Abstract: Human salmonellosis is an important food-borne disease and S. Typhimurium is the most common serotype attributed to pork products. Under a farm-to-fork strategy, reducing the levels of Salmonella-positive pigs entering the slaughterhouse is an important goal. A framework model was developed, where the effect of dynamic (infection characteristics) and non-dynamic (cleanness and disinfection, biosecurity measures, etc.) factors were considered. Four baseline scenarios were created, corresponding to different levels associated with national Salmonella monitoring programs, and sensitivity analyses were run for the non-dynamic factors. Moreover, the option of vaccination was incorporated into the model, in order to provide with a tool for the formulation of an optimum vaccination strategy depending on the characteristics of the vaccine.
Abstract: A self-regulated epidemic model was developed to describe the dynamics ofâSalmonellaâTyphimurium in pig farms and predict the prevalence of different risk groups at slaughter age. The model was focused at the compartment level of the pig farms and it included two syndromes, a high and a low propagation syndrome. These two syndromes generated two different classes of pigs, the High Infectious and the Low Infectious, respectively, which have different shedding patterns. Given the two different classes and syndromes, the Infectious Equivalent concept was used, which reflected the combination of High and Low Infectious pigs needed for the high propagation syndrome to be triggered. Using the above information a new algorithm was developed that decides, depending on the Infectious Equivalent, which of the two syndromes should be triggered. Results showed that the transmission rate ofâS. Typhimurium for the low propagation syndrome is around 0.115, pigs in Low Infectious class contribute to the transmission of the infection by 0.61-0.80 of pigs in High Infectious class and that the Infectious Equivalent should be above 10-14% of the population in order for the high propagation syndrome to be triggered. This self-regulated dynamic model can predict the prevalence of the classes and the risk groups of pigs at slaughter age for different starting conditions of infection.
Abstract: Some attempts have been made to elucidate the association between positive serology and Salmonella detection by bacterial culture in individual pigs and pig herds. This study aimed to appraise whether the existing knowledge on such association provides grounds for the utilization of serology monitoring data for predicting Salmonella subclinical infection of pigs entering the abattoir. Serology test results of pig carcasses (taken at abattoirs) originating from 436 representative active herds in Ireland were utilized to estimate the overall cecal Salmonella carriage of Irish slaughter pigs. To this effect, two separate simulations were conducted using (i) herd-level regression data and (ii) animal-level sensitivity (0.2890) and specificity (0.8895) data, which were extracted from published articles. The herd-level approach estimated a moderate prevalence of cecal Salmonella carriage of 0.222 (sigma = 0.094; 95% confidence interval [CI]: 0.069 to 0.431), which matched closely the mean prevalence value from the surveys' validation data of Salmonella-positive cecal samples (n = 1,098) obtained at Irish abattoirs (0.215; 95% CI: 0.192 to 0.240). The animal-level simulation generated an output distribution with slightly more uncertainty (sigma = 0.102 and 95% CI: 0.146 to 0.537) and a higher estimate of cecal carriage (0.312), which was an effect of the low relative sensitivity of serology, common under field conditions. While the herd-level simulation appeared to be technically more appropriate, since its correlation is only moderate, further elucidation of other factors related to subclinical infection should be attained for their incorporation in prospective dynamic on-farm models, which would be useful in the ultimate goal of estimating the risk of carcass contamination during slaughter.
Abstract: In previous work a deterministic model for the compartment level was built, taking into account the two different syndromes with which Salmonella Typhimurium appears at pig farms. Based on this model, a stochastic one was built in this work that simulated different compartmental sizes, taking into account compartments of 200 to 400 pigs. Multiple scenarios of starting conditions of infection (SCI) ranging from 0.25 to 100% were tested for each population size. The effect of each of these two factors on the probability of disease extinctions and the prevalence of each of the classes of the model and the risk groups of pigs were estimated. The results showed that the compartment population had an inverse effect on the probability of disease extinction. On the other hand, low SCI resulted in high levels of early extinctions reaching 45%, while higher SCI led to high levels of late extinctions. Early extinctions resulted in the absence of the pathogen from the compartment, while late extinctions did not assure it. This effect shows that reducing the population of the compartment combined with appropriate cleaning and good farming practices could have a positive effect in the reduction of the risk of introducing S. Typhimurium into the slaughtering procedure. On the other hand, the profile of seroprevalence at slaughter age allows for risk characterization of the farm, given the relative stability and the small variation for higher SCI.
Abstract: This risk assessment study aimed to estimate the prevalence of Salmonella on pig carcasses and pork joints produced in slaughterhouses, on the basis that within groups of slaughter there is a strong association between the proportion of Salmonella-positive animals entering the slaughter lines (x) and the resulting proportion of contaminated eviscerated pig carcasses (y). To this effect, the results of a number of published studies reporting estimates of x and y were assembled in order to model a stochastic weighted regression considering the sensitivities of the diverse Salmonella culture methods. Meta-analysis was used to assign weights to the regression and to estimate the overall effect of chilling on Salmonella incidence on pig carcasses. The model's ability to produce accurate estimates and the intrinsic effectiveness of the modeling capabilities of meta-analysis were appraised using Irish data for the input parameter of prevalence of Salmonella carrier slaughter pigs. The model approximated a Salmonella prevalence in pork joints from Irish boning halls of 4.0% (95% confidence interval, 0.3 to 12.0%) and was validated by the results of a large survey (n = 720) of Salmonella in pork joints (mean, 3.3%; 95% confidence interval, 2.0 to 4.6%) carried out in four commercial pork abattoirs as part of this research project. Sensitivity analysis reinforced the importance of final rinsing (r = -0.382) and chilling (r = -0.221) as stages that contribute to reducing considerably the occurrence of Salmonella on the final product, while hygiene practices during jointing seemed to moderate only marginally the amount of contaminated pork joints. Finally, the adequacy of meta-analysis for integrating different findings and producing distributions for use in stochastic modeling was demonstrated.
Abstract: The concept of Adverse Outcome Pathways (AOP) and Source to Outcome Pathways (S2OP) provide a framework for characterisation of the impact of chemical exposure at different levels of biological organisation. Recent publications, such as the NRC reports on âToxicity Testing in the 21st Century and Exposure Science in the 21st Centuryâ and the EU Scientific Committees report on âAddressing the new challenges of risk assessmentâ, provide a vision of future chemical risk assessments where uncertainty is reduced through integrating hazard and exposure characterisation within mechanistically-based risk assessments. These pathways-based approaches aim to apply recent developments in bioinformatics, mathematical modelling and âomics technologies to take a systems biology approach to risk assessment of the impact of new chemicals on either human health or the environment.
Our vision is to apply these pathways-based risk assessment approaches to identify and characterise the key impacts of Home and Personal Care (HPC) ingredients on human health or the environment. Such an approach will require a greater integration of diverse expertise and long-term investment in the development and application of new capabilities for:
⢠defining consumer and environmental exposures across all levels of biological organisation
⢠identifying and characterising the molecular initiating events (MIEs) that will be triggered by exposure to HPC ingredients using bioinformatics and analytical chemistry approaches
⢠applying mathematical modelling to characterise the biological dose response of triggering MIEs at a given exposure
⢠defining protection goals at different levels of biological organisation to benchmark whether our exposure scenario would result in adverse outcomes
In so doing, we aim to remove our dependence on apical endpoint (eco)toxicological studies thereby reducing our overall uncertainty and better informing decisions on the use of chemicals within consumer products.
Abstract: An interest has developed in the global research community on unscrambling the dose-response relationships of micro-organisms and their hosts, and developing mechanistic understanding and modeling to better understand the sources of uncertainty and explain gaps in knowledge such as the threshold of infectious dose and inter-individual variability. This presentation will report on an initiative to scope the research needs that will enable the community to move forwards in this direction. Taking the case studies of an opportunistic pathogen (on eye), a gastrointestinal virus (GI) and a yeast (on scalp), we have focused on the host pathogen interaction in order to identify the key pathways that will drive the infection process. A second objective of this work was, by identifying these key pathways, to find the best approaches to model them in a mechanistic way that will enable more informed risk assessments. We have focused on the micro-organisms and their virulence mechanisms, the host defenses (including the microbiota), and the host pathogen interaction and key molecular events that will trigger host response. A number of different modeling approaches were evaluated at the systems, cell and molecular level, for both the microorganism and the host, as well as for their interaction. Overall, moving from the current dose response models to mechanistic understanding and modeling of the infection may be a long way off, and combines a number of different science areas and modeling approaches. Although the whole process of an infection is complex, by indentifying the key pathways and following a causal approach, the process can be simplified and the number of models needed to accurately describe this dose-response relationship could be reduced. On the other hand, the part of the uncertainty that it is attributed to the inter-individual variability among different strains and hosts can be theoretically understood, mathematically described and predicted.
Abstract: In many cases, microbial data may be characterised by the presence of large numbers of negative samples or zero counts. This occurs with some hygiene indicator organisms and pathogens and complicates statistical treatment of the data under the
assumption of a log normal distribution. The objective of the present work was to introduce an alternative distribution framework, capable of representing this type of data without incurring in loss of information. The negative binomial (NB), zeroinflated Poisson (ZIP), zero-inflated negative binomial (ZINB), hurdle Poisson (HP) and hurdle negative binomial (HNB) distributions were fitted to actual data consisting of total coliforms (n=590) and Escherichia coli (n=677) present on beef carcasses sampled from nine Irish abattoirs. Due to a heterogeneity parameter that accounts for the large variance of the data, improvement over the simple Poisson was shown by the NB, although it overestimated the zero counts and underestimated the first few counts. Whereas the ZIP could not cope with the data over-dispersion (p<0.001 for Ï2 GOF), the ZINB collapsed into a NB in both data sets due to the non-significance of its logit component. Addressing both the large variance and the excess of zero counts, the HNB predicted the observed count data slightly better than the NB, and was capable of depicting, with a comparable degree of accuracy, data of ~13% zero counts (Ï2=70.15<Ï2 t for coliforms) as well as of ~42% zero counts (Ï2=69.30<Ï2 t for Escherichia coli). Because a (two- component) hurdle model consists of a logit model that determines whether or not contamination on a carcass is detected (zero versus non-zero), and a count model that determines the numbers of CFUs, the HNB distribution has an interesting interpretation for stochastic risk assessment applications. Thus, bacterial data of pathogens in beef (Salmonella, verotoxigenic Escherichia coli) consisting of a considerable amount of negative samples can be accurately represented using modified count data distributions.
Abstract: A deterministic model was built and validated to simulate the propagation of S. Typhimurium at the compartment level of modern pig farms in order to predict the prevalence of caecal-, culture- and sero-positive pigs in slaughter age. Given that S. Typhimurium may appear in two different syndromes, a high and a low propagation syndrome, two models were built and combined. The two different syndromes are characterised by two types of infectious pigs, high infectious pigs, shedding the pathogen in large numbers and high frequency, and low infectious pigs shedding the pathogen in smaller numbers less frequently. The transmission parameter β for the low infectious pigs is considered as a fraction ε of the β of the high infectious pigs. Given that the two infectious classes have different effect on the transmission, the concept of Infectious Equivalent critical limit (IEcl) was introduced to model the combination of pigs needed for the high propagation to be triggered. A scenario analysis was run to estimate the parameters β, ε and IEcl that are simulating better the experimental results. The results suggest that for β from 0.14 to 0.19, for ε from 0.61 to 0.8 and for IEcl from 10% to 14%, the model predicts well the propagation of S. Typhimurium in the compartment and the prevalence of the groups of pigs in question.
Abstract: For the last years there has been an increasing interest in microbial food safety and some efforts have been put into modelling Salmonella Typhimurium at the preharvest part of the pig food chain, the farm. Transmission of S. Typhimurium at the farm level is a dynamic process and a good approach is to model it using epidemic models. In this paper we use two different types of model to describe the dynamics of S. Typhimurium at farm, one deterministic and one stochastic and try to validate we aim to validate them using the results of an experimental infection. The results of each of the modelling techniques are discussed and compared. Deterministic models are used for modelling infectious diseases at big populations while at pig farm and pen level stochastic models seem to be more appropriate. Demographic event-driven stochasticity gives variation to the results and seems to explain the actual fluctuations of Salmonella prevalence of batches of pigs from the same farms arriving to slaughterhouse. The need for systematic sampling at slaughterhouse or farm level is drawn and future work towards enrichment of the models in order to predict most common real life scenarios is proposed.
Abstract: While serology and bacteriology tests measure distinct aspects of the Salmonella infection cycle in a pig, some research to date have indicated a relationship between caecal contents prevalence and seroprevalence at the herd level. Aiming to reduce Salmonella in pork, as part of a national monitoring programme, muscle samples from pig carcasses are taken at Irish abattoirs and corresponding pig herds are categorised according to meat juice serology. A Monte-Carlo simulation study was conducted on meat juice ELISA results of Salmonella from the most representative Irish pig herds (436), in order to estimate the overall prevalence of Salmonella in caecal contents of slaughter pigs in Ireland, as well as per serology category. The estimated prevalence of caecal Salmonella carriage was high, at an average of 0.222 (ÃÂ=0.094 and 95% CI: 0.069 â 0.431) and matched closely the mean prevalence value of a validation data set of Salmonella-positive caecal samples obtained at abattoirs (0.220), although with a comparatively smaller spread (95% CI: 0.197 â 0.246).
Abstract: The aim of this work was to develop a dynamic process for S. Typhimurium,
which can describe the dynamics of the pathogen in the farm predict the prevalence
of the pigs with the pathogen at slaughter age and be used for the evaluation of intervention
strategies for risk mitigation. Given that a model like this is one of the
first of its kind for short life production animals, the major challenges in modelling
infectious diseases were investigated, leading to some useful recommendations regarding
data collection and modelling issues. The first objective (Chapter 3) was to
compare the deterministic and stochastic approaches for the case of S. Typhimurium
in pig farms and to investigate if a stochastic model that uses demographic stochasticity
can produce results similar to the ones observed after sampling of pigs at the
slaughterhouse. The results show that especially for the case of Salmonella in pig
farms stochastic modelling and the event-driven approach appears to be more appropriate.
A second objective (Chapter 4) was to develop and to parameterise a baseline
deterministic model for S. Typhimurium at the level of the room/compartment
which is under an All-In-All-Out strategy, that will be able to describe the dynamics
of S. Typhimurium in pig farms and to serve as a base for further development
of a more integrated stochastic model. The model was developed using published
experimental data and existing theories on the transmission of S. Typhimurium, and
it mathematically described the propagation of the pathogen at the compartment
level of the pig farms. It included two syndromes, a high and a low propagation
syndrome and the selection of which syndrome to be triggered was based on the
force of infection of the closed population. It answers questions on the transmission
rate, the different effect of the infectious pigs on the transmission and the level
of infectious needed for the high propagation syndrome to be triggered. Because
of its frequency-dependent nature it can be applied to all the compartments, irrespective
to their population size. In Chapter 5 a stochastic model was built using
demographic stochasticity based on the previous deterministic model in order for
the effect of dynamic factors to be evaluated in terms of propagation of infection,
disease extinction and prevalence at slaughter age. From the results of the model,
it was concluded that compartments should be as small as possible, while strategies
aiming to the reduction of the prevalence of infectious pigs at the beginning of fattening
could increase the probability of disease extinctions up to 45%. On the other
hand, increased number of infectious pigs at the beginning of the fattening period,
either naturally or due to active immunisation, has an overall inverse effect to the
caecal-positive risk group, reducing the risk of introducing the pathogen into the
slaughterhouse. A fourth objective (Chapter 6) was to build a framework model,
where the effect of intervention strategies, including vaccination, on the prevalence
of S. Typhimurium at slaughter age was evaluated. The results show that the main
driver of the prevalence of each risk group is the biosecurity measures taken to
reduce the probability of primary infections from external sources. Cleaning and
disinfection as well as the transfer of pigs from room to room have limited effect,
while the effects of harvest age and animal husbandry strategies are more efficient,
although may differ depending on the level of infection of a farm. Regarding vaccination,
using an âoptimumâ vaccine, a double vaccination of the whole room population
at the 4th and the 14th week of the WGF cycle could lead to almost eradication
of the risk of introducing the pathogen in the food chain. Finally, a fifth objective
of this work (Chapter 7) was to identify data and modelling challenges of infectious
diseases in short life food production animals, exploring the potential alternatives
for different types of farms and pathogens with different infection characteristics.
The specific characteristics either of the farms or the pathogens pose new questions
that have to be answered on the modelling approach and the data collection process
to be followed. The modern farm infrastructure and the artificial way of living of
the animals make difficult the comparisons with human and wildlife epidemiology
and modelling should take into account these special characteristics and adapt the
models accordingly. There is no single answer in the question of what is the best
approach that should be followed.
Abstract: R is a free open-source implementation of the S statistical computing language and programming environment. The current status of R is a command line driven interface with no advanced cross-platform graphical user interface (GUI), but it includes tools for building such. Over the past years, proprietary and non-proprietary GUI solutions have emerged, based on internal or external tool kits, with different scopes and technological concepts. For example, Rgui.exe and Rgui.app have become the de facto GUI on the Microsoft Windows and Mac OS X platforms, respectively, for most users. In this paper we discuss RKWard which aims to be both a comprehensive GUI and an integrated development environment for R. RKWard is based on the KDE software libraries. Statistical procedures and plots are implemented using an extendable plugin architecture based on ECMAScript (JavaScript), R, and XML. RKWard provides an excellent tool to manage different types of data objects; even allowing for seamless editing of certain types. The objective of RKWard is to provide a portable and extensible R interface for both basic and advanced statistical and graphical analysis, while not compromising on flexibility and modularity of the R programming environment itself.
DOI 
