evidence: studies by Alkie et al. (2025), Caceres et al. (2024), Cui et al. (2024), Schafers et al. (2025) and Guan et al. (2024) all reported concentrations of infectious virus below the limit of detection after 30 minutes at 63°C. In addition, Nasheri et al. (2025) reported a point estimate of 5.41 log₁₀ reduction in just 15 s at 63 C. This observation is close to the model prediction, albeit slightly higher, suggesting a greater inactivation than the median estimate from the model. This same study also reported similar levels of viral inactivation at 60°C and 66°C, with reductions of 4.70 and 6.45 log₁₀ respectively, further supporting the consistency of the model across this relevant temperature range. At 72°C, the model predicts a median log₁₀ reduction of approximately 5 within 15 s of treatment. This prediction aligns well with the experimental results of Alkie et al. (2025), Caceres et al. (2024), Guan et al. (2024), Schafers et al. (2025) and Kaiser et al. (2024), who reported reduction of the infectivity close to 5 log₁₀, based on either direct measurements or lower‐bound estimates.

FIGURE E.2.

Observed and predicted inactivation (median and 95% CI) of HPAI H5N1 in milk at 63°C and 72°C.

APPENDIX F. Dose–response models

Hazard characterisation in microbiological risk assessment aims to assess the probability of adverse effects (infection, illness…) of a pathogen based on the dose to which individuals are exposed.

F.1. AVAILABLE MODELLING APPROACHES

Interagency Risk Assessment for the Public Health Impact of Highly Pathogenic Avian Influenza Virus in Poultry, Shell Eggs and Egg Products (Bauer et al., 2010)

FSIS in collaboration with FDA and APHIS published in 2010 a risk assessment aiming to evaluate the public health impact of HPAIv in poultry, shell eggs and egg products. The context of the assessment was growing concern over H5 and H7 HPAI strains, which had caused significant outbreaks in poultry and sporadic human cases worldwide at that period. The primary objectives were to estimate human exposure and illness risks from consuming contaminated poultry and eggs and to assess the effectiveness of mitigation strategies to reduce public health risks. Dealing with hazard characterisation, the goal was to estimate the dose–response (DR) relationship. As no published studies provided a human DR relationship for oral exposure to HPAI, alternative data sources were explored. Five types of datasets were identified: epidemiological studies, intranasal animal studies, intranasal human studies, human vaccine trials and animal ingestion/gavage studies. The most appropriate data considered came from a study by Beare and Webster (1991) on intranasal exposure of humans to avian influenza viruses. This choice eliminates uncertainties related to extrapolating from animal models to humans. However, extrapolating from intranasal exposure to oral exposure introduces significant, unquantifiable uncertainty.

The DR modelling relied on a classic exponential function:

$$p=1-e^{-\mathit{rD}}$$

where r represents the virulence parameter, and D is the administered dose. This model was chosen for its biological plausibility, its simplicity (having a single parameter measuring virulence) and its previous application in risk assessments, particularly in estimating human illnesses from HPAIv‐contaminated water ingestion.

The model parameters were estimated using Maximum Likelihood Estimation (MLE). Several datasets were tested, revealing significant variability in r‐values, reflecting differences in virulence among HPAIv strains. The authors concluded that this extrapolation carries substantial uncertainty, and the DR relationship for oral exposure remains hypothetical. However, their analysis suggests that the risk of infection from ingesting contaminated poultry products is likely lower than from respiratory exposure.

Koebel et al. ( 2024 )

The study by Koebel et al. (2024) aims to assess the risk of H5N1 infection through the consumption of contaminated milk. The research was prompted by recent detections of H5N1 in dairy products and the uncertainty regarding the virus's potential for oral transmission. The main objective is to quantify infection risk based on viral dose in milk while accounting for various factors related to purchasing and consumption of raw or pasteurised milk. The DR relationship was modelled using the exponential DR model.

r parameter was estimated using DR data from a ferret ingestion model (Lipatov et al., 2009), calibrated based on the median infectious dose (ID₅₀). The model was then adapted to reflect the specific conditions of this risk assessment. To account for uncertainty and variability, the study conducted a scenario analysis using different 𝑟 values:

• $r={10}^{-8}$ for baseline, considered the most representative value, derived from ferret ingestion data;

• $r={10}^{-6}$ for a high infectivity scenario; and

• $r={10}^{-10}$ for a lower infectivity scenario.

FDA Study (Chen et al., 2025 )

The study by Chen et al. (2025) confirms the use of the exponential DR model for H5N1 risk assessment in raw and pasteurised milk. It employs two complementary approaches: a ‘bottom‐up’ quantitative risk assessment and a ‘top‐down’ epidemiological analysis.

Bottom‐up approach. For inhalation exposure, the study used an available human DR model for H5N1 for the intranasal route of exposure, which is an exponential model with:

$r=1.35\times {10}^{-9}$ (uncertainty range: 2.42 × 10^{− 1}⁰ to 1.19 × 10⁻⁸)

This model predicts that an intranasal dose of 8.7 log₁₀ EID₅₀ is needed to infect 50% of individuals.

To modify the inhalation DR model to approximate an ingestion DR relationship more closely, the study considered the impact of gastric digestion on H5N1 inactivation. Based on temporal pH dynamics after milk consumption, milk gastric emptying observations and influenza inactivation at differing pH levels, they assumed that at least a 3‐log₁₀ reduction of viable virus occurs during gastric passage. This assumption was applied by reducing the r parameter of the inhalation exponential DR model to:

$r=1.35\times {10}^{-12}$ (uncertainty range: 2.42 × 10^{− 13} to 1.19 × 10^{− 11})

Top‐down approach. The top‐down approach works backwards from surveillance data to estimate detectable risk levels, asking ‘what's the minimum risk that would result in at least one detected case?’. Using a probabilistic framework, the study found that current surveillance systems would only detect 1.29% of H5N1 cases in urgent care settings and 8.79% in hospital settings, assuming severity similar to seasonal influenza. The calculated minimum detectable risk thresholds that would lead to 95% probability of observing at least one case was 0.051 cases per million servings after 30 days for pasteurised milk, decreasing to 0.017 after 90 days. For raw milk, thresholds were higher at 4.25 cases per million after 30 days and 1.42 after 90 days. These estimates decrease over time with no detected cases, based on the logic that if risk were truly high, cases should have appeared by now.

The top‐down approach serves as a check for bottom‐up estimates. While the bottom‐up model predicted 2.1 cases per million for raw milk consumption (suggesting one detectable case after ~2 months), no milk‐related H5N1 cases were detected after 5 months of exposure. This discrepancy suggests either raw milk herds were better protected than assumed, the ingestion DR model overestimates risk or other protective factors were not captured in the models.

F.2. EXPLORATION AND MODELLING OF ANIMAL TRIALS STUDIES

The dataset compiles results from experimental studies assessing H5N1 infectivity across multiple animal species. It includes data from eight records, covering five animal species: ferrets, mice, macaques, hamsters, pigs, Guinea pigs and cats (Bertran & Swayne, 2014; Eisfeld et al., 2024; Guan et al., 2024; Jones et al., 2025; Kwon et al., 2009; Lipatov et al., 2008; Lipatov et al., 2009; Octaviani et al., 2025; Reperant et al., 2012; Rosenke et al., 2025; Shinya et al., 2011; Tipih et al., 2024; Vahlenkamp et al., 2010). Each record corresponds to a unique combination of species, virus strain, inoculation route, exposure dose and symptom type (clinical infection or seroconversion), along with the number of animals exposed and the number that became infected.

The paper from Nooruzzaman, de Oliveira, et al. (2025), which evaluated the infectivity of HPAI‐contaminated raw milk, (solid) ripened cheese and cheese suspensions in ferrets (groups of 4 animals) through voluntary repeated feeding (days 0, 3, 4 and 5 of the experiment) was not considered for the modelling given the adopted administration scheme, and the possible accidental inhalation of virus during feeding, particularly for milk consumption. Noteworthy, this study showed no clinical signs, no virus detection in nasal, oropharyngeal and rectal swabs and no seroconversion in animals that were repeatedly fed with 4.70 ± 0.58 (solid cheese) or 3.04 ± 0.40 (cheese suspension) log₁₀ EID₅₀/animal infectious viral particles of HPAI H5N1 isolate TX2/24 (A/Cattle/Texas/063224–24‐1/2024), genotype B3.13. Contrary, two of the four ferrets that were fed 4.99 ± 0.42 log₁₀ EID₅₀/animal infectious viral particles through raw milk developed clinical signs (fever, weight loss) and shed virus (detection in the three swab types) but did not seroconvert before being euthanized.

To ensure consistency in dose comparison across studies, all exposure doses were subsequently converted to EID₅₀ units. This was done using the relationships between the original units (TCID₅₀, PFU, EID₅₀, etc.) from Appendix B. This normalisation allowed the pooling of data from various sources and ensured comparability of dose–response relationships across studies and species.

Figure F.1A displays the relationship between ingested H5N1 dose (expressed in EID₅₀) and the probability of clinical infection across all species. Each point represents an experimental group, with the size of the point proportional to the number of animals exposed. Vertical lines show exact binomial 95% confidence intervals. A wide range of susceptibility is evident across species, with mice exhibiting infection at relatively low doses, while others remain uninfected even at higher exposures. Figure F.1B presents the same structure but for seroconversion as the outcome, representing animals that mounted an immune response without necessarily displaying clinical signs. For DR modelling, only data on clinical infection were used. In studies reporting both clinical signs and seroconversion, only one endpoint was retained to avoid overlap: when clinical infection was observed in at least one group, the corresponding seroconversion data from the same study were excluded from model fitting. This ensured that each experimental group contributed only once to the modelling dataset and avoided inflating the apparent sample size. Consequently, seroconversion data are presented separately for descriptive purposes (Figure F.1B) but were not included in the estimation of DR relationships.

FIGURE F.1.

Observed infection outcomes following oral exposure to H5N1 across animal species. (A) Proportion of animals showing clinical signs after exposure. (B) Proportion of animals showing seroconversion. Each point represents an experimental group; point size reflects the number of animals per group. Vertical bars show 95% exact binomial confidence intervals. All doses were converted to EID₅₀ units using published conversion factors. Colours indicate animal species. Based on data extracted from Bertran & Swayne, 2014; Eisfeld et al., 2024; Guan et al., 2024; Jones et al., 2025; Kwon et al., 2009; Lipatov et al., 2008; Lipatov et al., 2009; Octaviani et al., 2025; Reperant et al., 2012; Rosenke et al., 2025; Shinya et al., 2011; Tipih et al., 2024; Vahlenkamp et al., 2010.

Based on the patterns observed in Figure F.1, two modelling approaches were undertaken. The first focuses on mice, which appear markedly more susceptible to H5N1 infection at lower doses. The second targets ferrets, for which the largest volume of experimental data is available. Other species were not included in the modelling due to limited data or inconsistent responses.

To account for variability across viral strains and experimental studie – as multiple studies were available for both mice and ferrets – the beta‐Poisson DR model was selected. This model is well‐suited to incorporate such heterogeneity, providing a flexible framework for representing inter‐study and inter‐strain variation in susceptibility.

Figure E.2 illustrates the DR relationships for H5N1 infection in mice and ferrets, modelled using a beta‐Poisson framework. In both cases, all doses were first converted to EID₅₀ units to allow cross‐study comparison. In mice, the estimated median infectious dose (ID₅₀) is 2.8 × 10³ EID₅₀, with a 95% confidence interval ranging from 1.5 × 10³ to 5.2 × 10³. For ferrets, the estimated ID₅₀ is higher, with a median value of 1.2 × 10⁸ EID₅₀ (95% CI: 4.0 × 10⁷ to 5.3 × 10⁸). The broader confidence band reflects greater variability across experiments and strains in this species. These results confirm that mice are more susceptible to H5N1 via oral exposure compared to ferrets. Therefore, the ferret model was adopted for further consideration and was considered together with the models by Bauer et al. (2010) and Chen et al. (2025) (see Figure 5 in the main text).  

FIGURE F.2.

Dose–response curves for H5N1 infection in mice (A) and ferrets (B), modelled using the beta‐Poisson framework. All doses are expressed in EID₅₀. Points represent observed infection probabilities, with vertical lines showing binomial confidence intervals. The solid red line indicates the median predicted probability of infection, and the shaded ribbon represents the 95% confidence band obtained via parametric bootstrap. Based on data extracted from Bertran & Swayne, 2014; Eisfeld et al., 2024; Guan et al., 2024; Jones et al., 2025; Lipatov et al., 2008; Octaviani et al., 2025; Shinya et al., 2011; Tipih et al., 2024).

APPENDIX G. Uncertainty analysis ToR 2e

TABLE G.1 Potential sources of uncertainty linked to specific AQ or SQ in the assessment of the risk of infection of dairy cattle in the EU with highly pathogenic avian influenza virus affecting dairy cows in the United States of America (H5N1 B3.13 genotype virus) – food safety assessment.

AQ or SQ	Source or location of the uncertainty	Nature or cause of the uncertainty	Impact of the uncertainty on the conclusions (e.g. over/underestimation)
SQ 1.1 SQ 1.2	Incomplete identification of production processes and selection of foodstuffs	The production processes were derived through literature and based on the knowledge of the WG experts.	The impact of this uncertainty is expected to be small as it is believed that the most relevant processes and foodstuffs have been included
	Selection for the risk assessment of key processing steps across different dairy product categories	Heat treatment, acidification/ coagulation, yield and ripening were considered in the assessment. No inactivation was assumed for the refrigeration storage step of raw milk before processing, and heating and brining of the curd was not specifically assessed. Also, the addition of preservatives or preservation cultures and use of moulds for surface ripening was not considered.	The impact of this uncertainty is described in SQ1.4
	Incomplete identification of production parameters (values)	The production parameters (heat treatment, acidification/coagulation, yield, ripening) were derived through literature and based on the knowledge of the WG experts. A selection of typical conditions was made for the risk assessment and included for most thermal treatments various conditions (e.g. HTST and LTST). As there are no legal requirements for thermisation, more uncertainty can be expected compared to pasteurisation.	The impact of this uncertainty is expected to vary depending on the parameter and foodstuff/dairy category. It is leading to an under or overestimation of the exposure
SQ 1.3 for levels in bulk milk	Dilution of contaminated bulk tank milk with milk from a non‐infected herd	In case the contaminated bulk milk from a herd is further mixed with milk from a non‐infected herd in the farm (not necessarily one bulk tank per farm per day) or at processing (as common in large‐scale production settings), the concentration of H5N1 B3.13 genotype virus in the milk will be decreased. The probability of this happening depends on the prevalence of infected herds.	The initial level in bulk milk is overestimated when mixing takes place which is leading to an overestimation of the exposure
	Average herd size in the EU	The average herd size in the EU was considered as 65 in the bulk tank model based on the size biased mean. The herd size in the EU varies. The impact of the herd size was evaluated using sensitivity analysis.	The impact of this uncertainty is expected to be small. It is leading to an under or overestimation of the exposure.
	Proportion of cows in a herd that is shedding H5N1 B3.13 genotype virus to the milk	The proportion of cows in a herd that is shedding H5N1 B3.13 genotype virus to the milk was considered as 1% in the model based on data from a single herd with 2500 cows in the USA. The proportion was considered a fixed value over time, although the estimate of the proportion was based on herd dynamics over time. The impact of the proportion of cows in a herd that is shedding H5N1 B3.13 genotype virus was evaluated using sensitivity analysis using a 5% value.	The impact of this uncertainty is expected to be high. It is leading to an under or overestimation of the exposure.
	Concentration of H5N1 B3.13 genotype virus in the milk from a cow contributing milk in the bulk tank	The concentration distribution was derived from a single study (Caserta et al., 2024) and was truncated based on the values derived from other studies.	The uncertainty from the dataset is captured in the model. The uncertainty derived from the single data source is not captured and is leading to an under or overestimation of the exposure.
	Relation between infectivity units (TCID50 and EID50)	Based on a single study, using both egg infection and cell culture (one line). Data may not reflect the variable sensitivity of different titration systems. Data originally expressed in TCID50 and converted into EID50 have larger associated uncertainty.	The uncertainty from the dataset is captured in the model. The uncertainty derived from the single data source is not captured and is leading to an under or overestimation of the exposure.
SQ 1.3 for levels in bulk colostrum	The same as for SQ1.3 for levels in bulk milk; in addition, no specific data on colostrum is available	The same as for SQ1.3 for levels in bulk milk; the uncertainty is expected to be higher. Mixing of colostrum may differ due to the different production volumes as compared to milk.	Exposure is over‐ or underestimated
	Initial level of contamination in bulk colostrum	The assessment for colostrum was based on the same parameters as those applied for milk (with the assumption that all bovines produce colostrum and that individual shedding of HPAI in colostrum is the same as in milk).	The initial level in bulk colostrum is underestimated
SQ 1.4	Impact of information not retrieved from the literature	It is possible that not all the information (reductions) may be retrieved using the search strategy	The impact of this uncertainty is expected to be small as it is believed that the relevant evidence has been identified by the literature review.
	Scarcity of data	Limited amount of data is available for the efficacy of certain time–temperature treatment conditions in the inactivation of HPAIv, and particularly H5N1 B3.13 genotype virus, in milk. No specific information is available for colostrum. The dataset was primarily built using D‐values compiled by Hessling (2022). Selection and exclusion criteria were applied to discard data judged as potentially biased – in particular, data from matrices such as high‐fat or dried products were removed. While this filtering increases confidence in the representativeness of the selected data for liquid milk products, it is important to note that the remaining studies may still involve matrix components (e.g. proteins, salt content) that could influence thermal resistance. Limited data are available for the effect of acidification and ripening during fermentation and cheesemaking on HPAIv, and particularly H5N1 B3.13 genotype virus, viability affect the log reductions applied. For ripening, the few data available (one study) on cheese from naturally contaminated milk display lower inactivation compared to data from experimentally contaminated cheese. Furthermore, variable temperature conditions in experimental studies on acidification may affect the log reduction applied. No information is available on the effect of freeze‐drying of viruses in milk/colostrum. Insufficient information is available on partition of HPAIv, and particularly H5N1 B3.13 genotype virus, in different phases (curd, whey) during cheesemaking. The inactivation of HPAI in colostrum (e.g. by heat treatment) was assumed to be the same as in milk	The impact of this uncertainty is expected to be varying according to the considered parameter. It is leading to an under or overestimation of the exposure
	Strain‐specific representativeness of reductions being found	Studies have been included using strains other than H5N1 B3.13 genotype virus Variability may occur even within strains belonging to this genotype, meaning that reductions observed may not always be fully representative. It cannot be excluded that the current strain may evolve, potentially leading to differences in its behaviour or inactivation characteristics.	It is leading to an under or overestimation of the exposure
	Model used for thermal inactivation	Depending on data availability, Dt‐values were either extracted from the published articles or estimated from the raw inactivation data using a log‐linear inactivation model. The classical Bigelow model was considered applicable to the dataset in which log10 (D) was assumed to vary linearly with temperature. A satisfactory agreement was found between observed D‐values in milk and predicted values (the model predicting higher D‐values compared to the seven observed ones). Uncertainty around the parameter estimates was quantified together with variability.	The uncertainty was not included in the model therefore part of the uncertainty is ignored. It is leading to an overestimation of the variation of the exposure.
	Heating during cheese making not considered as heat treatment of milk	Highly variable and procedural guideline‐specific heating conditions are used in the curding process. No inactivation was assumed for the refrigeration storage step of raw milk before processing, heating and brining of the curd was not specifically assessed. Also, the addition of preservatives or preservation cultures and use of moulds for surface ripening was not considered.	The impact of this uncertainty is expected to vary depending on the foodstuff/dairy category. It is leading to an over estimation of the exposure
	Relation between infectivity units (TCID50 and EID50)	D‐values were used as reported, established by counts expressed as TCID50 or EID50. As the slope in the Figure A.1 showing the regression line used to derive log EID50/mL values from log TCID50/mL does not equal 1, it is expected that the D‐values could be different depending on the expression used.	The impact of this uncertainty is expected to be small
SQ 2.1	Distribution of H5N1 B3.13 genotype virus in bulk tanks is assumed to be homogeneous	As continuous stirring of the bulk milk at farm and of the silo milk at processing is applied, it is considered that there is not much heterogenicity between concentrations in samples from the bulk tank	The impact of this uncertainty is expected to be small as continuous stirring in bulk tanks is applied
	Variability in serving sizes between consumers and between products in a category; not all serving sizes known	Variability in serving sizes results in a little uncertainty and variability in the doses	The impact of this uncertainty is expected to be small. It is leading to an overestimation of the variation of the exposure
SQ 2.2	Relation between infectivity units (TCID50 and EID50)	See SQ 1.2. Scarcity of studies comparing quantification of HPAIv, and particularly H5N1 B3.13 genotype virus, using different titration methods	It is leading to an overestimation of the variation of the exposure
	Scarcity of animal models data on H5N1 B3.13 genotype virus Heterogeneity and scarcity of data for animal models.	Few studies addressing infectious dose of H5N1 B3.13 genotype virus in animal models. Susceptibility (in terms of symptoms development and severity) of different animal models to HPAIv infection, and particularly H5N1 B3.13 genotype virus infection, is highly variable. A limited number of studies with experimental designs including multiple HPAIv doses are available.	A DR relation was not applied
	Inaccuracy of data reporting	Limited information is available for some experimental settings and for interpretative criteria applied to assess response.	A DR relation was not applied
	Representativity of DR data for H5Nx strains and for non‐H5 strains to the H5N1 B3.13 genotype virus	Scarcity of data are available comparing infection doses of H5N1 B3.13 genotype virus and of other H5Nx strains in the same animal model.	A DR relation was not applied
	Representativity of different animal models to humans and of human inhalation models to oral ingestion	Scarcity of DR data for human infection with H5Nx influenza viruses (one model considering exclusively inhalation and multiple influenza virus subtypes)	A DR relation was not applied
AQ3	Available data and model used to assess the impact of thermal treatments	See listed above under SQ 1.4
	Scarcity of studies testing the impact of non‐thermal treatments on viable AIV and viruses in general	No study testing the impact of the listed non‐thermal technologies on viable AIv in milk is available. For HPP, HPH, UV‐C, PEF and HPCD few studies are available that have evaluated the log10 reduction of some treatment conditions on viruses in general, especially for treatment of human milk.	This uncertainty source may lead to an over‐ or under‐estimation of the efficacy (log10 reduction) of specific treatment conditions of non‐thermal technologies towards H5N1 B3.13 genotype virus.

Abbreviations: AIv, avian influenza virus; D, decimal reduction time or the time required at a given temperature to kill 90% of the exposed microorganisms; DR, dose–response; EID₅₀, egg infectious dose 50%; HPAIv, highly pathogenic avian influenza virus; HPCD, high‐pressure carbon dioxide; HPH, high‐pressure homogenisation; HPP, high‐pressure processing; HTST, high temperature for a short time; LTST, low temperature for a long time; PEF, pulsed electric fields; SQ, sub‐question; TCID₅₀, tissue culture infectious dose 50%; UV‐C, ultraviolet C irradiation.

Annex A. Incursion risk of highly pathogenic avian influenza H5N1 Eurasian lineage goose/Guangdong clade 2.3.4.4b genotype B3.13 for the European Union: A risk assessment based on the L'ORA tool

Annex A is available under the Supporting Information section on the online version of the scientific output.

Annex B. Protocol to answer ToR 2e

Annex B is available under the Supporting Information section on the online version of the scientific output.

EFSA AHAW Panel ( EFSA Panel on Animal Health and Welfare) , Alvarez, J. , Boklund, A. E. , Dippel, S. , Dórea, F. , Figuerola, J. , Herskin, M. S. , Michel, V. , Miranda Chueca, M. Á. , Nannoni, E. , Nielsen, S. S. , Nonno, R. , Riber, A. B. , Stegeman, J. A. , Ståhl, K. , Thulke, H.‐H. , Tuyttens, F. , Winckler, C. , Bortolami, A. , … Gervelmeyer, A. (2025). Risk of infection of dairy cattle in the EU with highly pathogenic avian influenza virus affecting dairy cows in the United States of America (H5N1, Eurasian lineage goose/Guangdong clade 2.3.4.4b. genotype B3.13). EFSA Journal, 23(12), e9801. 10.2903/j.efsa.2025.9801