The Danish broiler industry, declared Salmonella-free by the EU since 2018, faces continuous Salmonella pressure. Parent flock infections pose a risk of vertical transmission to broiler flocks. This study addresses the need for a robust model to evaluate the effectiveness of the current Salmonella surveillance program and potential alternatives. The existing program relies on a top-down control approach, monitoring parent flocks and hatcheries, and testing broiler flocks before slaughter using environmental samples (boot swabs and dust). The conventional broiler production system operates as a pyramid, with parent flocks supplying eggs to hatcheries, which then provide chicks to broiler farms. All stages utilize an all-in/all-out biosecurity procedure. While Salmonella incidence is low, vertical transmission from infected parent flocks to broiler flocks remains a concern. Detecting infected parent flocks promptly before hatching is crucial to prevent widespread contamination. The decision of how many eggs to destroy from an infected parent flock (eggs produced in the week before detection or three weeks) balances minimizing transmission risk with economic compensation to breeders. The study aims to estimate: (i) the likelihood of detecting infected parent flocks within three weeks; (ii) the likelihood of detecting vertically infected broiler flocks; and (iii) the effect of enhanced environmental sampling in parent flocks on detection time. This information will help guide decision-making by the poultry industry and Danish food authorities.

Previous research highlights the historical increase in human salmonellosis linked to broiler chickens in Denmark (Wegener et al., 2003). Voluntary and mandatory Salmonella control programs have been implemented (Bisgaard, 1992). Studies on Salmonella sources and movement through integrated poultry operations have also been conducted (Bailey et al., 2001; Carrique-Mas et al., 2008; Soria et al., 2017). Rosenquist et al. (2003) and Pedersen et al. (2003) contributed research on risk assessment and organic broiler production. The authors reference work on the transmission of Salmonella within flocks, noting that the infectiousness of infected hens declines over time after infection (Thomas et al., 2009; Braden, 2006; Holt et al., 2007; Gast & Holt, 1998; Foley et al., 2011). The study also mentions Collineau et al. (2020) who used a similar SIR model approach for Salmonella transmission in Canadian broiler chickens. Several other studies on Salmonella transmission within poultry flocks are cited, emphasizing the variability in transmission rates between Salmonella strains (Liljebjelke et al., 2005; Thiagarajan et al., 1994; Cason et al., 1994). The methodology for estimating prevalence-dependent sensitivity of the monitoring systems draws on previous work done in UK's non-cage egg-layer production (Arnold et al., 2014, 2010; Arnold et al., 2014). The impact of sampling methods on Salmonella detection in broiler flocks has also been examined (Skov et al., 1999).

The study employed stochastic dynamic modeling to simulate Salmonella spread and detection within poultry flocks. For parent flocks, a modified SIR model (SI model with multiple infected compartments) was used, incorporating a decline in transmission rate over time for each infected hen. The model considers several factors: the number of initially infected hens, the transmission rate between hens, and the sensitivity of environmental sampling. The transmission rate decline was modeled using an equation from Thomas et al. (2009). The model assumes homogeneous mixing within the flock. The parameters for the dynamic model were derived from previous work by Thomas et al. (2009) and considered uncertainties in transmission rate and decline in environmental infectiousness. The prevalence-dependent sensitivity of the monitoring system was calculated using a function derived from Arnold et al. (2014), incorporating parameters for intercept and prevalence dependence. Monte Carlo simulation with 1001 iterations was used to account for uncertainty in the transmission rate and sensitivity parameters. Different scenarios were simulated, varying flock size (6000 and 12000 hens) and initial number of infected hens (10 and 100). For broiler flocks, the model assumed an initial infection of one chick and used a transmission rate from Heres et al. (2004), simulating the sampling procedure at days 16 and 26 post-insertion. The likelihood of detecting an infected flock was calculated based on the mean sensitivity of detection at different time points after introduction, considering the three weekly sampling points. The model was coded in the mc2d package in R. The current and alternative sampling methods are described in detail. The current method involves collecting two pairs of boot swabs per flock, while the alternative method involves collecting five pairs or a combination of boot swabs and dust samples. The study detailed the calculation of overall sensitivity, taking into account the time of sampling after Salmonella introduction. The uncertainty in overall sensitivity was estimated using Monte Carlo simulation, considering the uncertainty in transmission rates and the sensitivity function.

The study's key findings include: (1) The likelihood of detecting an infected parent flock within the first three weeks after infection is significantly influenced by the sampling method. Using five boot swabs (95% CI 70–100) is far more effective than using two (95% CI 40–100) or two boot swabs supplemented by a dust sample (95% CI 43–100). The likelihood is also strongly influenced by the initial number of infected hens. (Table 2 illustrates the median and 95% confidence intervals for various scenarios). (2) The likelihood of detecting a vertically infected broiler flock is very high, estimated at 100% (95% CI 99–100), even with only one initially infected chick in a flock of 40,000 (Figure 3). (3) The model demonstrates that with the current sampling approach, Salmonella can spread undetected in a parent flock for several weeks before testing positive, posing a risk of vertical transmission to broiler flocks. (Figure 1 and 2 graphically represent the estimated sensitivity to detect Salmonella using different sampling methods and initial infection levels). (4) Alternative sampling schedules (more frequent sampling and more samples) consistently show higher sensitivity at any given prevalence. (5) The model results align with observed data from the 2017–2018 surveillance program, where Salmonella was initially detected in broiler flocks before the parent flocks.

The results underscore the critical role of early detection in managing Salmonella outbreaks. The model's high sensitivity for broiler flocks suggests the current surveillance program is effective in preventing infected broilers from reaching slaughter. However, the low sensitivity for parent flocks highlights the need for improvements. The findings emphasize the benefits of increasing the frequency and intensity of environmental sampling in parent flocks to detect infections earlier, minimizing the risk of vertical transmission. The use of a stochastic model allowed for the incorporation of uncertainties in transmission rates and sensitivity, providing a more realistic assessment of surveillance program performance. The study's approach can be generalized to assess surveillance programs for other infectious diseases in various populations. Although the model parameters are based on S. Enteritidis, the study acknowledges potential variations in transmission rates for other strains and the impact of farm management practices. While the estimated sensitivity represents an average, it offers valuable support for decision-making regarding sampling strategies. Future work could explore the integration of farm-specific variations in transmission rates and investigate other sampling methodologies.

This study provides valuable insights into the effectiveness of Salmonella surveillance programs in broiler production. The model highlights the critical need to improve early detection in parent flocks by enhancing sampling strategies, reducing the risk of vertical transmission. The high sensitivity in broiler flocks suggests the program is effective in the later stages of the production chain. This work demonstrates a robust methodology that can be applied to other infectious diseases and populations. Future research should investigate the impact of farm-specific factors on transmission rates and explore the effectiveness of additional sampling methods.

The model's accuracy relies on the assumptions made regarding transmission rates and the sensitivity of environmental sampling methods. The parameters used might not fully capture the variations seen in real-world farm settings. While Monte Carlo simulation addresses parameter uncertainty, other factors, like human intervention, could affect the results. The study focuses on specific Salmonella strains, and the findings might not fully generalize to all strains. Further research could focus on model validation with larger datasets and explore the incorporation of other environmental factors affecting Salmonella transmission. The generalizability of the prevalence-dependent sensitivity equation to parent flocks from egg-layer production needs further investigation.