The results of our analysis, drawn from WOAH WAHIS outbreak notification data, offer valuable insights into the variability of surveillance performance in HPAI and ASF outbreak notification within and between countries/territories, and the key factors associated with this variability.

While several attributes have been developed to evaluate animal infectious disease surveillance systems, assessing these attributes has been constrained by a lack of standardised data, as well as the absence of frameworks and tools tailored to analyse such data^7,8,12,13. Therefore, our framework provides unique insights into the evaluation of animal infectious disease surveillance systems by incorporating three fatality metrics that explicitly account for the outbreak progression expected following pathogen introduction until notification. Furthermore, using outbreak notification data collected under WOAH guidelines within our framework enables a systematic evaluation of surveillance performance across countries/territories, which is crucial for controlling transboundary infectious diseases. This approach can also be applied at sub-country/territory levels, over time, and potentially to other diseases, and its robustness could be further improved through ongoing efforts to standardise such data.

In our modelling framework, each fatality metric has distinct—though not mutually exclusive—implications for surveillance performance. However, we emphasise that these fatality metrics should be assessed in light of pathogen characteristics: HPAI viruses typically spread rapidly with short latent period and generation time, leading to high fatalities within a short period¹⁴, whereas ASF viruses spread more slowly, with a more gradual onset of clinical signs and deaths¹⁵. Importantly, data completeness, accuracy, and bias—commonly considered attributes of surveillance evaluation^8,12,13—require continuous assessment. Additionally, while our study is a foundational step in assessing the variability of surveillance performance, follow-up studies on eco-social, veterinary, and agricultural factors are needed to better understand the observed variability and develop country-/territory-specific strategies.

For both HPAI and ASF, the number of susceptible animals and the country/territory of an outbreak were identified as the most significant factors. Regarding the number of susceptible animals, this result was expected as pathogen transmission likely increases with larger susceptible populations, and it was why we constructed the baseline model with the number of susceptible animals as a key variable, assuming primarily density-dependent transmission¹⁶. More specifically, the zero-fatality probability was estimated to decrease as the number of susceptible animals increased, suggesting that early detection becomes more challenging in larger herds, likely due to factors including higher transmission risk, background mortality, and husbandry conditions. However, beyond a certain population size, fatalities would not increase linearly with a larger number of susceptible animals when surveillance is in place, and our modelling framework effectively accounted for and demonstrated this dynamic.

The results regarding the country/territory of an outbreak indicate that the number of fatalities reported at initial notification is highly heterogeneous across countries/territories. Such heterogeneity could stem from a variety of country/territory-level factors not considered in our analysis, including the capacity of veterinary services and the level of viral circulation within countries/territories. In our analysis, a few countries/territories were estimated to have experienced significantly more or fewer fatalities than expected assuming the average performance across all analysed countries/territories. For these countries/territories, rather than simply attempting to transplant practices deemed successful in one setting to another, gathering detailed information on the eco-social, veterinary, and agricultural factors and assessing it in relation to differences in fatality metrics could help identify the key drivers of surveillance performance. In this process, it is important to keep in mind that the fatality metrics capture only a limited dimension of animal health systems and therefore cannot be used as the sole measure of their overall performance. This approach would then help understand how different contexts influence the actual operation of surveillance, providing a critical foundation for developing context-specific strategies to strengthen animal infectious disease surveillance at national, regional, and global levels.

Additionally, this assessment should also include scrutiny of potential differences in reporting practices and accuracy. The only country/territory (ID: 37) for which the best-fitting model did not accurately predict the number of fatalities from ASF outbreaks in villages serves as a good example. This country/territory was estimated to perform significantly better than expected based on the average performance. However, it had an exceptionally high zero-fatality probability compared to other countries/territories and also exhibited a high fatality slope. This pattern contrasted with other countries/territories also estimated to perform significantly better, since they tended to be associated with both a high zero-fatality probability and low fatality slope and threshold estimates, like country/territory 33 for both HPAI and ASF outbreaks. This suggests that the outlier behaviour of the country/territory 37 warrants further investigation to ensure that the observed trends genuinely reflect surveillance capacity or characteristics rather than being artefacts of data reporting issues. Such investigations could therefore begin by assessing potential systematic differences in reporting practices, for example, how village outbreaks were defined and reported. Although outbreaks were notified using a standardised format, there is no consensus on the definition of the premise types, which may lead to inconsistencies in classification, particularly between outbreaks in backyard farms and villages. If no significant differences are identified, the investigation could then focus on the epidemiological context of these villages and how surveillance systems operated there, compared to those in other countries/territories.

After controlling for the number of susceptible animals and the country/territory of an outbreak, other variables, including premise type, season, and the cluster occurrence of outbreaks, were significantly associated with the number of fatalities reported at initial notification. However, it is important to note that these factors did not contribute much to explaining the overall variability in the data, potentially due to the large size of the dataset: a large statistical power may have allowed detection of relatively small effects. That is, although these variables were found to have statistically significant associations, the practical relevance of their impact on the number of fatalities appears to be less important than that of the number of susceptible animals and the country/territory of an outbreak.

Regarding premise type, HPAI outbreaks on backyard farms and in villages were associated with a lower zero-fatality probability and a higher fatality slope compared to those on commercial farms. This pattern likely stems from a combination of relatively poor biosecurity and surveillance measures in these settings, as well as the inherent difficulty of early detection in HPAI outbreaks. The virus’s rapid spread and high lethality could lead to fatalities before detection through other clinical signs or active surveillance is possible, particularly in backyard and village settings. On the other hand, ASF outbreaks on backyard farms, while showing a higher fatality slope, were significantly more likely to be reported with zero fatalities. However, this does not necessarily indicate that backyard farms are inherently better at early detection of ASF outbreaks, given their typically poorer surveillance compared to commercial farms^17,18. A more plausible explanation is that surveillance efforts, when present, were intensified toward backyard farms due to their known susceptibility to ASF virus introduction, particularly from infected wild boar populations¹⁹. This, combined with the relatively slow progression of ASF outbreaks, may have made early detection easier.

Regarding season, our analysis revealed that the probability of reporting zero fatalities at initial notification was lowest in autumn for HPAI outbreaks and in spring for ASF outbreaks. Notably, these seasons occur just before the major transmission periods—winter for HPAI⁴ and summer/autumn for ASF¹⁰. One possible explanation is that farmers’ risk awareness, and potentially surveillance sensitivity, might have been lower before these major transmission seasons. Conversely, the seasons following the major transmission periods (that is, spring for HPAI and winter for ASF) showed similar levels of zero-fatality reporting as during the peak transmission seasons. This suggests that the enhanced surveillance and heightened risk awareness established during the high-risk periods likely persisted into the subsequent season. On the other hand, the fatality slope and fatality threshold exhibited notable variations in several, but not all, seasons, compared to the high-risk periods for both HPAI and ASF outbreaks. When outbreaks advance to the point of causing deaths, detection of HPAI and ASF outbreaks may primarily rely on noticeable changes in overall mortality compared to expected background mortality, along with other clinical signs. Therefore, seasonal fluctuations in background mortality and possibly population sizes could have influenced the fatality slope and fatality threshold^20,21,22. However, our results reflect the average trends across the countries/territories analysed, and therefore, different countries/territories may exhibit unique trends due to different climates and animal production cycles.

In lower-middle income countries/territories, the lower zero-fatality probability for HPAI outbreaks occurring outside of clusters suggests that, compared to high-income countries/territories, the ability to detect these outbreaks early—before fatalities occurred—was already limited, even during periods of lower outbreak intensity. In high-income countries/territories, signs of surveillance burden were evident during periods of intense outbreak activity, as indicated by the lower zero-fatality probability for HPAI outbreaks occurring within clusters. For upper-middle and lower-middle income countries/territories, outbreaks within clusters did not show a significant decrease in the zero-fatality probability, likely due to an already low zero-probability during periods of lower outbreak intensity. In contrast, for ASF outbreaks, the zero-fatality probability did not differ significantly between clustered and non-clustered outbreaks, suggesting that early detection of ASF outbreaks was relatively consistent, regardless of their occurrence within or outside clusters. On the other hand, compared to high-income countries/territories, ASF outbreaks in upper-middle-income countries/territories, both within and outside clusters, exhibited a significantly higher fatality threshold. ASF outbreak detection may have been delayed in these countries/territories during intense outbreak periods, leading to larger fatalities reported at notification.

It is important to note that outbreak clusters were identified based on both HPAI and ASF outbreaks, under the assumption that they exert equal pressure on veterinary authorities. However, the pressure exerted by HPAI and ASF outbreaks, as well as other pathogens, on veterinary authorities likely varies between countries/territories depending on multiple factors, including their surveillance and control capacities. We therefore used 180 days as the maximum temporal window to allow for the detection of all relevant clusters within the study period (e.g., short, intense clusters as well as long, protracted clusters), while still considering seasonal HPAI and ASF transmission patterns. Shorter maximum windows might fail to capture long, protracted clusters that place significant pressure on veterinary services.

For both HPAI and ASF outbreaks, the best-fitting model indicated high overdispersion in fatalities at initial outbreak notification, after accounting for the variables included. This suggests that the performance in outbreak notification is highly variable amongst premises within the same countries/territories. While this could arise from the stochasticity of viral transmission within premises (e.g., fatality patterns could be more heterogeneous depending on the course of viral transmission and the route of viral introduction), other important factors not considered in this study could also have played a role behind this variability, including differences in the level of on-premise biosecurity and surveillance, animal density, host breeds and species, transmissibility and virulence by viral types, and surveillance capacity between regional veterinary services. These emphasise the need to identify critical areas to improve surveillance capacity within as well as across countries/territories.

Finally, although our modelling framework was not specifically designed to compare differences in the outbreak dynamics between the HPAI and ASF viruses, certain observations suggest important surveillance implications arising from these differences. Compared to commercial farms, ASF outbreaks on backyard farms were associated with a higher zero-fatality probability, while HPAI outbreaks on backyard and village settings were linked to a lower zero-fatality probability. As discussed earlier, this high zero-fatality probability for backyard ASF outbreaks may have been influenced by surveillance efforts targeted at this setting. However, despite this, detection of ASF outbreaks before observing fatalities would likely be possible largely due to the relatively slow progression of ASF¹⁵. In contrast, HPAI outbreaks are characterised by rapid viral spread and resulting deaths¹⁴, which can be exacerbated by a lack of biosecurity in backyard and village settings, making early detection particularly challenging. From a surveillance perspective, this suggests that HPAI outbreaks, despite causing large fatalities, are relatively easier and faster to detect compared to ASF outbreaks, potentially leading to better surveillance sensitivity and timeliness. The slow progression of ASF outbreaks, on the other hand, allows for extensive viral circulation within premises and provides a larger window of opportunity for the virus to spread to other pig populations by the time mortality levels rise enough to trigger detection. This factor needs to be considered when improving ASF outbreak detection sensitivity and timeliness.

This study has limitations. Firstly, while the study period for HPAI outbreaks was designed to cover the global spread of wild bird-adapted HPAI H5N1 viruses, other HPAI viruses also circulated concurrently, affecting different regions. These other subtypes may have contributed to the observed heterogeneities in the number of fatalities due to their varying levels of transmissibility and virulence.

Secondly, the host animals in HPAI outbreaks were categorised as birds in the WOAH WAHIS. HPAI viruses, however, are known to pose varying levels of transmissibility and virulence to different bird species. Notably, chickens are generally more susceptible than ducks to HPAI viruses^23,24. This suggests that our fatality metrics could be influenced by the composition of bird species within a given country. In chicken farms, a sudden rise in mortality can serve as an early warning sign of an outbreak, but by the time it is detected, substantial losses may have already occurred, potentially leading to a low zero-fatality probability and a high fatality slope. In contrast, HPAI outbreaks in duck farms may spread widely before detection with a few or no deaths, potentially resulting in a high zero-fatality probability and a low fatality slope. Therefore, having information on host species could help improve the assessment of surveillance performance in HPAI outbreak notification, ensuring that differences in species susceptibility and transmissibility are accounted for. For ASF outbreaks, accounting for viral genotypes and swine breeds would also allow more robust parameter estimation, although these concerns may be less relevant compared to the HPAI host/viruses, given the relatively stable viral genetic structure and the less variable transmissibility and virulence across different swine breeds^25,26.

Thirdly, surveillance capacity could have changed dramatically during the study periods within countries/territories as the level of viral circulation varied, veterinary services gained experience with newly circulating viruses, and socioeconomic and geopolitical contexts evolved. Such changes may have occurred in either direction, making it difficult to incorporate into our framework given limited data availability. Although we used the number of veterinarians per a million terrestrial animal biomass²⁷ as a proxy for veterinary service capacity, this metric does not reflect potential temporal shifts in the reporting behaviour of key stakeholders, including veterinarians, farmers, and other value chain actors. Therefore, future studies should aim to complement quantitative data with detailed descriptions of the contextual factors, such as politics, socioeconomic, and veterinary system dynamics, that may have influenced these behaviours and, ultimately, surveillance performance²⁸.

Fourth, while WOAH member countries/territories are required to notify outbreaks of listed diseases in a standardised format, gaps in the data existed due to variations in how outbreaks were reported. For example, in some countries/territories, the number of dead animals was not available for a relatively large proportion of ASF and, particularly, HPAI outbreaks. This could reflect differences in reporting practices within countries/territories or pressure on veterinary services during certain periods (e.g., insufficient resources to collect detailed outbreak data when they occur in clusters). The extent of this missing information was significant in a few countries/territories, and its impact would depend on the degree to which the outbreaks excluded from the analysis due to missing data had different notification dynamics. Additionally, although not widespread, there were instances where the number of dead animals was reported as greater than the number of susceptible animals or equal to the number of culled animals in several outbreaks, making it challenging to clearly interpret these numbers. These data inconsistencies underscore the need for efforts to further improve and standardise reporting practices, as well as more precise definitions of the variables to be collected from notification reports.

Considering the limitations discussed above, future research would benefit from incorporating interviews with individual veterinary services to better understand differences in surveillance and reporting practices that may introduce biases into the data. These interviews could also help obtain more detailed outbreak notification data—such as host species, viral lineages, and other key epidemiological variables, including temporal changes in surveillance capacity—collected in a way that allows for meaningful comparisons within and between countries or territories. Not only would such efforts enable more robust comparisons between countries/territories, but they could also facilitate the application of our framework at the sub-national level, helping to identify heterogeneities within countries/territories and develop more tailored, context-specific recommendations. Crucially, these activities should be conducted in close collaboration with the veterinary services involved to ensure accurate interpretation of findings and to establish a constructive feedback loop that supports the continuous strengthening of surveillance and response systems.

In conclusion, our results highlight significant variability in animal health surveillance capacities for HPAI and ASF outbreaks, both within and between countries/territories, based on WOAH WAHIS outbreak notification data. These disparities may arise from specific drivers shaped by diverse eco-social, veterinary, and agricultural contexts across countries/territories, which were not fully captured by the macro-level variables used in this study. A key next step would thus be to focus global and regional efforts on identifying the specific drivers that influence the functioning of HPAI, ASF, and other animal disease surveillance systems. By doing so, we can develop more targeted, effective, and efficient recommendations to improve animal health surveillance capacities.