Skip to main content
NCBI home page
The Canadian Veterinary Journal logoLink to The Canadian Veterinary Journal
. 2020 Aug;61(8):853–859.

Descriptive network analysis of a Standardbred horse training facility contact network: Implications for disease transmission

Tanya M Rossi 1,✉, Rachael M Milwid 1, Alison Moore 1, Terri L O’Sullivan 1, Amy L Greer 1
PMCID: PMC7350062  PMID: 32741991

Abstract

Infectious respiratory disease is a common cause of morbidity among racehorses. Quantification of contact patterns in training facilities could help inform disease prevention strategies. The study objectives were to: i) describe the contact network among horses, locations, and humans at a Standardbred horse training facility in Ontario; ii) describe the characteristics of highly influential individuals; and iii) investigate how management changes alter the network metrics and discuss the potential implications for disease transmission. Proximity loggers detected contacts among horses, staff, and locations (n = 144). Network metrics and node centrality measures were described for a 2-mode and horse-only contact network. The 2-mode network density was 0.16. and the median node degree was 20 [interquartile range (IQR) = 12 to 27]. Yearlings and floating staff were most influential in the network suggesting biosecurity programs should emphasize reducing contacts in these groups. Removing highly influential staff or co-housing of age groups resulted in changes to network diameter and density.

Introduction

Infectious respiratory disease is a common morbidity in equine populations and is a frequent cause of missed training days in young racehorses (1–3). In addition, acute and chronic complications such as bronchopneumonia, equine asthma, and chronic obstructive pulmonary disease increase veterinary costs and wastage of horses (4–6). Therefore, the implementation of evidence-based disease prevention programs is essential to both horse welfare and economic returns.

Horse to horse transmission of respiratory disease occurs primarily through direct contact and inhalation of respiratory droplets from infected animals (7). Indirect spread through fomites by humans and barn locations may further facilitate disease spread (7). Therefore, knowledge of the within-facility contact network, including contacts that occur among horses, barn staff, and locations, is an important component of disease prevention and control protocols. Assuming that all individuals in a network mix homogeneously and have similar contact rates may lead to less efficient use of resources and less effective disease prevention programs. In addition, identifying groups of animals, or clusters, that are highly interconnected but have few connections to other horses within the facility may support management changes which would effectively isolate these clusters and limit disease spread.

Contact network analysis has been used to identify highly influential members within a population or high-traffic locations in a facility which can be targeted for disease prevention and surveillance strategies (8–11). For example, vaccine coverage can be optimized in subpopulations with the highest contact rates, and high traffic locations can be targeted for more frequent or stringent disinfection. Previous equine contact network studies have concentrated on the transport of horses among rather than within facilities (8,12–15). These studies have relied on owner surveys and data from government movement registries. Recent papers by Milwid et al (16,17) described the use of radio frequency identification tags to characterize contact patterns within pleasure and competition stables. This approach directly quantified the frequency and duration of contacts among network members thereby avoiding common limitations of owner reported data. However, differences in management practices and facilities among sectors of the equine industry may represent a limitation when extrapolating these results to racehorse populations. For example, in contrast to pleasure horses, Standardbred racehorses in Ontario spend most of their time housed in stalls, are usually turned out alone, and are managed by multiple staff members. These differences may impact the pattern and extent of disease spread. In addition, Milwid et al (16,17) did not identify the demographics of highly influential sources of indirect disease transmission, such as staff or barn location. Analysis of a 2-mode network including sources of direct (horses) and indirect (humans and locations) transmission, and inclusion of demographic data will allow identification of appropriate targets for disease prevention strategies.

The objectives of this study were to i) describe the 2-mode contact network of horses, locations, and humans, and the 1-mode horse only network in a multi-barn Standardbred training facility in southern Ontario; ii) describe the characteristics of highly influential nodes within the 2-mode network which could be targeted for additional biosecurity measures; and iii) investigate how management changes alter the network metrics and to discuss the potential implications for disease transmission.

Materials and methods

Study population

A multi-barn Standardbred training facility in southern Ontario was recruited by convenience sampling. The 300-stall facility consisted of a free-standing main barn, and 12 shed-row style barns with paired shed-rows attached through a shared wash stall. The property included a half-mile dirt training track, electronic hot-walker, and 10 paddocks for individual turnout. This facility is typical of Standardbred training facilities in southern Ontario where horses are usually housed in multiple adjacent barns or shed-rows, and amenities such as wash stalls, paddocks and training tracks, are shared among trainers.

All horses managed by the participating owner were included in this study (n = 82) and were housed in the main barn and shed-rows 3, 4, and 9. This was an open population consisting of yearling racehorses (16 to 22 mo old) purchased by private sale or auction, mature (2 to 5 y old) active racehorses, and a companion pony. Horse names, age, sex, gait, and stall location were provided through owner/trainer history and research team observation. Yearlings were housed in the main barn and shed-row 3, and mature racehorses were stabled in shed-rows 3, 4, and 9. The companion pony was housed both with a yearling and alone during the data collection period. Horses were occasionally moved between stalls and barns to accommodate management logistics.

The remaining shed-rows were either vacant or leased to independent trainers who did not participate in the study. These trainers managed approximately 40 to 50 horses separately from the study population. As each trainer had access to separate feed and water sources, horse harnessing areas, staff and equipment, contact between the study population and non-participating horses was minimal and limited to the training track and hot-walker.

In addition to the participating horses, staff members and owners working with study horses were recruited to participate in the study. Staff were given a written description of the study aims, data collection process (how contact data were collected, processed, and securely stored), and potential risks involved with study participation. A member of the research team reviewed the study information with participants and answered questions. Staff members gave written, informed consent and were able to withdraw from the study at any time. The study protocol and animal use were approved by the University of Guelph’s Research Ethics Board (REB #16AP009) and Animal Care Committee (AUP #3518).

Network data collection

Open Beacon radiofrequency identification (RFID) tags (Bitmanufactory, Cambridge, United Kingdom) were used to detect contacts within a 2-meter proximity between members of the study network. A detailed description of the validation and use of RFID technology in horses has been reported by Milwid et al (16). Briefly, tag firmware was modified so that when 2 tags came in contact, the unique tag IDs, timing, and duration of the contact were recorded and stored on the tags’ 8 MB flash drive. The RFID tags sense the proximity to other active tags between 0.5 to 6 m; however, detection distance can be set by the researcher to suit the research question. For this study the detection distance was set to ≤ 2 m, as we were interested in the contact network in the context of respiratory disease transmission. Infectious respiratory disease is transmitted primarily through direct contact or through inhalation of respiratory droplets, necessitating close physical (face to face) proximity between horses (7).

The RFID tags were activated following the insertion of a CR2032 coin battery immediately before deployment and were placed in a water-resistant plastic envelope (Clip style name tag #74541; Avery Products, Whitby, Ontario). Each participating individual (node) in the network was assigned a unique 32 integer tag ID to facilitate horse, location, and staff identification. Tags were secured to the nose band of horses’ halters with Vetrap (3M United States, St. Paul, Minnesota, USA) and left to run for the duration of the study (23 d). Staff attached tags to their work clothes using a clip style name tag holder (Avery Products). In addition to horses and staff, tags were placed at high-traffic facility locations including: doors (n = 9), wash-stalls (n = 5), crossties (n = 17), paddocks (n =10), the hot walker (n = 1), and the track entrance (n = 1).

Tags were removed after 23 days and contact data were uploaded to a MySQL database (Oracle Corporation, Redwood Shores, California, USA) using a proximity tag programmer-device. Data were separated by study day and exported as comma separated value (CSV) files.

Data analysis

Established terminology was used to describe the networks in this study. Individuals within a contact network are referred to as “nodes” and the connection between nodes as an “edge” (18). Networks are described as “directed” when the relationship between nodes is directional. For example, the transmission of disease from an infected animal to a susceptible animal. If edges are not directional the network is described as “undirected.” Observed networks in this study were considered to be undirected as edges represented contact and the potential to transmit disease between nodes was bi-directional. Networks can be 1-mode, in which all nodes belong to the same “set” (e.g., horses); or 2-mode, where nodes can belong to different sets [e.g., horses and sources of indirect disease transmission (staff and locations)] (19).

A 2-mode network including horse, staff, and location nodes, and a 1-mode horse-only network were generated from the data using the igraph package in R Statistical Software (https://www.r-project.org) (20). Location and staff nodes were grouped together as sources of indirect disease transmission. Contact data for the study period were summed and contacts of less than 30 s duration were removed as the probability of disease transmission during these brief contacts was considered low.

The relative importance of nodes in the network was assessed by measures of centrality, including: degree, defined as the number of connections for a given node; strength, defined as the total duration in minutes of contact for a given node; betweenness, defined as the frequency a node is on the shortest path between 2 nodes in the network; and eigenvector centrality, which “measures the importance of a node in the network by assigning its score relative to its connections to others, so that high-scoring neighbors of a node will contribute more to its individual score” (15,20). Highly influential nodes were identified by ranking nodes by centrality scores and determining the demographic information of the 10 highest ranking. The same method was used to identify nodes of low importance in the network.

Network density, defined as: “the proportion of connections among nodes in the network relative to the total number of possible connections,” and clustering coefficient, which represents the proportion of connected nodes which are also connected to one another, were calculated to determine the level of cohesion in the network (18). Modularity analysis was performed using the “cluster fast greedy” function in igraph, which identifies node clusters within the network by adding or removing nodes from a potential cluster to optimize the modularity score or edge density within the cluster (21). Network diameter, defined as the greatest number of nodes on the shortest path between 2 non-adjacent nodes, was also determined (18).

These network metrics have important implications for infectious disease dynamics. For example, the speed of disease spread is typically faster in highly clustered networks with small diameters (18,22), while the magnitude of epidemics (i.e., the number of clinical cases in an outbreak) is influenced by the interconnectivity represented by network density and node degree (18).

Modified networks

Modification of the existing networks as a result of policy or management changes can have implications for the magnitude and speed of disease spread within the population by increasing or decreasing the opportunities for transmission. Therefore, we tested the impact on the 1- and 2-mode networks of 3 management changes: i) removing the companion pony from the network; ii) removing co-housing of yearlings, the companion pony, and mature racehorses; and iii) eliminating highly influential floating staff (node degree > 85). For the first and third scenarios, the nodes of interest (the pony or staff members) and all associated edges were removed from the network and the network and node metrics were recalculated. For the second scenario, horses in shed-row 3 (the only barn in which both age groups were housed together) and the companion pony, were excluded from the existing networks and the network and node metrics were recalculated.

Results

The horse population in our study consisted of 37.2% mares, 14.1% geldings, and 48.7% stallions. Horse age ranged from 1 to 5 y, with a mean of 1.6 y ± 1 standard deviation (SD).

A static 2-mode undirected network, consisting of 140 nodes (84 horses, 12 staff, and 44 locations) and 1572 edges was generated from the contact data. Data from 1 horse were excluded as its halter and attached RFID tag were removed and moved through the barns by a staff member. Damage to 6 tags (3 horse and 3 location tags) prevented researchers from downloading data; however, contacts with these tags were captured by other active tags before damage occurred. Tag failure occurred in 1 horse tag which precluded data collection. In 3 cases staff members lost or did not return their assigned tags.

Network analysis was limited to the initial 7 d of the data collection period as a sudden decrease in ambient temperature on study day 8 resulted in battery failure in 15% of the RFID tags. Further battery failure occurred over the course of the study until all tag batteries were fully drained by study day 23.

Two-mode network

The unweighted median degree was 20 contacts per node [interquartile range (IQR) = 12 to 27]. Eigenvector centrality for each node ranged from 0 to 0.41 (median = 0.002). Median betweenness score was 50 (IQR = 5.8 to 177.3). Measures of node centrality were consistently highest for staff members (floating staff ), trainers, and owners who worked in multiple barns (supplementary tables available from the author). Similarly, among static tags, crossties were the most influential in the network while paddocks had the least influence. Network density was 0.16 and the clustering coefficient was 0.44 for the whole network. The network diameter was 7. Modularity analysis revealed 10 clusters within the whole network with a modularity score of 0.8. A network diagram of a 2-mode network of horses and staff is shown in Figure 1. Static tags were removed from the diagram to increase clarity of the graph.

Figure 1.

Figure 1

Open in a new tab

A 2-mode undirected contact network of horses and sources of indirect disease transmission (staff and barn locations) at a Standardbred horse training facility in southern Ontario. Connections between members of the network represent physical contact or proximity within ≤ 2 m of ≥ 5 min duration within a 7-day period. Static tags (barn locations) were removed from the network to increase graph clarity.

Horse-only network

Removing the staff and location nodes (nodes representing indirect vectors for potential disease transmission) from the previous network to generate a horse-only network, resulted in significantly lower network metrics. The unweighted median degree was 10 contacts per horse (IQR = 8 to 14). The median betweenness and eigenvector centrality scores for the horse-only network were 46 (IQR = 0 to 172.8) and 1.8e−5 (IQR = 2e−6 to 6e−4), respectively. Yearlings, especially those housed next to doors or wash stalls, were more influential in the network than were mature racehorses (supplementary tables available from the author). The horse-only network density was 0.13 and the clustering coefficient was 0.48. The network diameter was 5. Modularity analysis revealed 9 clusters within the horse-only network with a modularity score of 0.74. The horse-only network is shown in Figure 2.

Figure 2.

Figure 2

Open in a new tab

An undirected contact network of yearling and mature racehorses at a Standardbred horse training facility in southern Ontario. Connections between horses in the network represent physical contact or proximity within ≤ 2 m of ≥ 30 s duration within a 7-day period.

Modified networks

The effect of possible management changes on network metrics and their implications for disease transmission are summarized in Table 1. No changes in network or node metrics were observed when the companion pony was removed from the 1- and 2-mode networks. Removing co-housing of age groups changed the network diameter of the 1 network. The diameter of the modified 1-mode network was 9 (versus 5 for the unmodified network). The clustering coefficient and modularity of the networks were similar between the modified and unmodified networks. An increase in 2-mode network density was observed in the network without co-housing from 0.16 to 0.18.

Table 1.

A comparison of network metrics from observed and modified contact networks within a Standardbred horse training facility in southern Ontario and the implications for the spread of directly transmitted diseases.

Whole network Horse-only network Implications for disease transmission
Observed network Network density = 0.16
Network diameter = 7		 Network density = 0.13
Network diameter = 5
Network modifications
 Remove companion pony No change in any metrics examined No change in any metrics examined None
 Remove co-housinga No change in any metrics examined Network density = 0.13
Network diameter = 9		 Increasing network diameter slows the spread of a directly transmitted disease within the population (22).
 Remove highly influential staff b Network density = 0.14
Network diameter = 7		 N/A Decreased opportunities for disease transmission, thereby reducing the magnitude of an epidemic.
Open in a new tab
a

Removed co-housing of yearlings, mature racehorses, and a companion pony.

b

Removed 3 highly influential staff members (with degree > 85) who worked in multiple barns within the training facility.

N/A — Not available.

Three horses acted as bridges between barns and shed-rows in the 1-mode network. A mature racehorse acted as the single bridge between shed-row 4 (housing mature horses) and barns 1 and 2 (housing only yearlings). Two mature racehorses acted as bridges between shed-rows 4 and 9, which both housed only mature racehorses. As there were no direct connections between horses in shed-row 9, and barns 1 and 2, removal of these bridges fragmented the network into 3 subpopulations corresponding to barns 1 and 2, shed-row 4, and shed-row 9. Removing highly influential floating staff decreased the network density from 0.16 to 0.14 and median node degree decreased to 19 (IQR = 11 to 23.2), implying there was less interconnectivity in the network.

Discussion

This study described a 2-mode contact network between horses and sources of indirect disease transmission (staff and locations) at a multi-barn Standardbred horse training facility in November 2017. In addition, a 1-mode network of horse-to-horse contacts was generated, and the network metrics reported. The study population consisted of horses at different stages of their active racing career, from the initial yearling training period to veteran racehorses participating in weekly races. In addition, both trotters and pacers, considered to be distinct subpopulations within the Standardbred breed, were included in the study which increased the external validity and ability to extrapolate the results to other racehorse facilities. Unfortunately, due to battery failure, network analysis was limited to the initial 7 d of the data collection period to prevent surviving tags from having inflated centrality measures compared to tags that failed early. The facility’s training and competition schedule was consistent over the study period; therefore, the impact of restricting the analysis period was likely minor.

The network densities of the 2-mode and 1-mode networks were 16% and 13%, respectively. These results are on the lower end of the range previously described in pleasure and performance barns using RFID technology (9% to 62%) (17). These results are expected as racehorse facilities are managed differently than pleasure or performance horses. For instance, the stables described in Milwid et al (17) pastured horses in groups for several hours a day, while racehorses are typically pastured individually for short periods. In addition, Standardbred racehorse facilities commonly house horses in multiple barns, decreasing the opportunity for some horses in the network to come in contact, while barns with pleasure horses usually house all horses in a single barn.

The occurrence of network clustering was assessed at the 1-mode and 2-mode level through calculation of the coefficient of clustering and identification of observed clusters using the fast-greedy method. The coefficient of clustering is calculated by determining the likelihood that a horse’s neighbors are neighbors of each other (18). A coefficient of 0 suggests there is no clustering in the network and edges occur homogenously throughout the network. A coefficient of 1 means that nodes in the network are separated into discrete clusters. Our 2-mode network had a clustering coefficient of 0.5 which shows moderate clustering with contact occurring between different clusters. The horse-horse network had a higher clustering coefficient of 0.84 which indicates horses had fewer connections between clusters. These results suggest that infectious disease has the potential to spread to different barns in the facility, and that humans may play an important role in spreading disease between clusters. Similarly, when analyzing the networks using the fast-greedy method the 2-mode and 1-mode network were organized into 10 and 9 clusters, respectively. The density of edges within clusters throughout the entire network is represented by a modularity score, which was 0.8 and 0.74 for the 2-mode and 1-mode networks, respectively. These scores suggest cluster density was similar for the 2-mode and 1-mode networks and that location and staff nodes had little impact on node connectivity within clusters. These results show that the horses in the network did not mix in a homogeneous manner and that focusing biosecurity efforts towards specific groups for increased disease prevention protocols may lead to better biosecurity and a healthier equine population.

As equine respiratory disease transmission occurs primarily through direct contact, spread within a training center may be facilitated by mixing yearlings with mature racehorses, as it provides greater opportunities for contact between age groups (7). Therefore, one approach to disease control would be to cohort horses by age, thereby potentially fragmenting the network and reducing the spread of disease throughout the population. Cohorting by age group has been used effectively in all-in-all-out swine production and has successfully reduced transmission of respiratory infections to younger stock (23,24). A similar strategy has been proposed for sheep attending agricultural shows, with show eligibility limited by geographic region resulting in increased fragmentation of the network (25).

In the 2-mode network, eliminating co-housing from the network did not affect most network metrics. Staff acted as bridges between clusters and prevented any increase in fragmentation of the network. However, removing mixed housing increased the diameter of the 1-mode network, suggesting that although this management change would not decrease the number of cases in an epidemic, the speed of disease spread would be slowed. The effect of network structure on disease transmission has been previously investigated using mathematical models of epidemics which show that disease spreads more rapidly in small diameter networks (22). This change in epidemic dynamics (e.g., a similar number of total clinical cases but spread out over a longer period of time) would allow more time for other disease control interventions (e.g., isolating cases, or vaccinating at-risk populations) to be initiated and thereby have a greater impact on reducing the overall magnitude of an outbreak.

When co-housing was eliminated, a single mature racehorse acted as a bridge between the yearling and racehorse populations. The interaction between this horse and the yearling population may have occurred during use of shared amenities, such as the training track or co-transport in the same trailer. Further efforts to separate the 2 age groups, such as timing use of amenities to prevent co-mingling of age groups, may result in additional fragmentation of the network, limiting the risk of disease transmission to a smaller sub-population within the facility.

Node influence within the unmodified 2-mode network was assessed using a ranking of node centrality measures. The nodes with the highest degree and eigenvector centrality belonged to staff members, especially staff who worked in multiple barns in the facility like owners, trainers, and drivers. Therefore, staff members act as important vectors of disease across the network, assuming humans can act as mechanical vectors for a particular pathogen; for example Streptococcus equi subsp. equi and influenza H3N8 (7,26). Therefore, measures to reduce disease transmission could include cohorting staff by barn or to a smaller group of horses to reduce contacts and prevent disease spreading to other barns in the facility. Cohorting of staff and/or patients has been used effectively in human hospitals to prevent the spread of endemic or epidemic disease (27–29).

Removing highly influential floating staff from the 2-mode network reduced the density and median node degree, thereby reducing interconnectivity. As the nodes in this network have fewer connections there are less opportunities for disease transmission and the magnitude of an epidemic may be reduced (18). Since floating staff, such as trainers and owners, are essential to the running of training facilities, removing them entirely is not a realistic management strategy. However, there are strategies that may minimize their impact on disease spread. Biosecurity standards in swine production recommend that staff care for groups of pigs in order from youngest to oldest, as the risk of contracting infectious disease is greatest in young pigs versus adults (30). A similar practice would be feasible in Standardbred training facilities. Yearlings could be trained and cared for first, and then mature racehorses, reducing the spread of infection from older to younger animals. If this is not possible then sanitation protocols (e.g., use of foot baths, hand washing, and changing equipment between barns) should be put in place for floating staff to reduce their role in disease transmission (30).

Crossties and doors had a high influence on the network due to heavy use by horses and staff within a barn, which may also increase disease spread by fomite contamination. Including frequent disinfection of crossties or restricting crosstie use to smaller groups of horses into a biosecurity plan is therefore recommended. Centrality measures among horses were highest among yearlings in training. This result is likely due to the intensive management and high housing density of horses in barns 1 and 2. Yearlings often require frequent interactions with barn staff upon arrival and during the initial training period. Previous reports, both in this population and others, have found that young horses (≤ 2 y old) are more susceptible to respiratory infections than older horses, likely due to an immature immune system and the stress of early training (8,31–32). Our findings suggest that the high contact rate in yearling training barns may also contribute to the higher probability of disease in this age group (32). Grouping yearlings by birth month or entry into the facility from sales may help to buffer the impact of contact rates on disease transmission. Horses stabled next to doors and wash stalls had higher centrality measurements which was not surprising as these are high traffic locations. Placing isolation stalls away from these areas may reduce the likelihood of introducing new infections from recently introduced animals into the barn population.

Interestingly, the companion pony at the facility had the 7th highest betweenness score among all horses in this network. The pony acted as a companion to several yearlings in barns 1, 2, and 3 during the study period and may have acted as a bridge for disease transmission between barns and horse clusters. However, removing the pony from the population in our modified network did not change the network metrics as there was high interconnectivity and many other bridges existed between clusters. As only 1 pony was included in our study, further interpretation of the influence of companion animals on the network was not possible. However, a study by Morley et al (31) found that exercise ponies were at increased risk for influenza-mediated respiratory disease in a Thoroughbred training yard. The authors hypothesized that this may be due to the number and duration of contact between these horses and young racehorses. The role of companion animals in disease transmission in racehorse facilities should be further explored to determine the biosecurity risk posed by these animals. However, this risk should then be weighed against the potential welfare benefits companion animals provide, such as environmental enrichment and opportunities for socialization. In the future, the network composition of equine facilities should be considered when developing biosecurity programs.

Acknowledgments

The authors thank Karen Richardson, Wendy Xie, Roksolana Hovdey, Cassandra Campbell, and Dr. Emma Gardner for their help with study deployment and data collection. We are also grateful for the participation of the Standardbred horse training center’s owners, trainers, and staff, without whom this study would not have been possible. CVJ

Footnotes

Use of this article is limited to a single copy for personal study. Anyone interested in obtaining reprints should contact the CVMA office (hbroughton@cvma-acmv.org) for additional copies or permission to use this material elsewhere.

Funding for this project was generously provided by the Ontario Ministry of Agriculture, Food and Rural Affairs (OMAFRA) and the Natural Sciences and Engineering Research Council of Canada (NSERC).

References

  • 1.Bailey CJ, Rose RJ, Reid SWJ, Hodgson DR. Wastage in the Australian Thoroughbred racing industry: A survey of Sydney trainers. Aust Vet J. 1997;75:64–66. doi: 10.1111/j.1751-0813.1997.tb13836.x. [DOI] [PubMed] [Google Scholar]
  • 2.Powell DG, Burrows R, Goodridge D. Respiratory viral infections among Thoroughbred horses in training during 1972. Equine Vet J. 1974;6:19–24. doi: 10.1111/j.2042-3306.1974.tb03922.x. [DOI] [PubMed] [Google Scholar]
  • 3.Gilkerson JR, Bailey KE, Diaz-Méndez A, Hartley CA. Update on viral diseases of the equine respiratory tract. Vet Clin North Am Equine Pract. 2015;31:91–104. doi: 10.1016/j.cveq.2014.11.007. [DOI] [PubMed] [Google Scholar]
  • 4.Timoney PJ. Equine influenza. Comp Immunol Microbiol Infect Dis. 1996;19:205–211. doi: 10.1016/0147-9571(96)00006-9. [DOI] [PubMed] [Google Scholar]
  • 5.Wood JLN, Newton JR, Chanter N, Mumford JA. Association between respiratory disease and bacterial and viral infections in British racehorses. J Clin Microbiol. 2005;43:120–126. doi: 10.1128/JCM.43.1.120-126.2005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Willoughby R, Ecker G, McKee S, et al. The effects of equine rhinovirus, influenza virus and herpesvirus infection on tracheal clearance rate in horses. Can J Vet Res. 1992;56:115–121. [PMC free article] [PubMed] [Google Scholar]
  • 7.Myers C, Wilson WD. Equine influenza virus. Clin Tech Equine Pract. 2006;5:187–196. [Google Scholar]
  • 8.Sanchez-Matamoros A, Martinez-Lopez B, Sanchez-Vizcaino F, Sanchez-Vizcaino J. Social network analysis of equidae movements and its application to risk-based surveillance and to control of spread of potential equidae diseases. Transbound Emerg Dis. 2013;60:448–459. doi: 10.1111/j.1865-1682.2012.01365.x. [DOI] [PubMed] [Google Scholar]
  • 9.Frössling J, Ohlson A, Björkman C, Håkansson N, Nöremark M. Application of network analysis parameters in risk-based surveillance — Examples based on cattle trade data and bovine infections in Sweden. Prev Vet Med. 2012;105:202–208. doi: 10.1016/j.prevetmed.2011.12.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Christley RM, Pinchbeck GL, Bowers RG, et al. Infection in social networks: Using network analysis to identify high-risk individuals. Am J Epidemiol. 2005;162:1024–1031. doi: 10.1093/aje/kwi308. [DOI] [PubMed] [Google Scholar]
  • 11.Hayama Y, Kobayashi S, Nishida T, Muroga N, Tsutsui T. Network simulation modeling of equine infectious anemia in the non-racehorse population in Japan. Prev Vet Med. 2012;103:38–48. doi: 10.1016/j.prevetmed.2011.09.011. [DOI] [PubMed] [Google Scholar]
  • 12.Rosanowski SM, Cogger N, Rogers CW, Bolwell CF, Benschop J, Stevenson MA. Analysis of horse movements from non-commercial horse properties in New Zealand. N Z Vet J. 2013;61:245–253. doi: 10.1080/00480169.2012.750571. [DOI] [PubMed] [Google Scholar]
  • 13.Bell IG, Drury-Klein C. Analysis of horse movements and events during the 2007 outbreak of equine influenza in New South Wales, Australia. Aust Vet J. 2011;89(SUPPL1):63–68. doi: 10.1111/j.1751-0813.2011.00775.x. [DOI] [PubMed] [Google Scholar]
  • 14.Spence KL, O’Sullivan TL, Poljak Z, Greer AL. Descriptive and network analyses of the equine contact network at an equestrian show in Ontario, Canada and implications for disease spread. BMC Vet Res. 2017;13:191. doi: 10.1186/s12917-017-1103-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Spence KL, O’Sullivan TL, Poljak Z, Greer AL. A longitudinal study describing horse demographics and movements during a competition season in Ontario, Canada. Can Vet J. 2018;59:783–790. [PMC free article] [PubMed] [Google Scholar]
  • 16.Milwid RM, Sullivan TLO, Poljak Z, Laskowski M, Greer AL. Validation of modified radio-frequency identification tag firmware, using an equine population case study. PLoS One. 2019;14:e0210148. doi: 10.1371/journal.pone.0210148. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Milwid RM, O’Sullivan TL, Poljak Z, Laskowski M, Greer AL. Comparison of the dynamic networks of four equine boarding and training facilities. Prev Vet Med. 2019;162:84–94. doi: 10.1016/j.prevetmed.2018.11.011. [DOI] [PubMed] [Google Scholar]
  • 18.Dubé C, Ribble C, Kelton D, McNab B. A review of network analysis terminology and its application to foot-and-mouth disease modelling and policy development. Transbound Emerg Dis. 2009;56:73–85. doi: 10.1111/j.1865-1682.2008.01064.x. [DOI] [PubMed] [Google Scholar]
  • 19.Borgatti SP, Everett MG. Network analysis of 2-mode data. Soc Networks. 1997;19:243–269. [Google Scholar]
  • 20.Csárdi G, Nepusz T. The igraph software package for complex network research. InterJournal Complex Syst. 2006. [Last accessed February 14, 2020]. p. 1695. Available from: https://pdfs.semanticscholar.org/1d27/44b83519657f5f2610698a8ddd177ced4f5c.pdf.
  • 21.Newman MEJ. Fast algorithm for detecting community structure in networks. Phys Rev E. 2004;69:1–5. doi: 10.1103/PhysRevE.69.066133. [DOI] [PubMed] [Google Scholar]
  • 22.Watts D, Strogatz SH. Collective dynamics of ‘small-world’ networks. Nature. 1998;393:440–442. doi: 10.1038/30918. [DOI] [PubMed] [Google Scholar]
  • 23.Scheidt AB, Cline TR, Clark LK, et al. The effect of all-in-all-out growing-finishing on the health of pigs. J Swine Health Prod. 1995;3:203–205. [Google Scholar]
  • 24.Davies PR, Morrow WEM, Deen J, Gamble HR, Patton S. Seroprevalence of Toxoplasma gondii and Trichinella spiralis in finishing swine raised in different production systems in North Carolina, USA. Prev Vet Med. 1998;36:67–76. doi: 10.1016/s0167-5877(98)00072-5. [DOI] [PubMed] [Google Scholar]
  • 25.Webb C. Investigating the potential spread of infectious diseases of sheep via agricultural shows in Great Britain. Epidemiol Infect. 2006;134:31–40. doi: 10.1017/S095026880500467X. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Boyle AG, Timoney JF, Newton JR, Hines MT, Waller AS, Buchanan BR. Streptococcus equi infections in horses: Guidelines for treatment, control, and prevention of strangles — revised consensus statement. J Vet Intern Med. 2018;32:633–647. doi: 10.1111/jvim.15043. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Handwerger S, Beth Raucher B, David Altarac D, et al. Nosocomial outbreak due to Enterococcus faecium highly resistant to vancomycin, penicillin, and gentamicin. Clin Infect Dis. 1993;16:750–755. doi: 10.1093/clind/16.6.750. [DOI] [PubMed] [Google Scholar]
  • 28.Rosenberger LH, Hranjec T, Politano AD, et al. Effective cohorting and “superisolation” in a single intensive care unit in response to an outbreak of diverse multi-drug-resistant organisms. Surg Infect (Larchmt) 2011;12:345–350. doi: 10.1089/sur.2010.076. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Laurent C, Rodriguez-Villalobos H, Rost F, et al. Intensive care unit outbreak of extended-spectrum beta-lactamase-producing Klebsiella pneumoniae controlled by cohorting patients and reinforcing infection control measures. Infect Control Hosp Epidemiol. 2008;29:517–524. doi: 10.1086/588004. [DOI] [PubMed] [Google Scholar]
  • 30.Canadian Swine Health Board Technical Committee on Biosecurity. National Swine Farm-Level Biosecurity Standard. Canadian Swine Health Board. 2010. [Last accessed June 5, 2020]. Available from: http://www.swinehealthontario.ca/Portals/15/documents/resources/Biosecurity/Biosecurity%20Guides/CSHB_Biosecurity_StandardE.PDF?ver=2018-10-01-172057-063.
  • 31.Morley PS, Townsend HGG, Bogdan JR, Haines DM. Risk factors for disease associated with influenza virus infections during three epidemics in horses. J Am Vet Med Assoc. 2000;216:545–550. doi: 10.2460/javma.2000.216.545. [DOI] [PubMed] [Google Scholar]
  • 32.Rossi TM, Moore A, O’Sullivan TL, Greer AL. Equine Rhinitis A Virus infection at a Standardbred training facility: Incidence, clinical signs, and risk factors for clinical disease. Front Vet Sci. 2019;6:71. doi: 10.3389/fvets.2019.00071. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from The Canadian Veterinary Journal are provided here courtesy of Canadian Veterinary Medical Association

RESOURCES