Risk factors for the herd-level bacteriologic prevalence of Salmonella in Belgian slaughter pigs

Herd-level risk factors for salmonellosis in pigs were investigated in a cross-sectional study on 62 Belgian farrow-to-finish pig herds belonging to one slaughterhouse cooperative. Data concerning housing and ventilation, management, hygiene, biosecurity, production parameters, feeding, disease control and transport to the slaughterhouse were collected during a herd visit by means of a questionnaire. The percentage of positive animals in a slaughterhouse delivery, as determined by qualitative Salmonella isolation in the mesenteric lymph nodes taken from 30 slaughter pigs, was the outcome variable. All samples were taken in 4 different slaughterhouses. Variables first were submitted to a univariable analysis using a logistic mixed regression model, with herd as random effect. Variables which were related to the Salmonella prevalence (P < 0.05) were analysed further in a multivariable model. The clustering of Salmonella infection within a pen also was studied in a generalised mixed model with pen as random effect. Salmonella isolates were identified by serotype. In 57 (92%) of the herds, at least one sample was found positive for Salmonella. The median percentage of positive Salmonella samples per delivery was 64% (range: 0-100%). In the multivariable model, only type of floor was related significantly to the prevalence: 100% (95% CI 88-100) for herds with <50% slatted floors to 54% (36-70) for herds with fully slatted floors. The results from the analysis should be interpreted with care because only 62 herds were included in the study. Clustering between pigs from the same pen could not be demonstrated (variance +/- S.D.: 0.11 +/- 0.16). S. typhimurium (30%) and S. derby (20%) were most common among the 23 different serotypes that were found.     

Strategies for two-stage sampling designs for estimating herd-level prevalence

We propose a herd-level sample-size formula based on a common adjustment for prevalence estimates when diagnostic tests are imperfect. The formula depends on estimates of herd-level sensitivity and specificity. With Monte Carlo simulations, we explored the effects of different intracluster correlations on herd-level sensitivity and specificity. At low prevalence (e.g. 1% of animals infected), herd-level sensitivity increased with increasing intracluster correlation and many herds were classified as positive based only on false-positive test results. Herd-level sensitivity was less affected at higher prevalence (e.g. 20% of animals infected). A real-life example was developed for estimating ovine progressive pneumonia prevalence in sheep. The approach allows researchers to balance the number of herds and the total number of animals sampled by manipulating herd-level test characteristics (such as the number of animals sampled within a herd).     

Bayesian modeling of animal- and herd-level prevalences

We reviewed Bayesian approaches for animal-level and herd-level prevalence estimation based on cross-sectional sampling designs and demonstrated fitting of these models using the WinBUGS software. We considered estimation of infection prevalence based on use of a single diagnostic test applied to a single herd with binomial and hypergeometric sampling. We then considered multiple herds under binomial sampling with the primary goal of estimating the prevalence distribution and the proportion of infected herds. A new model is presented that can be used to estimate the herd-level prevalence in a region, including the posterior probability that all herds are non-infected. Using this model, inferences for the distribution of prevalences, mean prevalence in the region, and predicted prevalence of herds in the region (including the predicted probability of zero prevalence) are also available. In the models presented, both animal- and herd-level prevalences are modeled as mixture distributions to allow for zero infection prevalences. (If mixture models for the prevalences were not used, prevalence estimates might be artificially inflated, especially in herds and regions with low or zero prevalence.) Finally, we considered estimation of animal-level prevalence based on pooled samples.     

Policy-driven development of cost-effective, risk-based surveillance strategies

Animal health and residue surveillance verifies the good health status of the animal population, thereby supporting international free trade of animals and animal products. However, active surveillance is costly and time-consuming. The development of cost-effective tools for animal health and food hazard surveillance is therefore a priority for decision-makers in the field of veterinary public health. The assumption of this paper is that outcome-based formulation of standards, legislation leaving room for risk-based approaches and close collaboration and a mutual understanding and exchange between scientists and policy makers are essential for cost-effective surveillance. We illustrate this using the following examples: (i) a risk-based sample size calculation for surveys to substantiate freedom from diseases/infection, (ii) a cost-effective national surveillance system for Bluetongue using scenario tree modelling and (iii) a framework for risk-based residue monitoring. Surveys to substantiate freedom from infectious bovine rhinotracheitis and enzootic bovine leucosis between 2002 and 2009 saved over 6 million € by applying a risk-based sample size calculation approach, and by taking into account prior information from repeated surveys. An open, progressive policy making process stimulates research and science to develop risk-based and cost-efficient survey methodologies. Early involvement of policy makers in scientific developments facilitates implementation of new findings and full exploitation of benefits for producers and consumers.     

The consequences of risk-based surveillance: Developing output-based standards for surveillance to demonstrate freedom from disease

Risk factors associated with high or low long-term incidence of displaced abomasum (DA) or clinical ketosis were studied in 60 Swedish dairy herds, using multivariable logistic regression modelling. Forty high-incidence herds were included as cases and 20 low-incidence herds as controls. Incidence rates were calculated based on veterinary records of clinical diagnoses. During the 3-year period preceding the herd classification, herds with a high incidence had a disease incidence of DA or clinical ketosis above the 3rd quartile in a national database for disease recordings. Control herds had no cows with DA or clinical ketosis. All herds were visited during the housing period and herdsmen were interviewed about management routines, housing, feeding, milk yield, and herd health. Target groups were heifers in late gestation, dry cows, and cows in early lactation. Univariable logistic regression was used to screen for factors associated with being a high-incidence herd. A multivariable logistic regression model was built using stepwise regression. A higher maximum daily milk yield in multiparous cows and a large herd size (p = 0.054 and p = 0.066, respectively) tended to be associated with being a high-incidence herd. Not cleaning the heifer feeding platform daily increased the odds of having a high-incidence herd twelvefold (p < 0.01). Keeping cows in only one group in the dry period increased the odds of having a high incidence herd eightfold (p = 0.03). Herd size was confounded with housing system. Housing system was therefore added to the final logistic regression model. In conclusion, a large herd size, a high maximum daily milk yield, keeping dry cows in one group, and not cleaning the feeding platform daily appear to be important risk factors for a high incidence of DA or clinical ketosis in Swedish dairy herds. These results confirm the importance of housing, management and feeding in the prevention of metabolic disorders in dairy cows around parturition and in early lactation.     

Bayesian estimation of true between-herd and within-herd prevalence of Salmonella in Danish veal calves

Specialised veal producers that purchase and raise calves from several dairy herds are potentially at high risk of delivering Salmonella-infected animals to slaughter. However, the true prevalence of Salmonella infected veal producing herds and the prevalence of infected calves delivered to slaughter from infected herds are unknown in Denmark. Due to uncertainties about test sensitivity and specificity, these prevalences are not straightforward to assess. The objective of this study was to estimate the within-herd- and between-herd prevalence of Salmonella in veal calves delivered for slaughter to abattoirs in Denmark. Furthermore, it was investigated to which extent the estimates differed between a setup using both serological tests and faecal culture, compared to just serological tests, and whether the applied sampling scheme in the national surveillance programme in Denmark was sufficient to establish high posterior estimates of freedom from infection in individual herds. We used Bayesian analysis to avoid bias as a result of fixed test validity estimates. Serological test results from 753 animals and faecal culture from 1233 animals from 68 randomly selected Danish veal producing herds that delivered more than 100 calves to slaughter per year were used to estimate the prevalences and estimates of freedom from Salmonella. Serological test results of 7726 animals from 185 herds were used to compare the difference in prevalence estimates between serology alone vs. faecal culture combined with serology. We estimated that 34-57% of specialised veal producing herds were infected with Salmonella. Within the infected herds, 21-49% of the animals were infected. Few herds obtained high posterior estimates for the probability of freedom from infection given the collected data, with only six of 68 herds obtaining posterior probability of being infected less than 10%. Furthermore, this study indicated that serology is sufficiently sensitive and specific to be used for estimating the prevalence of Salmonella-infected specialised veal producing herds.     

Simulation model estimates of test accuracy and predictive values for the Danish Salmonella surveillance program in dairy herds

The Danish government and cattle industry instituted a Salmonella surveillance program in October 2002 to help reduce Salmonella enterica subsp. enterica serotype Dublin (S. Dublin) infections. All dairy herds are tested by measuring antibodies in bulk tank milk at 3-month intervals. The program is based on a well-established ELISA, but the overall test program accuracy and misclassification was not previously investigated. We developed a model to simulate repeated bulk tank milk antibody measurements for dairy herds conditional on true infection status. The distributions of bulk tank milk antibody measurements for infected and noninfected herds were determined from field study data. Herd infection was defined as having either >or=1 Salmonella culture-positive fecal sample or >or=5% within-herd prevalence based on antibody measurements in serum or milk from individual animals. No distinction was made between Dublin and other Salmonella serotypes which cross-react in the ELISA. The simulation model was used to estimate the accuracy of herd classification for true herd-level prevalence values ranging from 0.02 to 0.5. Test program sensitivity was 0.95 across the range of prevalence values evaluated. Specificity was inversely related to prevalence and ranged from 0.83 to 0.98. For a true herd-level infection prevalence of 15%, the estimate for specificity (Sp) was 0.96. Also at the 15% herd-level prevalence, approximately 99% of herds classified as negative in the program would be truly noninfected and 80% of herds classified as positive would be infected. The predictive values were consistent with the primary goal of the surveillance program which was to have confidence that herds classified negative would be free of Salmonella infection.     

Towards a risk-based surveillance for Trichinella spp. in Danish pig production

Increasing demands for cost-effectiveness in surveillance for human health hazards from animal origins can be met by introducing risk-based principles. This e.g. implies targeting subpopulations with higher risk of infection compared to the whole population. Furthermore, historical data from surveillance can be collated and used to assess future risk of infection. To demonstrate the effectiveness of combining these two approaches, we used a model called “Discounting historical evidence”. It depends mainly on the annual risk of introduction (PIntro) and the surveillance system sensitivity (SSe) (ability to detect infection if present). The model implies simulations that reiterate for a number of years. For each year the output is updated with the confidence on absence in infection. Trichinella spiralis infection in pigs is used as an example. In Denmark, more than 20 million pigs are tested annually. Despite more than 70 years of testing no pigs have been found positive for Trichinella. Hence, PIntro is low. SSe can be estimated from the maximum number of infected carcasses expected under the specified design prevalence, and the sensitivity of the test applied. According to our assessment, the current prevalence of Trichinella in Danish pigs is less than one case per million, which we interpret as negligible risk. Based on this, a risk-based surveillance programme for Trichinella is designed that targets all out-door reared pigs as well as all sows and boars (current total 610,000 slaughtered annually). These subpopulations are judged to have higher risk of getting Trichinella. Again, SSe and PIntro are estimated and the model results show that risk-based surveillance can be applied without jeopardizing human health. Finally, we incorporate wildlife surveys and test quality assurance in the programme. The results of the simulation model were included in an application to the European Commission concerning Denmark's status as a region with negligible risk of Trichinella. In July 2007, the European Commission granted status as “negligible risk” to Danish pigs and pork.     

Change in lameness risk estimates in piglets due to the modelling of herd-level variation.

In a previous study (Christensen, 1996. Prev. Vet. Med. 26, 107-118), an effect parameter changed from positive to negative depending on the model used. The study considered lameness in suckling piglets and the dataset included 7632 litters from 35 herds from the Health and Production Surveillance (HEPS) system database. To further investigate the change in the effect parameter, we performed a simple test to demonstrate the presence of herd-level variation. Finding substantial herd-level variation, we have re-analysed the data with special attention to the herd-level variation, and how to account for it. When herd-level variation was not accounted for in the model, a detrimental effect of prior treatment of the sow was seen but this turned into a beneficial effect when we accounted for the herd-level variation. The treatment of sow refers to any therapeutic treatment given to the sow after farrowing but before lameness or weaning occurred. This is, therefore, an example where not only the variances but also the effect parameters changed when we accounted for herd-level variation.     

Application of portfolio theory to risk-based allocation of surveillance resources in animal populations

Distribution of finite levels of resources between multiple competing tasks can be a challenging problem. Resources need to be distributed across time periods and geographic locations to increase the probability of detection of a disease incursion or significant change in disease pattern. Efforts should focus primarily on areas and populations where risk factors for a given disease reach relatively high levels. In order to target resources into these areas, the overall risk level can be evaluated periodically across locations to create a dynamic national risk landscape. Methods are described to integrate the levels of various risk factors into an overall risk score for each area, to account for the certainty or variability around those measures and then to allocate surveillance resources across this risk landscape. In addition to targeting resources into high risk areas, surveillance continues in lower risk areas where there is a small yet positive chance of disease occurrence. In this paper we describe the application of portfolio theory concepts, routinely used in finance, to design surveillance portfolios for a series of examples. The appropriate level of resource investment is chosen for each disease or geographical area and time period given the degree of disease risk and uncertainty present.     

A Bayesian model for spatial wildlife disease prevalence data

The analysis of the geographical distribution of disease on the scale of geographic areas such as administrative boundaries plays an important role in veterinary epidemiology. Prevalence estimates of wildlife population surveys are often based on regional count data generated by sampling animals shot by hunters. The observed disease rate per spatial unit is not an useful estimate of the underlying disease prevalence due to different sample sizes and spatial dependencies between neighbouring areas. Therefore, it is necessary to account for extra-sample variation and spatial correlations in the data to produce more accurate maps of disease incidence. The detection of spatial patterns is complicated by missing data in many of the geographical areas as the complete coverage of all areas is nearly impossible in wildlife surveys. For this purpose a hierarchical Bayesian model in which structured and unstructured over dispersion is modelled explicitly in terms of spatial and non-spatial components was implemented by Markov chain Monte Carlo methods. The model was empirically compared with the results of a non-spatial beta-binomial model using surveillance data of pseudorabies virus infections of European wild boars (Sus scrofa scrofa L.) in the Federal State of Brandenburg, Germany.     

Influence of diagnostic tests for estimating prevalence during surveillance programs

The influence of diagnostic tests on the estimation of the prevalence and the calculation of sample sizes with respect to different sampling schemes are presented in this paper. These sampling schemes are used for the implementation of surveillance programs. Assuming "perfect tests" (i.e. sensitivity = specificity = 100%) the calculated sample sizes, e.g. for an IBR/IPV-surveillance or a paratuberculosis survey, are half of the sample sizes considering sensitivity and specificity of the diagnostic tests. However the probability not to identify infected livestocks may be 4 times higher neglecting the test characteristics and assuming perfect tests.     

Environmental surveillance for Salmonella enterica in a veterinary teaching hospital

OBJECTIVE: To evaluate the extent of environmental contamination with Salmonella enterica in a veterinary teaching hospital. DESIGN: Longitudinal study. SAMPLES: Environmental samples obtained from 69 representative locations within a veterinary teaching hospital by use of a commercially available electrostatic wipe. PROCEDURE: Environmental samples were obtained for bacteriologic culture, and antimicrobial susceptibility testing was performed on each environmental isolate. Environmental isolates were compared with isolates obtained from animals during the same period to investigate potential sources of environmental contamination. RESULTS: 54 S. enterica isolates were recovered from 452 (11.9%) cultured environmental samples. Five different serotypes were recovered; the most common serotypes were S. Newport and S. Agona. Within the 5 serotypes recovered, 10 distinguishable phenotypes were identified by use of serotype and antimicrobial susceptibility patterns. Of the environmental isolates, 41 of 54 (75.9%) could be matched to phenotypes of isolates obtained from animal submissions in the month prior to collection of environmental samples. CONCLUSIONS AND CLINICAL RELEVANCE: Results indicated that environments in veterinary hospitals can be frequently contaminated with S. enterica near where infected animals are managed and fecal specimens containing S. enterica are processed for culture in a diagnostic laboratory. Bacteriologic culture of environmental samples collected with electrostatic wipes is an effective means of detecting contamination in a veterinary hospital environment and may be beneficial as part of surveillance activities for other veterinary and animal-rearing facilities.     

Designing and evaluating risk-based surveillance systems: potential unwarranted effects of applying adjusted risk estimates

Risk-based surveillance systems reveal occurrence of disease or infection in a sample of population units, which are selected on the basis of risk factors for the condition under study. The purpose of such systems for supporting practical animal disease policy formulations and management decisions are: A: to detect an emerging disease or infection, if it becomes introduced into a population; or B: to substantiate freedom from a condition in a population; or C: to detect cases and estimate the prevalence of an endemic condition in a population. In risk-based surveillance these aims should be met with prudent use of resources while maintaining acceptable system performance. High-risk category units are selected for testing by identification of the presence of specific high-risk factor(s), while disregarding other factors that might also influence the risk. On this basis we argue that the most applicable risk estimate for use in designing and evaluating a risk-based surveillance system would be a crude (unadjusted) relative risk, odds ratio or apparent prevalence. Risk estimates found in the published literature, however, are often the results of multivariable analyses implicitly adjusting the estimates for confounding from other risk factors. We describe some potential unintentional effects when using adjusted risk estimates in evaluating the efficacy and sensitivity of risk-based surveillance systems (SSe). In two examples, we quantify and compare the efficacy and SSe using adjusted and crude risk estimates. The examples use Danish surveillance data from previously published studies to evaluate systems aimed at risk-based detection of new cases of an endemic infection, i.e. Salmonella in dairy cattle herds (Example 1), and for substantiating the absence of a specific infection, i.e. Trichinella in the national slaughter pig population (Example 2), respectively.     

Use of simulation modeling to estimate herd-level sensitivity, specificity, and predictive values of diagnostic tests for detection of tuberculosis in cattle

OBJECTIVE: To estimate herd-level sensitivity (HSe), specificity (HSp), and predictive values for a positive (HPVP) and negative (HPVN) test result for several testing scenarios for detection of tuberculosis in cattle by use of simulation modeling. SAMPLE POPULATION: Empirical distributions of all herds (15,468) and herds in a 10-county area (1,016) in Michigan. PROCEDURE: 5 test scenarios were simulated: scenario 1, serial interpretation of the caudal fold tuberculin (CFT) test and comparative cervical test (CCT); scenario 2, serial interpretation of the CFT test and CCT, microbial culture for mycobacteria, and polymerase chain reaction assay; scenario 3, same as scenario 2 but specificity was fixed at 1.0; and scenario 4, sensitivity was 0.9 (scenario 4a) or 0.95 (scenario 4b), and specificity was fixed at 1.0. RESULTS: Estimates for HSe were reasonably high, ranging between 0.712 and 0.840. Estimates for HSp were low when specificity was not fixed at 1.0. Estimates of HPVP were low for scenarios 1 and 2 (0.042 and 0.143, respectively) but increased to 1.0 when specificity was fixed at 1.0. The HPVN remained high for all 5 scenarios, ranging between 0.995 and 0.997. As herd size increased, HSe increased and HSp and HPVP decreased. However, fixing specificity at 1.0 had only minor effects on HSp and HPVN, but HSe was low when the herd size was small. CONCLUSIONS AND CLINICAL RELEVANCE: Tests used for detecting cattle herds infected with tuberculosis work well on a herd basis. Herds with < approximately 100 cattle should be tested more frequently or for a longer duration than larger herds to ensure that these small herds are free of tuberculosis.     

Update: prevalence of Salmonella in pork sausage

One hundred seventy-five samples of fresh pork sausage representing thirty-five different commercial brands from six different retail stores were examined for the presence of salmonellae by standard enrichment, plating, biochemical and serological techniques. Contamination levels varied from 0 to 50% among stores and from 0 to 28% among brands. Prior research implied reduced prevalence of salmonellae in fresh pork sausage; however, these results indicate no variation in prevalence since 1969.     

Establishing a cost-effective national surveillance system for Bluetongue using scenario tree modelling

Vector-borne diseases pose a special challenge to veterinary authorities due to complex and time-consuming surveillance programs taking into account vector habitat. Using stochastic scenario tree modelling, each possible surveillance activity of a future surveillance system can be evaluated with regard to its sensitivity and the expected cost. The overall sensitivity of various potential surveillance systems, composed of different combinations of surveillance activities, is calculated and the proposed surveillance system is optimized with respect to the considered surveillance activities, the sensitivity and the cost. The objective of this project was to use stochastic scenario tree modelling in combination with a simple cost analysis in order to develop the national surveillance system for Bluetongue in Switzerland. This surveillance system was established due to the emerging outbreak of Bluetongue virus serotype 8 (BTV-8) in Northern Europe in 2006. Based on the modelling results, it was decided to implement an improved passive clinical surveillance in cattle and sheep through campaigns in order to increase disease awareness alongside a targeted bulk milk testing strategy in 200 dairy cattle herds located in high-risk areas. The estimated median probability of detection of cases (i.e. sensitivity) of the surveillance system in this combined approach was 96.4%. The evaluation of the prospective national surveillance system predicted that passive clinical surveillance in cattle would provide the highest probability to detect BTV-8 infected animals, followed by passive clinical surveillance in sheep and bulk milk testing of 200 dairy cattle farms in high-risk areas. This approach is also applicable in other countries and to other epidemic diseases.     

Bayesian methods for estimating pathogen prevalence within groups of animals from faecal-pat sampling

Pathogens such as Escherichia coli O157:H7 and Campylobacter spp. have been implicated in outbreaks of food poisoning in the UK and elsewhere. Domestic animals and wildlife are important reservoirs for both of these agents, and cross-contamination from faeces is believed to be responsible for many human outbreaks. Appropriate parameterisation of quantitative microbial-risk models requires representative data at all levels of the food chain. Our focus in this paper is on the early stages of the food chain-specifically, sampling issues which arise at the farm level. We estimated animal–pathogen prevalence from faecal-pat samples using a Bayesian method which reflected the uncertainties inherent in the animal-level prevalence estimates. (Note that prevalence here refers to the percentage of animals shedding the bacteria of interest). The method offers more flexibility than traditional, classical approaches: it allows the incorporation of prior belief, and permits the computation of a variety of distributional and numerical summaries, analogues of which often are not available through a classical framework. The Bayesian technique is illustrated with a number of examples reflecting the effects of a diversity of assumptions about the underlying processes. The technique appears to be both robust and flexible, and is useful when defecation rates in infected and uninfected groups are unequal, where population size is uncertain, and also where the microbiological-test sensitivity is imperfect. We also investigated the determination of the sample size necessary for determining animal-level prevalence from pat samples to within a pre-specified degree of accuracy.     

National surveillance of Salmonella enterica in food-producing animals in Japan

A total of 518 fecal samples collected from 183 apparently healthy cattle, 180 pigs and 155 broilers throughout Japan in 1999 were examined to determine the prevalence and antimicrobial susceptibility of Salmonella. The isolation rates were 36.1% in broilers, 2.8% in pigs and 0.5% in cattle. S. enterica Infantis was the most frequent isolate, found in 22.6% of broiler fecal samples. Higher resistance rates were observed against oxytetracycline (82.0%), dihydrostreptomycin (77.9%), kanamycin (41.0%) and trimethoprim (35.2%). Resistance rates to ampicillin, ceftiofur, bicozamycin, chloramphenicol and nalidixic acid were <10%. CTX-M-2 β-lactamase producing S. enterica Senftenberg was found in the isolates obtained from one broiler fecal sample. This is the first report of cephalosporin-resistant Salmonella directly isolated from food animal in Japan.