A stochastic simulation model of African swine fever transmission in domestic pig farms in the Red River Delta region in Vietnam

Abstract The main objectives of this study were to model various scenarios of African swine fever (ASF) virus transmission among farms in Vietnam and to evaluate the impact of control strategies using North American Animal Disease Spread Model (NAADSM). A total of 7,882 pig farms in the Red River Delta (RRD) region were obtained from the General Statistics Office, and then, random points corresponding to the number of farms in each province were generated as exact farm locations were not available. A total of 10 models were developed, including movement control scenarios. In addition, we conducted sensitivity analysis to assess the impact of indirect contact transmission probability (TP). Overall, the indirect contact exhibited an important role in transmitting the ASF virus. In order to minimize ASF transmission between farms, we found that movement restriction needed to reach a certain level (approximately between 50% and 75%) and that the restriction had to be applied in a timely manner. This study offers valuable insight into how ASF virus can be transmitted via direct and indirect contact and controlled among farms under the various simulation scenarios. Our results suggest that the enforcement of movement restriction was an effective control measure as soon as the outbreaks were reported. In addition, this study provided evidence that high standards of biosecurity can contribute to the reduction of disease spread.


| INTRODUC TI ON
African swine fever (ASF) is a highly contagious virus and classified as a notifiable disease by the World Organization for Animal health (OIE) (Tulman, Delhon, Ku, & Rock, 2009). It is a double-stranded DNA virus of the Asfarviridae family, genus Asfivirus (Dixon et al., 2005). The disease causes acute haemorrhagic fever with mortality of up to 100% depending on the virulence of the isolate, dose and route of exposure to the virus (Costard, Mur, Lubroth, Sanchez-Vizcaino, & Pfeiffer, 2013). Pigs are infected via contact with infected animals (including free-ranging and wild pigs), fomites, premises, vehicles, clothes, consumption of contaminated feed and bites of infected ticks (OIE, 2020a). The disease can have serious economic consequences, through reduced international trade and a decrease in pig populations, which can in turn pose a huge threat to food security (Blome, Gabriel, & Beer, 2013). The disease is endemic in sub-Saharan African countries, Caucasus, Eastern Europe and Baltic countries (OIE, 2020b). In Asia, the first outbreak was confirmed in northeastern China in August 2018 (Zhou et al., 2018), and then, the virus rapidly spread to other Asian countries (Dixon, Sun, & Roberts, 2019;FAO, 2020).
In Vietnam, the first ASF outbreak was reported in February 2019 in backyard pig farms in Hung Yen province (Van Phan Le et al., 2019). Since then, ASF outbreaks have been reported in all 63 provinces, resulting in approximately 6 million pigs (20% of pig production) that have been either culled or killed by the disease (FAO, 2020). As of May 2020, outbreaks have not been reported for more than 30 days in 35 of the 63 provinces.
The use of simulation models for infectious diseases is an important tool for decision-makers to evaluate the impact of outbreaks and to identify cost-effective control strategies (e.g. vaccination, movement control and depopulation) (Francis, Klotz, Harvey, & Stacey, 2010;Keeling, 2005;Morris, Wilesmith, Stern, Sanson, & Stevenson, 2001). The North American Animal Disease Spread Model (NAADSM) is a computer software used to develop simulation models of the spread of highly infectious animal diseases (Harvey et al., 2007). It provides a flexible framework with user-established parameters to define the disease spread by direct, indirect contact, airborne and local spread as well as assess the impact of various control measures.
In Vietnam, the first study using NAADSM explored porcine reproductive and respiratory syndrome (PRRS) transmission via direct and indirect contacts across different farm types (Lee et al., 2019). With the ongoing outbreak and impact of ASF on the pig sector, the main objectives of this study were to model various scenarios of ASF virus transmission among farms and to evaluate the impact of control strategies using NAADSM.

| Study location and population
The Red River Delta (RRD) region is the smallest region in Vietnam, located in the north (Figure 1), yet it has the highest concentration of human population (22 million) in the country (GSO, 2020). It also has the largest pig population (7.2 million), which accounts for 28%-29% of the pig population in Vietnam (GSO, 2018). In order to develop the ASF transmission models, farm locations and characteristics were necessary. The number of livestock farms at provincial level was obtained from the General Statistics Office of Vietnam. As exact farm locations were not available, random points corresponding to the number of farms in each province were generated using QGIS (Quantum GIS development 2012, QGIS version 3.12.2) ( Figure 1). The coordinates were then extracted and imported into NA ADSM. A total of 7,882 farms were recorded and used for the simulation model. These farms were categorized into 3 production types: a) small (<100 pigs), b) medium (≥100 pigs) and c) large (>1,000 pigs) (Nga, Ninh, Van Hung, & Lapar, 2014). The proportion of each production type was 70% (a), 25% (b) and 5% (c), respectively (Lapar, Binh, & Ehui, 2003)

| Model parameters
The NAADSM requires three key parameters: (a) transmission probability (TP) related to each contact (it is a probability between 0 and 1, and representing the likelihood of the contact herd will become infected given the exposure to an infected herd); (b) distance distribution associated with contact between farms; and (c) mean contact rates (estimated number of contacts per week) (Harvey et al., 2007). These parameters were estimated based on previous studies (Guinat et al., 2016;Lee et al., 2019) and assumptions (expert opinions) ( Table 1).
Transmission of ASF in Vietnam is mainly through indirect contact (e.g. swill feeding and human/vehicle movements). Therefore, TPs for direct and indirect contacts were considered as the same value (0.6) (expert opinions) for small and medium farms, whereas large farms were parametrized to have the value of 0.006 for indirect contact and 0.6 for direct contact TP due to comparatively higher levels of biosecurity (Table 1). These values were internally discussed among local experts. We assumed that the different infectious durations were determined by farm type (small farm: 52 weeks; medium farm: 10-12 weeks; and large farm: 4 weeks) as a result of different levels of biosecurity. Because, we assumed that a continuous flow (CF) system was used in small farms as they are replacing pigs continuously from different sources (with unknown disease status), while the all-in-all-out system (AIAO) was followed in large and some medium farms where these farms introduce new pigs in batches and mostly from farms with high biosecurity and known infection status. Our estimates of infectious duration for the three farm types were based on these considerations, where large and medium farms were allowed to remain infectious for a relatively shorter duration. In contrast, small farms due to the continuous reintroduction of animals were allowed to remain infectious for the entire simulation duration.
A PERT distribution was defined for contact distances between farms, with a minimum of 0.5 km, a most likely value of 30 km and a maximum of 300 km (

| Simulation model structure and sensitivity analysis
In Vietnam, most large farms are commercialized and contract farms, while medium farms are mainly suppliers and have high connectivity to small farms ( Figure 2). It is very rare that pigs from small farms move to other sized farms locally. We assumed that none of the pigs had resistance to ASF virus. If a single pig became infected, then the whole farm was considered to be infectious. The baseline scenario was that one medium farm was infected and the same farm-initiated infection in the following iterations. We assumed that the rest of the farms were susceptible at the beginning of the simulation and remained infectious until the end of the study period for small farms or remained infectious for a specified period for medium and large farms and became susceptible again (Table 1). Especially, medium and large farms were allowed to be infected multiple times during the simulation.
The model was run over 500 iterations for 52 weeks, which was long enough to cover at least one complete pig production cycle (6-8 months in Vietnam). Since there is no vaccine for ASF, we evaluated the effectiveness of movement control on contact rates by 25%, 50%, 75% and 100% reduction, which was applied to both direct and indirect contact rates. It was hypothesized that movement restrictions were imposed for the baseline scenario within 4 weeks of detection of outbreaks. We conducted sensitivity analysis to assess the impact of infectious duration for small farms (basic: 52 weeks) and indirect contact TP for the small/medium farms (basic: 0.6) by −25%, −50% and −75%, respectively. In addition, the timing of movement restriction was imposed at 2 weeks, 6 weeks and 8 weeks after detection, respectively, under the 25% and 50% movement restriction scenarios.

| RE SULTS
The baseline scenarios (A1: including both direct and indirect contacts) showed that a total of 7,640 (5 and 95 percentiles: 6,729-7,790) median farms were infected, while A2 (only including indirect contact) presented a slightly lower median number of infected farms (7,544, 5 and 95 percentiles: 5,890-7,685) (

| D ISCUSS I ON
This was the first study in Vietnam to assess the transmission of ASF virus among swine farms using NAADSM. The weekly mean contact rates by farm types were obtained from the previous study in Vietnam (Lee et al., 2019), which made our model more realistic in terms of applicability to local farms. In the model, indirect contact had a predominant role in the transmission of the ASF virus between farms. It has been suggested that the various means of indirect contact (e.g. swill feeding, human/transport-associated routes and improper disinfection) account for 70%-80% of the transmission of ASF virus among farms in Vietnam (DAH, 2020). In fact, it is still a common practice to give swill feeding in small pig holders in Vietnam even after ASF outbreaks have occurred. In addition, it is well known that wild boars and soft tickets could be the main source of infection in other countries (Galindo & Alonso, 2017;Thomson, 1985).

In Asia, infected wild boars have been reported in China and South
Korea ( Our simulation models showed that a decrease of indirect contact for TP resulted in a reduction of the number of infected farms when it reached a certain low level; otherwise, it was not effective. The main implication was that strict enforcement of high levels of biosecurity measures was the effective way to prevent the introduction of disease into pig farms. In Vietnam, poor biosecurity in small-and medium-scale farms has already been identified as the main risk factor for disease transmission (Lee et al., 2020). Indeed, the absence of disinfection mattresses, no or rare use of protective boots and clothes, irregular disinfection of farm premises and the use of left-over food for feedings are very common. One study showed that the biosecurity scores (it evaluates both external biose-

curity [reduce the introduction of diseases] and internal biosecurity
[reduce the spread of diseases]) in pig farms were between 53.68% and 55.05% based on percentage grade (0%-100%) (Tuan, Dewulf, Postma, Cuc, & Dinh, 2019). It is therefore very important to establish regular training programmes to educate farmers on biosecurity practices.
However, we acknowledge that in the absence of available data, the indirect contact TP, for small-and medium-scale farms (0.6), is used in the baseline scenario and other movement control scenarios, which was based on our assumption owing to the above considerations. Our indirect contact TP is larger than the probabilities used for this parameter in ASF spread models in different jurisdictions. Our sensitivity analysis supported our assumption that indirect contact had a larger role in ASF spread in Vietnam as smaller indirect contact TP had resulted in nominal spread of the virus, contrary to what had been observed during the initial ASF outbreaks (much rapid/wider spread) in Vietnam. Although our simulation provided some guidance on the probable range of this probability, the uncertainty in this parameter estimate was still not resolved, and future field studies may help to provide better estimates.
It was assumed that the ASF virus was introduced from China as the virus strain was 100% identical to China strains ( Fasina et al., 2012;Olesen et al., 2017). In addition, our results may be influenced by the local-area spread if some farms are in closer proximity. Therefore, in order to evaluate the impact of randomly created farm locations using QGIS, two more data were generated to make a comparison. We found that their impacts on the outcomes were negligible. In the study, the actual number of farms (especially, smallholders) in the RDD region may be much higher than the national data. Indeed, it is not easy to identify the number of smallholder farms (e.g. less than 10 pigs) in very remote rural and high mountainous areas unless farmers are willing to register. Therefore, it may be possible that the transmission of ASF virus was much faster than it was in our models.
This study offers valuable insight into how ASF virus can be transmitted via direct and indirect contact and controlled among farms under the various simulation scenarios. Our results suggest that the enforcement of movement restriction was an effective control measure as soon as the outbreaks were reported. In addition, this study provided evidence that high standards of biosecurity can contribute to the reduction of disease spread. This simulation model can be applied to other regions or countries with modified parameters. In addition, it may be useful for assessing the cost-effective infection control and prevention strategies in the Vietnamese context through running the 'what-if' scenarios related to ASF virus transmission.

ACK N OWLED G EM ENTS
This study was funded by the CGIAR Research Program on Livestock and we thank donors who support its work through their contributions to the CGIAR Trust Fund. CGIAR is a global research partnership for a food-secure future. Its science is carried out by 15 Research Centers in close collaboration with hundreds of partners across the globe. www.cgiar.org. We also would like to give special thanks to Mireille Ferrari for English editing.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no competing interests.

E TH I C A L A PPROVA L
Ethics approval was not required for this study as any sample collection or questionnaires from animal/human has not been gathered.

DATA AVA I L A B I L I T Y S TAT E M E N T
All datasets supporting our findings are available from the corresponding author on reasonable request.