Evaluation of Agricultural Interventions on Human and Poultry-Related Salmonella Enteritidis in British Columbia

Introduction

Between 2008 and 2012, over 1000 laboratory-confirmed human cases of Salmonella Enteritidis (SE) with a matching outbreak pulsed-field gel electrophoresis pattern or phage type (PT) (PFGE SENXAI.0003 or PT 8) were investigated in British Columbia (BC), Canada (population 4.5 million). Laboratory-confirmed cases of salmonellosis are only a small proportion of total cases occurring in the population, and this outbreak led to an estimated 25,000 cases in the BC population taking underreporting into consideration (Thomas et al., 2006). This is the largest enteric disease outbreak documented in recent BC history. During this period, an increase in the same strain of SE among chicken-related isolates was also identified (Taylor et al., 2012). Eggs were the most likely source of illness, and poor-quality ungraded broiler hatching eggs were identified in food service establishments implicated in the outbreak (Taylor et al., 2012). The eggs were thought to be sourced from hatching egg farms so the interventions were aimed at this segment of the chicken industry (Taylor et al., 2012). Farms that produce hatching eggs, also called broiler breeder farms, send their eggs to hatcheries to be incubated and hatched into chicks that are primarily raised for meat (broiler birds). In contrast, table egg farms send their eggs to grading stations to be graded for quality and packed into cartons for sale in retail and restaurants. Table egg monitoring over this time period showed a low and stable rate of SE. During this time period, it was thought that eggs from hatching egg farms were sold for consumption as eggs without being graded and packaged, so called “farm-gate” sales.

Collaboration among public health, animal health authorities, and industry led to a number of interventions implemented through the farm-to-fork chain (Table 1). Subsequent to these interventions, human and poultry incidence declined during 2012 and 2013 (BC Integrated Surveillance of Foodborne Pathogens, 2012, 2013). Evaluation of such interventions is important for understanding their impact in reducing illness or outbreaks, particularly to identify those that are most effective and advocate for future outbreak control resources. In this article, we report on a study to quantify the impact of three interventions on the incidence of SE in animals and humans.

Materials and Methods

Analysis

The minimum number of months between when interventions were implemented and when they would have impacted human and poultry incidence were determined based on known production timelines (Fig. 1). Months were the smallest scale of analysis. Separation and vaccination had an estimated lag time of 6 months to impact animal incidence and 8 months to affect human incidence. The ban of broiler hatching eggs from farm-gate sales had a lag time of 1 month to impact human incidence. We focused our statistical analyses on these periods as expected windows of time, in which an association between intervention and SE incidence might reasonably occur. Poisson regressions were conducted to assess the association between interventions and human and animal incidence. Monthly numbers of SE positives (incident human cases or infected breeder flocks, as described above) were treated as our dependent variable and intervention (monthly number of flocks, by hatchery, receiving live vaccine, inactive vaccine, or under chick separation regime) and hatchery (A, B, or C) were treated as explanatory factors. Model fit and variable selection were performed using standard backward elimination and Akaike information criterion; extra-Poisson variation was modeled using an overdispersion parameter. We also used time-series vector autoregressive models to assess the lagged association between human and animal incidence and to test for causality. Stationarity was assessed using Dickey–Fuller tests, and differencing was applied to nonstationary series. Model fit was examined using standard diagnostics (autocorrelation function, prediction error, and Q–Q plots). The Granger causality test determines if animal SE incidence is significantly predictive of human SE incidence and therefore more likely to be a truly causal influence as opposed to simply an association in time. All analyses were carried out using SAS version 9.4.

OPEN IN VIEWER

FIG. 1.

Poultry food chain/process with interventions and time periods.

Results

In all three hatcheries, breeder chick separation coincided with decreases in the number of SE-positive breeder flocks (overall separation effect p = 0.03). Indeed, the proportion of SE-positive breeder flocks dropped by more than half when comparing the year following full implementation of chick separation with the year before separation (17.9% ± 5.7% vs. 8.0% ± 5.0%, all hatcheries, N = 130 flocks). In contrast, vaccination (live or inactive) did not consistently affect the number of positive flocks. Inactivated vaccine was associated with a significant reduction in numbers of positive breeder flocks in one of the three hatcheries (inactive vaccine × hatchery interaction p < 0.001), from 20.7% ± 0.07% SE-positive flocks in the year before vaccination to 3.0% ± 3.5% positive flocks in the year following vaccination (N = 22 flocks). Live vaccine programs were not associated with reduced numbers of SE-positive breeder flocks in any hatchery.

However, neither the hatchery interventions considered here (live and inactive vaccine programs and breeder chick separation) nor the ban on farm-gate sales significantly affected the numbers of human cases in our data (p > 0.2 in all cases).

A time series relationship was identified between SE-positive breeder flocks and human cases. Specifically, increases in positive breeder flocks were associated with significant increases in human cases 8 months later (p = 0.04; Fig. 2). Further analysis suggests that this is indeed a Granger causal effect of breeder flocks on human cases and not simply an association in time (Granger causality test, p = 0.048).

OPEN IN VIEWER

FIG. 2.

Time series showing numbers of Salmonella Enteritidis (SE)-positive breeder flocks (animal) and incident SE cases (human), British Columbia, 2008–2013. Observed counts (points) are shown with regression model fit (solid line) and 95% confidence interval (colored ribbon). Dashed arrows represent the timing of three interventions (breeder chick separation, inactive or live vaccine, and ban on farm-gate egg sales), beginning with the earliest known start dates.

Discussion

In this study, we evaluated the impacts of specific interventions on both human and chicken SE incidence in BC between 2008 and 2013. Evaluating the impact of interventions across human and animal sectors, including time series analysis, are novel features of our work.

The 8-month lag observed between human and poultry incidence is consistent with our a priori expectations based on known timelines of poultry production and human exposure through chicken meat consumption. The demonstration of a Granger causal association between poultry incidence and human incidence reinforces the importance of the animal sector in determining human illness. Our analysis also showed a significant association across hatcheries between declines in SE incidence among chickens and the separation of breeder chicks during hatch. Although we did not detect a consistent effect of vaccination overall, inactivated vaccine was associated with reduced SE in one of the three hatcheries. Such differences between hatcheries might reflect the timing of interventions. In two hatcheries, for example, separation and inactive vaccine occurred together and could have combined in their effects.

Despite the effect of breeder chick separation on reduced SE-positive flocks, we were not able to detect an association between any of the interventions and human SE incidence. This may be due, in part, to the limitations of our data, including the time periods we selected for the intervention to have impact, to the nature of the interventions themselves, or to the fact that human illness was partly controlled by other means. We used available data collected routinely for other purposes and not specific to the evaluation of these interventions. The interventions were most directly targeted at reducing SE in poultry and were not directly related to when and where human exposure would occur. Thus, we could not easily attribute human illness to a particular hatchery or to particular hatchery-level interventions, which were applied at differing times across facilities. Our analysis was, therefore, better suited to detect impacts on poultry and quite limited for detecting impacts on human incidence. Detecting the downstream impacts of agricultural interventions on human illness remains an important challenge for future evaluations.

In the United Kingdom, a temporal relationship between the introduction of vaccination in broiler-breeder and egg-laying flocks and decreases in human salmonellosis has been observed (O'Brien, 2013). The comprehensive nature of this program may explain why impacts related to human illness were detectable. Separation during hatch has not been highlighted as a primary intervention for reduction of salmonellosis; however, many control programs internationally highlight the importance of on-farm interventions, which may include restricting or preventing movement (Wegener et al., 2003; Braden, 2006; O'Brien, 2013). Many countries have used interventions that include intensive testing and depopulation for infected flocks. These programs have been successful at decreasing animal and human salmonellosis, but have come with notable costs for both industry and government (Wegener et al., 2003).

A large number and variety of interventions were implemented from farm-to-fork to control the SE outbreak in BC (Table 1). The use of multiple interventions throughout the farm to food chain is often highlighted as best practice in mitigating SE as well as other enteric pathogens (Wegener et al., 2003; Braden, 2006; Poirier et al., 2008; Sears et al., 2011; O'Brien, 2013). No other interventions were directed at the poultry industry or the public during this time period. Therefore, the decreases in SE associated with the interventions described are believed to represent causal impacts, although other factors such as changes in strains of Salmonella, human consumption/exposure patterns, or behavioral changes cannot be excluded.

Certain limitations are noteworthy in our study. First, two interventions occurred almost simultaneously (vaccination and separation) in two of the hatcheries, hindering our ability to assess each intervention separately. The association between separation and chicken SE incidence may have been influenced by vaccination and vice versa. Second, we assumed consistent implementation of interventions (e.g., constant number of flocks vaccinated/separated over time); however, unknown variations, such as changes in vaccine or periods of time when separation was not applied, may have occurred. Furthermore, positive flocks were determined based on samples taken in the hatchery, and a flock was considered positive for its entire life. This may have overestimated the number of positive breeder flocks, as eggs from multiple different breeder flocks would frequently be represented in a single hatcher. However, the process of designating positives was consistent, so any bias would be equally represented. Finally, only 1–3 years of post-intervention data were available for this evaluation. This relative scarcity of data following implementation of interventions makes before–after comparisons more challenging and hinders the assessment of long-term impacts on poultry and human incidence. This study provides an assessment of the impact of interventions up to a point in time. Further assessment of these interventions would be required to better understand their impact.

Our work identified limitations and considerations for others planning similar evaluations. We recommend, when interventions are being planned, to consider the establishment of an evaluation protocol, including what data will be required to measure the impacts. Identification and collection of additional data variables may improve the ability to demonstrate the effectiveness and impact of interventions, but may require additional resources. Ongoing measurements of the rollout or intensity of an intervention is key for establishing time-series effects on health outcomes. A priori identification of critical time periods between interventions and their expected impacts was valuable and helped to guide analyses specific to an intervention, as well as to reinforce conclusions.

Unfortunately, after 2 years of low incidence, the incidence of SE in poultry and humans increased again starting in 2014. Although it is not clear why this occurred, it may be due to modification in the implementation of some of these interventions. A more comprehensive, multisectoral approach is now being considered and will also need to be evaluated.

Conclusion

This evaluation demonstrated that some interventions had detectable impacts and others did not. This information will help inform what mitigation procedures and future interventions should be considered in BC and elsewhere. This work adds to our experience that collaborative and integrated assessment and investigation by public health and animal health is essential to develop and refine approaches for interventions throughout the food chain.