A within-flock model of Salmonella Heidelberg transmission in broiler chickens

Preventive Veterinary Medicine 174 (2020) 104823

Contents lists available at ScienceDirect

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

A within-flock model of Salmonella Heidelberg transmission in broiler chickens

T

Lucie Collineaua, Charly Phillipsa,b, Brennan Chapmana,c, Agnes Agunosd, Carolee Carsond, Aamir Fazila, Richard J. Reid-Smithc,d, Ben A. Smitha,* a

Public Health Risk Sciences Division, National Microbiology Laboratory, Public Health Agency of Canada, Guelph, ON, Canada Department of Biomedical Engineering, University of Waterloo, Waterloo, ON, Canada c Department of Population Medicine, Ontario Veterinary College, University of Guelph, Guelph, ON, Canada d Centre for Food-Borne, Environmental and Zoonotic Infectious Diseases, Public Health Agency of Canada, Guelph, ON, Canada b

ARTICLE INFO

ABSTRACT

Keywords: Compartmental model SEIR model Infection dynamics Risk assessment Public health

As part of the development of a quantitative microbial risk assessment (QMRA) model of third-generation cephalosporins (3GC)-resistant Salmonella Heidelberg, a compartmental (SEIR) model for S. Heidelberg transmission within a typical Canadian commercial broiler chicken flock was developed. The model was constructed to estimate the within-flock prevalence and the bacterial concentration in the barn environment at pre-harvest, and to assess the effect of selected control measures. The baseline scenario predicted an average within-flock prevalence of 23.5 % (95 % tolerance interval: 15.7–31.4) and an average bacterial concentration of 3.579 (0–4.294) log CFU/g of feces in the barn environment at pre-harvest (on the day the flock is sent to slaughter). Because vertical introduction of S. Heidelberg into the barn was already uncommon in the baseline scenario, vaccination of broiler parent flocks appeared to have a negligible effect, while vaccination of broiler chicken flocks substantially reduced the bacterial concentration at pre-harvest. Cleaning and disinfection between batches markedly reduced the within-flock prevalence at preharvest, but the effect on bacterial concentration was limited outside of the beginning of the production period. Extending downtime between batches by 7 days had little effect on within-flock prevalence or bacterial concentration of S. Heidelberg when compared to the baseline scenario. This study provides a basis to describe S. Heidelberg dynamics within a broiler chicken flock and to predict the within-flock prevalence and bacterial concentration at pre-harvest, and includes a description of the limitations and data gaps. The results of these analyses and associated uncertainties are critical information for populating QMRA models of the downstream impacts on public health from on-farm and other food-chain practices. Specifically, the study findings will be integrated into a broader farm-to-fork QMRA model to support the risk-based control of S. Heidelberg resistant to 3GC in broiler chicken in Canada.

1. Introduction Non-typhoidal salmonellosis represents a significant human health burden in Canada, with an incidence rate of 21.4 laboratory confirmed cases per 100,000 people in 2016 (Government of Canada, 2018a). An estimated 63 % of non-typhoidal Salmonella cases in Canada are foodbornerelated, and of those, 34%–42% are attributable to consumption of poultry products (Butler et al., 2015; Davidson et al., 2011), with ingestion occurring either as a result of incomplete cooking, or via cross-contamination of ready-to-eat products (e.g., lettuce) during food preparation (O’Mahoney et al., 1990). The Salmonella serovar Heidelberg was the third most frequently isolated serovar in the Canadian population in 2016 (Government ⁎

of Canada, 2018a), and it is more frequently reported in North America than in other regions of the world (Vieira et al., 2009). S. Heidelberg is likewise commonly identified in chicken samples collected at farm, abattoir, and retail by the Canadian Integrated Program for Antimicrobial Resistance Surveillance (CIPARS) (Government of Canada, 2017a). In humans, infection with S. Heidelberg generally results in a mild to moderate illness, but severe complications such as septicemia, myocarditis, extra-intestinal infections, and death can occur, especially in vulnerable subpopulations such as pregnant women and children (Kennedy et al., 2004; Vugia et al., 2004). Third-generation cephalosporin (3GC) resistant infections are of particular concern, as 3GCs are the antimicrobials of choice for treating severe or invasive salmonellosis in the aforementioned

Corresponding author at: Public Health Risk Sciences Division, Public Health Agency of Canada, 370 Speedvale Avenue West, Guelph, N1H 7M7, Canada. E-mail address: [email protected] (B.A. Smith).

https://doi.org/10.1016/j.prevetmed.2019.104823 Received 18 April 2019; Received in revised form 10 September 2019; Accepted 1 November 2019 0167-5877/ Crown Copyright © 2019 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

populations (Kariuki et al., 2015; Shane et al., 2017). The occurrence of 3GC-resistance among S. Heidelberg of human and chicken origin was shown to temporally mirror the extra-label (i.e., off-label) injection of ceftiofur (a 3GC antimicrobial agent) in-ovo at Canadian broiler hatcheries (Dutil et al., 2010). This was a relatively common practice to prevent Escherichia coli omphalitis in broiler chickens prior to 2014, when the Canadian chicken industry voluntarily stopped its use. In chickens, infections with S. Heidelberg are usually asymptomatic, although morbidity (including enteritis, peritonitis and hepatitis) and mortality have been reported in chicks experimentally inoculated at one day of age (Borsoi et al., 2011; Roy et al., 2001). Birds typically become infected through the ingestion of bacteria from their environment. Both vertical (i.e., upstream in the production pyramid such as breeder flocks and hatcheries) and horizontal (i.e., between-birds or between flocks/farms) transmission of Salmonella contributes to broiler contamination at pre-harvest and further downstream in the production chain (Liljebjelke et al., 2005). Carry-over of Salmonella from one batch to another within the same barn, despite thorough cleaning and disinfection, has also been described (Kloska et al., 2017). As part of the development of a quantitative microbial risk assessment (QMRA) model of S. Heidelberg resistant to 3GCs in broiler chicken meat from farm to consumption, two key parameters related to the on-farm contamination of broilers are required: i) the within-flock prevalence of S. Heidelberg at pre-harvest, i.e., the proportion of birds colonized with S. Heidelberg on the day the flock is sent to slaughter, and ii) the concentration of S. Heidelberg in the barn environment at pre-harvest, i.e., the number of colony-forming units (CFU) per gram of feces on the floor of the barn. This information can be acquired through results of surveillance or research (should adequate data exist), modelling, or a combination thereof. On-farm surveillance for Salmonella is usually based on the collection of environmental pooled samples (Rajan et al., 2017); data characterizing within-flock prevalence and concentration at pre-harvest is sparse and often not serovar specific. In Canada, the on-farm surveillance program for Salmonella at pre-harvest conducted by CIPARS and FoodNet Canada is based on the collection of four pooled fecal samples, i.e., one pooled sample per floor quadrant, where potentially the entire flock has been walking and shedding feces (Government of Canada, 2017a; Government of Canada, 2017b). While this sampling strategy provides useful information on the types of strains circulating and on the flock status towards Salmonella, it is impossible to infer an accurate estimate of the within-flock prevalence at pre-harvest from these data. In addition, no enumeration of Salmonella concentration in the barn environment or on the birds is performed. As an alternative, an infectious disease model may be used to estimate the within-flock prevalence and environmental concentration at pre-harvest. Several models for Salmonella transmission during live production stages have been developed, including probabilistic transmission models (Nauta et al., 2000; Ranta and Maijala, 2002; Van der Fels-Klerx et al., 2008) and QMRA models (Maijala et al., 2005; Parsons et al., 2005). However, they were developed using the flock as the unit of interest. Within-flock Salmonella transmission models have been developed in laying flocks (Prévost et al., 2006; Zongo et al., 2010); however, Salmonella transmission in laying flocks is likely to differ from transmission in broiler flocks (caged birds versus free-moving birds, differences in length of production cycle and genetics). In addition, none of the above models were

specific to S. Heidelberg. An auxiliary benefit of the use of an infectious disease model is the ability to assess the efficacy of on-farm interventions in reducing withinflock prevalence and environmental Salmonella concentration at pre-harvest. Bucher et al. (2012) conducted a scoping review of control measures implemented by the broiler chicken industry to mitigate the occurrence of Salmonella (Bucher et al., 2012). At farm level, these include measures to prevent Salmonella vertical transmission (e.g., eradication of positive flocks, breeder flock vaccination, treatment of hatching eggs), horizontal transmission (e.g., broiler flock vaccination, use of competitive exclusion products or feed and water additives) and carry-over between batches (e.g., strict cleaning and disinfection procedures, extended downtime periods between production cycles). The effects of vaccination, competitive exclusion and feed and water additives, as well as the treatment of hatching eggs have been assessed via systematic review and meta-analysis (Kerr et al., 2013; Totton et al., 2012a,b). However, the effect of vaccination on Salmonella concentrations was not described in these studies (Totton et al., 2012b). To our knowledge, the effects of strict cleaning and disinfection and extended downtime between batches have only been described in individual studies (Kloska et al., 2017; Luyckx et al., 2015; Maertens et al., 2017). Currently, there is no mandatory control strategy specific to S. Heidelberg in broiler chicken production in Canada. However, the guidance provided by the ‘National Strategy for the Control of Poultry-Related Human Salmonella Enteritidis Illness in Canada’ is likely to be effective in controlling other Salmonella serovars, including S. Heidelberg (Government of Canada, 2015). The objectives of the present study were to i) develop a model for the transmission of S. Heidelberg in a typical Canadian commercial broiler chicken flock to inform the within-flock prevalence and the environmental concentration in the barn at pre-harvest, and ii) assess the effect of selected measures for the control of S. Heidelberg during broiler chicken production, including broiler breeder and broiler chicken vaccination, as well as cleaning and disinfection and extended downtime between batches. 2. Materials and methods 2.1. Model description The hypothetical population under study is a closed batch of N broiler chickens placed in a barn at day 0 (as day-old chicks) and sent to slaughter at day 36 (i.e., an all-in all-out system, typical of Canadian commercial chicken production) (Agunos et al., 2017). A density-dependent compartmental model was developed, in which birds were classified into mutually exclusive health states (Keeling and Rohani, 2008): susceptible S (i.e., with no S. Heidelberg in their digestive tract), infected-non-infectious E (i.e., in a latent state where they have ingested and been colonized in one or both of intestinal and systemic sites but are not yet excreting S. Heidelberg), infected-infectious I (i.e., where they are colonized in one or both of intestinal and systemic sites and are excreting S. Heidelberg) and recovered and immunized R (i.e., where they have cleared the bacteria and acquired immunity against future S. Heidelberg infections). No feedback loop from the R compartment to the S compartment was included, as infection with Salmonella is known to confer long-lasting immunity of approximately 500 days in laying hens (considerably longer than the duration of broiler production) (Protais et al., 1996). Fig. 1. Flow diagram of Salmonella Heidelberg transmission in a broiler chicken flock, representing transition between health states (grey boxes) and interaction with environmental compartments (white boxes). S: susceptible; E: infected-non-infectious; I: infected-infectious; R: recovered and immunized, C: bacterial population (#CFU) in the barn environment, D: amount of feces (grams) in the barn environment. Solid arrows represent transitions between health states; dashed arrows represent dependencies in the model and dotted arrow represents CFU removal from the C compartment. The transmission function f(C,D) is defined in Eq. (2). The transmission parameters γ, α, δ, λ and ω are described in Table 1. 2

Preventive Veterinary Medicine 174 (2020) 104823

Expert opinion (Dr. Agnes Agunos, DACVP, PHAC) Expert opinion (Dr. Agnes Agunos, DACPV, PHAC) (Gast et al., 2005), Expert opinion (Dr. Richard K. Gast, U.S. National Poultry Research Center) (Borsoi et al., 2011; Jiang et al., 2010) In-vivo farm experiment

(Prévost et al., 2006) CFU.day−1 Mortality rate of S. Heidelberg in the barn environment λ

where f(C(t),D(t)) = 1 − exp[−ρ × (η / D(t)) × C(t)]

DACPV = Diplomate of the American College of Poultry Veterinarians; PHAC = Public Health Agency of Canada. a The outcomes of the exponential distributions were bound to the reciprocal of the durations of the infectious and recovery periods. b t represents the age of the birds (in days).

Amount of feces ingested daily by a random bird Infectious ratea Recovery ratea Bacteria excretion rate in the feces Amount of feces shed in the environment by a random bird η γ α δ ω

20,000 36 day 1–2: 0.00110976 day 3–7: 0.000112252 day 8–36: 1.07098x10^(-9) ∼ Pert (0, 0.05, 2) ∼ Exponential (Uniform (1, 2)) ∼ Exponential (Triangular (21, 23, 28)) ∼ 10^Pert (1.78, 3.87, 5.97) day 1–3: 0.3057tb + 0.7738 day 4–36: 9.4464 ln(tb) - 4.1504 0.1 Size of the population Duration of the production period Probability of infection given ingestion of 1 CFU of S. Heidelberg N d ρ

dS(t)/ dt = f (t)S(t) dE(t)/ dt = f (t)S(t) E(t) dI(t)/ dt = E(t) I(t) dR(t)/ dt = I(t) dC(t)/ dt = I(t) C(t) dD(t)/ dt = [S(t) + E(t) + I(t) + R(t)]

Value

(Chicken Farmers of Canada, 2018a) (Agunos et al., 2017) See supplementary material

birds days – – – gram.day−1 day−1 day−1 CFU.day−1 gram.day−1

Two environmental compartments C and D were included to track, respectively, the bacterial population (number of CFU) of S. Heidelberg and the amount of feces (grams) in the barn environment over time (Fig. 1). For a given time point, the estimated concentration (number of CFU per gram of feces) of S. Heidelberg in the barn environment is estimated by the ratio of C to D. Flock mortality was not considered, as this production metric is multifactorial in nature and S. Heidelberg infections are usually asymptomatic in chicken. Therefore, mortality (due to factors and infections other than S. Heidelberg) was assumed to affect each health state equally and to have no influence on the withinflock prevalence or the bacterial concentration of S. Heidelberg. The model is run in discrete time steps with a time interval of one day. Each day, the transition between states occurs in accordance with the following set of differential equations:

Parameter Definition

Table 1 Parameters used in the Salmonella Heidelberg within-flock transmission model.

Unit

Source

L. Collineau, et al.

(1) (2)

Table 1 provides a summary of the model input parameters. The transmission function f(C,D) follows a typical exponential dose-response function that describes the probability of infection of a bird given ingestion of a fraction (η / D) of S. Heidelberg CFU present in the barn environment C, assuming the received dose is Poisson distributed (Vose, 2008). In accordance with De Silva et al. (2017), η / D corresponds to the ratio of the amount η of feces ingested daily by a random bird to the total amount of feces D present in the farm environment (De Silva et al., 2017). The infectious dose of Salmonella in chicken is known to increase with the age of the bird (Poppe, 2000; Sadler et al., 1969). A literature review and meta-analysis of chicken infection experiments with Salmonella was conducted to inform the probability ρ of infection given ingestion of one CFU (see Supplementary material). Based on the distribution of the age of the birds at the time of challenge among included studies, ρ was defined separately from days 1–2, days 3–7 and days 8–36 of age. The Solver function of Microsoft Excel 2010 was used to estimate the optimal ρ parameter fitting the raw data within each time period (Table 1). Similarly, the amount of feces daily shed in the environment by a random bird ω was believed to increase with the age of the birds. In the absence of literature quantifying such an increase, the quantity of feces shed by a flock of broiler chickens was systematically recorded. Specifically: a batch of 42 straight-run (unsexed birds – males and females combined) broiler chickens hatched on April 3rd 2018 was followed over a 36 day period. Chickens were kept in a coop which was cleaned out twice daily by scooping out all of the chicken feces present on the floor. From day 21 onwards, chickens had outdoor access to a fenced area where feces were also collected. Each day, collected feces were weighed and the total amount of feces shed over 24 h was used to estimate the average amount of feces daily shed per bird. A logarithmic distribution provided good fit to the raw data (R² = 0.859). Data were missing from day 1 to day 3, so extrapolation was performed using a linear distribution. The infectious and recovery rates γ and α were defined by assigning an exponential distribution to the duration of the infectious and recovery periods, respectively, with the outcome of the exponential distribution bound to the reciprocal of the durations of the infectious and recovery periods. The bacteria excretion rate δ was informed using the distribution of S. Heidelberg fecal excretion rates among unvaccinated birds between two and 20 days post-inoculation as reported by Jiang 3

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

et al. (2010) and Borsoi et al. (2011). The amount of feces η ingested daily by a random bird, as well as the duration of the infectious period (1/γ) were informed by the opinion of a poultry production expert in the research team, given the lack of available literature. The model was implemented in the open source environment R version 3.5.2 using Monte Carlo simulation with 25,000 iterations (R Development Core Team, 2018). Generally, a conservative approach was used to parameterize the model; where data were sparse, inputs were selected to provide a worst-case scenario. This reflects the approach of downstream, human-health risk modelling activities. The model is available at: https://doi.org/10.5281/zenodo.3351611.

which collects random fluff samples from all of the hatchers of seven participating hatcheries (i.e., all Ontario commercial hatcheries affiliated with the Canadian Hatcheries Federation; http://www.cpepc. ca) every six weeks (OHSFP, 2017). Within a positive hatcher, the probability p_dolly that a dolly is contaminated was estimated using the OHSFP breeder flock prevalence of S. Heidelberg as a surrogate value. Within a positive dolly, the most likely proportion p_chick of infected chicks was estimated using the OHSFP proportion of positive environmental samples in broiler breeder flocks as a surrogate, whereas the minimum and maximum numbers of infected chicks were defined, respectively, as the tray and dolly capacities, assuming cross-contamination occurs within trays, but not between trays, in a positive dolly. OHSFP data from 2011 (i.e., before the introduction of S. Heidelberg vaccination in January 2013) were used as a baseline scenario (Table 3). Therefore, the number of positive hatchers was defined as:

2.2. Model assumptions The model was based on the following assumptions: i S. Heidelberg can be introduced in the barn via two routes only: 1) introduction of day-old chicks already infected-infectious at placement, or 2) carry-over of S. Heidelberg in the barn environment from the previous batch - other potential sources of S. Heidelberg which could occur later in the production period (e.g., via contaminated feed, wildlife, pests, farm visitors or equipment) were not considered due to limited data; ii birds are only infected via ingestion of S. Heidelberg CFU from the environment - direct bird-to-bird transmission was not considered, as the fecal-oral route is believed to be the main route of Salmonella transmission (Foley et al., 2013; Heres et al., 2004); iii there is homogenous mixing of feces and S. Heidelberg CFU, and therefore a homogenous concentration (CFU/g feces) of S. Heidelberg in the barn environment; and iv each bird has the same probability of ingestion of S. Heidelberg CFU in the barn environment.

n_hatch_pos ∼ Binomial (N / N_hatch, p_hatch) as:

(3)

The number of positive dollies within a positive hatcher was defined

n_dolly_pos ∼ Binomial (N / N_dolly, p_dolly) as:

(4)

The number of infected chicks within a positive dolly was defined

n_chick_pos ∼ Pert (N_tray, p_chick x N_dolly, N_dolly)

(5)

The number I0 of day-old chicks already infected-infectious at placement was subsequently calculated as: I0 = n_hatch_pos x n_dolly_pos x n_chick_pos

(6)

The occurrence of new infections during the transport of day-old chicks was not considered because transport duration, under normal operational context, is usually short (< 24 h, within province transport) (Farm and Food Care Ontario and the Poultry Service Association, 2015).

2.3. Initial conditions: I0 The number I0 of day-old chicks already infected-infectious at placement of the batch on the farm was estimated using available data on the occurrence of S. Heidelberg in the earlier stages of production, namely in broiler parent flocks and hatcheries. A typical Canadian commercial hatchery was modeled, where fertilized eggs are collected from multiple broiler parent flocks. Shortly after reception, eggs are placed in incubators for approximately 18 days, and later transferred into a hatcher where pipping occurs between days 19 and 21. Day-old chicks are then transferred into transport boxes and loaded in trucks for transport to broiler farms, usually within 24 h (Farm and Food Care Ontario and the Poultry Service Association, 2015). Although egg mixing can occur upon their transfer into incubator, hatcher and transport boxes, it is recommended that eggs from each broiler parent flock source are set together (Mauldin and Morrison, 2002). Within a hatcher, eggs are placed on trays stacked on multiple dollies. The number of hatchers and dollies required to reach the flock size N was estimated using capacity data of hatchers (N_hatch) and dollies (N_dolly) from the three main commercial brands used in Canada (Table 2). The probability p_hatch that a hatcher is contaminated with S. Heidelberg was assessed using data from the Salmonella monitoring programme of the Ontario Hatchery Supply Flock Policy (OHSFP),

2.4. Initial conditions: C0 The initial number C0 of S. Heidelberg CFU in the barn environment at placement was defined as the load carried over from the previous batch due to inadequate cleaning and disinfection procedures between production cycles. Canadian chicken broiler farms are required to implement a full cleaning and disinfection in every barn at least once per year, and dry cleaning (including the complete removal of litter and other debris) between each flock cycle (Chicken Farmers of Canada, 2014). No literature could be found on the number of S. Heidelberg CFU carried over in such conditions. Existing literature has focused on the detection or scoring of Salmonella in the barn environment after cleaning and disinfection (Kloska et al., 2017; Maertens et al., 2017). However, these methods rely on semi-quantitative quality monitoring approaches using culture plates that sample a specified area (e.g., direct contact on specified surfaces within the barn or by swab technique), followed by incubation and enumeration, making it difficult to infer the total number of CFU present in the barn environment at the time of sampling. Consequently, an alternative approach was needed to obtain an estimate of C0. An initial version of the model was run with 25,000

Table 2 Capacity of hatchers, dollies and trays used in Canadian broiler hatcheries (sources: (Hatchtech, 2017; Jamesway, 2017; Petersime, 2017). Parameter

Definition

Value

Unit

N_hatch N_dolly N_tray

Hatcher egg capacity Dolly egg capacity Tray egg capacity

∼ Discrete Uniform (7040, 12960, 15120, 16800, 19200, 21120, 28800, 42240) ∼ Discrete Uniform (2160, 2520, 3640, 4160, 6000, 7040) ∼ Discrete Uniform (77,88,130,144,168)

Number of eggs Number of eggs Number of eggs

4

Preventive Veterinary Medicine 174 (2020) 104823 (Government of Canada, 2018b) ∼ 10^Pert (0.78, 2.87, 4.97) (Jiang et al., 2010) Expert opinion (Dr. Agnes Agunos, PHAC) ∼ 10^Pert (1.78, 3.87, 5.97) (Borsoi et al., 2011; Jiang et al., 2010) (Government of Canada, 2018b) ∼ 10^Pert (1.78, 3.87, 5.97)

(Borsoi et al., 2011; Jiang et al., 2010)

(Government of Canada, 2018b) ∼ 10^Pert (1.78, 3.87, 5.97)

(Borsoi et al., 2011; Jiang et al., 2010)

(Government of Canada, 2018b) ∼ 10^Pert (1.78, 3.87, 5.97)

(Borsoi et al., 2011; Jiang et al., 2010)

(Government of Canada, 2018b) ∼ 10^Pert (1.78, 3.87, 5.97) (Borsoi et al., 2011; Jiang et al., 2010) Source

Source

δ (CFU.day-−1)

Fecal excretion rate

C0

C0 = C36_initial × 10^(−Dcleaning) × 10^(−Ddisinf) × (1 − λ)^tdown

(7)

In the absence of literature describing the effect of dry cleaning on Salmonella bacterial populations, the study of Luyckx et al. (2015) describing the effect of water cleaning on the reduction of total aerobic flora was used to define Dcleaning using a Pert distribution (Dcleaning (log10CFU) ∼ Pert (0.8, 1.3, 1.8)) (Luyckx et al., 2015). In the baseline scenario, Ddisinf was set to zero (i.e., no disinfection), and tdown was defined by fitting a distribution to the downtime data obtained from an annual on-farm survey conducted by CIPARS among 656 broiler chicken flocks across Canada between 2013 and 2017 (Government of Canada, 2018b). The possible accumulation of S. Heidelberg CFUs over time (i.e., over multiple batches produced successively on the same farm) was not considered in the present model, as an explorative simulation over five successive batches showed no significant effect on C0 (data not shown). 2.5. Model outputs and sensitivity analysis The model provided three types of outputs: i) the distribution of the chicken population between health states over time, ii) the within-flock prevalence over time (i.e., the proportion of infected-infectious birds out of the total flock population N), and iii) the evolution of the bacterial concentration (CFU per g of feces) in the barn environment over time. Within-flock prevalence and bacterial concentration at pre-harvest (day 36) were compared with existing literature for model validation. Additionally, a sensitivity analysis was performed using partial rank correlation coefficients with 100 bootstrap replicates in order to identify the input parameters that have the largest influence on the within-flock prevalence and the bacterial concentration in the barn environment at pre-harvest. The range of values used for each input parameter included in the sensitivity analysis is presented in Table 4. 2.6. Assessment of possible interventions In addition to the baseline scenario, six intervention scenarios were defined to explore the effect of broiler parent flock vaccination (Scenario 1), the effect of various levels of cleaning and disinfection between batches (Scenarios 2–4), the effect of extended downtime between batches (Scenario 5), as well as the effect of broiler chicken flock vaccination (Scenario 6) (Table 3). The effect of broiler parent flock vaccination was assessed by varying the initial condition I0 of the model, using OHSFP data from 2015, i.e., with reduced p_hatch, p_dolly and p_chick following the introduction of S. Heidelberg vaccination (Scenario 1). More precisely, S. Heidelberg vaccination was part of a Salmonella vaccination strategy implemented in January 2013 in all broiler parent flocks in Ontario, that included live attenuated S. Typhimurium vaccination followed by killed autogenous multivalent vaccination against S. Enteritidis, S. Heidelberg, S. Kentucky and S. Typhimurium (OHSFP, 2017). The effect of cleaning and disinfection between batches was evaluated by varying the initial condition C0 of the model, using different values for Ddisinf as defined in the literature (Payne et al., 2005) (Scenario 2), or using a theoretical enhanced disinfection programme that would result in a 5 (Scenario 3) or 7 (Scenario 4) log reduction of the bacterial population. The effect of extended downtime between batches was assessed using a theoretical

No disinfection was considered in this scenario.

∼ Beta (31,1665) ∼ Beta (36,88) ∼ Beta (129,3468) OHSFP 2011 0 -a ∼ Logistic (16.4,3.7) ∼ Beta (31,1665) ∼ Beta (36,88) ∼ Beta (129,3468) OHSFP 2011 0 -a ∼ Logistic (23.4,3.7) ∼ Beta (31,1665) ∼ Beta (36,88) ∼ Beta (129,3468) OHSFP 2011 7 Hypothetical scenario ∼ Logistic (16.4,3.7) ∼ Beta (31,1665) ∼ Beta (36,88) ∼ Beta (129,3468) OHSFP 2011 5 Hypothetical scenario ∼ Logistic (16.4,3.7) ∼ Beta (31,1665) ∼ Beta (36,88) ∼ Beta (129,3468) OHSFP 2011 ∼ Pert (2.22, 3.67, 4.26) (Payne et al., 2005) ∼ Logistic (16.4,3.7) ∼ Beta (2,1409) ∼ Beta (16,113) ∼ Beta (16,3724) OHSFP 2015 0 -a ∼ Logistic (16.4,3.7)

∼ Beta (31,1665) ∼ Beta (36,88) ∼ Beta (129,3468) OHSFP 2011 0 -a ∼ Logistic (16.4,3.7) (Government of Canada, 2018b) ∼ 10^Pert (1.78, 3.87, 5.97) (Borsoi et al., 2011; Jiang et al., 2010) p_hatch p_dolly p_chick Source Ddisinf (log10CFU) Source tdown (days) I0

iterations to simulate the effect of a new introduction of S. Heidelberg in a barn via vertical introduction (i.e., I0_initial defined as in the baseline scenario, see Table 3) but in the absence of carry-over (i.e., C0_initial = 0). The predicted distribution of the bacterial population (#CFU) in the barn environment at pre-harvest (C36_initial) was used to inform the parameters of a Pert distribution describing C0, after applying the effect of cleaning Dcleaning, disinfection Ddisinf, as well as the effect of S. Heidelberg decay in the barn environment during the downtime period between production cycles tdown:

a

Scenario 6 Baseline + broiler chicken flock vaccination Scenario 5 Baseline + extended downtime between batches Scenario 2 Baseline + disinfection between batches Scenario 1 Baseline + broiler parent flock vaccination Baseline Scenario with cleaning only between batches Scenario

Table 3 Definitions of the model variables altered in Scenarios 1–6 in comparison to the baseline scenario.

Scenario 3 Baseline + enhanced disinfection between batches (5 log reduction)

Scenario 4 Baseline + enhanced disinfection between batches (7 log reduction)

L. Collineau, et al.

5

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

Table 4 Range of input parameters used in the sensitivity analysis. Parameter

Definition

Range (min ; max)

Unit

ρ η γ α δ ω λ I0 C0

Probability of infection given ingestion of 1 CFU of S. Heidelberg Amount of feces ingested daily by a random bird Infectious rate Recovery rate Bacteria excretion rate in the feces Amount of feces shed in the environment by a random bird Mortality rate of S. Heidelberg in the barn environment Number of day-old chicks already infected-infectious at placement Initial number of S. Heidelberg in the barn environment at placement

0 ; 0.4 0;2 0.5 ; 1 1/28 ; 1/21 1 ; 10^7 1 ; 40 0.05 ; 0.15 0 ; 20,000 0 ; 10^10

– gram.day−1 day−1 day−1 CFU.day−1 gram.day−1 CFU.day−1 – CFU

extension of the baseline downtime by 7 days, based on expert opinion of the maximum downtime that can potentially be applied in a typical Canadian commercial broiler chicken flock (Scenario 5). It is worth noting that this extended downtime was longer than the one currently recommended by the Canadian chicken industry (14 days, (Chicken Farmers of Canada, 2018b)). The effect of broiler chicken flock vaccination was evaluated by reducing the fecal excretion rate δ as described by Jiang et al. (2010) observations of the reduction of S. Heidelberg excretion following vaccination with a live attenuated strain of S. Typhimurium (Scenario 6) (Jiang et al., 2010). All the other model input parameters remained equal as described in Tables 1 and 2.

[95 % tolerance interval: 3140–6277] birds still infected-infectious at pre-harvest (day 36). 3.2. Evolution of within-flock prevalence The evolution of within-flock prevalence over time in the baseline and intervention scenarios, as well as the predicted within-flock prevalence at pre-harvest, are presented in Fig. 3 and Table 5, respectively. In agreement with the changes in health states (Fig. 2), Fig. 3 shows that the within-flock prevalence increases sharply to a peak of 0.759 (95 % tolerance interval: 0.57 –0.947) at day 6 in the baseline scenario, and later decreases to 0.235 (95 % tolerance interval: 0.157–0.314) at pre-harvest (day 36). The application of different levels of disinfection (shown as reductions of C0 in Scenarios 2, 3 and 4) in addition to cleaning (as per the baseline Scenario) substantially reduced the peak of the epidemic curve, as well as the within-flock prevalence at pre-harvest. A reduction in within-flock prevalence was also observed with broiler chicken flock vaccination (Scenario 6). Broiler parent flock vaccination (Scenario 1) had no effect on the within-flock prevalence, likely because vertical transmission rarely occurred, even in the baseline scenario (Table 3). Similarly, extension of the downtime between

3. Results 3.1. Distribution of health states over time Fig. 2 represents the distribution of the chicken population among health states over time in the baseline scenario. Following placement at day 0, most birds quickly become infected-non-infectious (state E) and subsequently infected-infectious (state I) within a week of age. Recovery occurs at a relatively slow rate, leading to an average of 4708

Fig. 2. Distribution of the chicken population by health states over time in the baseline scenario. S: susceptible; E: infected-non-infectious; I: infected-infectious; R: recovered and immunized. Solid lines represent mean distributions and dotted lines represent 95 % tolerance intervals around the mean distribution from 25,000 iterations of the model. Closed population of N = 20,000 birds from placement (day 0) until pre-harvest (day 36). 6

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

Fig. 3. Average within-flock prevalence of Salmonella Heidelberg infection in the baseline and intervention scenarios (25,000 iterations). Closed population of N = 20,000 birds from placement (day 0) until pre-harvest (day 36). Baseline and intervention scenarios are defined in Table 3. The curve for Scenario 1 overlaps with the baseline scenario.

batches (Scenario 5) had virtually no impact on the epidemic curve.

3.4. Sensitivity analysis The parameters with the largest influence on the within-flock prevalence and the bacterial concentration at pre-harvest (day 36) are displayed in Fig. 5a and b, respectively. The within-flock prevalence was mostly influenced by the recovery rate α and the initial number of infected-infectious birds at placement I0. The bacterial concentration was mostly influenced by the bacterial excretion rate in feces δ, the daily mortality rate λ of S. Heidelberg in the barn environment, and to a lesser extent, by the recovery rate α.

3.3. Evolution of bacterial concentration The evolution of bacterial concentration over time in the baseline and intervention scenarios, as well as the predicted bacterial concentration at pre-harvest are presented in Fig. 4 and Table 6, respectively. In the baseline scenario and in Scenarios 1, 5 and 6, a peak in bacterial concentration is observed at day 1. This is because the CFU being carried over from the previous batch are ‘diluted’ in a small amount of feces, assumed to be negligible (one gram, as to avoid an undefined expression). After day 1, once bacteria have started to die and feces have started to accumulate in the barn environment, the bacterial concentration follows the epidemic curve as described in Fig. 3 (i.e., the CFU shed by the birds become more influential than the CFU carried over from the previous batch). No such peak is observed in Scenarios 2–4, where disinfection reduces C0. Broiler parent flock vaccination (Scenario 1) and extended downtime between batches (Scenario 5) had no effect on the bacterial concentration over time. Bacterial concentrations following disinfection between batches (Scenarios 2–4) increase rapidly up until days 8–9 to approach the baseline scenario results, and follow the epidemic curve from day 9 onwards (Fig. 4); at this point, the bacterial concentration was more influenced by the epidemic curve than by the CFU carried over from the previous batch. A marked reduction of the bacterial concentration at pre-harvest was only observed with broiler chicken flock vaccination (Scenario 6).

4. Discussion A within-flock model of S. Heidelberg transmission in a typical Canadian commercial broiler chicken flock was developed to estimate difficult-to-obtain parameters, and evaluate on-farm intervention strategies. The inclusion of two environmental compartments, in addition to more traditional SEIR compartments, made it possible to track not only the within-flock prevalence but also the bacterial concentration in the barn environment. Both outcomes are critically lacking in existing Salmonella risk assessment models and can be used as input variables to estimate the prevalence of contaminated birds and the concentrations of exterior contamination, respectively, throughout the transport, processing, and further stages of a farm-to-fork QMRA.

Table 5 Within-flock prevalence at pre-harvest (day 36) in the baseline and intervention scenarios. Scenario Baseline Scenario Scenario Scenario Scenario Scenario Scenario

1 2 3 4 5 6

Description

Mean within-flock prevalence at day 36 (95% tolerance interval)

Scenario with cleaning only between batches Baseline + broiler parent flock vaccination Baseline + disinfection between batches Baseline + enhanced disinfection between batches (5 log reduction) Baseline + enhanced disinfection between batches (7 log reduction) Baseline + extended downtime between batches Baseline + broiler chicken flock vaccination

0.235 0.235 0.091 0.045 0.018 0.229 0.191

7

(0.157–0.314) (0.157–0.314) (0–0.285) (0–0.214) (0–0.134) (0.140–0.318) (0.071–0.311)

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

Fig. 4. Average bacterial concentration (log CFU per g of feces) in the baseline and intervention scenarios (25,000 iterations). Closed population of N = 20,000 birds from placement (day 0) until pre-harvest (day 36). Baseline and intervention scenarios are defined in Table 3. The curve for Scenario 1 overlaps with the baseline scenario.

4.1. Model validation

intentionally included birds arriving from both Salmonella positive and negative farms. Only two studies were found from existing literature that provided an indirect estimate of the within-flock prevalence at pre-harvest by testing individual fecal samples (i.e., individual droppings assumed to be representative of individual birds). One study conducted by the U.S. Department of Agriculture on a broiler chicken farm positive for S. Heidelberg showed that 18.0 % (95 % confidence interval: 9.8–30.8) of individual fecal samples were positive at day 35 (Bailey et al., 2002). Another study conducted among nine conventional broiler chicken farms in Wisconsin, U.S. reported that 21.7 % (95 % confidence interval: 18.0–25.9) of individual fecal samples were positive for Salmonella at day 34 (Siemon et al., 2007). Our estimate falls within the higher end of those estimates, which is in agreement with the conservative approach followed in our model.

4.1.1. Within-flock prevalence estimates The model baseline scenario estimated that an average of 23.5 % (95 % tolerance interval: 15.7–31.4) of birds were infected-infectious on day 36. This estimate was difficult to validate given the lack of data available in the literature, which was one of the reasons for developing this model. For comparison, 7.5 % (95 % confidence interval: 5.2–10.6) of the 374 farms tested at pre-harvest by CIPARS over 2013–2015 were positive for S. Heidelberg, and within the positive farms (i.e., those having at least one positive sample), 72.3 % (95 % confidence interval: 63.4–79.8) of 112 pooled fecal samples were positive for S. Heidelberg (Government of Canada, 2017a). However, direct comparison with the outcome of this model is not possible due to the sampling strategy employed by CIPARS. On the one hand, the use of pooled fecal samples by CIPARS (i.e., samples covering a quadrant of the barn floor) increases the probability of detecting Salmonella, and therefore overestimates the prevalence of infected birds. On the other hand, only one isolate per sample is serotyped by CIPARS, which plausibly decreases the probability of detecting S. Heidelberg in favour of more abundant serovars, such as S. Enteritidis, in samples with more than one serovar. Additionally, a national microbiological baseline study conducted by the Canadian Food Inspection Agency (CFIA) in 2013 showed that an overall 5.3 % (95 % confidence interval: 4.7–6.0) of pooled caecal samples collected from birds at their arrival at slaughter (one pool containing 20 caeca from individual birds) were positive for S. Heidelberg (CFIA, 2016; Dr. Pablo Romero Barrios, personal communication). Again, direct comparison with our results is difficult, as the CFIA

4.1.2. Bacterial concentration estimates The baseline scenario predicted an average bacterial concentration at pre-harvest of 3.579 (0–4.294) log CFU per g of feces. Again, this finding is difficult to validate because of a lack of data from the literature. Berghaus et al. (2013) reported an average concentration of 1.56 (standard deviation: 1.13) log MPN (Most Probable Number) of Salmonella per g of fecal samples and 1.19 (standard deviation: 0.83) log MPN per g of litter samples collected at pre-harvest among 55 commercial broiler chicken flocks in Georgia, U.S. (Berghaus et al., 2013). However, Salmonella serovars recovered by Berghaus et al. (2013) did not include S. Heidelberg. S. Kentucky comprised 61.0 % of isolates recovered by Berghaus et al. (2013), which is excreted at a

Table 6 Bacterial concentration at pre-harvest (day 36) in the baseline and intervention scenarios. Scenario Baseline Scenario Scenario Scenario Scenario Scenario Scenario

1 2 3 4 5 6

Description

Mean bacterial concentration at day 36 (95% tolerance interval) (log CFU per g of feces)

Scenario with cleaning only between batches Baseline + broiler parent flock vaccination Baseline + disinfection between batches Baseline + enhanced disinfection between batches (5 log reduction) Baseline + enhanced disinfection between batches (7 log reduction) Baseline + extended downtime between batches Baseline + broiler chicken flock vaccination

3.579 3.579 3.472 3.365 3.140 3.575 2.505

8

(0–4.294) (0–4.294) (0–4.273) (0–4.245) (0–4.165) (0–4.293) (0–3.242)

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

Fig. 5. (a) Partial rank correlation coefficients (PRCC) between model input parameters and within-flock prevalence at pre-harvest (100 bootstrap replicates). Input parameters are described in Table 1. I0 : number of day-old chicks already infected-infectious at placement; C0 : initial number of S. Heidelberg CFU in the barn environment at placement. Dots represent PRCC indices and bars represent 95 % confidence intervals around indices. (b) Partial rank correlation coefficients (PRCC) between model input parameters and the bacterial concentration in the barn environment at pre-harvest (100 bootstrap replicates). Input parameters are described in Table 1. I0 : number of day-old chicks already infected-infectious at placement; C0: initial number of S. Heidelberg CFU in the barn environment at placement. Dots represent PRCC indices and bars represent 95 % confidence intervals around indices.

lower rate compared to S. Heidelberg (1.15 vs 2.00 log MPN per g of caecal content among non-vaccinated broilers challenged at 1 day of age) (Young et al., 2007). The infectious and recovery rates, as well as

the infectious dose may also differ between serovars, but serovar-specific data are lacking to be able to evaluate this aspect. Other studies that quantified Salmonella concentrations on broiler chicken exteriors 9

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

prior to processing (i.e., upon their arrival at the processing plant) reported diverse levels of contamination, varying from 2.0 (standard deviation: 0.8) to 7.2 (standard deviation: 0.2) log CFU per g on feathers, and from 5.8 (standard deviation: 0.2) to 6.3 (standard deviation: 0.2) log CFU per g on chicken skins (Cason et al., 2007; Kotula and Pandya, 1995). Direct comparison with the level of contamination observed on-farm may, however, be hampered by the occurrence of cross-contamination during transport to slaughter.

Heidelberg specific, and it is unknown whether S. Heidelberg would have higher or lower survival capacity in the barn environment. 4.4. The effect of interventions We explored the effect of selected interventions on the within-flock prevalence and bacterial concentration of S. Heidelberg at pre-harvest. No evidence was found from the peer-reviewed literature that vaccination of broiler parent flocks had an effect on the level of immunity of their progeny; following a conservative approach, we assumed no such effect exists. Should new evidence become publicly available, the doseresponse (ρ parameter) and/or the bacterial excretion rate (δ parameter) could be adjusted accordingly. In the current model, broiler parent flock vaccination was assumed to reduce the bacterial contamination of their eggs, and consequently the prevalence of infectedinfectious chicks at day 0. Pseudo-vertical transmission of S. Heidelberg to day-old chicks (i.e., infection of day-old chicks via contact with contaminated feces on the egg shell or equipment at the hatchery) is likely the main transmission route, although true vertical transmission (via eggs contaminated internally as a result of colonized reproductive organs) has also been described for S. Heidelberg (European Food Safety Authority (EFSA), 2009). In the current model, broiler parent flock vaccination appeared to have a negligible effect on the average within-flock prevalence and bacterial concentration of S. Heidelberg at pre-harvest. This is because the probability of vertical introduction occurring was informed using a surrogate, p_hatch (the probability that a hatcher is contaminated with S. Heidelberg). Based on OHSFP data from fluff sampling at hatcheries, p_hatch was very low; consequently, vertical introduction of S. Heidelberg rarely occurred in the present model, including in the baseline scenario. However, the effect of broiler parent flocks vaccination was likely underestimated by the fact that cross-contamination within incubators and hatchers was not modelled. Broiler parent flocks vaccination likely improves the microbial quality of incubators, hatchers and general hatchery environment (e.g., by reducing the contamination originating from eggs exploding during incubation), but no data were available to quantify this aspect. Further data on S. Heidelberg transmission from broiler parent flocks to hatcheries and from hatcheries to broiler flocks, as well as data on cross-contamination with S. Heidelberg at hatcheries are needed to confirm our findings. It may also be that vaccinating broiler parent flocks against S. Heidelberg is not as effective as vaccinating against other Salmonella serovars such as S. Enteritidis and S. Typhimurium, as other factors may be contributing to the efficacy of the current vaccination scheme (e.g., immunogenicity of the strain used in the autogenous vaccine against the diverse strains of S. Heidelberg present in the field). In the absence of data on the number of CFU remaining in a barn environment after cleaning and/or disinfection (existing studies focus on detecting or scoring Salmonella concentrations after a given incubation period), an alternative approach was used, applying the effects of cleaning with or without disinfection to the bacterial concentration predicted at pre-harvest by our model. No data could be found on the effect of dry cleaning, which is the minimum on-farm requirement in Canadian commercial farms. Therefore, the least effective of the four cleaning protocols studied by Luyckx et al. (2015) (i.e., simple cold water cleaning with a cleaning product) was used as a surrogate, but this likely overestimates the effect of dry cleaning. Additionally, the cleaning effect was assessed on the total aerobic flora, not on Salmonella in particular (Liljebjelke et al., 2005). Similarly, the effect of disinfection (Scenario 2) was based on a laboratory trial mimicking the on-farm effect of four disinfectants applied on nalidixic acid-resistant S. Typhimurium (Payne et al., 2005). It is unknown whether S. Heidelberg would respond similarly to those disinfectants. A lower initial bacterial concentration C0 (obtained via increasing levels of disinfection in Scenarios 2–4) resulted in a more pronounced reduction in within-flock prevalence and bacterial concentration.

4.2. Initial conditions Data were lacking to accurately define I0, so surrogate values from existing surveillance programs in broiler parent flocks and hatcheries had to be used. The probability of a hatcher being contaminated with S. Heidelberg (p_hatch) was low in the baseline scenario, therefore limiting the occurrence of vertical introduction. For those instances where vertical introduction did occur, however, I0 was relatively high because of the high probability of dolly contamination (p_dolly). The rarity of vertical introduction of S. Heidelberg in the barn seems to be in line with CIPARS observations: out of 451 chick pads tested at placement of day-old chicks on 150 farms over 2013–2015, only 0.4 % (95 % confidence interval: 0.12–1.6) were positive for S. Heidelberg (Government of Canada, 2017a). 4.3. Interpreting the sensitivity analysis The sensitivity analysis showed that the within-flock prevalence at pre-harvest was mainly influenced by the recovery rate α and the initial number of infected-infectious birds at placement I0. There is uncertainty surrounding the duration of the Salmonella infectious period in chicken, due to the tendency for intermittent excretion to occur (Van Immerseel et al., 2004). In an experimental study for S. Typhimurium, bacteria were either completely eliminated from systemic organs or reduced to very low counts (i.e., below detection limits) within 21 days after the first instance of excretion (Bjerrum et al., 2003). Other studies on paratyphoidal serovars reported that it takes 28 days for the number of chickens excreting bacteria to gradually decline (Poppe, 2000). Consultation with an expert on Salmonella in poultry confirmed that Salmonella in feces become virtually undetectable 21–28 days following the first instance of excretion (Dr. Richard K. Gast, U.S. National Poultry Research Center). Our triangular distribution, defined by a range of 21–28 days and a most likely value of 23 days (as reported by Gast et al. (2005) for S. Heidelberg infection in hens) therefore appeared reasonable. The sensitivity analysis further showed that the bacterial excretion rate in feces δ had a large influence on the bacterial concentration at pre-harvest. Two studies were used to provide data to inform this parameter (Borsoi et al., 2011; Jiang et al., 2010). Jiang et al. (2010) reported two oral inoculation experiments of 5 broiler chickens with 106 CFU of S. Heidelberg observed from day 3 until day 14 post-inoculation as part of a vaccination assay (the control group being used to inform our model), while the Borsoi et al. (2011) study was based on an oral inoculation experiment of 15 broiler chickens with 105 CFU of S. Heidelberg, observed from day 2 until day 20 post-inoculation. The bacterial excretion rate observed by Borsoi et al. (2011) was relatively stable over time, while the rate observed by Jiang et al. (2010) showed no clear temporal pattern. A single distribution of δ was therefore used from placement until pre-harvest in the present model and aimed to capture the most likely values between both studies. Additional data on S. Heidelberg fecal excretion rates in chicken are critically needed to improve the precision of the model. The daily mortality rate λ of Salmonella in the barn environment also had a large impact on the bacterial concentration at pre-harvest. We used the daily mortality rate proposed by Prévost et al. (2006), who had tried to recreate the bacterial concentration diminution observed by Hollinger et al. (2000). However, this mortality rate was not S. 10

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

However, existing literature suggested that a 7 log reduction of C0 might be difficult to achieve in practice. Luyckx et al. (2015) reported maximums of 3.9 and 6.2 log reductions in total aerobic flora and Enterococcus spp., respectively, following thorough cleaning and disinfection procedures in 12 broiler houses from 5 Belgian farms. Further data are needed to characterize the maximum achievable reduction in S. Heidelberg. Additionally, routine cleaning and disinfection is not practical under Canadian conditions, especially during cold seasons when extended periods of time are needed to dry the full barn before placement of a new batch of birds. Extending downtime between batches by 7 days (Scenario 5) had little effect on within-flock prevalence or bacterial concentration of S. Heidelberg when compared to the baseline scenario. This was surprising, as the sensitivity analysis showed a significant influence of the daily mortality rate λ of Salmonella in the barn environment on the bacterial concentration at pre-harvest. It may be that the extended downtime period considered in Scenario 5 was still too short to be effective against S. Heidelberg, although it is believed to be ideal for most other on-farm pathogens such as viral agents, fungal and protozoal agents (Dr. Agnes Agunos, PHAC, personal communication). Combining an extended downtime period with enhanced cleaning and disinfection procedures (e.g., combining Scenarios 2 and 5 together) may prove to be more beneficial, as suggested for the control of Campylobacter spp. (Agunos et al., 2014). Broiler flock vaccination (Scenario 6) substantially reduced the bacterial concentration at pre-harvest, although the effect on within-flock prevalence was more limited. This is likely because broiler flock vaccination reduced the bacterial excretion rate δ, which was shown, via sensitivity analysis, to be the most influential parameter on the bacterial concentration at pre-harvest. Broiler flock vaccination against Salmonella is, however, not common practice in Canada. Only 3.7 % (5 out of 136) of the chicken broiler flocks surveyed by CIPARS in 2016 were vaccinating against Salmonella, while none of the 373 flocks surveyed over 2013–2015 were doing so (Government of Canada, 2018b) Other measures to control horizontal transmission of S. Heidelberg (e.g., competitive exclusion, and feed and water additives) were not included in the present study, as data were already available from existing systematic reviews and meta-analyses. Their effect will be considered in the overall QMRA.

assumed that transmission occurred exclusively via ingestion of S. Heidelberg CFU in the barn environment, which necessitated the development of our own dose-response model for S. Heidelberg in chicken. This dose-response model was based on experiments with various Salmonella serovars (including S. Enteriditis, S. Heidelberg and S. Typhimurium) given the paucity of S. Heidelberg specific data. A simple exponential dose-response model was used, given the sparse data available, but other dose-response relationships are described in the literature, and could be considered provided additional data become available ?(Vose, 2008). The hypotheses of homogenous distribution and mixing of CFU in the barn environment, as well as the equal probability of CFU ingestion by all birds are obvious simplifications of the transmission dynamics inherent to this type of model. The model focused on positive flocks, for which carry-over between batches systematically occurred (although at different levels), and vertical introduction sporadically occurred. It does not imply that every Canadian farm is positive for S. Heidelberg. As mentioned earlier, only 7.5 % (95 % confidence interval: 5.2–10.6) of the 374 farms tested at pre-harvest by CIPARS over 2013–2015 were positive for S. Heidelberg (Government of Canada, 2017a). However, it was considered a waste of computational power to model those farms where no transmission was occurring. The proportion of positive and negative flocks will therefore be integrated into future QMRA modelling. In addition, the assumption that carry-over between batches systematically happens in a positive flock seemed reasonable considering the type of cleaning and disinfection procedures typically used in Canadian broiler farms (i.e., dry cleaning only), and the fact that no on-farm control programme specific to S. Heidelberg is currently in place. Canadian data were used to populate the model whenever possible. However, data used to inform the occurrence of vertical transmission of S. Heidelberg and the associated number I0 of day-old chicks already infected-infectious at placement originated from Ontario only. Although monitoring programs for Salmonella in broiler parent flocks and hatcheries exist in other Canadian provinces, these data are not publicly available and not specific to S. Heidelberg. Since 2013, Ontario has been the Canadian province with the highest prevalence of S. Heidelberg on broiler chicken farms (Government of Canada, 2017a). The use of Ontario data may therefore have introduced an overestimation of I0, which is acceptable considering the conservative approach used in our model. Additionally, under the 2017 tariff rate quotas of the North American Free Trade Agreement (NAFTA), 17.4 % of hatching eggs used in Canada originate from US broiler parent flocks, and 3.7 % of day-old chicks placed in Canadian broiler chicken farms originate from US hatcheries. This practice was ignored in the present model in the absence of public data on the Salmonella status of those eggs and birds (Agriculture and Agri-Food Canada, 2018). The present study highlighted a number of major data gaps, namely the amount of feces ingested daily by a random bird (parameter η) and the infectious rate (parameter γ), as well as the bacteria excretion rate in the feces (parameter δ, for which highly variable results are reported in the literature). These data gaps should be addressed, for example via field experiments, for future improvement of the model.

4.5. Limitations Typical of models of this nature, our model had both data gaps (as mentioned) and limitations. First, a limited number of routes for the introduction of S. Heidelberg into the barn were considered. Contaminated feed is also known to be a risk factor for Salmonella in chicken broiler production (Marin et al., 2011), but S. Heidelberg has not been found by CIPARS in feed/feed ingredients (Government of Canada, 2017a). Salmonella can also enter into a barn via multiple other routes related to poor biosecurity practices, such as contaminated wildlife, personnel, visitors or equipment (Arsenault et al., 2007; Le Bouquin et al., 2010). For example, a study conducted among 81 broiler chicken flocks in Québec showed that chicken flocks that failed to systematically lock the chicken house had 2.6 higher odds of colonization with Salmonella compared to those that did not (Arsenault et al., 2007). However, while such introductions likely occur, it is very difficult to know how often, at which time point, and at which concentration Salmonella are introduced into the barn via these routes. Moreover, introductions via these routes are not necessarily controllable. In a similar within-flock transmission model developed for Campylobacter where various chicken ages at first exposure to the bacteria were considered, it was recognized that age at first exposure was a rather difficult parameter to control, therefore limiting its value for intervention purposes (Hartnett et al., 2001). The fecal-oral route is believed to be the main route of Salmonella transmission between broilers (Foley et al., 2013; Heres et al., 2004) although direct transmission between birds is possible. Our model

5. Conclusions A transmission model for S. Heidelberg was developed to inform the within-flock prevalence and the bacterial environmental concentration in a positive farm at pre-harvest. Both outcomes are needed to link farm-related practices to the occurrence of S. Heidelberg downstream in the farm-to-fork continuum in future quantitative risk assessments. While the within-flock prevalence could be estimated with good precision, the precision around bacterial environmental concentration estimates was poorer due to data gaps. With acknowledgment of the current constraints and limitations of our model, vaccination of broiler parent flocks against S. Heidelberg had a negligible effect given the rare occurrence of S. Heidelberg vertical introduction, while vaccination of 11

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al.

broiler flocks considerably reduced the bacterial environmental concentration at pre-harvest. Cleaning and disinfection likewise reduced the within-flock prevalence at pre-harvest, but the effect on bacterial environmental concentration was limited to the beginning of the production period. Extending downtime between batches by 7 days showed no reduction of within-flock prevalence or bacterial environmental concentration of S. Heidelberg when compared to the baseline scenario. The public health impact of these measures will be assessed by integrating the results from this study into a farm-to-fork QMRA model.

food/chemical-residues-microbiology/food-safety-testing-bulletins/2016-08-17/ december-2012-december-2013/eng/1471358115567/1471358175297?chap=0# s7c11 (Accessed January 2019). . Chicken Farmers of Canada, 2014. On-Farm Food Safety Assurance Program. (Accessed January 2019). https://www.chickenfarmers.ca/wp-content/uploads/2014/07/ OFFSAP-Manual-2014.pdf. Chicken Farmers of Canada, 2018a. Chicken Data Booklet. (Accessed January 2019). https://www.chickenfarmers.ca/wp-content/uploads/2018/07/2018-Data-Booklet_ NEW.pdf. Chicken Farmers of Canada, 2018b. Raised by a Canadian Farmer On-Farm Food Safety Program (2014) and Animal Care Program (2018) Flock-Specific Records V6.0. (Accessed January 2019). https://www.chickenfarmers.ca/wp-content/uploads/ 2014/07/Flock-Specific-Records-v.-6.0_-EN_2018-singles.pdf. Davidson, V.J., Ravel, A., Nguyen, T.N., Fazil, A., Ruzante, J.M., 2011. Food-specific attribution of selected gastrointestinal illnesses: estimates from a Canadian expert elicitation survey. Foodborne Pathog. Dis. 8, 983–995. De Silva, K.R., Eda, S., Lenhart, S., 2017. Modeling environmental transmission of MAP infection in dairy cows. Math. Biosci. Eng. 14, 1001–1017. Dutil, L., Irwin, R., Finley, R., Ng, L.K., Avery, B., Boerlin, P., Bourgault, A., 2010. Ceftiofur resistance in Salmonella enterica serovar Heidelberg from chicken meat and humans, Canada. Emerg. Infect. Dis. 16, 48–54. European Food Safety Authority (EFSA), 2009. Quantitative estimation of the impact of setting a new target for the reduction of Salmonella in breeding hens of Gallus gallus. EFSA J. 7, 1036. Farm & Food Care Ontario and the Poultry Service Association, 2015. Ontario Poultry Handling and Transportation Manual. (Accessed January 2019). http://www. poultryserviceassociation.com/uploads/2/7/9/6/27967763/2017_poultry_handling_ and_transportation_manual.pdf. Foley, S.L., Johnson, T.J., Ricke, S.C., Nayak, R., Danzeisen, J., 2013. Salmonella pathogenicity and host adaptation in chicken-associated serovars. Microbiol. Mol. Biol. Rev. 77, 582–607. Gast, R.K., Guard-Bouldin, J., Holt, P.S., 2005. The relationship between the duration of fecal shedding and the production of contaminated eggs by laying hens infected with strains of Salmonella Enteritidis and Salmonella Heidelberg. Avian Dis. 49, 382–386. Government of Canada, 2015. National Strategy for the Control of Poultry-Related Human Salmonella Enteritidis Illness in Canada. Bureau of Microbial Hazards, Food Directorate, Health Products and Food Branch. Government of Canada, 2017a. Canadian Integrated Program for Antimicrobial Resistance Surveillance (CIPARS) 2015 Annual Report. Public Health Agency of Canada, Guelph, Ontario. Government of Canada, 2017b. Foodnet Canada. Short report 2015. Public Health Agency of Canada, Ottawa. Government of Canada, 2018a. National Enteric Surveillance Program Annual Summary 2016. Public Health Agency of Canada, Guelph, Ontario. Government of Canada, 2018b. Canadian Integrated Program for Antimicrobial Resistance Surveillance (CIPARS) 2016 Annual Report. Public Health Agency of Canada, Guelph, Ontario. Hartnett, E., Kelly, L., Newell, D., Wooldridge, M., Gettinby, G., 2001. A quantitative risk assessment for the occurrence of Campylobacter in chickens at the point of slaughter. Epidemiol. Infect. 127, 195–206. Hatchtech, 2017. Hatchtech Incubation Technology. Technical Specifications. (Accessed January 2019). https://hatchtechgroup.com/. Heres, L., Urlings, H.A.P., Wagenaar, J.A., de Jong, M.C.M., 2004. Transmission of Salmonella between broiler chickens fed with fermented liquid feed. Epidemiol. Infect. 132, 107–116. Hollinger, K., 2000. Epidemiology and salmonellosis. In: Wray, C., Wray, A. (Eds.), Salmonella in Domestic Animals. CABI Publishing, New York, N.Y, pp. 341–354. Jamesway, 2017. JAMESWAY Incubator Company Inc. Platinum2.0. Single Stage Incubators and Hatchers. (Accessed January 2019). https://www.jamesway.com. Jiang, Y., Kulkarni, R.R., Parreira, V.R., Poppe, C., Roland, K.L., Prescott, J.F., 2010. Assessment of 2 Salmonella enterica serovar Typhimurium-based vaccines against necrotic enteritis in reducing colonization of chickens by Salmonella serovars of different serogroups. Can. J. Vet. Res. 74, 264–270. Kariuki, S., Gordon, M.A., Feasey, N., Parry, C.M., 2015. Antimicrobial resistance and management of invasive Salmonella disease. Vaccine 33 (Suppl. 3), C21–C29. Keeling, M., Rohani, P., 2008. Modeling Infectious Diseases in Humans and Animals. Princeton University Press, Princeton, New Jersey. Kennedy, M., Villar, R., Vugia, D.J., Rabatsky-Ehr, T., Farley, M.M., Pass, M., Smith, K., Smith, P., Cieslak, P.R., Imhoff, B., Griffin, P.M., 2004. Hospitalizations and deaths due to Salmonella infections, FoodNet, 1996 -1999. Clin. Infect. Dis. 38, S142–S148. Kerr, A.K., Farrar, A.M., Waddell, L.A., Wilkins, W., Wilhelm, B.J., Bucher, O., Wills, R.W., Bailey, R.H., Varga, C., McEwen, S.A., Rajic, A., 2013. A systematic reviewmeta-analysis and meta-regression on the effect of selected competitive exclusion products on Salmonella spp. prevalence and concentration in broiler chickens. Prev. Vet. Med. 111, 112–125. Kloska, F., Casteel, M., Kump, F.W., Klein, G., 2017. Implementation of a risk-orientated hygiene analysis for the control of Salmonella java in the broiler production. Curr. Microbiol. 74, 356–364. Kotula, K.L., Pandya, Y., 1995. Bacterial contamination of broiler chickens before scalding. J. Food Prot. 58, 1326–1329. Le Bouquin, S., Allain, V., Rouxel, S., Petetin, I., Picherot, M., Michel, V., Chemaly, M., 2010. Prevalence and risk factors for Salmonella spp. contamination in French broilerchicken flocks at the end of the rearing period. Prev. Vet. Med. 97, 245–251. Liljebjelke, K.A., Hofacre, C.L., Liu, T., White, D.G., Ayers, S., Young, S., Maurer, J.J., 2005. Vertical and horizontal transmission of Salmonella within integrated broiler production system. Foodborne Pathog. Dis. 2, 90–102.

Funding sources This study was funded by the Government of Canada as part of the Genomics Research and Development Initiative on Antimicrobial Resistance (GRDI-AMR). Acknowledgments The authors would like to thank Dr. Rachel Ouckama (Maple Lodge Hatcheries Ltd) for sharing data on the Ontario broiler breeder Salmonella surveillance program and provincial vaccination program for breeders and Dr. Richard K. Gast (U.S. National Poultry Research Center) for providing expertise on Salmonella pathogenicity in chicken. Thank you to Dr. Pablo Romero Barrios (Canadian Food Inspection Agency) for providing S. Heidelberg data from the national microbiological baseline study. We gratefully acknowledge Dorica and Silviu Popa for providing data on the mass of feces produced by their chickens. 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.2019. 104823. References Agriculture and Agri-Food Canada, 2018. Canada’s Poultry Import Regime. (Accessed January 2019). http://www.agr.gc.ca/eng/industry-markets-and-trade/canadianagri-food-sector-intelligence/poultry-and-eggs/poultry-and-egg-market-information/ imports-and-exports/canada-s-poultry-import-regime/?id=1384971854404#trq. Agunos, A., Waddell, L., Léger, D., Taboada, E., 2014. A systematic review characterizing on-farm sources of Campylobacter spp. for broiler chickens. PLoS One 9, e104905. Agunos, A., Léger, D.F., Carson, C.A., Gow, S.P., Bosman, A., Irwin, R.J., Reid-Smith, R.J., 2017. Antimicrobial use surveillance in broiler chicken flocks in Canada, 2013–2015. PLoS One 12, e0179384. Arsenault, J., Letellier, A., Quessy, S., Normand, V., Boulianne, M., 2007. Prevalence and risk factors for Salmonella spp. and Campylobacter spp. caecal colonization in broiler chicken and turkey flocks slaughtered in Québec, Canada. Prev. Vet. Med. 81, 250–264. Bailey, J.S., Cox, N.A., Craven, S.E., Cosby, D.E., 2002. Serotype tracking of Salmonella through integrated broiler chicken operations. J. Food Prot. 65, 742–745. Berghaus, R.D., Thayer, S.G., Law, B.F., Mild, R.M., Hofacre, C.L., Singer, R.S., 2013. Enumeration of Salmonella and Campylobacter spp. in environmental farm samples and processing plant carcass rinses from commercial broiler chicken flocks. J. Appl. Environ. Microbiol. 79, 4106–4114. Bjerrum, L., Engberg, R.M., Pedersen, K., 2003. Infection models for Salmonella Typhimurium DT110 in day-old and 14-day-old broiler chickens kept in isolators. Avian Dis. 47, 1474–1480. Borsoi, A., do Santos, L.R., Rodrigues, L.B., de Souza Moraes, H.L., Salle, C.T.P., do Nascimento, V.P., 2011. Behavior of Salmonella Heidelberg and Salmonella Enteritidis strains following broiler chick inoculation: evaluation of cecal morphometry, liver and cecum bacterial counts and fecal excretion patterns. B. J. Microbiol. 42, 266–273. Bucher, O., Fazil, A., Rajic, A., Farrar, A., Willis, R., McEwen, S.A., 2012. Evaluating interventions against Salmonella in broiler chickens: applying synthesis research in support of quantitative exposure assessment. Epidemiol. Infect. 140, 925–945. Butler, A.J., Thomas, M.K., Pintar, K.D.M., 2015. Expert elicitation as a means to attribute 28 enteric pathogens to foodborne, waterborne, animal contact, and person-to-person transmission routes in Canada. Foodborne Pathog. Dis. 12, 335–344. Cason, J.A., Hinton Jr, A., Northcutt, J., Buhr, R., Ingram, K.D., Smith, D.P., Cox, N., 2007. Partitioning of external and internal bacteria carried by broiler chickens before processing. J. Food Prot. 70, 2056–2062. CFIA, 2016. Canadian Food Inspection Agency, National Microbiological Baseline Study in Broiler Chicken. December 2012 – December 2013. http://www.inspection.gc.ca/

12

Preventive Veterinary Medicine 174 (2020) 104823

L. Collineau, et al. Luyckx, K.Y., Van Weyenberg, S., Dewulf, J., Herman, L., Zoons, J., Vervaet, E., Heyndrickx, M., De Reu, K., 2015. On-farm comparisons of different cleaning protocols in broiler houses. Poult. Sci. 94, 1986–1993. Maertens, H., De Reu, K., Van Weyenberg, S., Van Coillie, E., Meyer, E., Van Meirhaeghe, H., Van Immerseel, F., Vandenbroucke, V., Vanrobaeys, M., Dewulf, J., 2017. Evaluation of the hygienogram scores and related data obtained after cleaning and disinfection of poultry houses in Flanders during the period 2007 to 2014. Poult. Sci. 97, 620–627. Maijala, R., Ranta, J., Seuna, E., Pelkonen, S., Johansson, T., 2005. A quantitative risk assessment of the public health impact of the Finnish Salmonella control program for broilers. Int. J. Food Microbiol. 102, 21–35. Marin, C., Balasch, S., Vega, S., Lainez, M., 2011. Sources of Salmonella contamination during broiler production in Eastern Spain. Prev. Vet. Med. 98, 39–45. Mauldin, J.M., Morrison, T., 2002. Equipment for hatcheries. In: Bell, D.D., Weaver, W.D. (Eds.), Commercial Chicken Meat and Egg Production. Springer US, Boston, MA, pp. 685–705. Nauta, M.J., Giessen, V.D., Henken, ï.A.M., 2000. A model for evaluating intervention strategies to control Salmonella in the poultry meat production chain. Epidemiol. Infect. 124, 365–373. O’Mahoney, M., Barnes, H., Stanwell-Smith, R., Dickens, T., Jephcott, A., 1990. An outbreak of Salmonella Heidelberg infection associated with a long incubation period. J. Public Health 12, 19–21. OHSFP, 2017. Ontario Hatchery Supply Flock Policy. Efficacy of Ontario Broiler Breeder Salmonell Avaccination Program. Parsons, D.J., Orton, T.G., D’Souza, J., Moore, A., Jones, R., Dodd, C.E.R., 2005. A comparison of three modelling approaches for quantitative risk assessment using the case study of Salmonella spp. in poultry meat. Int. J. Food Microbiol. 98, 35–51. Payne, J.B., Kroger, E.C., Watkins, S.E., 2005. Evaluation of disinfectant efficacy when applied to the floor of poultry grow-out facilities. J. Appl. Poult. Res. 14, 322–329. Petersime, 2017. Conventional Series Multi- and Single-Stage Incubators. (Accessed January 2019). http://www.petersime.com/images/uploads/widgets/LR_PET_ 03960_ConventionalBrochure_UK_update.pdf. Poppe, C., 2000. Chapter 7: Salmonella infections in the domestic fowl. In: Wray, C., Wray, A. (Eds.), Salmonella in Domestic Animals. CABI Publishing, New York, N.Y, pp. 107–132. Prévost, K., Magal, P., Beaumont, C., 2006. A model of Salmonella infection within industrial house hens. J. Theor. Biol. 242, 755–763. Protais, J., Colin, P., Beaumont, C., Guillot, J.F., Lantier, F., Pardon, P., Bennejean, G., 1996. Line differences in resistance to Salmonella Enteritidis PT4 infection. Br. Poult. Sci. 37, 329–339. R Development Core Team, 2018. R: a Language and Environment for Statistical Computing. ISBN 3-900051-07-0. Retrieved 01/08, 2017, from. R foundation for statistical computing, Vienna, Austria. http://www.R-project.org). Rajan, K., Shi, Z., Ricke, S.C., 2017. Current aspects of Salmonella contamination in the US poultry production chain and the potential application of risk strategies in

understanding emerging hazards. Crit. Rev. Microbiol. 43, 370–392. Ranta, J., Maijala, R., 2002. A probabilistic transmission model of Salmonella in the primary broiler production chain. Risk Anal. 22, 47–58. Roy, P., Dhillon, A.S., Shivaprasad, H.L., Schaberg, D.M., Bandli, D., Johnson, S., 2001. Pathogenicity of different serogroups of avian Salmonellae in specific-pathogen-free chickens. Avian Dis. 45, 922–937. Sadler, W.W., Brownell, J.R., Fanelli, M.J., 1969. Influence of age and inoculum level on shed pattern of Salmonella Typhimurium in chickens. Avian Dis. 13, 793–803. Shane, A.L., Mody, R.K., Crump, J.A., Tarr, P.I., Steiner, T.S., Kotloff, K., Langley, J.M., Wanke, C., Warren, C.A., Cheng, A.C., Cantey, J., Pickering, L.K., 2017. 2017 Infectious Diseases Society of America clinical practice guidelines for the diagnosis and management of infectious diarrhea. Clin. Infect. Dis. 65, e45–e80. Siemon, C.E., Bahnson, P.B., Gebreyes, W.A., 2007. Comparative investigation of prevalence and antimicrobial resistance of Salmonella between pasture and conventionally reared poultry. Avian Dis. 51, 112–117. Totton, S.C., Farrar, A.M., Wilkins, W., Bucher, O., Waddell, L.A., Wilhelm, B.J., McEwen, S.A., Rajic, A., 2012a. The effectiveness of selected feed and water additives for reducing Salmonella spp. of public health importance in broiler chickens: a systematic review, meta-analysis, and meta-regression approach. Prev. Vet. Med. 106, 197–213. Totton, S.C., Farrar, A.M., Wilkins, W., Bucher, O., Waddell, L.A., Wilhelm, B.J., McEwen, S.A., Rajić, A., 2012b. A systematic review and meta-analysis of the effectiveness of biosecurity and vaccination in reducing Salmonella spp. in broiler chickens. Food Res. Int. 45, 617–627. Van der Fels-Klerx, H.J., Tromp, S., Rijgersberg, H., van Asselt, E.D., 2008. Application of a transmission model to estimate performance objectives for Salmonella in the broiler supply chain. Int. J. Food Microbiol. 128, 22–27. Van Immerseel, F., De Buck, J., Pasmans, F., Bohez, L., Boyen, F., Haesebrouck, F., Ducatelle, R., 2004. Intermittent long-term shedding and induction of carrier birds after infection of chickens early post hatch with a low or high dose of Salmonella Enteritidis. Poult. Sci. 83, 1911–1916. Vieira, A., Jensen, A.R., Pires, S., Karlsmose, S., Wegener, H.C., Wong, D.L.F., 2009. WHO global foodborne infections network Country databank – a resource to link human and non-human sources of Salmonella. In: 12th Symposium of the International Society for Veterinary Epidemiology and Economics- Durban. South Africa. Vose, D., 2008. Risk Analysis: A Quantitative Guide. John Wiley & Sons. Vugia, D.J., Samuel, M., Farley, M.M., Marcus, R., Shiferaw, B., Shallow, S., Smith, K., Angulo, F.J., 2004. Invasive Salmonella infections in the United States, FoodNet, 1996 -1999: Incidence, serotype distribution, and outcome. Clin. Infect. Dis. 38, S149–S156. Young, S.D., Olusanya, O., Jones, K.H., Liu, T., Liljebjelke, K.A., Hofacre, C.L., 2007. Salmonella incidence in broilers from breeders vaccinated with live and killed Salmonella. J. Appl. Poult. Res. 16, 521–528. Zongo, P., Viet, A., Magal, P., Beaumont, C., 2010. A spatio-temporal model to describe the spread of Salmonella within a laying flock. J. Theor. Biol. 267, 595–604.

13