This thesis is concerned with the application of recently developed epidemiological and statistical tools to inform the optimisation of a national surveillance strategy of considerable importance to human health. The results of a series of epidemiological investigations of surveillance strategies for zoonotic Salmonella are presented. Salmonella are one of the most common and serious zoonotic foodborne pathogenic bacteria globally. These studies were motivated by the increasing focus on the cost-effectiveness of surveillance while maintaining consumer confidence in food supply. Although data from the Danish Salmonella surveillance and control programme has been used in these investigations, the techniques may be readily applied to other surveillance data of similar quality. The first study describes the spatial epidemiological features of Danish Salmonella surveillance and control programme data from 1995 to 2004, using a novel method of spatially adaptive smoothing. The conditional probability of a farm being a case was consistently high in the the south-west of Sonderjylland on the Jutland peninsula, identifying this area for further investigation and targeted surveillance. The identification of clustering of case farms led into the next study, which closely examines one year of data, 2003, for patterns of spatial dependency. K-function analyses provided evidence for aggregation of Salmonella case farms over that of all farms at distances of up to six kilometres. Visual semivariogram analyses of random farm-level effects from a Bayesian logistic regression model (adjusted for herd size) of Salmonella seropositivity, revealed spatial dependency between pairs of farms up to a distance of four kilometres apart. The strength of the spatial dependency was positively associated with slaughter pig farm density. We describe how this might inform the surveillance programme by potentially targeting herds within a four kilometre radius of those with high levels of Salmonella infection. In the third study, farm location details, routinely recorded surveillance information, and industry survey data from 1995 were combined to build a logistic seroprevalence model. This identified wet-feeding and specific pathogen free herd health status as protective factors for Salmonella seropositivity, while purchasing feed was a risk factor. Once adjusting for these covariates, we identified pockets of unexplained risk for Salmonella seropositivity and found spatial dependency at distances of up to six km (95% CI: 2–35 km) between farms. A generalised linear spatial model was fitted to the Jutland data allowing formal estimation of the range of spatial correlation and a measure of the uncertainty about it. There was a large within-farm component to the variance, suggesting that gathering more farm level information would be advantageous if this approach was to be used to target surveillance strategy. The fourth study again considers data from the whole study period, 1995 to 2004. A detailed temporal analysis of the data revealed there was no consistent seasonal pattern and correspondingly no benefit in targeting sampling to particular times of the year. Spatiotemporal analyses suggested a local epidemic of increased seroprevalence occured in west Jutland in late 2000. Lorelogram analyses showed a defined period of statistically significant temporal dependency, suggesting that there is little value in sampling more frequently than every 10 weeks on the average farm. The final study uses findings from the preceding chapters to develop a zero-inflated binomial model which predicts which farms are most at risk of Salmonella, and then preferentially samples these high-risk farms. This type of modelling allows assessment of similarities and differences between factors that affect herd infection status (introduction) and those that affect the seroprevalence in infected herds (persistence and spread). The model suggested that many of the herds where Salmonella was not detected were infected but at a low prevalence. Using cost and sensitivity, we compared the results with those under the standard sampling scheme based on herd size, and the recently introduced risk-based approach. Model based results were less sensitive, but showed significant cost savings. Further model refinements, sampling schemes, and the methods to evaluate their performance are important areas for future work, and should continue to occur in direct consultation with Danish authorities.

This thesis is concerned with the application of recently developed epidemiological and statistical tools to inform the optimisation of a national surveillance strategy of considerable importance to human health. The results of a series of epidemiological investigations of surveillance strategies for zoonotic Salmonella are presented. Salmonella are one of the most common and serious zoonotic foodborne pathogenic bacteria globally. These studies were motivated by the increasing focus on the cost-effectiveness of surveillance while maintaining consumer confidence in food supply. Although data from the Danish Salmonella surveillance and control programme has been used in these investigations, the techniques may be readily applied to other surveillance data of similar quality. The first study describes the spatial epidemiological features of Danish Salmonella surveillance and control programme data from 1995 to 2004, using a novel method of spatially adaptive smoothing. The conditional probability of a farm being a case was consistently high in the the south-west of Sonderjylland on the Jutland peninsula, identifying this area for further investigation and targeted surveillance. The identification of clustering of case farms led into the next study, which closely examines one year of data, 2003, for patterns of spatial dependency. K-function analyses provided evidence for aggregation of Salmonella case farms over that of all farms at distances of up to six kilometres. Visual semivariogram analyses of random farm-level effects from a Bayesian logistic regression model (adjusted for herd size) of Salmonella seropositivity, revealed spatial dependency between pairs of farms up to a distance of four kilometres apart. The strength of the spatial dependency was positively associated with slaughter pig farm density. We describe how this might inform the surveillance programme by potentially targeting herds within a four kilometre radius of those with high levels of Salmonella infection. In the third study, farm location details, routinely recorded surveillance information, and industry survey data from 1995 were combined to build a logistic seroprevalence model. This identified wet-feeding and specific pathogen free herd health status as protective factors for Salmonella seropositivity, while purchasing feed was a risk factor. Once adjusting for these covariates, we identified pockets of unexplained risk for Salmonella seropositivity and found spatial dependency at distances of up to six km (95% CI: 2–35 km) between farms. A generalised linear spatial model was fitted to the Jutland data allowing formal estimation of the range of spatial correlation and a measure of the uncertainty about it. There was a large within-farm component to the variance, suggesting that gathering more farm level information would be advantageous if this approach was to be used to target surveillance strategy. The fourth study again considers data from the whole study period, 1995 to 2004. A detailed temporal analysis of the data revealed there was no consistent seasonal pattern and correspondingly no benefit in targeting sampling to particular times of the year. Spatiotemporal analyses suggested a local epidemic of increased seroprevalence occured in west Jutland in late 2000. Lorelogram analyses showed a defined period of statistically significant temporal dependency, suggesting that there is little value in sampling more frequently than every 10 weeks on the average farm. The final study uses findings from the preceding chapters to develop a zero-inflated binomial model which predicts which farms are most at risk of Salmonella, and then preferentially samples these high-risk farms. This type of modelling allows assessment of similarities and differences between factors that affect herd infection status (introduction) and those that affect the seroprevalence in infected herds (persistence and spread). The model suggested that many of the herds where Salmonella was not detected were infected but at a low prevalence. Using cost and sensitivity, we compared the results with those under the standard sampling scheme based on herd size, and the recently introduced risk-based approach. Model based results were less sensitive, but showed significant cost savings. Further model refinements, sampling schemes, and the methods to evaluate their performance are important areas for future work, and should continue to occur in direct consultation with Danish authorities.

This thesis is concerned with the application of recently developed epidemiological and statistical tools to inform the optimisation of a national surveillance strategy of considerable importance to human health. The results of a series of epidemiological investigations of surveillance strategies for zoonotic Salmonella are presented. Salmonella are one of the most common and serious zoonotic foodborne pathogenic bacteria globally. These studies were motivated by the increasing focus on the cost-effectiveness of surveillance while maintaining consumer confidence in food supply. Although data from the Danish Salmonella surveillance and control programme has been used in these investigations, the techniques may be readily applied to other surveillance data of similar quality. The first study describes the spatial epidemiological features of Danish Salmonella surveillance and control programme data from 1995 to 2004, using a novel method of spatially adaptive smoothing. The conditional probability of a farm being a case was consistently high in the the south-west of Sonderjylland on the Jutland peninsula, identifying this area for further investigation and targeted surveillance. The identification of clustering of case farms led into the next study, which closely examines one year of data, 2003, for patterns of spatial dependency. K-function analyses provided evidence for aggregation of Salmonella case farms over that of all farms at distances of up to six kilometres. Visual semivariogram analyses of random farm-level effects from a Bayesian logistic regression model (adjusted for herd size) of Salmonella seropositivity, revealed spatial dependency between pairs of farms up to a distance of four kilometres apart. The strength of the spatial dependency was positively associated with slaughter pig farm density. We describe how this might inform the surveillance programme by potentially targeting herds within a four kilometre radius of those with high levels of Salmonella infection. In the third study, farm location details, routinely recorded surveillance information, and industry survey data from 1995 were combined to build a logistic seroprevalence model. This identified wet-feeding and specific pathogen free herd health status as protective factors for Salmonella seropositivity, while purchasing feed was a risk factor. Once adjusting for these covariates, we identified pockets of unexplained risk for Salmonella seropositivity and found spatial dependency at distances of up to six km (95% CI: 2–35 km) between farms. A generalised linear spatial model was fitted to the Jutland data allowing formal estimation of the range of spatial correlation and a measure of the uncertainty about it. There was a large within-farm component to the variance, suggesting that gathering more farm level information would be advantageous if this approach was to be used to target surveillance strategy. The fourth study again considers data from the whole study period, 1995 to 2004. A detailed temporal analysis of the data revealed there was no consistent seasonal pattern and correspondingly no benefit in targeting sampling to particular times of the year. Spatiotemporal analyses suggested a local epidemic of increased seroprevalence occured in west Jutland in late 2000. Lorelogram analyses showed a defined period of statistically significant temporal dependency, suggesting that there is little value in sampling more frequently than every 10 weeks on the average farm. The final study uses findings from the preceding chapters to develop a zero-inflated binomial model which predicts which farms are most at risk of Salmonella, and then preferentially samples these high-risk farms. This type of modelling allows assessment of similarities and differences between factors that affect herd infection status (introduction) and those that affect the seroprevalence in infected herds (persistence and spread). The model suggested that many of the herds where Salmonella was not detected were infected but at a low prevalence. Using cost and sensitivity, we compared the results with those under the standard sampling scheme based on herd size, and the recently introduced risk-based approach. Model based results were less sensitive, but showed significant cost savings. Further model refinements, sampling schemes, and the methods to evaluate their performance are important areas for future work, and should continue to occur in direct consultation with Danish authorities.