The use of the Walktrap community detection algorithm appeared to be a powerful tool as it was able to identify trade communities that match known production systems and areas. For example, intensive production systems in north-west of France, north-east of Spain, and north of Italy were clearly identified. Similarly, the commercial outdoor production systems in Extremadura – Iberian pigs – or in southwest of France, and the small-scale pig production systems in center and southern Italy were also identified. Considering that the movement of animals is the main source of disease introduction/spread into new areas, the use of these methods may help to more cost-effectively trace the sources of infection in case of an epidemic and define zones or compartments that prevent the spread of infectious diseases while maximizing business continuity. In counterpart, as Walktrap algorithm was used only on live animal movements, small-scale production systems that have or report few exchanges of pigs, such as the Corsican, Sardinian, or the East Balkan pigs, will not emerge. Moreover, in these small scale production systems, the role of contaminated fomites in the spread of infectious diseases may be more important than the movement of live animals compared to intensive farms, given the absence or low bio-security levels. Therefore, other methods, such as farmer interviews, should be used to complement the information regarding trade patterns and to identify and describe other high-risk contacts associated with fomites . In addition, reader should notice that Walk trap algorithm treats directed networks as undirected. Unfortunately,rolling grow tables the few algorithms that implicitly consider directionality of the movements are computationally intensive and usually do not work for large networks.
For example, we were able to use InfoMap for Bulgaria, but the algorithm was not working for France, Italy, or Spain. Nevertheless, for Bulgaria, we had a good agreement using both methods: the number and characteristics of the communities were similar when comparing InfoMap and Walktrap algorithm . Therefore, we assumed that Walktrap algorithm was performing well and was a good choice in this case to describe the modularity of our directed networks. Report of movements is quite similar and mandatory in the four countries included in the study since at least 2009, and the official veterinarians were quite confident of the reporting compliance for the year 2011, except for some specific areas of the Islands of France and Italy . It is important to note that although authorities are usually not very open to share animal movement records with this level of detail due to confidentiality issues, there is an extraordinary value of accessing and analyzing this information to unravel the complexity of the trade network structure and characteristics and better inform policies. In fact, thanks to the high quality of movement data and the availability of full datasets from four European countries during the same period, several network measures and proxies of pig trade patterns could be computed and compared in detail. Results highlighted that some proxies can be used whatever the systems considered, whereas other are specific to a country or even a production system . Indeed, the scale-free topology was observed for every trade network, whatever the country or the production system considered, as previously reported in countries with a predominantly intensive pig production system . This means that most premises have few connections while few premises have many connections .
These premises were mostly trade operators and multipliers but other production types, such as farrow-to-finish or finishers, could also have a lot of connections . They may play an important role in the spread of infectious diseases and could be targeted to more efficiently detect and control them . A closer look at the degree distributions also revealed that trade operators behaved differently according to the production system, acting as collectors in industrial systems, and as dispatchers in countries with a lot of SP. Trade operators may thus play different roles as “super-receivers” or “super-spreaders” in disease epidemics and may be good candidates to target risk-based surveillance or control strategies, respectively. Future studies aiming to evaluate weather or not the preferential attachment observed in those scale-free networks can be explained differently in each country will be valuable. Small-world properties, which had been previously reported for pig trade networks , were not observed in this study when considering the whole countries, except for France. They were however observed when considering trade communities, particularly when these communities contained trade operators. Trade operators were present in all communities with indoor commercial producers but were rarely observed in communities mostly comprised of small-scale or outdoor producers. Infectious diseases will thus spread differently by trade movements according to the production system. They will spread quicker, more remotely, and extensively in intensive production systems and slower, more locally, in extensive or small-scale production systems . These results suggest that to simulate realistic networks based on network topology , modelers should consider that pig trade networks have both scale-free and small-world properties in intensive production systems, but only scale-free properties in outdoor or small-scale production systems. Further analyses using data from other countries could be useful to confirm these results.
Shipment distances were similar between countries, with most of movements occurring within 100 km as previously described in other European countries . As expected, the greatest shipment distances where observed in countries with the longest territories and might have been even greater if movements from/to foreign countries had been included in the analysis. However, the distances appeared to be linked with the type productions systems, whatever the country considered . The contrast was particularly marked in Bulgaria and Italy, with short shipment distances for communities of small-scale producers and long distances for communities with commercial farms . These short shipment distances might be due to the fact that small-scale producers are usually located in remote areas, such as the less developed areas or mountains, with limited access to expressways or trains. Thus, they tend to trade with neighbors, which are mostly also small scale producers. They might also be less likely to form connection with geographically and network distant premises, which could explain why the small-world properties were not observed in small-scale production systems. The impact of premises location and transport facilities on shipment distances and network topology could be further investigated to more accurately model pig movements. For all countries, shipment rates were much lower than those described in recent studies from Canada, even when considering only commercial farms . This might be due not only to the higher specialization and inherent more integrated, multi-site, structure of commercial premises in North America but also to differences in data sources and data representativeness and quality as, for example, Dorjee et al. obtained data only from one major pig company and Thakur et al. from volunteer farmers. Shipment rates were particularly low in Bulgaria, illustrating the lower degree of specialization for Bulgarian pig farms,cannabis growing equipment and thus a less need to exchange pigs between premises. This certainly have important implications in terms of disease prevention and control and should be considered when defining zones or compartments to mitigate disease spread while allowing business continuity. Shipment sizes were also not homogeneous between countries. Pig batches sent to finishers were the largest, as previously observed ; however, those sent to finishing farms in Spain were larger than those sent to finishing farms in France or Italy, likely due to the differences in farm sizes. As expected, pig batches sent from or to SP were of small size, whatever the country considered. Mixing patterns by premise types are also useful to more realistically simulate pig trade networks . Commercial pig production is usually considered to have a pyramidal organization with multipliers sending reproductive pigs to farrowing or farrow-to-finish farms and these ones sending piglets to finishers [e.g., Ref. ]. Mixing patterns measured in this study do partially reflect this organization but also highlight some unexpected trade patterns. Indeed, results reveal the major role played by trade operators in France, which tended to proportionally receive most of shipments no matter the type of farm sending pigs, whereas in Spain or Italy, multipliers also played a central role in the pig trade organization. They also highlighted that SP were not isolated, and not only receive but also sent pigs to commercial producers. These mixing patterns are thus important to consider for surveillance or control strategies or when modeling disease spread. Thus, even if shipments rates and shipment distances seem to be linked with the type of production of the premises sending and receiving pigs, mixing patterns could depend on economic or organization rules that are country-dependant.
Modelers using models based on statistics on shipments rates, shipment distances, and mixing patterns between production types should thus consider this information. Other methods, such as exponential random graph models , have been recently used to better capture the complex topology and mixing patterns of pig trade networks . Our study can be used to select ERGM parameters. For example, the existence of long distance shipments suggest that geographical information should be used in ERGMs to adequately capture the spatial patterns of pig trade at country level. Community detection methods have been suggested as a useful tool to identify compartments or zones that could be used in the design of diseases surveillance and control programs to preserve business continuity and minimize trade disruption . Results of this study suggest that, in general, for disease prevention and control, the most cost effective strategy in intensive production systems would be the compartmentalization, due to the extensive areas covered, whereas for small-scale production systems, such as in southern Italy or in north-western Bulgaria, zoning would be more effective. The specific topology of pig trade in such areas could also be used to implement risk-based interventions for disease prevention or better control in case of an epidemic. Indeed, only few premises create a bridge between communities , and these premises could be targeted to implement control and surveillance measures . Results presented in this study have been obtained considering complete pig trade networks of four different EU countries. The aim of this study was to better understand the complex pig network organization, topology, and structure of the most representative pig production systems present in the EU, including small-scale and outdoor. However, we used data only from 1 year. We do agree that seasonality and reproducibility of results over different years is the key to be able evaluate the validity of our results and its usefulness to inform disease spread models and risk-based interventions . Those aspects might be particularly sensible in small-scale production systems where production is known to be seasonal . For that reason, we did check the reproducibility between years with some additional information we had available for France, Italy, and Bulgaria , and we found that results were similar among years for the following parameters: shipment sizes, distances, contact matrix per type of premises, and network topology . The largest communities selected in this paper were also stable over the years, covering globally the same geographical area in the different years . We also observed that industrial premises did not show a strong seasonality, whereas in small-scale pig production systems , pig movements had strong seasonal variations clustered in specific time periods but with seasonal patterns repeated yearly . Therefore, we believe that even with information from 1 year, results are valuable to inform disease spread models and risk-based interventions. Moreover, most of disease spread models usually inform their parameter values using year-level data . Nevertheless, other studies should be conducted to address more in detail the seasonality and temporal patterns and characteristics of the pig movement network of different production systems. Particularly, we recommend to explore whether or not the frequency of shipments, the geographical dispersion of the communities and the premises that create bridge between communities are concordant for different years.Cardiovascular disease makes up 34% of smoking-related mortality and smoking accounts for 16% of all cardiovascular mortality . These estimates do not include the effects of passive smoking on cardiovascular disease ; the relative risk of death from coronary heart disease for passive smokers is 1.25 .