A descriptive analysis of swine movements in Ontario (Canada) as a contributor to disease spread

A descriptive analysis of swine movements in Ontario (Canada) as a contributor to disease spread

Preventive Veterinary Medicine 159 (2018) 211–219 Contents lists available at ScienceDirect Preventive Veterinary Medicine journal homepage: www.els...

945KB Sizes 0 Downloads 11 Views

Preventive Veterinary Medicine 159 (2018) 211–219

Contents lists available at ScienceDirect

Preventive Veterinary Medicine journal homepage: www.elsevier.com/locate/prevetmed

A descriptive analysis of swine movements in Ontario (Canada) as a contributor to disease spread

T



Dylan John. Melmer , Terri L. O’Sullivan, Zvonimir Poljak Department of Population Medicine, University of Guelph, ON, N1G 2W1, Canada

A R T I C LE I N FO

A B S T R A C T

Keywords: Network analysis Weak component Animal movement Swine Ontario Canada

In recent times, considerable efforts have been made to develop infrastructure and processes of tracing livestock movements. One of common use of this type of data is to assess the potential for spread of infections in source populations. The objectives of this research were to describe Ontario pig movements in 2015, and to understand the potential for disease transmission through animal movement on a weekly and yearly basis. Swine shipments from January to December 2015 represented 224 production facilities and a total of 5398 unique animal movements. This one-mode directed network of animal movements was then analyzed using common descriptive network measures. The maximum yearly (y) weak component (WCy) size and maximum weekly (w) weak component size (WCw) was 224 facilities, and 83 facilities, respectively. The maximum WCw did not change significantly (p > 0.05) over time. The maximum strong component (SC) consisted of two facilities both on a weekly, and on a yearly basis. The size of the maximum ingoing contact chain on a yearly basis (ICCy) was 173 nodes with one abattoir as the end point, and the maximum ICCw consisted of 53 nodes. The size of the maximum outgoing contact chain (OCCy) contained 79 nodes, with one sow herd as a starting point. The maximum OCCw was 6 nodes. Regression models resulted in significant quadratic associations between weekly count of finisher facilities with betweenness > 0 (p = 0.02) and weekly count of finisher facilities with in-degree and out-degree > 0 (p = 0.01) and week number. Higher weekly counts of nursery and finisher facilities with betweenness > 0 and in-degree and out-degree both > 0 values occurred during summer months. All study facilities were connected when direction of animal movement was not taken into consideration in the yearly network. As such, yearly networks are potentially representative of infections with long incubation periods, subclinical infections, or endemic infections for which active control measures have not being taken. When the direction of animal movement was considered, such infection could still spread substantially and affect 35% of the study population (79/224). In the study population, finisher sites were proportionally and consistently most represented in WCw (min = 51%, max = 78%), which reflects current Ontario herd demographics. However, abattoirs were overrepresented when the number of facilities in the study population was taken into consideration. This, and the size of the maximum ICCw both suggest that abattoirs could be, at least for some infectious diseases, suitable establishments for targeted sampling.

1. Introduction Improvements in livestock health management strategies over the past few decades have resulted in the elimination of certain swine diseases, however, despite these efforts the risk of emerging infections remains (Davies, 2012). In addition to ongoing disease challenges faced by swine populations, the movement of animals and animal by-products between countries has also increased. This increased movement has also increased the potential for disease transmission world wide



(Ayudhya et al., 2012; Davies, 2012; Wernike et al., 2013). Globally, three new species of swine infectious agents per year were reported between 1985 and 2010 (Fournié et al., 2015). For emerging and for endemic swine diseases, various pathways of dissemination have been confirmed or suspected in different populations, with animal movement documented as an important contributor to transmission (Dee et al., 2004; Dubé et al., 2009; Guinat et al., 2016; Pasick et al., 2014). Network analysis has been utilized in studies to investigate how animal movements have contributed to disease spread (Bigras-Poulin et al.,

Corresponding author at: University of Guelph, 50 Stone Road East, Population Medicine Building, Bld #174, ON, N1G 2W1, Canada. E-mail address: [email protected] (D.J. Melmer).

https://doi.org/10.1016/j.prevetmed.2018.09.021 Received 12 March 2018; Received in revised form 18 August 2018; Accepted 19 September 2018 0167-5877/ © 2018 Elsevier B.V. All rights reserved.

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

from unique facilities within the production system. Company Export was used to describe an identical type of destination for animal movement as CI, except that the destination was outside of Canada. Company External was used to describe entities that, over a period of the entire year, served as source of animals for nodes within the production system, but could also be destination for animals from the production system.

2007; Dubé et al., 2008; Natale et al., 2009; Thakur et al., 2016). Such studies vary with respect to the completeness of the network, time periods over which the networks were constructed, and measures utilized to make recommendations. In Canada, two recent studies utilized swine movement data to investigate network structure of swine populations (Dorjee et al., 2013; Thakur et al., 2016). In a study based on swine movements in four Canadian provinces, Thakur et al (2016) reported that swine farms showed high indirect connectivity via shared trucks. It has been argued that there is a high potential for disease transmission in swine populations due to its small-scale, and scale-free network topology (Dorjee et al., 2013). Despite useful information, the studies were conducted on data collected more than ten years ago. Significant disease events that recently occurred, and evolving nature of swine production systems could both have influenced the movement patterns over time. The primary objective of this study was to describe the contact structure, based on 2015 data collected from an Ontario swine data management company containing information on movements between individual farms and other production facilities located in Ontario. By addressing this objective, the potential for disease transmission through animal movement on a weekly and yearly level is hypothesized.

2.2. Data management Data management and analysis was conducted with R version 3.3.2 (R Core Team, 2015). Packages doBy (Højsgaard and Halekoh, 2016) and lubridate (Grolemund and Wickham, 2011) were used for data aggregation and date organization. One-mode networks were constructed from the movement data among all nodes using functionality available in the package igraph (Csardi and Nepusz, 2006), and were initially analyzed as a directed yearly network. Following this, weekly networks were constructed, and selected network and node characteristics were examined descriptively and analytically over time. S.Table 1 provides definitions of the descriptive network measures analyzed within this study (Dubé et al., 2009).

2. Materials and methods 2.3. Descriptive network analysis 2.1. Study population 2.3.1. Yearly network A directed one-mode network was generated for 2015. Heatmaps were constructed to visualize the number of animal movements throughout the entire year between facility types. Accompanying dendrograms were constructed using hierarchical cluster analysis based on Euclidean distances between outgoing or ingoing number of movements and the Unweighted Pair Group with Arithmetic Mean (UPGMA) methods. Network level measures of interest included the size of the strong (SCy) and weak components (WCy), density, diameter of the network, and the largest ingoing contact chain (ICCy) and outgoing contact chain (OCCy). The density measure was calculated based on a directed network. Contact chains are used to establish the temporal sequence of direct and indirect connections (i.e., animal movements) that were associated with the facility of interest (ingoing), or after the facility (outgoing) (Noremark et al., 2011); and were determined using functionality available in the package EpiContactTrace (Noemark and Widgren, 2014). Node centrality measures of interest were: in & out degree, and betweenness.

The data was obtained from an Ontario swine data management company and included all animal movements between the facilities in the study population. The data analyzed consisted of animal movements associated with commercial swine production related facilities (nodes) located in southwestern Ontario, with some movements to locations in other provinces and exports to other countries in North America. These facilities were categorized by the stages of production of the animals shipped and received and were assigned to a facility type which is described below. The study period was between January 1st and December 31st, 2015. The raw data consisted of 5398 unique movements (edges) between the nodes, with 16 descriptive variables defining edge attributes. Nodes were categorized by facility type and included traditional facility designations: sow, nursery, finisher facilities, abattoir sites. These facility names are based on the typical production and life cycle of pigs within the Ontario swine production system. When abattoirs were referred to on an individual basis, the unique facility was assigned an arbitrary designation from A1-A10.The typical swine facility types in a three-site production system can be divided into the following: (i) a sow site houses breeding animals (sows) and their offspring until the 3–4 weeks of age, when they are moved to nursery facilities, (ii) nursery facility where animals are housed, typically for up to 8 weeks, after which period they are moved to finisher sites, and (iii) finisher sites where they grow until market weight, which in this source population, typically lasts for up to 16–17 weeks. More specific classification of herd types is possible, depending which production classes are located together on the same premises. There were additional non-traditional facility classifications that were based on whether the node was a destination or a source that was external to the company’s database. The non-traditional facilities were designated a unique identifier (UI) that represented business-related entities and these entities were further categorized as: Company Internal (CI), Company External (CE), and Company Export (CEX). The details about the specific geographic locations of these entities were not part of the dataset, and consequently, each UI could represent one or more nodes grouped within it. The definitions of these three categories of entities were made based on the direction of animal movements over the period of the entire year and were then consistently applied for both yearly and weekly networks. Company Internal described nodes that are within Canada and are only a destination for movements of animals

2.3.2. Weekly network Weekly network level measures considered were strong and weak components, order, density and diameter, and largest in- (ICCw) and out-going contact chain, (OCCw). Proportions of facility types within the weekly weak components (WCw) were calculated, as well as the proportion of facility types in the entire study population that were observed in each WCw. Additionally, node centrality measures calculated were: in- and outdegree, and betweenness. Following the calculation of weekly nodeand network-level measures; selected measures were further aggregated to a weekly level to evaluate any trends over time. Specifically, the size of the largest weak and strong components, demographic characteristics of the nodes involved in the largest weak components, number of nursery and finisher sites with the betweenness > 0, and number of finisher and nursery sites where both in-degree and out-degree were > 0 in a given week were determined. 3. Statistical analysis Different types of regression were used to assess development of several measures indicative of network structure over time. Specifically, 212

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

linear regression was used to explore development of maximum weekly (weeks 1–52) weak component and total number of animals shipped over the study period. Week 53 data was dropped from all analysis a priori, with the rationale that it was an incomplete week (spanning into 2016). Nursery and finisher sites were considered separately by their location within the swine production chain, and the number of nodes with betweenness values > 0 and number of nodes with in- and outdegree > 0 in a given week were evaluated using Poisson regression. In all analyses, linear and quadratic effect of week number were used as explanatory variables. All statistical analyses were conducted through stats version 3.3.2 (R Core Team, 2015). Model fit, normality and homoscedasticity were tested where applicable.

represented 35.7% of the study population (Table 1). 4.3. Yearly Network: number of distinct animal movements

The network (total nodes n = 224) consisted of 33 sow herds (∼15%), 46 nursery facilities (∼21%), 119 finisher facilities (∼53%), and 10 abattoir sites (∼4%). Two nodes were classified as CI, ten nodes as CE and four nodes as CEX which cumulatively accounted for ∼7% of the facilities. The number of pigs shipped between facilities ranged from 1 pig to 2805 pigs with an average of 294 pigs moved within a single shipment.

The number of individual shipments between different types of facilities over the entire year are displayed in Fig. 1. The highest number of unique shipments occurred from finisher facilities to abattoir sites (n = 3208), followed by movements from sow to nursery facilities (n = 998), and from nursery to finisher facilities (n = 751). Abattoirs also received shipments from nursery facilities, although with lower frequency (n = 109), and sow facilities received 18, 28, and 63 shipments of pigs from other sow herds, nursery sites, and finisher sites, respectively. Recorded movements from external companies mostly occurred to finisher sites (n = 52), while exports outside Canada occurred from finisher (n = 14) and nursery facilities (n = 12). Abattoirs, designated A1-A10, were used at various frequencies throughout the year. The frequencies ranged from a maximum usage of 406 times with regards to A1, to a minimum usage of 1 with regards to A10. Abattoirs, Company Export and Company Internal did not send shipments of pigs to any facility type (Fig. 1); thus, in terms of shipments sent they are more closely related based on the lack of shipments sent. Contrary to this, Company Internal (CI), Company Export (CEx), and Company External (CE), received shipments at a lesser extent from other facilities and are therefore more closely related in terms of the frequency of shipments received (Fig. 1, top dendrogram).

4.2. Yearly Network: network measures

4.4. Yearly Network: node centrality

The yearly network order, or total number of unique facilities was 224 and, graph density, or total number of possible connections made was 0.02 (2%). The network diameter or the longest series of connections was also 224 nodes. Similarly, only one weak component was detected in the yearly network and it consisted of all 224 nodes. When direction of edges was considered, there were only two strong components identified in the yearly network and both of the components consisted of two nodes. The first strong component consisted of two nursery facilities, and the second strong component consisted of one nursery and one finisher facility. Abattoirs had the largest yearly ingoing contact chain with a maximum ICCy of 173 nodes associated with A5 as the terminal node. Sow herds had the largest yearly outgoing contact chain and one sow herd had the maximum OCCy of 79 nodes which also indirectly shipped to all abattoirs recorded in this study population. Both ICCy and OCCy had a median value of 5, while the interquartile range varied from 2 to 10 nodes and 2 to 15 nodes, respectively. The facility demographics of the largest OCCy and ICCy are displayed in Table 1. The largest ICCy represented 77.7% of the study population, and the largest OCCy

Mean degree and betweenness values for the yearly directed network are presented in Table 2. Abattoirs had the highest average indegree (mean = 48.7, max = 110.0, SD = 41.0), while the highest average out-degree was detected in nursery facilities (mean = 11.0, max = 18.0, SD = 4.1). Betweenness was highest for nursery facilities followed by finishers (Table 2).

4. Results 4.1. Study population

4.5. Weekly Network: network measures Mean density of the weekly networks was 0.002 (SD = 0.0003) and ranged from 0.0003 to 0.0021. The order of the networks was 224, and weekly network diameter ranged across all 53 weeks between one and four nodes. The linear regression model using the total animals received per week as the dependent variable, demonstrated that across the entire year there was a weekly increase of the number of pigs transported (β = 91.8, SE = 26.39, Intercept = 27,237.38, p = 0.001). The normality assumption of residuals was satisfied (Shapiro-Wilks, p = 0.71), along with homoscedasticity (Breusch-Pagan test, p = 0.14).

Table 1 Total count of facility types observed in the largest yearly outgoing contact chain (OCC = 79) and in the largest ingoing contact chain (ICC = 173) during 2015 among 224 swine establishments, based on data obtained from an Ontario swine data management company. Total facility count represents largest yearly OCC and ICC with origin and destination sites, respectively, included in the count. Facility Type

Count within OCC

OCC percent occurrencea

OCC relative percent occurrenceb

Count within ICC

ICC percent occurrencea

ICC relative percent occurrenceb

Total in study population

Abattoir CEc CEXc CIc Finisher Nursery Sow Total (OCC/ICC)

10 0 1 1 52 15 1 (start) 80

12.5% 0.0% 1.25% 1.25% 65.0% 18.75% 1.25% 100%

100.0% 0.0% 25.0% 50.0% 43.7% 32.6% start 35.7%

1 (end) 3 0 0 108 44 18 174

0.57% 1.72% 0.0% 0.0% 62.06% 25.30% 10.35% 100%

end 30.0% 0.0% 0.0% 90.8% 95.7% 54.5% 77.7%

10 10 4 2 119 46 33 224

a b c

Calculated as % of all facility types in the OCC or ICC. Calculated as % of all facility types in the study population. CE=Company External, CEX= Company Export, CI=Company Internal. 213

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

Fig. 1. Heatmap of the number of distinct movements between different types of facilities for the 2015 year. Right row* names shipping to bottom column names, Fin = Finisher, Sow = Sow Barn/Facility, Nurs = Nursery, CE = Company External, CI = Company Internal, CEx = Company Export. * Dendrograms displayed on both axes depict relatedness of each facility type on the basis of ingoing (top) or outgoing (left) number of movements.)

maximum weak components, a proportion which could vary between a minimum of 40.0% and a maximum of 80.0% (Table 3). Weekly Network: node centrality Only finisher, nursery and sow sites had a betweenness value larger than zero. Of those, nursery sites had the highest average betweenness, followed by finisher and sow sites (Table 4), although finisher sites had the highest calculated maximum betweenness in a specific week. Descriptive statistics based on weekly number of direct connections, in- and out-degree, are summarized in Table 4. Abattoirs had a maximum in-degree of 38 direct connections, in a given week. The largest maximum in-degree of 38 was in A5. Finisher and nursery sites had the next highest maximum in- and out-degree values (Table 4). Number of finisher and nursery sites with both in- and out-degree values > 0 in each week appeared to change over time with higher values observed in mid-year (Figs. 5 and 6). Subsequent analysis of the number of finisher and nursery facilities with a high in- and out-degree measurements in the same week indicated a quadratic association with week number. For finisher facilities specifically, a significant quadratic relationship for the effect of week was found (p = 0.01) (Table 5) and resulted in a model that fit the data well (deviance chi-square GOF p = 0.9). When weeks 1 and 52 (highest leverage values) were removed, coefficient for quadratic effect of the week number decreased (min p = 0. 051, max p = 0.16); however, the shape of the association remained similar with peak number of facilities expected during mod-year. Additionally, week 43 was removed, which had a high Pearson residual, and the significant relationship was still

The weekly maximum weekly ingoing contact chain (ICCw) ranged from 4 to 53 nodes and had an average of 30.2 nodes (SD = 6.21). The largest ICCw was associated with a single abattoir site (A5). The maximum weekly outgoing contact chain (OCCw) ranged between 1 and 6 nodes with an average of 4.2 nodes (SD = 0.9). The maximum OCCw was identified with several origin nodes across several weeks. Weekly strong components had a max of 2 nodes at week 26. This occurred due to a bidirectional movement of animals occurring between 2 nodes in that specific week. The minimum weekly weak component size was 2 and a maximum was 83 nodes. When only considering the largest weak weekly components, weak components ranged from 37 nodes in weeks 20 and 52, to a maximum of 83 nodes in week 43, with mean maximum weak component size of 54.2 (SD = 9.5), or 24.2% of the study population (Fig. 2). Furthermore, when examined with linear regression, the size of the maximum weak component did not significantly change over time (p = 0.14). Table 3 displays the proportion of each facility type in the maximum weak weekly component (MWCw) when summarized over the entire year. For example, finisher sites on average contributed to 62.0% of all nodes in the MWCw; over the entire year; this varied from 51.0% to 78.0% (Table 4). Overall weekly weak component composition overtime can be seen in Fig. 3. However, when considering percent occurrence within a maximum weak component as a proportion of the total number of each facility type within the network (Table 3, Fig. 4), abattoirs occurred more frequently than any other facility type. On average, 53.0% of the total abattoirs in the network were part of weekly

Table 2 Descriptive statistics of node centrality measures obtained from the 2015 yearly network of swine movements among 224 swine establishments, derived from data obtained from an Ontario swine data management company. Facility Type

Abattoir CEa CEXa CIa Finisher Nursery Sow a

In-degree

Out-degree

Betweenness

Minimum

Mean

Maximum

Minimum

Mean

Maximum

Minimum

Mean

Maximum

1.0 0.0 1.0 1.0 0.0 1.0 0.0

48.7 0.6 2.3 7.0 4.1 2.7 0.7

110.0 2.0 3.0 13.0 10.0 6.0 3.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 1.5 0.0 0.0 4.3 11.0 3.5

0.0 10.0 0.0 0.0 10.0 18.0 16.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 5.3 19.7 0.1

0.0 0.0 0.0 0.0 25.3 74.4 2.0

CE=Company External, CEX= Company Export, CI=Company Internal. 214

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

Fig. 2. Development of the size of the maximum weak component over 52 weeks of the 2015 year, on the basis of data obtained from an Ontario swine production and data management company.

Statistically significant trends for the number of finisher facilities with betweenness measures > 0 over time were identified in the Poisson regression model (deviance chi-square GOF p = 0.9). Number of finisher facilities with high betweenness values had a significant quadratic relationship with time, i.e. week (p = 0.021) (Table 5), whereas such association was not statistically significant in the population of nursery facilities; a finding that was consistent with in- and out degree count values (Table 5).

Table 3 Demographic composition of the maximum weekly weak component (MWCw) calculated from the 2015 weekly networks of animal movements among 224 Ontario swine establishments and expressed as descriptive statistics of the proportion of sites in the maximum weak component, and as descriptive statistics of the proportion of sites in the study population (relative proportion). Facility Type

Mean

Maximum

Minimum

Standard Deviation

MWC Proportion Finisher Nursery Abattoir Sow CEa CEXa CIa

0.62 0.16 0.10 0.10 0.01 0.00 0.00

0.78 0.23 0.16 0.17 0.04 0.02 0.02

0.51 0.04 0.06 0.00 0.00 0.00 0.00

0.07 0.04 0.02 0.04 0.01 0.01 0.00

Relative Proportion Finisher 0.28 Nursery 0.20 Abattoir 0.53 Sow 0.17 CEa 0.08 CIa 0.10 CEXa 0.01

0.35 0.41 0.80 0.39 0.20 0.50 0.50

0.21 0.04 0.40 0.00 0.00 0.00 0.00

0.04 0.08 0.08 0.08 0.07 0.20 0.08

a

5. Discussion 5.1. Network measures Output from network analyses utilizing network descriptive measures have been previously reported as estimators of risk for disease transmission (Arruda et al., 2016; Benfeild et al., 2000; Büttner et al., 2013a; Dubé et al., 2008a,b; Noremark et al., 2011; Rossi et al., 2017). When such investigations are conducted, the time scale used to construct a network should be related to disease epidemiology and relevant infection control options. For instance, weekly networks may be more useful for examining diseases with a short incubation period and with apparent clinical signs which could lead to quick identification of disease. In contrast, results based on the yearly networks could be more relevant for infectious agents that do not show obvious clinical signs and have a very long incubation period (Kao et al., 2007). Alternatively, such measures could be relevant for infections that do not initiate any infection control measures at the population level. Thus, investigation

CE=Company External, CEX= Company Export, CI=Company Internal.

present (p = 0.017). Similar patterns were observed for nursery facilities; however, no statistical quadratic relationship was identified with week number (Table 5).

Table 4 Descriptive statistics of node betweenness, in-degree and out-degree obtained from the 2015 weekly network of swine movements among 224 swine establishments, based on data obtained from an Ontario swine production and data management company. Facility Type

Abattoir CEa CEXa CIa Finisher Nursery Sow a

In-degree

Out-degree

Betweenness

Minimum

Mean

Maximum

Minimum

Mean

Maximum

Minimum

Mean

Maximum

0.0 0.0 0.0 0.0 0.0 0.0 0.0

4.6 0.0 0.1 0.1 0.1 0.3 0.1

38.0 2.0 2.0 2.0 4.0 3.0 2.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.1 0.0 0.0 0.4 0.3 0.5

0.0 2.0 0.0 0.0 5.0 4.0 3.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.1 0.3 0.0

0.0 0.0 0.0 0.0 16.0 12.0 1.0

CE=Company External, CEX= Company Export, CI=Company Internal. 215

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

Fig. 3. Area map of the demographics in maximum weak component (MWC) calculated from the 2015 weekly networks of animal movements among 224 Ontario swine establishments and expressed as proportion of sites* (percentage within MWC) in the maximum weak component in a given week. *Other = Company External, Company Export, Company Internal

Therefore, yearly epidemic size estimated in the maximum yearly weak component (MWCy) most likely is an over estimation due to the fact that directionality is ignored within weak component measurements (Dubé et al., 2008). On a weekly level, there was considerable variability in the size of the maximum weak component, but the weekly trend in the size over the period of one year could not be detected. This finding is expected when considering that swine production is a long-term process, and the source population did not experience major contraction or expansion during the defined study period. Thus, the number of sites that were connected through weak links considered in this study should not vary either. Proportion of facility types included in the MWCw was stable over time, with the finisher facilities consistently forming the majority of sites, which on average slightly exceeded their presence in the study

of both time periods is justifiable, particularly from the perspective of infectious agents causing production-limiting disease. Such diseases do not necessarily initiate regulatory response, yet they represent a bulk of health challenges in swine populations in North America (Holtkamp et al., 2007). Being able to enhance our understanding of the potential spread of these types of diseases is necessary, along with the nationally reportable diseases with obvious trade implications. Component size has been used as a potential estimator of the upper bound of an epidemic depending on an appropriate time scale and structure of the network (Kao et al., 2007; Dubé et al., 2008; Guinat et al., 2016). Within our yearly network and when considering weak connections, all 224 facilities were included in the weak component. However, the use of this statistics as an outbreak estimator is limited because direction of animal movements is not taken into consideration.

Fig. 4. Area map displaying demographic composition of the maximum weak component calculated from the 2015 weekly networks of animal movements among 224 Ontario swine establishments and expressed as relative proportion of sites* based on the total number of each facility type in the study population in a given week. *Other = Company External, Company Export, Company Internal 216

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

Fig. 5. Observed weekly count of finisher facilities that had value of in-out degree > 0 in a given week year 2015, and the expected values based on the Poisson regression (p < 0.05). Coefficients from the Poisson regression model are provided in Table 5.

network. This finding is important because, at least for productionlimiting diseases, active surveillance of finisher sites in the source population may not be as frequent as it is for sow and nursery sites. Previous papers have used the maximum strong component to estimate epidemic size (Guinat et al., 2016). When considering only strong connections, the maximum yearly and weekly strong component consisted of only two nodes in this study. The Ontario swine industry mostly practices a one-way flow of animals, where the requirements for strong component are difficult to satisfy. Consequently, their use as estimators of outbreak size in this population are of limited value. When number of sites in the weekly maximum weak components was compared to the number of each facility type within the network, abattoirs became the predominant facility type. This supports the use of abattoirs for targeted surveillance, at least for diseases that could be efficiently monitored through serological assays. Abattoirs have previously been considered as surveillance sites for a number of production limiting diseases (Anonymous, 2015, 2012; Cameron et al., 2014). An investigation into the practicality of sampling abattoirs for disease surveillance demonstrated that, targeted surveillance should be continued and that abattoir personnel provide an effective means of

Table 5 Poisson regression for weekly count of finisher and nursery facilities with betweenness > 0 and both in-degree and out-degree > 0 in a given week, based on 2015 swine movement data obtained from an Ontario swine production and data management company. Outcome

Facility Type

Variable

Coefficient

Betweenness

Finisher

Intercept Week Week^2 Intercept Week Week^2 Intercept Week Week^2 Intercept Week Week^2

1.128 4.462e-02 −7.455e-04 8.825e-01 4.416e-02 −6.397e-04 1.119 4.608e-02 −7.634e-04 0.9213571 4.218e-02 −6.000e-04

Nursery

In-Out Degree

Finisher

a

Nurseryb

Standard error

p

0.018 0.0003

0.013 0.021

0.02 0.0003

0.024 0.065

0.02 0.0003

0.01 0.01

0.019 0.0003

0.029 0.079

a Raw data and predicted values from Poisson regression model are depicted in Fig. 5. b Raw data are depicted in Fig. 6.

Fig. 6. Observed weekly count of nursery facilities that had value of in/out degree > 0 (in/out degree count) in a given week year 2015. 217

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

study conducted in Germany, which suggested that farrowing and multiplier facilities had the highest betweenness values (Büttner et al., 2013b). This discrepancy may be due to differences in production systems, resulting in network structure, or incomplete data in the current study, which did not include genetic herds with sufficient details. Facilities with a larger betweenness value may be associated with high animal flow as these locations will be used more frequently in animal movement and thereby could be suitable nodes where to focus disease control interventions (Dorjee et al., 2013). Specifically, the number of finisher facilities with weekly betweenness > 0 and with both in-degree > 0 and out-degree > 0 in the warmer period of the year suggest potential for higher degree of animal contacts between different age groups (Table 5 & Fig. 5). The number of facilities with in- and outdegree > 0 and betweenness > 0 in a given week was evaluated because it was of interest to explore whether the number of facilities that receive and ship animals within the same week changes over the calendar year. The expected number of facilities with such practice peaked during mid-year, and was lower near the start and the end of the calendar year. There are few possible explanations for such finding: (i) it is possible that producers have different perception of risk of disease transmission during warmer months when cleaning, disinfection and drying between batches is easier than in colder months, (ii) economics of swine production cycle and animal movements between nursery and finisher facilities, (iii) possible disease outbreak or application of infection control measures which dictate different scheduling of pig movement in different parts of the network, or (iv) accidental finding. In order to try to exclude the accidental finding, we have conducted statistical analysis without observations that resulted in large residuals. Although statistical significance changed, the shape of the relationship remained very similar. Regardless of possible reasons, such finding to our knowledge, has not been reported before and warrants further investigations to determine its repeatability over multiple years, and therefore investigate seasonality, and explore possible reasons. This study was subject to several noteworthy limitations. A main limitation of this paper is the limited nature of the study network. The data that were studied represented only a fraction of premises in the source population, and contacts beyond the study network were largely missed. When the impact of such selection bias was recognized, it was described in the appropriate sections. Furthermore, this limitation also had impact on possible misclassification of some nodes. In particular, Company Internal (CI), Company Export (CEx), and Company External (CE) could have represented more than one node, but such information was not available to the authors. It is possible that some of the latter nodes, or even nodes that were classified as abattoirs actually represented animal traders or assembly points. Animal traders are important types of nodes whose characteristics and roles in current production systems needs to be better described. Possible identification and characterization of such nodes could only be possible if data from a larger network at the animal establishment level from the source population were available. Herd size was not available as a part of dataset to make comparison with the source population. However, herd demographics indicate that this study population was reflective of the three-site swine production system, which is the dominant production practice in the target population of Ontario swine herds.

harvesting tissue samples for diagnostic purposes (O’Sullivan, 2011), despite some potentials for selection bias (O’Sullivan, 2011). Furthermore, within this study, the role of abattoirs was likely underestimated because although complete information about the movements between nodes in this production system were included, only a partial information was included about the abattoirs that receive pigs from other production systems. It is likely that other production systems use these same abattoirs, thus the study results have likely underestimated the potential role of abattoirs in the entire source population. As an alternative to both the weak and strong components, outgoing contact chain (OCC) was used in this study as estimate of the potential size of an outbreak. This measure, alongside ingoing contact chain, considered the directionality of animal movement within a discrete time interval. Dubé et al., 2008, noted the OCC provided a better estimation of an outbreak chain than component size (Dubé et al., 2008). As such, it has been frequently used in studies of animal movement (Büttner et al., 2013a; Dorjee et al., 2013; Dubé et al., 2008; Noremark et al., 2011). The maximum yearly OCC consisted of 79 nodes, or 36% of the study population with a single sow location as its origin. This sow herd was connected to ∼44% of the finisher facilities indirectly and ∼33% of the nursery facilities directly over the course of the entire year. The largest ICC for the yearly network had a total of 173 different nodes and consisted of ∼78% of the study population and terminated at one of the abattoirs. This provides further support of the importance of the role of abattoirs in targeted sampling as the majority of the facilities within this network terminated their shipment chain at one abattoir. In addition to investigations of groupings such as network components and contact chains, a possible approach to planning infectious disease control is the identification of central nodes in the network. It is possible to propose theoretical biosecurity intervention actions corresponding to node removals within a network across various time scales of interest, with the objective to limit the outbreak size. Selective removal of central nodes was investigated by Büttner et al. (2013a, 2013b) and Lebl et al. (2016), and both studies demonstrated a more fragmented network directly affecting outbreak probability and its size (Büttner et al., 2013b; Lebl et al., 2016). Fragmented networks naturally create transmission barriers as the potential transmission pathways (e.g., animal movement) are interrupted. This also aids in outbreak investigations as the network of movements has already been isolated from the rest of the system. 5.2. Node centrality In- and out-degree have been used in various production animal networks as indicators of potential for direct disease transmission (Guinat et al., 2016; Lebl et al., 2016; Natale et al., 2011; SánchezMatamoros et al., 2013). It has also been suggested that alongside OCC size, degree and betweenness can be used to pinpoint high risk facilities (Büttner et al., 2013b). In this study, on a yearly level, nursery facilities had the highest average and maximum out-degree value, yet this pattern changed to sow facilities when considering a weekly time period. This is likely a consequence of the duration of time that groups of animals spend in each facility type. The majority of sow herds in the source population had weekly batch farrowing systems, which resulted in weekly movement of animals to one or more nursery sites. Once received in the nursery, animals remained for a number of weeks, and in the case of the all-in all-out (AIAO) flow do not receive or ship to any other facilities. Because of the typical requirements of AIAO in the finisher stage and longer duration of stay, nursery herds shipped into a larger number of finishers, but with lower frequency, which is inversely related to the duration of stay in the finisher stage. Nursery facilities also had the highest betweenness values across weekly and yearly time scales. This finding is consistent with results of other studies from Canada and the US (Dorjee et al., 2013; Lee et al., 2017; Valdes-Donoso et al., 2017). However, it is inconsistent with a

6. Conclusion In conclusion, across the entire year there was a high degree of indirect connectivity within this swine population. Additionally, the MWCw displayed a relatively constant size across 2015, suggesting that the upper bound of a theoretical outbreak due to spread through weak links is expected to be relatively constant throughout the year, at least in the initial phase of the outbreak. The proportional contribution of facility types was also relatively constant over time; however, sites classified as abattoirs were over-represented when their frequency in the study population was accounted for. Similarly, the ICCw and ICCy 218

Preventive Veterinary Medicine 159 (2018) 211–219

D.J. Melmer et al.

both terminated in abattoirs and included a substantial proportion of the study population for the two-time periods that were studied. For some infectious diseases, this confirms that abattoirs could be considered as suitable sampling sites for disease monitoring. The OCCy and OCCw involved 35% and ∼3% (2.6%) of study population, suggesting a potential upper reach of infections through animal movement in a year and a week, respectively, in this specific system. Counts of finisher sites with high betweenness and with in- and out-degree > 0 increased in the mid-year and suggests that degree of animal contacts could be higher in warmer period of the year and warrants further investigation. Furthermore, in combination with weekly betweenness counts this suggest more finisher facilities to have high animal flow rates during the same time span, giving a higher chance for disease transmission or endemic circulation for this swine system. Further investigation of community structures via animal movements along with spatial information may provide useful insight into high-risk populations for endemic and possibly epidemic diseases.

respiratory syndrome virus to susceptible pigs. Can. J. Vet. Res. 68, 128–133. Dorjee, S., Revie, C.W., Poljak, Z., McNab, W.B., Sanchez, J., 2013. Network analysis of swine shipments in Ontario, Canada, to support disease spread modelling and riskbased disease management. Prev. Vet. Med. 112, 118–127. Dubé, C., Ribble, C., Kelton, D., McNab, B., 2008. Comparing network analysis measures to determine potential epidemic size of highly contagious exotic diseases in fragmented monthly networks of dairy cattle movements in Ontario, Canada. Transbound. Emerg. Dis. 55, 382–392. https://doi.org/10.1111/j.1865-1682.2008. 01053.x. Dubé, C., Ribble, C., Kelton, D., McNab, B., 2009. A review of network analysis terminology and its application to foot-and-mouth disease modelling and policy development. Transbound. Emerg. Dis. 56, 73–85. https://doi.org/10.1111/j.1865-1682. 2008.01064.x. Fournié, G., Kearsley-Fleet, L., Otte, J., Pfeiffer, D.U., 2015. Spatiotemporal trends in the discovery of new swine infectious agents. Vet. Res. 46, 1–9. https://doi.org/10.1186/ s13567-015-0226-8. Grolemund, G., Wickham, H., 2011. Dates and times made easy with lubridate. J. Stat. Softw. 3, 1–25. Guinat, C., Relun, A., Wall, B., Morris, A., Dixon, L., Pfeiffer, D.U., 2016. Exploring pig trade patterns to inform the design of risk-based disease surveillance and control strategies. Sci. Rep. 6, 28429. https://doi.org/10.1038/srep28429. Højsgaard, S., Halekoh, U., 2016. doBy: Groupwise Statistics, LSmeans, Linear Contrasts, Utilities. . Holtkamp, D., Rotto, H., Garcia, R., 2007. The economic cost of major health challenges in large US swine production systems. Am. Assoc. Swine Pract. 85–90. Kao, R.R., Green, D.M., Johnson, J., Kiss, I.Z., 2007. Disease dynamics over very different time-scales: foot-and-mouth disease and scrapie on the network of livestock movements in the UK. J. R. Soc. Interface 4, 907–916. https://doi.org/10.1098/rsif.2007. 1129. Lebl, K., Lentz, H.H.K., Pinior, B., Selhorst, T., 2016. Impact of network activity on the spread of infectious diseases through the German pig trade network. Front. Vet. Sci. 3, 48. https://doi.org/10.3389/fvets.2016.00048. Lee, K., Polson, D., Lowe, E., Main, R., Holtkamp, D., Martínez-López, B., 2017. Unraveling the contact patterns and network structure of pig shipments in the United States and its association with porcine reproductive and respiratory syndrome virus (PRRSV) outbreaks. Prev. Vet. Med. 138, 113–123. https://doi.org/10.1016/j. prevetmed.2017.02.001. Natale, F., Giovannini, A., Savini, L., Palma, D., Possenti, L., Fiore, G., Calistri, P., 2009. Network analysis of Italian cattle trade patterns and evaluation of risks for potential disease spread. Prev. Vet. Med. 92, 341–350. https://doi.org/10.1016/j.prevetmed. 2009.08.026. Natale, F., Savini, L., Giovannini, A., Calistri, P., Candeloro, L., Fiore, G., 2011. Evaluation of risk and vulnerability using a Disease Flow Centrality measure in dynamic cattle trade networks. Prev. Vet. Med. 98, 111–118. https://doi.org/10.1016/j.prevetmed. 2010.11.013. Noemark, M., Widgren, S., 2014. EpiContactTrace: an R-package for contact tracing during livestock disease outbreaks and for risk-based surveillance. BMC Vet. Res. 10, 71. https://doi.org/10.1186/1746-6148-10-71. Noremark, M., Hakansson, N., Lewerin, S.S., Lindberg, A., Jonsson, A., 2011. Network analysis of cattle and pig movements in Sweden: measures relevant for disease control and risk based surveillance. Prev. Vet. Med. 99, 78–90. https://doi.org/10.1016/ j.prevetmed.2010.12.009. O’Sullivan, T.L., 2011. An Investigation of Various Sources of Health Data for Swine Disease Surveillance. ProQuest Diss. Theses. University of Guelph. Pasick, J., Berhane, Y., Ojkic, D., Maxie, G., Embury-Hyatt, C., Swekla, K., Handel, K., Fairles, J., Alexandersen, S., 2014. Investigation into the role of potentially contaminated feed as a source of the first-detected outbreaks of porcine epidemic diarrhea in Canada. Transbound. Emerg. Dis. 61, 397–410. https://doi.org/10.1111/ tbed.12269. R Core Team, 2015. R: a Language and Environment for Statistical Computing. Rossi, G., De Leo, G.A., Pongolini, S., Natalini, S., Zarenghi, L., Ricchi, M., Bolzoni, L., 2017. The potential role of direct and indirect contacts on infection spread in dairy farm networks. PLoS Comput. Biol. 13, 1–19. https://doi.org/10.1371/journal.pcbi. 1005301. Sánchez-Matamoros, A., Martínez-López, B., Sánchez-Vizcaíno, F., Sánchez-Vizcaíno, J.M., 2013. Social network analysis of equidae movements and its application to riskbased surveillance and to control of spread of potential equidae diseases. Transbound. Emerg. Dis. 60, 448–459. https://doi.org/10.1111/j.1865-1682.2012. 01365.x. Thakur, K.K., Revie, C.W., Hurnik, D., Poljak, Z., Sanchez, J., 2016. Analysis of swine movement in four canadian regions : network structure and implications for disease spread. Transbound. Emerg. Dis. 63, e14–e26. https://doi.org/10.1111/tbed.12225. Valdes-Donoso, P., VanderWaal, K., Jarvis, L.S., Wayne, S.R., Perez, A.M., 2017. Using machine learning to predict swine movements within a regional program to improve control of infectious diseases in the US. Front. Vet. Sci. 4, 1. https://doi.org/10.3389/ fvets.2017.00002. Wernike, K., Hoffmann, B., Beer, M., 2013. Single-tube multiplexed molecular detection of endemic porcine viruses in combination with background screening for transboundary diseases. J. Clin. Microbiol. 51, 938–944. https://doi.org/10.1128/JCM. 02947-12.

Acknowledgments Funding provided by, Ontario Ministry of Agriculture Food and Rural Affairs. The authors are grateful to the Ontario management and data company for enabling access to data. Appendix A. Supplementary data Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.prevetmed.2018.09. 021. References Anonymous, 2012. CFIA: Swine Disease Surveillance - Fact Sheet [WWW Document]. URL. (Accessed 6.10.17). http://www.inspection.gc.ca/animals/terrestrial-animals/ diseases/surveillance/swine-disease-surveillance/eng/1329691307451/ 1329691665112. Anonymous, 2015. Public Health Agency of Canada: Antimicrobial Resistance [WWW Document]. URL. (Accessed 6.10.17). http://www.phac-aspc.gc.ca/cipars-picra/ 2012/chap1-section-2-eng.php. Arruda, A.G., Friendship, R., Carpenter, J., Hand, K., Poljak, Z., 2016. Network, cluster and risk factor analyses for porcine reproductive and respiratory syndrome using data from swine sites participating in a disease control program. Prev. Vet. Med. 128, 41–50. https://doi.org/10.1016/j.prevetmed.2016.03.010. Ayudhya, S.N., Assavacheep, P., Thanawongnuwech, R., 2012. One world - one health: the threat of emerging swine diseases. An Asian perspective. Transbound. Emerg. Dis. 59, 9–17. https://doi.org/10.1111/j.1865-1682.2011.01309.x. Benfeild, D., Nelson, J., Rossow, K., Nelson, C., Steffen, M., Rowland, R., 2000. Diagnosis of persistent or prolonged porcine reproductive and respiratory syndrome virus infections. Vet. Res. 31, 71. Bigras-Poulin, M., Barfod, K., Mortensen, S., Greiner, M., 2007. Relationship of trade patterns of the Danish swine industry animal movements network to potential disease spread. Prev. Vet. Med. 80, 143–165. https://doi.org/10.1016/j.prevetmed.2007.02. 004. Büttner, K., Krieter, J., Traulsen, A., Traulsen, I., 2013a. Efficient interruption of infection chains by targeted removal of central holdings in an animal trade network. PLoS One 8. https://doi.org/10.1371/journal.pone.0074292. Büttner, K., Krieter, J., Traulsen, A., Traulsen, I., 2013b. Static network analysis of a pork supply chain in Northern Germany-Characterisation of the potential spread of infectious diseases via animal movements. Prev. Vet. Med. 110. https://doi.org/10. 1016/j.prevetmed.2013.01.008. Cameron, A., Njeum, D., Chibeu, D., Martin, T., 2014. . FAO. 2014. Risk-Based Disease Surveillance – A Manual for Veterinarians on the Design and Analysis of Surveillance for Demonstration of Freedom From Diseaease. FAO Animal Production and Health Manual No 17, Rome, Italy. Csardi, G., Nepusz, T., 2006. The igraph software package for complex network research. Int. J. Complex Syst. 1695. Davies, P.R., 2012. One World, One Health: The Threat of Emerging Swine Diseases. A North American Perspective. Transbound. Emerg. Dis. 59, 18–26. https://doi.org/10. 1111/j.1865-1682.2012.01312.x. Dee, S.A., Deen, J., Otake, S., Pijoan, C., 2004. An experimental model to evaluate the role of transport vehicles as a source of transmission of porcine reproductive and

219