Analysis of the baseline survey on the prevalence of Salmonella in

Transcription

The EFSA Journal / EFSA Scientific Report (2008) 206, 1-111
SCIENTIFIC REPORT
Analysis of the baseline survey on the
prevalence of Salmonella in slaughter pigs,
in the EU, 2006-2007
Part B: factors associated with
Salmonella infection in lymph nodes,
Salmonella surface contamination of carcasses,
and the distribution of Salmonella serovars1
Report of the Task Force on Zoonoses Data Collection
(Question N° EFSA-Q-2006-042B)
Adopted on 14 November 2008
1
For citation purposes: Report of the Task Force on Zoonoses Data Collection on the Analysis of the baseline survey on the
prevalence of Salmonella in slaughter pigs, Part B, The EFSA Journal (2008) 206, 1-111
© European Food Safety Authority, 2008
1
Summary
A European Union-wide baseline survey was carried out to determine, at the point of slaughter,
the prevalence of pigs infected with Salmonella, in order to provide the scientific basis for setting
a Community reduction target for Salmonella in slaughter pigs. The sampling of slaughter pigs
took place between October 2006 and September 2007. The pigs were randomly selected from
those slaughterhouses that together accounted for 80% of the pigs slaughtered within each
Member State. All participating Member States and Norway sampled ileocaecal lymph nodes
from the selected slaughtered pigs. Moreover, 13 Member States additionally sampled the
corresponding pigs’ carcasses by swabbing in order to appreciate the external contamination of the
carcasses. A total of 19,159 slaughter pigs with validated results from the European Union and
Norway were included in the survey analyses, corresponding to information on 19,025 lymph
node samples (from 25 Member States and Norway) and 5,736 carcass swab samples
(from 13 Member States).
The analysis of Salmonella prevalence was carried out earlier and was published by the European
Food Safety Authority on 30 May 2008 in the Part A report. The Community observed prevalence
of Salmonella-positive slaughter pigs was 10.3%, whereas data from the group of 13 Member
States showed that the observed prevalence of carcasses contaminated with Salmonella was 8.3%
overall. In both cases, prevalence varied among Member States.
In the risk factor analysis, an association between the prevalence of slaughter pigs infected with
Salmonella in their lymph nodes and the frequency of Salmonella surface contamination of the pig
carcasses was observed. A Salmonella infected pig was twice as likely to yield a Salmonella
contaminated carcass. However, contaminated carcasses could also derive from uninfected pigs,
suggesting potential for cross-contamination in the slaughterhouse environment. The risk of
carcasses becoming contaminated with Salmonella varied significantly between slaughterhouses
even when other associated factors, such as the prevalence of infected slaughter pigs, were
accounted for. Moreover, in some slaughterhouses the risks of producing a contaminated carcass
both from a Salmonella infected pig and from a non-infected pig were significantly higher than in
some other slaughterhouses. This indicates that certain slaughterhouses are more capable of
controlling and preventing Salmonella contamination than others.
The delay between sampling and the start of laboratory testing was found to have an impact on the
likelihood of detecting Salmonella from the samples. The bacterium was most likely to be
detected from lymph nodes and carcass swabs when the sample was tested 3-4 or 1-2 days after
sampling, respectively. Also the probability of detecting Salmonella from a lymph node sample
augmented when the weight of the sample increased.
At the European Union level, the carcasses were less at risk of being contaminated during the first
months of the survey, October 2006 to March 2007, compared to the rest of the survey period,
from April to September 2007.
The analyses also revealed that there is considerable variation between the significant factors
associated with Salmonella infection in slaughter pig’s lymph nodes, or Salmonella carcass
contamination, among Member States and also when compared to EU level.
2
A tendency towards Member State-specific clusters of Salmonella serovars was identified for
Salmonella infection in slaughter pigs, and spatial distribution of serovars was very
heterogeneous. S. Typhimurium and S. Derby were widespread and dominant in the Member
States. However, S. Enteritidis was relatively prevalent in some eastern EU Member States.
The descriptive analysis of the serovar distribution supported the notion that pig meat contributes
to human Salmonella infection. However, many serovars isolated from slaughter pigs in this
survey are also common in other food producing animal species and food thereof, indicating that
the potential for the contribution to human infections is shared between different sources.
It is recommended that Member States would consider the factors found to be associated with
Salmonella infection in slaughter pigs and carcasses in this survey when they are designing their
Salmonella control programmes for slaughter pigs. Control measures both at primary production
and at slaughterhouse level should be included in the programmes. In particular sampling and
testing procedures need standardisation to enhance sensitivity and comparability of monitoring
results.
Member States and the EU pig meat industry are encouraged to develop and enhance Salmonella
controls in primary production and at slaughterhouses in order to prevent and reduce the
contamination of pig carcasses with Salmonella. Member States are also invited to perform further
studies at national level to identify specifically the risk factors for Salmonella infection of
slaughter pigs and surface contamination of carcasses.
3
Table of contents
Summary.............................................................................................................................................2
1. Introduction ....................................................................................................................................6
2. Objectives .......................................................................................................................................7
3. Materials and methods....................................................................................................................8
3.1. Data description .....................................................................................................................8
3.2. Analysis of factors associated with Salmonella positivity.....................................................9
3.2.1. Definition of the outcome variables ..............................................................................9
3.2.2. Factors investigated.......................................................................................................9
3.2.3. Exploratory analysis of potentially associated factors ................................................10
3.2.4. Analysis of multicollinearity among potentially associated factors............................10
3.2.5. Identification of factors associated with Salmonella positivity ..................................11
3.2.5.1 Statistical model ...................................................................................................11
3.2.5.2 Model building for Salmonella lymph node infection at EU- and country level .12
3.2.5.3 Model building for Salmonella carcass surface contamination at the MS group
and MS level .........................................................................................................13
3.3. Analysis of the association between slaughter pigs’ lymph node Salmonella infection and
their carcass Salmonella contamination .............................................................................13
3.4. Analysis of the serovars and phage types distribution.........................................................13
3.4.1. Spatial distribution of reported Salmonella serovars in lymph nodes.........................13
3.4.2. Comparison between Salmonella serovar and phage type distribution in slaughter
pigs, other animal species, feed and human salmonellosis cases ................................14
4. Results ..........................................................................................................................................15
4.1. Analysis of factors associated with Salmonella infection in lymph nodes of slaughter pigs15
4.1.1. Descriptive analysis of factors potentially associated with Salmonella infection.......15
4.1.1.1 Factors related to the sensitivity of the sampling process ....................................15
4.1.1.2 Factors related to the lymph node infection .........................................................20
4.1.2. Analysis of multicollinearity among potential factors ................................................28
4.1.3. Multiple regression analysis at EU level.....................................................................28
4.1.4. Multiple regression analysis at the country level ........................................................29
4.2. Analysis of factors associated with surface contamination of carcasses with Salmonella ..33
4.2.1. Descriptive analysis of factors potentially associated with Salmonella
contamination ..............................................................................................................33
4.2.1.1 Factors related to the sensitivity of the sampling process ....................................33
4.2.1.2 Factors related to surface contamination of carcasses..........................................35
4.2.2. Analysis of multicollinearity among potential factors ................................................43
4.2.3. Multiple regression analysis at the 13-MS group level...............................................43
4.2.4. Multiple regression analysis at the MS level ..............................................................45
4
4.3. Analysis of the serovars and phage type distribution ..........................................................47
4.3.1. Spatial distribution of Salmonella serovars in lymph nodes .......................................47
4.3.2. Differences in serovar distribution between the reporting countries ..........................49
4.3.3. Comparison between serovar distributions in slaughter pigs, the animal species,
feed and human cases in the EU ..................................................................................49
4.3.4. Phage type distributions ..............................................................................................52
4.3.4.1 S. Enteritidis phage types......................................................................................52
4.3.4.2 S. Typhimurium phage types ................................................................................53
4.3.5. Comparison between phage type distribution in slaughter pigs and humans .............55
5. Discussion.....................................................................................................................................58
5.1. Analysis of factors associated with Salmonella infection in lymph nodes or surface
contamination of carcasses .................................................................................................58
5.1.1. Effect of sampling and testing procedures ..................................................................59
5.1.2. Effect of lymph node Salmonella infection on surface contamination of carcasses ...59
5.1.3. Effect of the slaughterhouse on the risk of carcass contamination .............................60
5.1.4. Effect of the time of sampling on Salmonella results .................................................61
5.2. Analysis of serovar and phage type distribution..................................................................61
5.2.1. Spatial distribution of Salmonella serovars in lymph nodes .......................................61
5.2.2. Comparison of serovar and phage type distribution in slaughter pigs, feed and
human salmonellosis cases ..........................................................................................62
6. Conclusions ..................................................................................................................................64
7. Recommendations ........................................................................................................................65
Task Force on Zoonoses Data Collection members .........................................................................66
Acknowledgements ..........................................................................................................................66
Abbreviations....................................................................................................................................67
List of Tables ....................................................................................................................................68
List of Figures...................................................................................................................................69
References ........................................................................................................................................71
List of Annexes.................................................................................................................................73
5
1. Introduction
This report describes the results of a baseline survey carried out in the European Union (EU) to
estimate the prevalence of Salmonella in slaughter pigs. This survey was the fourth in a series of
baseline surveys of Salmonella carried out within the EU. The objective of the surveys has been to
obtain comparable data for all Member States (MSs) through harmonised sampling schemes.
The European Commission has asked the European Food Safety Authority (EFSA) to analyse the
results at the survey. In EFSA the task was assigned to the Task Force on Zoonoses Data
Collection.
According to Regulation (EC) No 2160/2003 (EC, 2003) on the control of Salmonella and other
zoonotic agents, which aims to reduce the incidence of food-borne diseases in the EU, results of
the survey will enable the setting of the Community target for the reduction of the prevalence of
Salmonella infection in slaughter pigs.
A report from the Task Force on Zoonoses Data Collection on the “Analysis of the baseline survey
on the prevalence of Salmonella in slaughter pigs in the EU, 2006-2007, part A: Salmonella
prevalence estimates” (EFSA, 2008a) was issued on 30 May 2008. That report included the
analyses of the prevalence of Salmonella in slaughter pigs, the most frequent Salmonella serovars
reported and the impact of the sampling design.
The present Part B report contains analyses of the effects of potential risk factors for Salmonella
infection in pigs and contamination of pig carcasses. Further analyses of the distribution of the
serovars and phage types of Salmonella isolates are also included. Objectives, sampling frame,
diagnostic testing methods, as well as data collection, evaluation, reporting and timelines of the
baseline survey are specified in Commission Decisions 2006/668/EC (EC, 2006) and
2007/219/EC (EC, 2007) concerning a baseline survey on the prevalence of Salmonella in
slaughter pigs.
6
2. Objectives
The objectives of the EU-wide baseline survey on Salmonella in slaughter pigs are described in
detail in the Part A report.
The specific objectives related to this Part B report are:
•
to investigate the effect of factors, which may be associated with Salmonella infection of
slaughter pigs in the ileo-caecal lymph nodes, at the EU level and for each MS
individually,
•
to investigate the effect of factors, which may be associated with Salmonella surface
contamination of slaughter pig carcasses, at the level of a group of 13 MSs, that reported
the information, and for each of those MSs individually,
•
to investigate the association between the results from bacteriological test of lymph node
and the results from bacteriological test of carcass swab, with respect to Salmonella,
•
to investigate the Salmonella serovar distribution in slaughter pigs across the EU, and
•
to analyse the information submitted by MSs regarding S. Enteritidis and S. Typhimurium
phage types isolated from slaughter pigs.
The analyses of the antimicrobial susceptibility of Salmonella isolates from the survey will be
specifically addressed in a separate report to be published later by EFSA.
7
3. Materials and methods
A detailed description of the design of the baseline survey, sampling design, sample size and
bacteriological testing can be found in Annex I of Commission Decision 2006/668/EC of 29
September 2006 (EC, 2006) concerning a financial contribution from the Community towards a
baseline survey on the prevalence of slaughter pigs to be carried out in the MSs, and in the Part A
report.
3.1. Data description
A detailed description of the validation and cleaning of the dataset carried out is provided in the
Part A report. The final dataset contained data from 19,159 slaughter pigs (from 25 MSs and
Norway), together with information on 19,0711 lymph node samples (from 25 MSs and Norway)
and 5,736 carcass swab samples from 13 MSs.
In each participating country, a representative sample of carcasses (of market-age pigs weighing
between 50 and 170 kg) was randomly selected in slaughterhouses representing at least 80% of
domestic production. In order to assess the infection status of slaughter pigs, a 25 gr. sample from
an aggregate of ileo-caecal lymph nodes were collected from each carcass. A complementary
instruction further indicated that some additional lymph nodes of the distal jejunal chain were to
be sampled, if necessary, to complete the weight of the sample up to 25gr. At the laboratory, all
lymph nodes of the sample were pooled and analysed for the detection of Salmonella. In addition,
13 MSs sampled swabs from the surface of the same carcasses in order to determine the
Salmonella contamination at the end of the slaughterline. An area of 400 cm2 of the carcass
surface was swabbed in a standardised way.
Certain MSs also conducted serological examination of meat juice or blood samples from the
slaughter pigs. However, as explained in the Part A report, no meaningful analysis of this data
could be conducted because different assays and cut-off values were used by the MSs. As these
serological results were not comparable between MSs, no further analysis was carried out for the
Part B report.
In the analysis for this Part B report, Norway is included in the EU level analysis dataset.
1
In total, 46 lymph node samples originating from 5 countries were discarded due to missing crucial covariate
information.
8
3.2. Analysis of factors associated with Salmonella positivity
The general assumptions and framework of the statistical analysis carried out are reported in detail
in the Part A report. The observed prevalence1 of infected slaughter pigs and of contaminated
carcasses was defined as the proportion of positive slaughter pigs, or, as the proportion of positive
carcasses processed over the one-year period of the baseline survey in MSs.
The effect of potential factors on Salmonella positivity was analysed at slaughter pig/carcass level.
A slaughter pig was considered infected if microbiological culture of the lymph node sample
detected Salmonella, otherwise it was considered negative. A carcass was considered
contaminated, if microbiological culture of the carcass swab detected Salmonella, otherwise it was
considered negative.
3.2.1. Definition of the outcome variables
Data on slaughter pig Salmonella infection in lymph nodes and on Salmonella contamination of
carcasses were separately analysed and positivity for Salmonella spp. (hereafter Salmonella) was
the considered outcome.
In the Part A report, the prevalence of any Salmonella serovar was reported as Salmonella spp.
and in addition, the prevalence of S. Typhimurium, S. Derby, and Salmonella serovars other than
S. Typhimurium and S. Derby were analysed separately. The analyses for this Part B report also
examined each of these outcomes separately but no important differences were observed
compared to the results for the Salmonella spp. outcome. Therefore, the results of the analyses of
factors associated with the detection of S. Typhimurium, S. Derby or Salmonella serovars other
than S. Typhimurium and S. Derby are not presented.
3.2.2. Factors investigated
Information on factors potentially associated with Salmonella positivity was collected by the
competent authorities or under their supervision at the time of sampling. The mandatory fields in
the questionnaire included factors that could be associated with the outcome variables. The
following factors, described in detail in Annexes II and III, were considered:
Factors potentially associated with Salmonella infection in slaughter pigs:
• Factors related to the sensitivity of the sampling and testing process
1. Weight of the lymph node sample
2. Number of lymph nodes in the sample
3. Time between the date of sampling and the date of testing in the laboratory
• Factors related to lymph node infection
4. Month of sampling
5. Hour of sampling in the slaughterhouse
6. Weight of carcasses
1
In this report the observed prevalence means the prevalence estimate that accounts for clustering and weighting but
not for imperfect test sensitivity or specificity.
9
Factors potentially associated with Salmonella surface contamination of carcasses:
• Factors related to the sensitivity of the sampling and testing process
1. Time between the date of sampling and the date of testing in the laboratory
• Factors related to the surface contamination of carcasses
2. Status of the lymph node sample with respect to Salmonella infection
3. Month of sampling
4. Hour of sampling in the slaughterhouse
5. Weight of the carcasses
Some additional data and variables were collected on a voluntary basis by MSs. However, the
effects of these optional factors could not be evaluated due to the scarce data reported.
3.2.3. Exploratory analysis of potentially associated factors
The compulsory information that was recorded about each sample describes factors, or variables,
that might be associated with the presence of Salmonella in lymph node samples or on carcass
swab samples. Categorical variables were analysed through frequency tables and bar graphs.
Multiple bar graphs, by MS and for the EU global data, were produced by lattice package in the R
software. Quantitative variables were described through measures of central tendency and
dispersion such as mean and standard deviation as well as median and first and third quartiles.
Box plots were used for graphical visualisation.
The association between each factor and the status of the sample with respect to Salmonella
infection/contamination was visually explored by:
a) multiple bar graphs of weighted frequency counts of Salmonella positive and negative
slaughter pigs, by MS and different levels of categorical variables;
b) bar graphs of Salmonella prevalence and 95% confidence intervals, by different levels of
categorical variables;
c) box plots of quantitative variables for Salmonella positive and negative samples.
In the above bivariate analyses, the possible association between each of the individual factors and
Salmonella infection/contamination was considered.
In addition, the association between the proportion of Salmonella positive carcasses and the
proportions of positive pigs in lymph node samples in the same slaughterhouse was visually
investigated by Bland-Altman and box plot graphs. Only the slaughterhouses (n=146), for which
the number of pigs sampled was greater than 10, were included in this exploratory analysis.
3.2.4. Analysis of multicollinearity among potentially associated factors
The data were further analysed for evidence of association among potentially associated factors,
since they may be correlated with each other or one may completely explain the association of
another (collinearity). The Variance Inflation Factor (VIF) was used as a formal method to detect
correlation among risk factors (multicollinearity, explained in the section on regression analysis).
Essentially, each potential risk factor is used as the outcome in a regression analysis (described in
10
detail in Annex I, section I.3.1.). A VIF value that equals 1 indicates that there is no correlation
among risk factors, whereas VIF values higher than 1 indicate a correlation. A VIF value
exceeding 10 is interpreted as an indication of strong multicollinearity.
3.2.5. Identification of factors associated with Salmonella positivity
Multiple regression analysis was applied to obtain estimates of the association between each
factor, adjusted for the effect of other factors (potential for confounding1) and Salmonella
infection of the ileo-caecal lymph nodes of slaughter pigs or surface contamination of carcasses
with Salmonella. Multiple regression analyses were carried out at the EU level and separately by
MS.
3.2.5.1 Statistical model
Given the use of a binary outcome variable (Salmonella positive or negative status) taking only
two, mutually exclusive values (which were coded as 1 when the survey test was positive and
0 otherwise) logistic regression was the model of choice. However, as previously done in
prevalence estimation (Report part A), certain data characteristics needed to be taken into account
in the analysis.
Firstly, certain slaughter pigs/carcasses, which were the epidemiological units of the analysis,
were slaughtered at the same slaughterhouse. Therefore, they were exposed to the same conditions
and to certain same risk factors, including those on which no information was available in the
current survey but that might have been associated with Salmonella infection/contamination. Pigs
slaughtered in the same slaughterhouse are more likely to have been submitted to similar rearing
and pre-harvest processes, including comparable managerial and hygiene practices of farming,
transportation, and lairage. Similarly, carcasses processed in the same slaughterhouse are bound to
be exposed to similar risk factors for surface contamination associated with the slaughter process.
It was, therefore, reasonable to believe that slaughter pigs/carcasses processed at the same
slaughterhouse could not be considered as independent observations in statistical analysis.
Consequently, correlation among outcomes in those pigs/carcasses slaughtered at the same
slaughterhouse was taken into account in the regression models. Possible country confounding
effects were also taken account of in the analysis.
For the analysis of risk factors for slaughter pig infection a model was fitted where the effect of
slaughterhouse was included as random (random intercept logistic regression) and the effect of the
country as a country-specific fixed effect. The assumption underlying this type of model is that
each slaughterhouse, and consequently each slaughter pig processed in the slaughterhouse, is
1
In bivariate analysis, a potential risk factor might appear to be associated with Salmonella infection solely due to its
association with another risk factor for the infection. If, for example, slaughter pigs from MSs with high prevalence
were mostly sampled in summer months, summer could result as strongly associated with Salmonella when analysing
data at EU level. In this case, conclusions on a strong seasonality of the infection could be drawn, although it was just
the effect of unbalanced sampling. In fact, in this example, season may not have any real effect on Salmonella
infection. Confounding is, therefore, the over- or under- estimation of the effect of a potential risk factor due to its
association with other risk factors. In the example, the effect of season was overestimated due to the confounding
effect of MS. In order to eliminate confounding, and to obtain valid estimates of the effect of season, an adjustment
for MS is necessary, which can be achieved by multiple regression analysis. In certain cases, however, two or more
potential risk factors may be so strongly associated that separate estimates of their respective effects cannot be
obtained. In this case, the term collinearity or multicollinearity is used.
11
characterised by a certain baseline level of risk of infection, regardless of the exposure to risk
factors considered in the survey. The inclusion of a country-specific effect, which consists in
modelling a different parameter for each country in the model, is an attempt to correct
confounding between factors and country. A logistic mixed model, with a slaughterhouse random
effect on the intercept and country-specific fixed intercept, was therefore used to detect and assess
the effects of risk factors for Salmonella infection at slaughter pig level.
In a comparable way, to detect and assess the effects of risk factors for contamination of carcasses
with Salmonella a logistic mixed model was fitted at carcass level with a country-specific fixed
intercept, a slaughterhouse random intercept and random slope for predictor whose effect was
expected to vary across slaughterhouses. Inserting a random effect of the slaughterhouse on the
slope of a predictor allows the effect of that predictor on the risk of contamination to vary between
slaughterhouses, in addition to the baseline level of risk varying between slaughterhouses (random
intercept). More detailed explanations on analytical methods are given in Annex I.
Secondly, the sampling design of the survey was stratified. Slaughter pigs were sampled from
slaughterhouses that, in turn, were sampled from MSs. Slaughterhouse and MS can, therefore, be
considered as strata. The proportion of sampled slaughterhouses was not constant across MSs.
Similarly, the proportion of sampled pigs was not constant across slaughterhouses. Therefore, the
analysis had to be weighted in order to account for the stratified design and the varying proportion
of throughput from each slaughterhouse that was sampled, in order to obtain an unbiased estimate
of the association between possible risk factors and Salmonella infection of lymph nodes or
contamination of carcass surface. This approach was also followed when calculating prevalence
(Part A report). The weight to account for the sampling fraction of pigs within slaughterhouses
(WY2) was calculated as the ratio between the reported number of pigs produced in a
slaughterhouse during a year and the number of sampled pigs in the same slaughterhouse. The
weight to account for unequal sampling of slaughterhouses within a MS (WY1) was a proxyweight calculated as the ratio between 80 percent of the annual domestic throughput of slaughter
pigs in the MS and the sum of the annual throughputs of the sampled slaughterhouses in the same
MS (Annex I).
3.2.5.2 Model building for Salmonella lymph node infection at the EU- and country level
The investigation of the association between factors and the Salmonella infection in slaughter pigs
(lymph node samples) at EU level was done using a starting model that contained a global
intercept, a country-specific fixed effect, the factors of interest, and a random intercept for
slaughterhouse. This model was reduced by removing stepwise the most non-significant risk
factors until only covariates with P-values smaller than or equal to 0.05 remained in the final
model. Since no positive results were reported by Finland, it could not be considered in the EU
level analysis, as no country-specific effect could be estimated. Bulgaria was also excluded from
the global model building because its weight WY1 was not determined.
A similar model building exercise was carried out at country level: for each of the participating
countries a separate model was run. As in the EU model building, covariates were selected
through a backward selection procedure using random effect logistic regression. A
slaughterhouse-specific random intercept was incorporated into the model, which was fitted using
the GLIMMIX procedure in the SAS® System. The model for each country was then further
reduced so that only covariates with P-values smaller than or equal to 0.25 remained. Further, for
certain countries (Austria, Cyprus, Ireland, Sweden, The Netherlands), the slaughterhouse
12
random-effect was not taken into account in the logistic regression model, because of specific
model fitting obstacles.
3.2.5.3 Model building for Salmonella carcass surface contamination at the MS group and
MS level
The investigation of the association between factors and carcass Salmonella contamination at the
level of the group of 13 MSs was also carried out using a backward selection procedure. The
starting model contained a global intercept, a MS-specific fixed effect, all potentially associated
factors of interest and a random intercept for slaughterhouse. A slaughterhouse random slope was
also added for the “Lymph node infection” variable. The model was fitted using the GLIMMIX
procedure in the SAS® System. As Slovenia and Sweden had no Salmonella positive samples,
these MS data were not included in the analysis because no information was available to estimate
the country-specific effect. A similar model building exercise was performed on MS level: for
each of the participating MSs a separate model was fitted.
3.3. Analysis of the association between slaughter pigs’ lymph node
Salmonella infection and their carcass Salmonella contamination
The quantification of the association between the bacteriological culture of lymph node samples
and culture of carcass swabs with respect to Salmonella, was done by investigating the odds ratio
(OR) covered in the final EU level and MS-specific risk factor analyses models for carcass
Salmonella contamination as mentioned above, with the lymph node sample with respect to
Salmonella infection as an explanatory variable for the carcass swab outcome.
3.4. Analysis of the serovars and phage types distribution
3.4.1. Spatial distribution of reported Salmonella serovars in lymph nodes
The geographical analysis of the Salmonella serovar distribution was limited to country level, as
the location (coordinates) of the individual pig herds and/or slaughterhouse was not available. The
scan statistics (SaTScanTM) developed by Kulldorff was applied to detect spatial clusters of MSs
where each of the selected serovars was detected. The detection of clusters would allow
generating hypotheses on transmission or on common sources of Salmonella serovars in slaughter
pigs of neighbouring MSs. Moreover, SaTScan also allows for the detection of individual MSs
with a significant above EU average risk of Salmonella-specific serovar infection in slaughter
pigs.
SaTScan uses a circular window of different sizes to scan the study area. For each circle the
method computes the likelihood that the risk of infection is higher inside the circle compared to
outside the circle. The circle with the highest likelihood value is the one that has the highest
probability of containing a cluster. SaTScan accounts for multiple testing through the calculation
of the highest likelihood of occurrence for all possible cluster locations and sizes. The Poisson
model was chosen, which requires information about the number of estimated positive cases in
13
each MS and the population data. The estimated number of positive cases for each serovar was
calculated from the estimated prevalence. All estimated positive cases were geocoded to the
centroid of its respective country. The maximum window size was defined here as 50% of cases
and 999 replications were performed. It was set to look for spatial clusters of Salmonella spp., of
S. Derby, of S. Typhimurium, S. Enteritidis, S. Infantis and S. Rissen. Only the most likely cluster
and non-overlapping significant secondary clusters were displayed in this analysis. For the
analysis, the SaTScan output was imported into Arc GIS 9.1 to create cluster maps to visually
examine and compare identified clusters.
3.4.2. Comparison between Salmonella serovar and phage type distribution in
slaughter pigs, other animal species, feed and human salmonellosis cases
The serovar distribution found in ileo-caecal lymph nodes and on carcasses of slaughter pigs was
compared with the serovar distribution among MSs in animal feed and in human salmonellosis
cases as reported in the Community Summary Report on Zoonoses in 2006 (EFSA, 2006a). It was
also compared with serovar distribution among MSs in laying hen holdings, broiler and turkey
flocks as reported in previous baseline surveys (EFSA, 2007a; EFSA, 2007b; EFSA, 2008b).
Phage type distribution was described for S. Enteritidis and S. Typhimurium for lymph node and
carcass samples. The descriptive analysis of the serovar and phage type data was performed in
Microsoft Excel.
14
4. Results
4.1. Analysis of factors associated with Salmonella infection in lymph
nodes of slaughter pigs
In the following, the results are presented of the univariate description of potentially associated
factors and of the bivariate association between potentially associated factors, and Salmonella
infection in slaughter pigs, as determined by lymph node analyses. The graphs presenting the
bivariate associations must be considered as exploratory data analysis because these associations
have not been adjusted for the effect of other factors (potential for confounding) and for the MSs’
effects. Following the bivariate analysis, results from the multiple regression analysis are
presented, which are adjusted for the recorded confounding variables, notably country effect.
infection
4.1.1.1 Factors related to the sensitivity of the sampling process
•
Weight of the lymph node samples
A graphical display of the total weight of sampled lymph nodes by MS is presented in Figure 1.
This graph, as in similar ones presented hereafter, displays weighted frequencies (Annex II Tables II.1 and II.2). This means that the weighting of each pig was taken into account to show a
balanced prevalence within each month. Most lymph node samples (89%) belonged to the first
two weight categories: between 15 and 24gr and between 25 and 34gr, whereas only 11% of the
lymph node samples weighed more than 34gr. Salmonella positive lymph node samples belonged
mostly to the 15-24gr category, probably due to Spain’s strong contribution. The impact of the
weight of the lymph node samples on Salmonella detection in slaughter pigs has to be assessed
taking into account MS effect – refer to section 4.1.3.
15
Figure 1.
Bar plot of the weight of the lymph node samples tested, by country and for the EU, and by Salmonella status
y p
p
Salmonella negative
Salmonella positive
EU
The Netherlands The United Kingdom
Weighted number of pigs
1500
1000
500
0
Poland
1500
1000
500
0
Portugal
Slovakia
Slovenia
Spain
Sweden
Ireland
Italy
Latvia
Lithuania
Luxembourg
Norway
Estonia
Finland
France
Germany
Greece
Hungary
Austria
Belgium
Bulgaria
Cyprus
Czech Republic
Denmark
15
-2
4
25 gr
-3
4
35 gr
-4
4
>= gr
45
g
15 r
-2
4
25 gr
-3
4
35 gr
-4
4
>= gr
45
g
15 r
-2
4
25 gr
-3
4
35 gr
-4
4
>= gr
45
g
15 r
-2
4
25 gr
-3
4
35 gr
-4
4
>= gr
45
g
15 r
-2
4
25 gr
-3
4
35 gr
-4
4
>= gr
45
g
15 r
-2
4
25 gr
-3
4
35 gr
-4
4
>= gr
45
gr
1500
1000
500
0
Weight of the lymph node sample
16
10000
8000
6000
4000
2000
0
1500
1000
500
0
1500
1000
500
0
•
Number of lymph nodes in the sample
Figure 2 presents a graphical display of the number of lymph nodes per sample within MSs and
Norway, by means of box plots1. At EU level, the mean was 16.6 and the median (Q1; Q3) was 10
(6; 16). Medians were highest in Estonia (23), Ireland (20) and Slovakia (19), whereas the lowest
median (5) was encountered in Belgium, Hungary, Italy, Luxembourg and Poland. Descriptive
statistics of the number of lymph nodes in samples are presented in Annex II – Tables II.3 and
II.4.
Figure 2.
Box plot of the number of lymph nodes (LN) per sample, per country
The number of sampled slaughter pigs per MS is indicated between brackets.
Norway (408)
United Kingdom (599)
The Netherlands (1086)
Sweden (394)
Slovenia (429)
Slovakia (385)
Portugal (658)
Poland (1176)
Luxembourg (312)
Lithuania (461)
Latvia (392)
Italy (708)
Ireland (422)
Hungary (658)
Greece (345)
Germany (2567)
Finland (419)
Estonia (420)
Denmark (998)
Czech Republic (653)
Cyprus (359)
Bulgaria (176)
Belgium (601)
Austria (617)
0
20
40
60
80
Number of LN
The median number of collected lymph nodes per sample was not different for Salmonella
positive than for Salmonella negative ileo-caecal lymph node samples (Figure 3).
1
In the horizontal box plots, the left of the box represents the first quartile (Q1) of the distribution and the right the
third quartile (Q3), whereas the bar inside the box represents the median. Small circular symbols indicate extreme
values, differing from the box > 1.5 times the difference between the third and the first quartile (interquartile range).
17
Box plot of the number of lymph nodes per sample by Salmonella status of
40
0
20
Number of LN
60
80
Figure 3.
sample
negative (n=16443)
positive (n=2587)
Salmonella spp.
•
Time between the dates of sampling and testing in the laboratory
The time between the date of sampling and the date of testing in the laboratory varied among MSs
(Figure 4 and Annex II - Table II.5). Most lymph node samples (53%) were analysed for
Salmonella 1 day after sampling. Eighty-six percent of the samples were tested between 0 and 2
days after sampling.
18
Figure 4.
Frequency distribution of the weighted number of days between sampling and testing of lymph node samples, by country
and for the EU, and by Salmonella status
y p
p
Salmonella negative
Salmonella positive
EU
1500
10000
8000
6000
4000
2000
0
The Netherlands The United Kingdom
1000
500
0
1500
Poland
Portugal
Slovakia
Slovenia
Spain
Sweden
Ireland
Italy
Latvia
Lithuania
Luxembourg
Norway
Estonia
Finland
France
Germany
Greece
Hungary
1500
1000
500
0
1000
500
0
1500
1000
500
Austria
Belgium
Bulgaria
Cyprus
Czech Republic
Denmark
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
0
Days between sampling and testing
19
1500
1000
500
0
In general, there was an increase in Salmonella prevalence associated with an increased number of
days between sampling and testing up to a delay of 4 days in testing, followed by a decrease for a
delay of over 4 days (Figure 5, and Annex II – Table II.6) up to the accepted maximum of 7 days.
Figure 5.
Weighted Salmonella prevalence by number of days between sampling and
testing, with 95% confidence intervals, in the EU
30
20
10
3063
10132
3197
0
1
2
1691
619
168
52
43
3
4
5
6
7
0
% Prevalence of Salmonella spp. positive pigs
40
The number of sampled pigs is indicated inside each bar.
Time between sampling and testing
As no linear trend was observed, the “time between the date of sampling and testing” variable was
categorised into 3 levels for further analyses: 0-2 days, 3-4 days and 5-7 days. Categorisation
results are shown in Annex II – Tables II.7 and II.8.
4.1.1.2 Factors related to the lymph node infection
•
Month of sampling
A graphical display of the number of lymph node samples collected at MS-specific and at EU
level every month during the survey is presented in Figure 6 (see also Annex II – Tables II.9). The
collection of lymph node samples in slaughter pigs was homogeneous during the survey for most
participating countries. However, Bulgaria, Latvia, Lithuania and Portugal were delayed in the
start of sampling. The number of lymph node samples peaked at the EU level in September 2007
largely due to the contribution of Hungary, Poland and Spain, where most samples were taken in
that month. Denmark and France also contributed to the peak. Salmonella prevalence seems to be
lower during the first two months of the survey (October - November 2006) compared with the
following months of the survey. A slight increasing trend in prevalence is also suggested by the
graphical visualisation of the data, from January to the summer months of 2007 (Figure 7,
Annex II – Table II.10).
20
Figure 6.
Bar plot of the weighted number of lymph node samples collected by month and country, and for the EU, and by
Salmonella status
Months are ordered from October 2006 to September 2007. y p
p
Salmonella negative
Salmonella positive
EU
400
300
200
100
0
400
300
200
100
0
The Netherlands
The United Kingdom
Poland
Portugal
2000
1500
1000
500
0
Slovakia
Slovenia
Spain
Sweden
Ireland
Italy
Latvia
Lithuania
Luxembourg
Norway
Estonia
Finland
France
Germany
Greece
Hungary
Austria
Belgium
Bulgaria
Cyprus
Czech Republic
Denmark
Month of Sampling
21
Ju
l
O
ct
Ja
n
Ap
r
Ju
l
O
ct
Ja
n
Ap
r
Ju
l
O
ct
Ja
n
Ap
r
Ju
l
O
ct
Ja
n
Ap
r
r
Ju
l
Ap
O
ct
Ja
n
Ju
l
O
ct
Ja
n
Ap
r
400
300
200
100
0
400
300
200
100
0
400
300
200
100
0
Figure 7.
Weighted Salmonella prevalence by the month of sampling, with 95%
confidence intervals, in the EU
30
20
10
789
1204
1471
1639
1725
Oct
Nov
Dec
Jan
Feb
1762
1627
1680
1684
1696
1678
2076
Mar
Apr
May
Jun
Jul
Aug
Sep
0
40
The number of sampled pigs is indicated inside each bar.
Month of sampling
•
Sampling quarter
Treating the month of sampling as a categorical variable implies a nominal variable with
12 categories. Including all months as categories of a class variable may yield to overparameterisation of the multiple regression model, especially when countries are considered
separately. To remedy this problem and because a seasonal trend could be expected to occur, a
categorical variable “Sampling quarter” was created with the following four categories: OctoberDecember 2006, January-March 2007, April-June 2007, and July-September 2007. In order to test
for any seasonal effect on the risk of Salmonella infection in slaughter pigs, the four categories
were coded: 1 when the slaughter pig was sampled in the period October-December 2006, 2 when
sampled in the period January-March 2007, 3 in the period April-June, and 4 in the period JulySeptember 2007. A graphical display of the numbers of lymph node samples collected at MSspecific and at EU level in each quarter during the survey is presented in Figure 8 (see also Annex
II – Table II.11).
22
Figure 8.
Bar plot of the weighted number of tested lymph node samples, by sampling quarter and country, and for the EU, and by
Salmonella status
Quarters are ordered from October-December 2006 (1) to July-September
2007 (4).
y p
p
Salmonella negative
Salmonella positive
5000
4000
3000
2000
1000
0
Bulgaria
Cyprus
Czech Republic
Denmark
1
Quarter of Sampling
23
800
600
400
200
0
800
600
400
200
0
4
Belgium
3
Austria
800
600
400
200
0
2
Hungary
1
Greece
4
Germany
3
France
2
Finland
1
Estonia
4
Norway
3
Luxembourg
2
Lithuania
1
Latvia
4
Italy
3
Ireland
800
600
400
200
0
2
Sweden
1
Spain
4
Slovenia
3
Slovakia
2
Portugal
1
Poland
4
The United Kingdom
3
The Netherlands
2
800
600
400
200
0
EU
Generally, Salmonella prevalence in lymph nodes appears to increase towards the end of the
survey (Figure 9, see also Annex II – Table II.12). However, care must be taken in interpreting
this observation, as there were substantial differences among the MSs in the distribution of
samples across the quarters of the sampling period. Therefore, confounding is possible.
Figure 9.
Weighted Salmonella lymph node prevalence by sampling quarter, with 95%
confidence intervals, in the EU
30
20
10
3462
5124
4990
5449
1
2
3
4
0
40
Quarters are ordered from October – December 2006 (1) to July —September 2007 (4).
Number of sampled pigs represented inside each bar.
Quarter of sampling
•
Hour of sampling
A graphical display of the number of samples collected at country-specific and at EU level during
each hour of the working day in slaughterhouses is presented in Figure 10.
24
Figure 10.
Bar plot of the weighted number of lymph node samples collected, by hour of sampling and country, and for the EU, by
Salmonella status
Hours are ordered from 00 to 23.
y p
p
Salmonella negative
Salmonella positive
EU
The Netherlands
The United Kingdom
Poland
Portugal
400
300
200
100
0
2500
2000
1500
1000
500
0
Slovakia
Slovenia
Spain
Sweden
400
300
200
100
0
400
300
200
100
0
Ireland
Italy
Latvia
Lithuania
Luxembourg
Norway
Estonia
Finland
France
Germany
Greece
Hungary
Austria
Belgium
Bulgaria
Cyprus
Czech Republic
Denmark
Hour of Sampling
25
20
15
10
05
00
20
15
10
05
00
20
15
10
05
00
20
15
10
05
20
00
15
10
05
20
00
15
10
05
00
400
300
200
100
0
400
300
200
100
0
Figure 11 suggests a lower prevalence in lymph node samples during the day time (see also Annex
II – Table II.13). However, most samples were taken between 05:00 and 18:00.
Figure 11.
Weighted Salmonella prevalence by hour of sampling, with 95% confidence
intervals, in the EU
Number of sampled pigs indicated inside each bar.
80
60
40
20
100
Lymph node samples
52
57 153 310 10111755263824782787235616991006 844 515 418 252 154 135 93
2
3
56
28
70
0
111 53
0
1
4
5
6
7
8
9
10
12
14
16
18
20
22
Hour of sampling
Treating the variable “Hour of sampling” as a categorical variable in the modelling exercise
implies a nominal variable with 24 categories, and will result in an additional 23 parameters. As
presented in detail in Annex IV (section IV.1), a sine type of evolution of the prevalence of
infection can be modelled, with peaks during the night time. The new variable “Time of sampling
(sine)” was therefore used in the building of the model on slaughter pig infection. A significant
effect of this variable would imply that there is a sine trend such that day and night results differ
significantly.
•
Weight of carcasses
At EU level, the median carcass weight (Q1; Q3) was 85 kg (74; 95). The distribution of carcass
weights for slaughter pigs sampled for lymph nodes, at country level is shown in Figure 12
(Annex II – Table II.14). The heaviest carcasses were sampled in Italy (median=132 kg) and
Hungary (median=110 kg), whereas the medians were lowest in Cyprus (70 kg), Greece (70 kg),
Estonia (72 kg) and Latvia (72 kg). The median weight of carcasses in the group of infected
slaughter pigs is lower than the median weight of carcasses of negative slaughter pigs (Figure 13,
see also Annex II. – Table II.15). However, this observation must be adjusted for potential
confounding before any conclusions are made.
26
Figure 12.
country
Box plot of carcass weights for slaughter pigs sampled for lymph nodes, per
Norway (408)
The Netherlands (1087)
Sweden (394)
Spain (2619)
Slovenia (431)
Slovakia (385)
Portugal (658)
Poland (1176)
Luxembourg (313)
Lithuania (461)
Latvia (392)
Italy (709)
Ireland (422)
Hungary (658)
Greece (345)
Germany (2567)
France (1163)
Finland (419)
Estonia (420)
Denmark (998)
Cyprus (359)
Bulgaria (176)
Belgium (601)
Austria (617)
40
60
80
100
120
Weight of carcass
100
80
40
60
Weight of carcass
120
Figure 13.
Box plot of carcass weight for slaughter pigs sampled for lymph nodes, by
Salmonella status
negative (n=16443)
positive (n=2588)
Salmonella spp.
27
4.1.2. Analysis of multicollinearity among potential factors
The VIF values calculated for the multicollinearity analysis among the factors associated with
Salmonella prevalence in lymph node samples in the EU are presented in Table IV.1 of Annex IV.
This analysis showed that multicollinearity was not important for the global model.
The analysis was repeated focussing on each separate participating country and the VIF values are
displayed in Table IV.2 of Annex IV. This analysis showed that multicollinearity was neither an
issue for the MS models.
4.1.3. Multiple regression analysis at EU level
In this section, Norway is also included in the EU level analyses and results referring to it.
Overall, factors associated with Salmonella infection in lymph node samples of slaughter pigs are
presented in Table 1. The two factors retained in the final regression model were related to the
sensitivity of the sampling process. The model included country-specific effects (Annex IV –
Table IV.5). Therefore, each odds ratio1 (OR) was adjusted for MSs. A random intercept for the
slaughterhouses (Annex IV – Table IV.6) was also inserted in the model.
Table 1.
The final random effect logistic model for factors associated with Salmonella
infection in lymph nodes of slaughter pigs, in the EU, 2006-2007.
Random effect logistic model a, b
Variables
15-24gr
25-34gr
35-44gr
≥ 45gr
OR
95%CI
1
1.3
1.2
1.9
1.1, 1.6
0.8, 1.7
1.2, 3.0
1
1.2
0.99
1.04, 1.4
0.65, 1.5
c
Time (in days) between the date of sampling and
testing in the laboratory c
0 to 2 days
3 to 4 days
5 to 7 days
a
Estimates and standard errors were assessed using a mixed model with a slaughterhouse random effect on
the intercept and country-specific fixed intercept.
b
As the country-specific effect of Finland could not be estimated (no lymph node samples tested positive
for Salmonella), that Member State was not considered in the EU level analysis.
c
Significant at P-value < 0.05.
1
An OR of 1.0 implies that there is no association between a risk factor and Salmonella infection; an OR above 1.0
implies an increased risk of Salmonella infection among pigs exposed to that factor while an OR below 1.0 implies a
reduced risk of Salmonella infection among exposed pigs. In any study, it is possible that an OR different to 1.0 may
arise by chance and the level of significance (P-value) estimates this probability. Consequently, if the 95% confidence
interval of the OR does not comprise 1, meaning that both the lower and the upper limits are either greater, or less
than 1, it can be concluded that the association with a potential risk factor and Salmonella is statistically significant (P
< 0.05).
28
According to the analyses, the probability of Salmonella detection in lymph nodes increases as the
weight of the lymph node samples increases. For example, the odds of detecting Salmonella
infection in a lymph node sample weighing more than 45gr is 1.9 times higher than the odds for a
15-24gr sample. The effect of the time delay between sampling and the start of laboratory testing
is also identified as a factor associated with Salmonella detection in lymph nodes. A 3-4 day delay
in testing increases the likelihood of detection of Salmonella in lymph node samples by 20%
compared to a delay of less than 2 days. Conversely, the impact of a delay of over 4 days on the
risk of Salmonella detection was not significantly different from a delay of less than 2 days.
However, only few samples were examined more than 4 days after sampling.
4.1.4. Multiple regression analysis at the country level
Results of regression analysis by country are presented in Table 2, where rows correspond to
country and columns to potential associated factors. Cells in the table contain the OR measuring
the risk factor effect in the corresponding column and its 95% CI, in the country in the
corresponding row. In addition, the shading of the cells is used to show whether the OR obtained a
level of conventional statistical significance. Each level of significance (P-value of less than 1%,
5%, 10%, and 25%, respectively) is indicated by a different shade of grey, darker meaning there
was a greater probability that the observed association did not arise by chance alone. Empty cells
mean that there was insufficient evidence of an association between the risk factor and Salmonella
infection in that particular country and the risk factor was thus not included in the final model. For
some risk factors with more than two categories, data were not available for all categories in some
countries. For instance, for Belgium only categories “15-24gr” and “25-34gr” of the variable
“Weight of the lymph node samples” were available. Therefore, only these two levels are
available for comparison to obtain an OR estimate for this factor for the country.
The matrix presentation of results facilitates the identification of factors that may increase or
reduce the risk of Salmonella infection across countries, as the effects of these factors might vary
among countries. Indeed, a great variability between significant risk factors obtained for each
country was observed. Some factors even had contrasting effects depending on the country. In
addition, when these effects are studied at EU level, these results may average out so that no
significant effect is observed in the general model.
The final models fitted for Austria, Ireland, Latvia, Portugal, Slovenia, Sweden and the United
Kingdom did not identify factors among those tested that were significantly associated (at a level
of 5% or less) with Salmonella infection in lymph nodes. As the model fitted for Norway is based
on one positive pig only, the results of this analysis should thus be considered with caution.
Significant associations (P-value < 0.05) observed for each of the factors across the MSs are:
• Carcass weights (10 kg increments) – This factor was significantly associated with an
increased risk of Salmonella infection in three countries (Lithuania, Poland, and Norway);
where for every 10 kg increase in carcass weight, the risk increased by 20-60%. However, for
two countries (Belgium and Hungary), the increase of carcass weight appeared to be
associated with a lower risk of infection.
29
Table 2.
Random effect logistic models for factors associated with Salmonella infection in slaughter pigs in participating countries
Odds ratio estimates and 95%CI are presented for significant (at different levels of significance) risk factors obtained for each country separately. The shade of gray of the
cell illustrates the level of significance (P-value) of the association, according to the following scale:
Austria a
617
Belgium
601
Bulgaria
176
Cyprus a
359
Czech Republic
653
Denmark
998
Estonia
420
≥ 45gr
vs.
15-24gr
6.9
0.8, 62
11.5
1.2, 112
10.2
0.8, 125
0.6
N/A
N/A
0.3, 0.9
0.1 b
0.4 b
0.005, 1.7
0.03, 6.7
2.3
N/A
1.2, 4.5
Sampling quarter
Time between
sampling and testing
3 to 4 days
vs.
0 to 2 days
5 to 7 days
vs.
0 to 2 days
3.3
0.65, 16
N/A
Oct.-Dec.06
vs.
Jul.-Sep.07
Jan.-Mar.06
vs.
Jul.-Sep.07
Apr.-Jun.06
vs.
Jul.-Sep.07
Weight of carcasses
(10-kg increments)
35-44gr
vs.
15-24gr
0.10 < P < 0.25:
Sampling time (sine)
25-34gr
vs.
15-24gr
0.05 < P < 0.10:
Number of lymph nodes
in the sample
0.01 < P < 0.05:
Weight of the
lymph node samples
Number of carcasses
P < 0.01:
0.92
0.87
0.84, 1.01
0.76, 0.99
0.06
N/A
0.04, 0.11
N/A
4.3
1.8
2.6
0.62
1.5, 13
0.6, 5.5
0.9, 7.9
0.35, 1.1
2.2
1.05
0.9
1.03
1.8
0.12, 39.4
0.05, 24.3
0.05, 14.6
1.01, 1.05
0.67, 4.7
1.8
N/A
N/A
1.07, 2.9
1.5
0.7
1.01, 2.1
0.09, 5.5
0.05
0.01, 0.5
30
France
1,163
Germany
2,567
Greece
345
Hungary
658
Ireland a
422
Italy
708
Latvia
392
Lithuania
461
Luxembourg
312
≥ 45gr
vs.
15-24gr
1.3
0.82
4.2
0.94, 1.7
0.21, 2.9
1.3, 13
1.8
1.3
0.9
1.2, 2.7
0.45, 3.7
0.32, 2.3
Sampling quarter
Time between
3 to 4 days
vs.
0 to 2 days
5 to 7 days
vs.
0 to 2 days
Oct.-Dec.06
vs.
Jul.-Sep.07
Jan.-Mar.06
vs.
Jul.-Sep.07
Apr.-Jun.06
vs.
Jul.-Sep.07
Weight of carcasses
(10-kg increments)
35-44gr
vs.
15-24gr
25-34gr
vs.
15-24gr
in the sample
Weight of the
lymph node samples
Continued
Number of carcasses
Table 2.
N/A
0.90
0.79, 1.04
0.078
0.32
1.1
0.010, 0.64
0.14, 0.76
0.40, 2.9
1.2
0.93, 1.6
1.7
1.3
2.1
0.65
0.74, 4.0
0.49, 3.5
0.86, 5.1
0.46, 0.91
1.1 b
1.6 b
14 b
0.67, 1.7
0.75, 3.5
11, 17
0.98
2.6
3.0
1.9
0.94, 1.01
1.1, 6.3
1.3, 7.1
0.79, 4.6
2.2
0.98, 4.9
0.02
<0.001, 8.8
31
1.6
1.2, 2.1
2.1
5.9
4.5
0.61
1.1, 3.9
0.47, 74
0.36, 57
0.34, 1.1
Poland
1,176
Portugal
658
Slovakia
385
Slovenia
429
Spain
Sweden a
Netherlands a
2,619
≥ 45gr
vs.
15-24gr
1.1 b
0.56 b
0.52, 2.5
0.045, 6.9
Sampling quarter
Time between
3 to 4 days
vs.
0 to 2 days
5 to 7 days
vs.
0 to 2 days
Oct.-Dec.06
vs.
Jul.-Sep.07
Jan.-Mar.06
vs.
Jul.-Sep.07
Apr.-Jun.06
vs.
Jul.-Sep.07
8.7 b
0.64
1.8, 41
0.31, 1.3
0.43
0.20, 0.94
0.32, 1.4
NA
0.66
0.8
0.6
0.5, 1.3
0.3, 1.1
0.48
0.59
1.4
0.98
0.37
0.23
0.15, 1.5
0.19, 1.8
0.23, 8.3
0.38, 2.5
0.093, 1.4
0.098, 0.5
0.68
0.49, 0.94
N/A
N/A
N/A
1.2
1.03, 1.5
1.7
0.71
0.83
1.3
0.91, 3.0
0.19, 2.7
0.39, 1.8
0.99, 1.6
0.89
0.60
0.77
0.89
0.49, 1.6
0.49, 0.73
0.63, 0.93
0.75, 1.05
0.86
394
Weight of carcasses
(10-kg increments)
35-44gr
vs.
15-24gr
25-34gr
vs.
15-24gr
in the sample
Weight of the
lymph node samples
Continued
Number of carcasses
Table 2.
0.05
0.68, 1.1
<0.001, 3.3
0.47
0.19, 1.2
1,086
United Kingdom
599
Norway c
408
1.9
2.0
1.3
1.7
1.2, 3.2
0.95, 4.1
0.13, 13
0.96, 2.9
Results based on independent logistic regression model.
As the number of samples in some categories is less than 5% of the total number for this country, the significance of the variable should be considered cautiously.
c
The model for Norway is based on 1 positive slaughter pig only.
a
b
32
18
1.6
9.3, 36
1.3, 2.0
•
•
•
•
•
Sampling time – A sine function was used to describe the effect of time of day upon the risk of
Salmonella infection. A significant association, whatever is the direction, means that there is a
difference in the risk of infection between the pigs slaughtered at the end of the working day
compared to the beginning. In three countries (Bulgaria, Estonia, and Norway), there was
evidence of a significant association of this variable with the risk of infection.
Sampling quarter – Five countries showed a significant association between this factor and
detecting Salmonella infection. In Cyprus and Luxembourg the risk was higher in OctoberDecember 2006 compared to the summer quarter (July-September) of 2007. Conversely, in
Greece the risk in the summer quarter of 2007 was greater than that the first two quarters of
the study (October-December 2006 and January-March 2007). In Slovakia and Spain, the risk
was greater in the summer quarter of 2007 compared to that of April-June 2007, and to those
of quarters January-March 2007 and April-June 2007, respectively.
Time between sampling and testing – This factor was not significantly (P < 0.05) associated
with the outcome in any country1.
Weight of lymph node samples – in six MSs (Cyprus, Denmark, France, Germany, Italy2, and
Poland²), increased weight of lymph node samples generally increased the probability of
observing Salmonella infection (OR > 1.0), while in two MSs (Belgium and Bulgaria²), there
was a reduced probability of detecting Salmonella infection.
Number of lymph nodes in the sample – the probability of detecting Salmonella infection
increased significantly as the number of lymph nodes tested increased in the Czech Republic.
4.2. Analysis of factors associated with surface contamination of
carcasses with Salmonella
contamination
4.2.1.1 Factors related to the sensitivity of the sampling process
•
Time between the date of sampling and testing in the laboratory
The time between the date of sampling and testing in the laboratory varied among MSs (Figure 14,
Annex III – Table III.1) but being mostly one or two days. In general, Salmonella was more likely
to be detected from the sample when testing started after 2-4 days from sampling, compared to
testing on sampling day or on the following day (Figure 15, Annex III – Table III.2). There was no
evidence that testing 5 or more days after the sampling altered the probability of detecting
Salmonella, although there were fewer observations in this period and thus the results are more
uncertain. These results are not adjusted for the country effect. As no linear trend was observed,
the “time between the date of sampling and testing” variable was categorised into four levels: 0
day, 1 day, 2 days and 3-7 days (Annex III – Tables III.3 and III.4).
1
However, in Austria and Denmark, a delay of more than 2 days was associated with an increased probability of
observing Salmonella infection compared to samples that were tested after less than 2 days, at the significance level of
25%. In Denmark, testing 5 or more days after sampling was associated with a reduced risk of Salmonella infection
appearing. The time between sampling and testing was found to be significantly associated with Salmonella detection
at the EU level, probably due to higher statistical power because of a greater sample size.
2
As the number of samples in some categories is less than 5% of the total number for this country, the significance of
the variable should be considered cautiously.
33
Figure 14.
Frequency distribution of the time (days) between sampling and testing of carcass swabs, by MS and for the 13-MS group,
and by Salmonella status
Salmonella negative
Salmonella positive
13-MS
2500
2000
1500
1000
500
0
The United Kingdom
Weighted number of carcasses
400
300
200
100
0
Lithuania
Poland
Slovenia
Sweden
Denmark
France
Ireland
Latvia
400
300
200
100
0
400
300
200
100
0
Austria
Belgium
Cyprus
Czech Republic
Days between sampling and testing
34
7
6
5
4
3
2
1
0
7
6
5
4
3
2
1
0
7
6
5
4
3
2
1
0
7
6
5
4
3
2
1
0
400
300
200
100
0
Figure 15.
Weighted Salmonella prevalence of carcass contamination by number of days
between sampling and testing, with 95% confidence intervals, in the 13-MS group
30
20
10
0
40
The number of sample pigs is indicated inside each bar.
827
3146
1036
524
145
37
10
11
0
1
2
3
4
5
6
7
4.2.1.2 Factors related to surface contamination of carcasses
•
Lymph node infection
The bivariate analysis indicated that there may be an association between the Salmonella infection
status of slaughter pigs and the Salmonella contamination of carcasses, at the 13-MS group level.
The weighted prevalence of Salmonella contamination of carcasses was greater for slaughter pigs
with Salmonella infection in lymph nodes compared to the pigs with un-infected lymph nodes
(Figure 16).
30
20
10
5349
387
0
1
0
40
Figure 16.
Weighted prevalence of Salmonella contaminated carcasses by Salmonella
infection status of the slaughter pig, with 95% confidence intervals, in the 13-MS group a
a
0: negative, 1: positive
35
•
Analysis of the association between proportions of Salmonella contaminated carcasses and of
pigs positive in lymph node samples at the slaughterhouses (146 slaughterhouses where at
least 10 pigs were sampled)
In Figure 17, the proportion of Salmonella positive pigs (in lymph nodes) is represented on the
horizontal axis, whereas the difference between proportions of Salmonella contaminated carcasses
and of positive pigs (lymph nodes) is represented on the vertical axis. Each slaughterhouse is
represented by a circle, whose size is proportional to the number of pigs sampled in that
slaughterhouse. Circles above the horizontal line represent slaughterhouses where the proportion
of contaminated carcasses was greater than that of positive pigs, whereas circles below the line
correspond to slaughterhouses where proportion of contaminated carcasses was smaller than the
proportion of positive pigs. The proportions of Salmonella contaminated carcasses and of
Salmonella infected pigs varied importantly among slaughterhouses. Moreover, the proportion of
contaminated carcasses differed importantly between slaughterhouses having a similar proportion
of Salmonella infected pigs. For example, in the graph, slaughterhouses SH1 and SH2 have both a
proportion of Salmonella infected pigs of 13.4%, whereas their respective proportion of
Salmonella contaminated carcasses were 25.4% and 0.1%. For the majority of slaughterhouses,
the proportion of contaminated carcasses is lower than the proportion of positive pigs. However,
for 35 slaughterhouses (24%) contamination of carcasses is greater than that of pigs.
30
20
-10
0
10
SH1
SH2
-20
Difference between proportions of positive carcasses and pigs
Figure 17.
Bland-Altman graph of the slaughterhouse-specific proportion of Salmonella
positive pigs against the slaughterhouse-specific difference between proportions of
Salmonella contaminated carcasses and of Salmonella positive pigs
0
10
20
30
40
Proportion (%) of Salmonella spp. positive pigs
Figure 18 presents a box plot of the proportions of Salmonella contaminated carcasses in the
slaughterhouses (n=146) categorised by the proportion of Salmonella positive pigs. The widths of
the box plots are proportional to the number of slaughterhouses per category.
36
50
40
30
20
10
0
Proportion (%) of Salmonella spp. contaminated carcasses
60
Figure 18.
Box plot of slaughterhouse-specific proportion Salmonella contaminated
carcasses, by slaughterhouse-specific proportion Salmonella positive pigs (146
slaughterhouses)
1.25 (0.0, 6.7)
9.1 (6.9, 13.3)
16.7 (13.5, 20.0)
25.2 (20.3, 39.5)
Category of slaughterhouses by the proportion (%) of Salmonella spp. positive pigs
The median (minimum, maximum) of the proportion of positive pigs in each category is reported under the box.
The median proportion of contaminated carcasses (the horizontal line in boxes) and the
interquartile range (the height of boxes) appear to increase with the proportion of positive sampled
pigs in the slaughterhouse. For a given category (i.e. for slaughterhouses exposed to a comparable
ingress of Salmonella infected pigs), the proportion of contaminated carcasses varies importantly
between the slaughterhouses and this variation is higher for slaughterhouses in the categories
having higher proportion of positive pigs.
•
Month of sampling
A graphical display of the numbers of carcasses sampled at the MS-specific and at the 13-MS
group level in each month during the survey is presented in Figure 19 (Annex III – Tables III.5).
Sampling of carcasses of slaughter pigs was homogeneous during the survey for most
participating countries, although the number of carcasses tested increased progressively during the
first four months of the survey. The start of sampling was delayed in Latvia and Lithuania.
Generally, Salmonella prevalence on carcasses appeared to be lowest at the beginning of the
survey (Figure 20, Annex III – Table III.6).
37
Figure 19.
Bar plot of the weighted number of carcass swabs collected by month and MS,
and for the 13-MS group, by Salmonella status
Months are ordered from October 2006 to September 2007.
Salmonella negative
Salmonella positive
13-MS
The United Kingdom
600
60
40
20
0
Lithuania
400
200
0
Poland
Slovenia
Sweden
France
Ireland
Latvia
60
40
20
0
Denmark
Austria
Belgium
Cyprus
Czech Republic
Month of Sampling
Figure 20.
Weighted Salmonella prevalence of carcass contamination by month of
sampling, with 95% confidence interval, in the 13-MS group
30
20
10
288
415
370
480
536
Oct
Nov
Dec
Jan
Feb
530
515
504
516
529
514
539
Mar
Apr
May
Jun
Jul
Aug
Sep
0
40
Number of sampled pigs indicated inside each bar.
Month of sampling
38
l.
Ju
r.
Ap
n.
Ja
O
ct
.
l.
Ju
r.
Ap
n.
Ja
O
ct
.
l.
Ju
r.
Ap
n.
Ja
O
ct
.
l.
Ju
r.
Ap
n.
Ja
O
ct
.
60
40
20
0
60
40
20
0
•
Sampling quarter
The numbers of carcasses sampled at the MS-specific and at the EU level in each quarter during
the survey is presented in Figure 21 (see also Annex III – Table III.7). Some variation in the
number of samples was obvious between the quarters of the survey. Generally, the Salmonella
prevalence on the carcasses appears to increase with the quarter during the survey (Figure 22, see
also Annex III – Table III.8).
Figure 21.
Bar plot of the weighted number of carcass swabs collected by quarter and
MS, and for the 13-MS group, and by Salmonella status
Quarters are ordered from October-December 2006 (1) to July-September 2007 (4).
Carcass swabs
Salmonella negative
Salmonella positive
13-MS
The United Kingdom
1500
150
1000
100
500
50
0
0
Lithuania
Poland
Slovenia
Sweden
Denmark
France
Ireland
Latvia
150
100
50
0
150
100
50
0
Austria
Belgium
Cyprus
Czech Republic
150
100
50
4
3
2
1
4
3
2
1
4
3
2
1
4
3
2
1
0
Quarter of Sampling
39
Figure 22.
Weighted Salmonella prevalence of carcass contamination by sampling
quarter, with 95% confidence intervals, in the 13-MS group
Number of sampled pigs are indicated inside each bar.
30
20
10
1073
1546
1535
1582
1
2
3
4
0
40
Quarters are ordered from October-December 2006 (1) to July-September 2007 (4).
Quarter of sampling
•
Hour of sampling
A graphical display of the number of carcasses sampled at the MS-specific and at the 13-MS
group level in each hour of the working day is presented in Figure 23. The results shown in Figure
24 show that there were very few samples taken after 20:00 and before 05:00, thus the seemingly
higher prevalence observed during these times must be interpreted with caution (see also Annex
III. - Table III.9).
A new variable “Time of sampling (sine)” was created (Annex IV, section IV.1) and used in the
building of the model on carcass contamination. A significant effect of this variable would imply
that there is a sine trend such that day and night results differ significantly.
40
Figure 23.
Bar plot of the weighted number of carcass swabs collected by hour of sampling and MS, and for the 13-MS group, and by
Salmonella status
Hours are ordered from 00 to 23.
Salmonella negative
Salmonella positive
EU
2500
2000
1500
1000
500
0
The United Kingdom
150
100
50
0
Lithuania
Poland
Slovenia
Sweden
Denmark
France
Ireland
Latvia
150
100
50
0
150
100
50
0
Austria
Belgium
Cyprus
Czech Republic
150
100
50
Hour of Sampling
41
20
15
10
05
00
20
15
10
05
00
20
15
10
05
00
20
15
10
05
00
0
Figure 24.
Weighted Salmonella prevalence carcass contamination by hour of sampling,
with 95% confidence interval, in the 13 MS-group
80
60
40
20
10
3
3
1
5
40 290 522 811 728 765 696 547 383 362 184 154 79
54
0
1
2
3
4
5
18
44
22
21
6
6
0
100
Number of sampled carcasses indicated inside each bar.
6
7
8
9
10
12
14
16
20
22
Hour of sampling
•
Weight of carcasses
The distribution of carcass weights of slaughter pigs sampled for carcass swabs, at Member State
level is shown in Figure 25 (see also Annex III – Table III.10). At the 13-MS group level, the
median (Q1; Q3) was 84 kg (76; 92). On average, the heaviest carcasses were sampled in Austria
(median=95 kg), whereas the medians were lowest in Cyprus (70 kg), and Latvia (72 kg). The
median carcass weight in the group of contaminated carcasses is not different to the median
carcass weight of the group of carcasses tested negative for surface contamination (Figure 26 and
Annex III – Table III.11).
Figure 25.
Box plot of the weight of the carcasses sampled with carcass swabs per MS
Sweden (402)
Slovenia (441)
Poland (447)
Lithuania (461)
Latvia (391)
Ireland (422)
France (413)
Denmark (344)
Cyprus (359)
Belgium (381)
Austria (617)
40
60
80
100
120
Weight of carcass
42
100
80
40
60
Weight of carcass
120
Figure 26.
Box plot of the weight of carcasses sampled with carcass swabs, by Salmonella
status, in the 13-MS group
negative (n=5349)
positive (n=387)
Salmonella spp.
4.2.2. Analysis of multicollinearity among potential factors
The VIF values calculated for the analysis of multicollinearity among the factors associated with
Salmonella contamination of carcasses in the 13-MS group are presented in Table IV.3 of Annex
IV. This analysis showed that multicollinearity was not an issue for the global model.
The analysis was repeated focusing on each of the participating countries separately and the VIF
values are displayed in Table IV.4 of Annex IV. No issues regarding multicollinearity were
observed in the analysis carried out by country.
4.2.3. Multiple regression analysis at the 13-MS group level
The factors associated with Salmonella surface contamination of carcasses at the 13-MS level are
presented in Table 3. The three factors retained in the final model were related to the sensitivity of
the sampling and testing process, the infection of slaughter pig lymph nodes and the sampling
quarter. The model included significant MS-specific effects (Annex IV, Table IV.7) and therefore,
each OR was adjusted for country effect. Correlation among Salmonella infection/contamination
of pigs/carcasses slaughtered at the same slaughterhouse was taken into account in the model. The
slaughterhouse random effects inserted in the model on the intercept and on the slope of the
variable “Lymph node infection of the slaughter pigs” were also statistically significant, with Pvalue < 0.0001 and P-value = 0.0008, respectively (Annex IV, Table IV.8).
43
Table 3.
The final random effect logistic model for factors associated with the
Salmonella surface contamination of carcasses of slaughter pigs, in a group of 13 MSs, 20062007.
Variables
Random effect logistic model a, b
OR
95%CI
Time (in days) between the date of sampling and
testing in the laboratory c
0 day
1 day
2 days
3 to 7 days
0.51
1
1.009
0.70
0.28, 0.93
0.76, 1.3
0.52, 0.96
Lymph node infection of the slaughter pig c
No
Yes
1
1.8
1.1, 2.8
Sampling quarter c
Oct. – Dec. 2006
Jan. – Mar. 2007
Apr. – Jun. 2007
Jul. – Sept. 2007
0.51
0.58
1.002
1
0.35, 0.72
0.44, 0.77
0.77, 1.3
-
a
Estimates and standard errors were assessed using a mixed model with a slaughterhouse random effect on
the intercept (P-value < 0.0001) and on the slope of the “Lymph node infection of the slaughter pig” variable
(P-value = 0.0008) and country-specific fixed intercept.
b
As the country-specific effect of Slovenia and Sweden could not be estimated (no carcass swabs tested
positive for Salmonella), these countries were not considered in the MS group level analysis.
c
Significant at P-value < 0.05.
According to the analyses, a slaughter pig that was infected by Salmonella in the lymph nodes was
approximately twice more likely to result in a carcass that is contaminated with Salmonella on the
surface than a pig whose lymph nodes were not shown to be infected. Also the time of sampling
throughout the 1-year survey was shown to have an impact on Salmonella contamination of
carcasses. The carcasses were less likely to get contaminated in October-December and JanuaryMarch than in July-September. However, no significant difference was observed between the
likelihood of having a Salmonella positive carcass in April-June compared to the July-September
quarter.
The effect of the time delay between sampling and the start of laboratory testing was identified as
a factor associated with the detection of Salmonella in carcass swabs. Compared to a test carried
out 1-2 days after that of sampling, the likelihood of detecting Salmonella is reduced by
approximately 50% when the test is performed on the day of sampling and by 30% when delayed
by more than 3 days.
The significance of the slaughterhouse random intercept revealed that the baseline risk of
Salmonella carcass contamination varied between the slaughterhouses, even when other factors
such as lymph node infection, month of sampling and time before analysis were accounted for.
Moreover, the model indicated that the impact of Salmonella infection in pigs, as measured by
lymph nodes, on the carcass surface contamination varied between slaughterhouses (significant
44
slaughterhouse random slope). According to the results, the risk of getting a contaminated carcass
from a Salmonella infected pig differed between the slaughterhouses. In a similar way, the risk of
contaminating a carcass from a pig with Salmonella negative lymph nodes depended on the
slaughterhouse.
4.2.4. Multiple regression analysis at the MS level
The results of the analysis by country are displayed in Table 4. The different levels of significance
are indicated by different shades of grey, darker meaning more significant. Empty cells indicate
that the effect of the covariate was not sufficiently significant in that particular country to be
maintained in the final model. However, for some multi-level covariates not all categories were
available in all countries.
Variability between significant risk factors obtained for each country was observed and some
factors even had contrasting effects depending on the country. When these effects are studied at
the level of the 13 MS-group, these results may average out so that no significant effect is
observed in the 13 MS-group model.
The final models, respectively Austria, Cyprus, Czech Republic, Lithuania, and Poland did not
identify any factors of being significantly associated (at a level of 5% or less) with surface
contamination of carcasses.
The associations observed for each factor across the MSs are:
• Weight of carcasses (1-kg increments) – This factor was not significantly associated with the
outcome in any MS considered.
• Sampling time – A sine function was used to describe the effect of time of day upon the risk of
Salmonella infection. A significant association, whatever the direction, means that there is a
difference in the risk of infection between the pigs slaughtered at the end of the working day
compared to the beginning. This factor was not significantly associated with the outcome in
any MS.
• Sampling quarter – In three MSs there was a significant association between this factor and
detection of Salmonella contamination. In Belgium the risk was lower in October-December
2006, whereas it was lower in France in January-March 2007, compared to the summer quarter
(July-September) of 2007. Conversely, in Ireland, the risk in January-March 2007 and AprilJune 2007 was greater than that in the summer quarter of study (July-September 2007).
• Time between sampling and testing – In Denmark, the probability of detecting Salmonella in
carcass swabs is higher after a delay of 2 days compared to samples tested the day after that of
sampling.
• Salmonella detection in lymph node samples – In three MSs (France, Latvia, and the United
Kingdom), the infection of lymph nodes of the carcass was significantly associated with an
increased probability of observing Salmonella contamination on the surface of the carcass.
45
Table 4.
Random effect logistic models for factors associated with Salmonella surface contamination of carcasses of slaughter pigs
for the 13 participating MSs.
Austria
617
Belgium
381
Cyprus a
359
Czech Republic
417
3.1
0.50, 19
11
0.77, 162
13
0.93, 170
Denmark
344
0.14
<0.01, 999
18
2.6, 124
4.3
0.52, 36
3.3
0.64, 17
France
413
<0.01
<0.01, -
0.45
0.20, 0.97
0.43
0.20, 0.93
2.6
1.3, 5.1
Ireland
422
1.4
0.69, 2.9
0.33
0.12, 0.94
0.57
0.21, 1.5
Latvia
391
Lithuania
461
Poland
447
United Kingdom
641
a
0 day
vs.
1 day
2 days
vs.
1 day
3 to 7 days
vs.
1 day
Yes
vs.
No
Quarter
Oct.-Dec.06
vs.
Jul.-Sep.07
Jan.-Mar. 07
vs.
Jul.-Sep. 07
Apr.-Jun. 07
vs.
Jul.-Sep. 07
Carcass weight
(by 10-kg
increments)
No. of carcasses
Lymph
node
infection
Country
Sampling time
(sine)
Odds ratio estimates and 95%CI are presented for significant (at different levels of significance) risk factors obtained for each country separately. The colour of the cell
illustrates the degree of significance (P-value) of the association, according to the following scale:
0.01<P<0.05:
0.05<P<0.10:
0.10<P<0.25:
P<0.01:
0.06
0.001, 3.8
0.29
0.10, 0.81
1.4
0.66, 2.9
2.3
0.59, 8.9
0.93
0.85, 1.02
0.36
0.090, 1.4
0.29
0.14, 0.60
0.61
0.30, 1.2
2.0
0.72, 5.8
5.8
2.2, 15
3.6
1.4, 9.8
0.5
0.18, 1.3
5.6
1.2, 26
N/A
2.3
1.4, 3.8
Results are based on independent logistic regression.
0.84
0.37, 1.9
46
0.04
0.00, 1.32
4.3. Analysis of the serovars and phage type distribution
4.3.1. Spatial distribution of Salmonella serovars in lymph nodes
To investigate the spatial distribution of the most frequently reported serovars isolated from the
lymph nodes of slaughter pigs, the scan statistic was performed. Bulgaria was not included in the
analysis due to the lack of data on their actual production numbers. Table 5 shows the most likely
spatial clusters with their respective relative risk (RR) and level of significance (P-value).
Table 5.
Most likely clusters of Salmonella spp., S. Typhimurium, S. Derby,
S. Enteritidis, S. Infantis, and S. Rissen in MSs
a
Relative
Riska
Secondary
high risk MSs
Relative
Riska
PT, ES, IE, FR, UK, LU
2.6
GR
IT
1.8
1.2
S. Typhimurium
PT, ES, IE, FR, UK, LU, BE
2.5
-
-
S. Derby
PT, ES, IE, FR, UK, LU, BE, NL, IT
3.9
GR
1.3
S. Enteritidis
HU, SK, SI, CZ, PL
5.1
PT
3.0
S. Infantis
DK, DE
3.6
FR
2.1
S. Rissen
PT, ES
201.4
-
-
Serovar
Most Likely Cluster
Salmonella spp.
P-value = 0.001
Among slaughter pigs (ileo-caecal lymph node samples), spatial cluster analysis showed that the
most likely cluster for Salmonella included six countries (Portugal, Spain, Ireland, France, the
United Kingdom, Luxembourg). A significant RR of 2.6 suggested that slaughter pigs in these
countries are 2.6 times more likely to become infected than slaughter pigs outside those countries.
After detection of most likely clusters, the scan statistic also identified, for some serovars, other
single MSs (e.g. Greece and Italy for Salmonella) with a risk of Salmonella infection in slaughter
pigs significantly above the EU average but lower than the risk in the most likely clusters of MSs.
The most likely spatial cluster for S. Typhimurium included Portugal, Spain, Ireland, France, the
United Kingdom, Luxembourg, and Belgium. This cluster of countries presented a relatively high
RR for this serovar (RR = 2.5). A similar scenario was found for S. Derby, but this cluster
presented a larger radius and included also the Netherlands and Italy. S. Enteritidis clustered
spatially in Eastern Europe, whereas S. Infantis is clustered in Denmark and Germany. Finally,
Portugal and Spain, which were detected as the most likely cluster for S. Rissen, were also
included in the most likely clusters for three other of the proposed spatial scan analyses.
Maps of most likely and secondary clusters presented in Table 5 are displayed in Figure 27.
47
Figure 27.
Most likely clusters of Salmonella spp., S. Typhimurium, S. Derby,
S. Enteritidis, S. Infantis, and S. Rissen in slaughter pigs (ileo-caecal lymph nodes)
No data available from Bulgaria, Romania and Malta.
48
4.3.2. Differences in serovar distribution between the reporting countries
As reported in the Part A report, the diversity of isolated serovars in the ileo-caecal lymph node
samples differed considerably between MSs from two serovars in Sweden and Lithuania up to
more than 20 serovars in Germany, France, Greece and Spain. The diversity was particularly
pronounced in Greece, where only 345 pigs were sampled, whereas the sample sizes were three to
seven times larger in the three other MSs mentioned.
S. Typhimurium and S. Derby are widespread and dominant in slaughter pigs across most of the
MSs (Figure 28). However, S. Enteritidis had a relatively high prevalence in eight MSs.
Figure 28.
Relative frequency distribution of S. Typhimurium, S. Derby, S. Enteritidis,
S. Anatum, S. Infantis, S. Agona and other Salmonella serovars isolated in MSs from ileocaecal lymph nodes from slaughter pigs, 2006-2007.
100%
80%
60%
40%
20%
0%
AT BE BG CY CZ DE DK EE ES FR GR HU IE
S. Typhimurium
S. Derby
S. Enteritidis
S. Anatum
IT
LT LU LV NL PL PT SE
S. Infantis
S. Agona
SI SK UK NO
Other serovars
4.3.3. Comparison between serovar distributions in slaughter pigs, the animal
species, feed and human cases in the EU
Salmonella Enteritidis, the most frequent cause of human salmonellosis, was relatively rare in
slaughter pigs. Therefore, it is excluded from this visual analysis to allow an effective comparison
of frequencies of other serovars. S. Enteritidis in humans is broadly recognised to be principally
associated with the poultry food chains, particularly the consumption of table eggs and products
thereof (EFSA, 2007). For the same purpose, S. Rissen and S. 4,[5],12:i: were removed from the
pig data for comparison because these serovars highly clustered in Spain and Portugal (which both
accounted respectively for 98% and 89% of the lymph node isolates of these serovars) and were
not reported as associated with human cases of salmonellosis in Spain (data not available for
Portugal).
Disregarding the contribution of S. Enteritidis, S. Rissen and S. 4,[5],12:i:, Figure 29 and Figure
30 respectively compare relative Salmonella serovar distribution of the most frequent serovars
49
isolated from slaughter pigs in lymph node samples as well as carcass swabs with that of serovars
reported in human salmonellosis cases. There were only minor differences between the most
frequently isolated serovars in lymph nodes as compared to carcasses. Moreover, disregarding
S. Enteritidis, there appears to be some agreement between the most frequently reported serovars
in humans and those isolated in slaughter pigs. This is particularly the case for isolates from the
carcass surface. However, some discrepancies were also observed. For example, countries where
S. Derby was relatively dominant in carcasses did not demonstrate analogous proportions in
humans (e.g. Austria, Latvia). Conversely, some countries could report a significant serovar in
humans and did not observe analogous findings in carcasses (e.g. S. Infantis in Austria and the
United Kingdom or S. Derby in Cyprus).
Figure 29.
Comparison of the Salmonella serovar distribution in ileo-caecal lymph nodes
from slaughter pigs and humans (TESSy, 2006) in MSs and Norway.
Only the distribution of the most commonly isolated serovars in slaughter pigs is presented.
S. Enteritidis, S. Rissen, and S. 4,[5],12:i:- were excluded from this figure.
100%
Relative serovar distribution
80%
60%
40%
S. Typhimurium
S. Derby
Spain
S. Anatum
S. Infantis
The
Norw ay
Netherlands
S. London
United
Kingdom
S. Bredeney
50
Pigs
Pigs
Sw eden Slovenia Slovakia
Humans
Pigs
Humans
Pigs
Humans
Pigs
Humans
Pigs
Humans
Pigs
Hungary Luxemburg Latvia
Humans
Pigs
Humans
Pigs
Humans
Pigs
France
Humans
Humans
Pigs
Pigs
Estonia
Humans
Pigs
Czech Germany Denmark
Republic
Humans
Pigs
Humans
Pigs
Humans
Pigs
Cypres
Humans
Pigs
Austria
Humans
0%
Humans
20%
Figure 30.
Comparison of the Salmonella serovar distribution in carcass swabs from
slaughter pigs and humans (TESSy, 2006) in seven MSs.
Only the distribution of the most commonly isolated serovars in slaughter pigs is presented.
S. Enteritidis was excluded from this figure.
100%
Relative serovar distribution
80%
60%
40%
20%
0%
Humans
Pigs
Austria
Humans
Pigs
Cypres
S. Typhimurium
Humans
Pigs
Czech Republic
S. Derby
Humans
Pigs
Denmark
Humans
Pigs
France
S. Bredeney
Humans
Latvia
Pigs
Humans
Pigs
United Kingdom
S. Infantis
The comparison of the distribution of serovars isolated from feed and ileo-caecal lymph node
samples of slaughter pigs showed that many of the serovars detected in slaughter pigs are also
isolated from feed (Table 6). Serovars isolated from slaughter pigs have also been isolated from
broilers and laying hens (e.g. S. Infantis and S. Livingstone), and S. Derby, S. Bredeney and
S. London are shared with turkeys. In contrast, S. Rissen, S. Bovismorfibicans, S. Goldcoast, S.
Give, and S. Thompson were only isolated from slaughter pigs.
51
Table 6.
Number of isolations of Salmonella serovars from slaughter pigs (baseline
survey 2006-2007), pig feed (2006), humans (ECDC), broiler flocks (baseline survey 20052006), laying hen holdings (baseline survey 2004-2005), and turkey flocks (baseline survey
2006-2007).
Salmonella
serovars
S. Typhimurium
S. Derby
S. Rissen
S. Enteritidis
S. Anatum
S. Bredeney
S. Infantis
S. London
S. Brandenburg
S. Agona
S. Newport
S. Montevideo
S. Bovismorbificans
S. Goldcoast
S. Give
S. Livingstone
S. Thompson
S. Hadar
Slaughter
pigs
Humans
1,040
380
151
126
63
51
49
33
31
28
24
19
15
14
11
9
9
8
19,009
484
105
91,325
159
160
1,261
88
243
388
751
359
304
143
191
86
196
726
Feed (pigs, oil
seed and fruit)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
-
Broilers
flocks
Laying
hens
holdings
Turkeys
flocks
65
13
538
32
10
295
16
8
31
39
59
123
14
899
21
26
171
38
11
27
50
53
86
123
55
186
72
31
31
33
13
152
4.3.4. Phage type distributions
4.3.4.1 S. Enteritidis phage types
Data on S. Enteritidis phage types were only provided from the ileo-caecal lymph node isolates
from slaughter pigs by four MSs (Austria, Belgium, Hungary and the United Kingdom). Fifteen
MSs with S. Enteritidis isolates did not report phage typing information. Only one out of three
MSs (Austria) phage typed a S. Enteritidis isolate from the carcass swabs.
The four MSs providing information on S. Enteritidis phage types from lymph nodes reported a
total of 22 isolates out of which 17 (77%) were phage typed. This represented 14% of all 126
reported S. Enteritidis isolates from ileo-caecal lymph nodes of slaughter pigs in the EU baseline
survey. Reported phage types from lymph node and carcass swab samples are presented in
Table 7. In this table the ranking is based on the number of specific S. Enteritidis phage typepositive slaughter pigs in the four MSs.
52
Table 7.
Distribution of the S. Enteritidis phage types in slaughter pigs in four
reporting MSs, EU slaughter pigs baseline survey, 2006-2007.
Ileo-caecal lymph nodes
Phage type
PT 8
PT 4
PT 1
PT 4b
PT 5a
PT 6a
PT 9a
PT 13
PT 13a
PT 20
PT 21
Carcass-swabs
No. of strains
MSs reporting
phage types
No. of strains
5
3
1
1
1
1
1
1
1
1
1
AT, HU
AT, HU
AT
UK
AT
AT
BE
HU
UK
BE
BE
1
-
MS reporting
phage types
-
AT
-
4.3.4.2 S. Typhimurium phage types
Data on S. Typhimurium phage types was provided from ileo-caecal lymph node isolates by five
MSs (Belgium, Hungary, Sweden, the Netherlands and the United Kingdom), whereas the
remaining 19 MSs with S. Typhimurium isolates did not provide any phage typing information.
The MSs that reported information regarding S. Typhimurium phage types from lymph nodes
reported 217 isolates out of which 207 (95%) were phage typed. This represented 20% of the all
1,040 reported S. Typhimurium isolates in the EU baseline survey. The reported phage types in
Belgium, Hungary, Sweden, and the United Kingdom are presented in Table 8. The ranking is
based on the percentages of S. Typhimurium phage type-positive ileo-caecal lymph node isolates
in those four MSs. The results reported by the Netherlands are presented separately in Table 9,
because of a different phage typing system used in this country.
Phage type information on carcass swab isolates was reported from two (Belgium and the United
Kingdom) of the 13 MSs having undertaken the carcass swab study. A total of 80 out of 93
isolates (86%) in the two MSs was phage typed. In total, 191 isolates were reported from the ten
MSs isolating S. Typhimurium from the carcass swabs.
A total of 23 phage types (excluding RDNC) were recorded in lymph node samples and 13 in
carcass swabs, indicating a large diversity of S. Typhimurium phage types in pigs. There was
some overlap of phage types in lymph nodes and carcass swabs, but 10 phage types were only
isolated from lymph node samples, whereas four phage types were only found in carcass swabs.
Phage type U288 was the most frequently reported one but reported only in the United Kingdom.
DT 193 was the second most frequently recorded phage type and found in Belgium, Hungary, and
the United Kingdom in both lymph nodes and carcass swabs. FT 506 and 507 accounted for more
than half the S. Typhimurium phage types in the Netherlands. However, one third of the isolates
were non typeable.
53
Table 8.
Distribution of the S. Typhimurium phage types in slaughter pigs in four
reporting MSs, EU slaughter pigs baseline survey, 2006-2007.
Phage type
U 288
DT 193
U 302
DT 104
DT 104b
DT 120
DT 12
DT 135
DT 40
DT 56
DT 193w
DT 208
DT 2
DT 41
DT 56a
DT 85
DT 194
U 277
U 310
DT 110
DT 185
DT 35
DT 170
RDNC a
Non typeable
a
Carcass-swabs
No. of strains
MSs reporting
phage types
No. of strains
MSs reporting
phage types
28
26
19
16
8
7
4
3
2
2
2
2
1
1
1
1
1
1
1
17
11
UK
BE, HU, UK
BE, HU, UK
BE, HU, UK
UK
BE, HU
BE
HU
SE
UK
BE
UK
BE
SE
UK
UK
UK
SE
BE
BE, HU, UK
BE, HU, UK
16
13
2
4
3
13
2
1
4
2
2
1
1
5
11
UK
BE, UK
UK
BE, UK
UK
BE, UK
BE
BE
BE, UK
BE
BE
BE
UK
BE, UK
BE, UK
RDNC: ‘Reacts but Does Not Conform’.
Table 9.
Distribution of the S. Typhimurium phage types in slaughter pigs in the
Netherlands, 2006-2007.
Phage type
FT 506
FT 507
FT 510
FT 508
FT 2
FT 110
FT 292
FT 301
FT 401
Non typeable
No. of strains
13
12
3
2
1
1
1
1
1
18
54
4.3.5. Comparison between phage type distribution in slaughter pigs and
humans
In order to further investigate the role of pig meat as a source of human S. Enteritidis and
S. Typhimurium infection, the phage typing results from the slaughter pig baseline survey and
isolates from humans (Community Summary Report, 2006) were compared (Table 10, 11 and 12).
Phage type distributions in humans were only available from a fraction of the MSs and, as
reported earlier, only a minor proportion of the MSs reported the phage types of the isolates found
in the baseline surveys. Interpretation should consequently be carried out cautiously due to limited
numbers and lack of representativeness. Some phage types were shared by the human cases and
pigs in the reporting MSs.
Table 10.
Comparison of S. Enteritidis phage types isolated from humans and slaughter
pigs (ileo-caecal lymph nodes and carcass swabs).
No. of slaughter pigs as
reported in the EU baseline
survey, 2006-2007
No. of human S. Enteritidis phage types
reported in 2006 a
Phage type
PT 4
PT 8
PT 1
PT 21
PT 6
PT 14b
PT 6a
PT 1b
PT 13a
RDNC
PT 13
PT 56
PT 11
PT 3
PT 1c
PT 4b
PT 2
PT 23
PT 7
PT U
PT 19
PT 6c
Not typeable
Other
a
AT
CZ
HU
NL
PT
UK
Total
AT
BE
HU
UK
1,125
964
212
884
371
67
201
3
30
91
1
3
38
56
5
11
10
33
32
27
79
3
90
4
2
1
1
13
83
6
4
6
398
642
22
174
246
20
85
113
89
44
22
32
20
24
28
59
315
41
47
55
69
9
17
8
10
2
1
2
2
15
23
296
28
23
47
2,069
1,088
1,492
609
246
538
218
12
117
46
1
93
78
14
2
4
2
20
1,089
3,910
2,825
1,777
1,724
933
657
436
397
273
226
129
93
89
62
58
57
46
36
35
32
27
24
71
1,295
2
1
1
1
1
1
2
2
4
1
-
1
1
-
Source: European Centre for Disease prevention and Control (ECDC).
55
Table 11.
Comparison of S. Typhimurium phage types isolated from humans and
slaughter pigs (ileo-caecal lymph nodes and carcass swabs).
No. of slaughter pigs as reported
in the EU baseline survey, 2006-2007
No. of human S. Typhimurium phage
types reported in 2006 a
a
Carcass swabs
Phage type
AT
CZ
HU
UK
Total
BE
HU
SE
UK
BE
UK
DT 104
DT 46
DT 193
DT 104l
RDNC
DT 104b
DT 120
DT 8
DT 1
DT 41
U 302
DT 56
DT 135
U 311
U 288
DT U
DT 208
DT 35
DT 194
DT 125
Not typeable
Other
267
14
79
92
33
4
18
68
18
34
63
8
22
3
2
5
14
31
62
103
24
64
45
14
14
13
33
60
370
108
46
72
73
93
46
9
10
50
44
38
37
1
13
725
433
267
184
182
162
136
114
97
86
80
55
50
46
38
37
23
15
14
14
13
46
850
3
5
12
4
2
3
8
8
1
3
3
8
3
1
-
1
3
5
20
2
8
9
2
28
3
2
12
1
1
3
11
1
11
3
1
2
16
3
8
2
2
1
7
2
Source: European Centre for Disease prevention and Control (ECDC).
56
Table 12.
Comparison of S. Typhimurium phage types isolated from humans and
slaughter pigs (ileo-caecal lymph nodes and carcass swabs) in the Netherlands.
Phage type
No. of human S. Typhimurium
phage types reported from
NL in 2006
FT 561
FT 507
FT 506
FT 510
FT 296
FT 401
FT 508
FT 60
FT 3
FT 61
FT 80
FT 2
other
185
116
79
27
21
8
7
7
6
4
4
2
39
No. of slaughter pigs as reported
from NL in the EU baseline survey,
2006-2007
12
13
3
2
1
4
57
5. Discussion
Salmonella infections in pigs are often sub-clinical, although some animals may show clinical
signs varying from mild diarrhoea to acute septicaemia and death. There is convincing evidence
that some human cases of salmonellosis may derive from Salmonella infection of pigs or
contaminated pig products although the population attributable fraction has not been estimated
(EFSA, 2006b). The main motivation to control Salmonella infection in pigs is to protect public
health.
5.1. Analysis of factors associated with Salmonella infection in lymph
nodes or surface contamination of carcasses
This EU-wide baseline survey estimated the prevalence of fattening pigs infected with Salmonella
at slaughter within the 27 Member States (MSs) and these estimates were published in the Part A
report. In addition, 13 MSs sampled carcass surfaces and the prevalence of Salmonella positive
carcass swabs was also reported in Part A. During the conduct of the survey, some compulsory
complementary data was recorded. This Part B report considers whether any of these factors were
associated with the risk of isolation of Salmonella from either the ileo-caecal lymph nodes or from
the surface of carcasses. It should be noted that many potential factors of relevance to Salmonella
infection in slaughter pigs or surface contamination of carcasses such as - among others - hygiene
during slaughter and subsequent processing, slaughter techniques, the speed of the slaughter line
and the cleaning and disinfection procedures used, were not part of the present survey. MSs could
also report other optional information on a voluntary basis, but these data were too scarce to
enable an epidemiological analysis within the scope of this Part B report.
In the EU level multiple regression analyses, the structure of the statistical models took into
account the fact that slaughter pigs originating from the same country have a higher probability of
sharing similar domestic conditions, that pigs slaughtered in the same slaughterhouse were more
likely to have comparable rearing, transport and lairage conditions, and that carcasses from the
same slaughterhouse were submitted to similar processes. Furthermore, possible country
confounding effects were also taken account of in the analyses.
The analyses performed at MS level in this survey should be regarded as a preliminary attempt to
investigate effects of mandatory reported factors in each MS. Moreover, it allowed the assessment
of the variability of those effects between MSs. The results of these analyses showed that the
importance of the identified factors varied importantly between MSs, as indeed the exposure to
some factors was protective in some MSs but increased the risk in others. Consequently, these risk
factor analyses results at MS level should be regarded as indicative and need to be complemented
by specific studies carried out at national level, taking into account domestic conditions.
58
5.1.1. Effect of sampling and testing procedures
The multiple regression analyses carried out at the EU level demonstrate that variations in the
implementation of sampling and testing procedures had an impact on the probability of detecting
Salmonella in the samples.
The probability of detecting Salmonella from a lymph node sample was highest when the delay
between sample collection and laboratory testing was 3-4 days and then decreased once again. A
similar increase in the detection rate after 1-2 days from sampling was observed in carcass swab
samples. The significant impact of delayed testing on the likelihood of detecting the Salmonella
has previously been observed in stools of human origin (Poisson et al., 1993).
There might be several biological explanations to this phenomenon, such as a short-term die-off of
competing bacteria in the sample to such an extent that Salmonella identification is less likely to
fail due to the overgrowth of other bacteria in the two to three days following sampling.
Subsequently, the decline in numbers of Salmonella bacteria, which occurs from the first day,
becomes the major factor. It might also be that stress induced by a changed environment due to
sampling and subsequent refrigeration of the sample prevented the initial growth of Salmonella as
the bacterium might have needed to adapt to new conditions.
An increased probability in detecting Salmonella from heavier lymph node samples was observed
in the survey. This may simply be due to a greater chance of including any Salmonella or
harvesting a larger number of Salmonella bacteria in a heavier sample. A similar effect of the
weight of faecal samples on the probability of Salmonella detection has been demonstrated
previously (Funk et al., 2000; Champagne et al., 2005). However, it might also be that the larger
weight of the lymph nodes was in some cases due to an inflammatory reaction to Salmonella
infection, which would also increase the probability of detecting the bacterium.
These results demonstrate that the applied sampling and testing procedures had an impact on the
detection of Salmonella. It is possible that the slaughter pig Salmonella prevalence estimates
presented in the Part A report for MSs might have been partly different if all MSs had applied
exactly the same sampling and testing procedures. The results imply that the standardisation of
sampling instructions and of testing procedures is important in the forthcoming national
Salmonella control programmes in pigs. This standardisation would enhance data comparability,
at least within the country and also at the EU level.
5.1.2. Effect of lymph node Salmonella infection on surface contamination of
carcasses
In this survey, Salmonella surface contamination of the carcass was affected by the Salmonella
infection status of the pig as reflected by the lymph nodes. A Salmonella infected pig was twice
more likely to yield a Salmonella contaminated carcass. This can be regarded as an expected
finding, as slaughter pig infection and carcass contamination often arise from the same sources,
such as intestinal carriage of the pig or contamination from the slaughterhouse lairage
environment. Other studies have reported similar associations between the Salmonella infection
status of the pig before slaughtering and the surface contamination of the carcass (Hald et al.,
2003; Sorensen et al., 2004; Mc Dowell, 2007).
59
Lymph node infection can be considered as a marker of the asymptomatic intestinal carriage of
Salmonella. The infection can develop on the farm, during transport, or in slaughterhouse lairage.
This intestinal Salmonella carriage or infection may result in carcass contamination during the
slaughter process in the case of faecal leakage from the intestinal tract. Salmonella can also be
present on pig skin before slaughtering and may subsequently be recovered on the carcass. The
presence of Salmonella on pig skin, may on its behalf originate from intestinal infection of the pig
or from a contaminated environment on the farm, in the transport vehicle or in the slaughterhouse
lairage.
The analyses results indicate that processing slaughter pigs that are not infected with Salmonella
reduces the risk of subsequent carcass surface contamination. Therefore, controlling the
Salmonella prevalence in pigs during primary production (i.e. from farm to slaughtering) would
have a beneficial impact on Salmonella contamination of carcasses and pig meat. These controls
are also likely to reduce the overall Salmonella contamination of slaughterhouse environment,
since incoming pigs are the primary source of Salmonella ingress to slaughterhouses.
The survey results also underline the role of the slaughterhouse environment in Salmonella carcass
contamination. Even though a pig infected in the lymph nodes was more likely to yield a
contaminated carcass, there were many contaminated carcasses deriving from pigs with lymph
nodes tested negative. Some of these may be due to limited testing sensitivity to detect all
Salmonella positive pigs but others may result from cross-contamination from other carcasses or
through contact with contaminated surfaces or equipment within the slaughterhouses.
Consequently, good slaughter hygiene is also vital in the prevention of carcass Salmonella
contamination.
5.1.3. Effect of the slaughterhouse on the risk of carcass contamination
The effect of the slaughterhouse on carcass contamination was also considered in the analyses.
The results showed that the baseline risk of Salmonella carcass contamination varied considerably
between slaughterhouses, even when other factors such as lymph node infection, month of
sampling and delay in testing were taken into account in the statistical model. This slaughterhouse
effect can be interpreted as the combined results of all the other factors affecting the likelihood of
the carcass becoming contaminated with Salmonella within the individual slaughterhouse. Such
factors are likely to include - among other things – the Salmonella contamination of
slaughterhouse environment, hygiene during slaughter and subsequent processing, slaughter
techniques, the speed of the slaughter line and the cleaning and disinfection procedures used.
Furthermore, the analysis showed that, depending on the slaughterhouse, Salmonella infection in
pigs arriving on the slaughter line has either a stronger or weaker impact on carcass
contamination. In some slaughterhouses carcasses were more likely to become Salmonella
contaminated than in others, both when processing infected or non-infected slaughter pigs.
Apparently certain slaughterhouses were more capable of controlling and preventing Salmonella
contamination risk in the slaughter process. This implies that while slaughterhouse and the
processing steps offer a further opportunity for Salmonella risk mitigation in pig, they may also
contribute to increase the risk, notably in case of poor hygienic performances.
60
The survey did not collect information on slaughterhouse characteristics that could have
contributed to the contamination of carcasses. However, it may be in the interest of MSs to
investigate further these specific slaughterhouse factors in order to improve the control of
Salmonella and the protection of public health in their country.
5.1.4. Effect of the time of sampling on Salmonella results
Throughout the survey an association was found between Salmonella contamination of pig
carcasses and the time of sampling. Carcasses were less at risk of being contaminated during the
first months of the survey, October 2006 to March 2007, compared to the rest of the survey period,
from April to September 2007. This effect could be due to the season, but this hypothesis should
be studied further and confirmed by additional pluriannual studies conducted in MSs, particularly
since seasonal climatic conditions differ between MSs across the EU. The effect of the sampling
months on the risk of surface contamination of carcasses was previously described in the
literature. Mc Dowell et al. (2007) reported, in Northern Ireland, that the highest odds of carcass
contamination occurred in the quarter April to June and the lowest in October to December. A
study carried out in five slaughterhouses from three European countries also showed that the risk
of carcass contamination was significantly higher in the summer compared to the autumn months
(Hald et al., 2003).
In the analyses, there was no evidence that time of sampling during the day was associated with
the risk of Salmonella infection of lymph nodes or carcass contamination.
5.2. Analysis of serovar and phage type distribution
5.2.1. Spatial distribution of Salmonella serovars in lymph nodes
Spatial distribution analysis identified likely clusters of MSs, representing geographical areas
where infection with a particular serovar was significantly higher than in the general EU slaughter
pig population. The only geographical information available for the analysis was the MS and
therefore the smallest geographical unit for inclusion or exclusion in a cluster were individual
MSs. This analysis implied that all Salmonella isolations occurred at a central point within the
MS. This is obviously a gross simplification and the analysis thus investigated the adjacency of
MSs in which positive pigs occurred, rather than the true geographical relatedness of positive pigs.
Nevertheless, the outcomes were consistent with the visual appraisal of MS-specific prevalence
and show that various serovars isolated from slaughter pigs are not evenly distributed across the
EU. S. Typhimurium and S. Derby are clustered in western MSs, whereas S. Enteritidis is
clustered in eastern MSs. S. Infantis appeared to cluster in north-eastern MSs, while S. Rissen
clustered in the Iberian peninsula.
The clustering of Salmonella serovars in specific geographic areas may mirror common sources or
reservoirs of infection such as endemic wildlife species, specific raw feed ingredients, or indeed,
infected breeding herds of pigs. Geographic clustering is also consistent with the potential for the
clonal spreading of a particular Salmonella serovar among holdings following the introduction to a
region, e.g. through the movement of infected animals, or through feed or animal transport
vehicles. Clustering may also reflect a selection pressure for a specific serovar in a region.
61
The present survey design did not collect data on various relevant factors that could explain more
in-depth the identified clusters, such as feed ingredients, production managerial procedures, farm
characteristics and pig movements, as well as the Salmonella status of the suppliers of piglets or
breeding animals; nor holding bio-security such as access to wild or other domestic livestock.
However, one hypothesis may be that the differences in pig farming structure could partly explain
the observed differences, with larger and more industrialised productions in western MSs and
more extensive and mixed productions in eastern MSs. MSs where particular serovars are
prevalent should attempt to identify specific risk and/or protective factors enabling appropriate
control measures in their country.
5.2.2. Comparison of serovar and phage type distribution in slaughter pigs,
feed and human salmonellosis cases
There were substantial variations among MSs in Salmonella serovars detected from pigs in this
survey, as stated in Part A report. This serovar distribution was analysed and compared to those in
human salmonellosis cases, in other food production animal species, and in feed. The analyses of
S. Enteritidis and S. Typhimurium phage types was carried out as well, but it proved less useful
because the phage type data was only available from few MSs.
In this survey, S. Typhimurium dominated the serovars isolated from pigs at EU level, and this
serovar is also clearly the second most often reported serovar from human cases, following
S. Enteritidis. According to analyses, in most MSs, where S. Typhimurium presented an important
proportion of the serovars found in slaughter pigs, the serovar was also the dominant cause of
human non-Enteritidis Salmonella infections. This supports the notion that pig meat may
contribute to the human S. Typhimurium infection in EU.
There also appeared to be some agreement between the human and pig proportion of S. Derby and
S. Infantis serovars at MS level, even though some discrepancies were observed. This can be
expected as these human cases of these serovars are likely to represent infections from many
different sources and food chains. Generally, many serovars isolated from slaughter pigs in this
survey (such as S. Enteritidis, S. Infantis and S. Livingstone) have also been isolated from broilers
and laying hens, while S. Derby, S. Bredeney and S. London were shared with turkeys.
Whereas there is general acceptance of a substantial contribution of pig meat to Salmonella cases
in humans, particularly regarding S. Typhimurium infection (Berends et al., 1998; Hald et al.,
2004; EFSA, 2006), the true attribution of risk arising from pig meat remains unknown at EU
level. While it is known that also other food producing animal species (e.g. poultry and cattle) and
the food thereof are sources of S. Typhimurium and other Salmonella serovar infections in
humans, a more in-depth source attribution analysis is needed to examine the relative contribution
of the animal species. The on-going Quantitative Risk Assessment on Salmonella in pigs that is
carried out by EFSA’s Scientific Panel on Biological Hazard may contribute importantly in this
aspect. In addition, a thorough phage typing and a molecular typing of all Salmonella isolates
from humans, food and food producing animals would facilitate a better understanding of
attribution of risk to specific food chains.
When considering the sources of Salmonella in pigs, it was interesting to note, that those MSs,
which had a higher prevalence of S. Enteritidis in slaughter pigs in the survey, had reported also a
relatively high S. Enteritidis prevalence in laying hen holdings and/or in broilers flocks in the
62
previous baseline surveys. This might be indicative of either a common source of this serovar or
its circulation between these food animal sectors. Also, a number of serovars detected in the
slaughter pigs in the survey have been also isolated from feed. Feed is a plausible and wellrecognised source of introduction of Salmonella into pig herds, particularly in the case of new
serotypes that may be able to establish themselves in pig production
63
6. Conclusions
•
In the survey, a positive association between the frequency of slaughter pigs infected with
Salmonella in their lymph nodes and the frequency of Salmonella surface contamination of
pig carcasses was observed. A Salmonella infected pig was twice more likely to yield a
Salmonella contaminated carcass. However, contaminated carcasses could also derive from
uninfected pigs.
•
The risk of pig carcasses becoming contaminated with Salmonella also varied significantly
between slaughterhouses even when other associated factors, such as the frequency of
infected slaughter pigs, were accounted for. In some slaughterhouses the risks of producing
a contaminated carcass from a Salmonella infected pig or from a non-infected pig were
higher. This indicates that certain slaughterhouses were more capable of controlling and
preventing Salmonella contamination than others.
•
At EU level, pig carcasses were most likely to become contaminated with Salmonella in
the second half of the survey period, from April to September 2007. However, this possible
seasonal effect should be verified in further studies in individual MSs.
•
There was substantial variation between MSs in the factors found associated with
Salmonella infection of slaughter pigs and carcass contamination. Also the level of
importance of these factors varied, and while sometimes the same factor could be
protective in some MSs, it could increase the risk in others.
•
A number of factors such as those related to rearing and processing were not investigated
in the survey. Therefore, it was not possible to estimate the association of these factors
with Salmonella infection of pigs or contamination of carcasses and their potential
confounding role on the effect of factors on which data were available. However, results of
this analysis are useful starting points for more specifically aimed studies in the EU and in
individual MSs.
•
The manner in which sampling and testing procedures were applied in the survey affected
the likelihood of detecting Salmonella from the lymph node samples and the carcass
surface samples. The likelihood of detection was highest when there was some delay
between sampling and the start of laboratory testing. In addition, the probability of finding
Salmonella from the lymph nodes increased with the weight of the samples.
•
The analyses of the Salmonella serovar distribution revealed some agreement between the
most frequently reported serovars in human salmonellosis and those isolated in slaughter
pigs. This supports the notion that pigs and pig meat contribute to Salmonella infection in
humans, even though it is acknowledged that other food animal species and food thereof
also play a role as a source of these infections in humans.
•
Analysis of serovar distributions indicated the clustering of specific serovars in pig
production chains of geographic regions within the EU. This clustering may indicate
common sources of these serovars among the MSs in question.
64
7. Recommendations
•
MSs are invited to consider the factors found to be significantly associated with Salmonella
infection in slaughter pigs and contamination on carcasses at EU level in this survey, when
designing national Salmonella control programmes for slaughter pigs. The Salmonella
infection status of the pig (reflected by the lymph node infection) and the slaughterhouse
process were both shown to have an impact on the risk of carcass contamination. An integrated
control programme that addresses both the primary production and the slaughter process may
prove to be a feasible and cost-effective option.
•
MSs are specifically encouraged to guarantee Salmonella controls in primary production as in
the slaughterhouses in order to prevent subsequent contamination of the carcass surface and to
improve protection at public health.
•
The EU pig meat industry is invited on its part to pay increased attention to slaughter hygiene
and other factors in slaughterhouses that may affect Salmonella contamination of pig
carcasses.
•
It is recommended that MSs carry out further national studies to identify more closely the
factors that put slaughter pigs and carcasses at risk of becoming infected or contaminated with
Salmonella in their country, taking into account their national Salmonella prevalence, serovar
distribution and the characteristics of their slaughterhouses.
•
The harmonisation of sampling and testing procedures should be considered of importance by
MSs when designing national Salmonella control programmes, as well as by EU legislation
when defining the target for the reduction of Salmonella prevalence in slaughter pigs.
•
Since the probability of isolating Salmonella from a lymph node sample or a carcass swab
varied according to the delay between sample collection and laboratory testing, MSs are
invited to carry out studies on the survival rates of Salmonella in different relevant matrices.
65
Task Force on Zoonoses Data Collection members
Andrea Ammon, Alenka Babusek, Marta Bedriova, Karin Camilleri, Georgi Chobanov, Mariann
Chriel, Inrid Dan, Jürg Danuser, Kris De Smet, Marie Edan, Matthias Hartung, Birgitte Helwigh,
Merete Hofshagen, Sarolta Idei, Patrícia Inácio, Elina Lahti, Lesley Larkin, Peter Much, Edit
Nagy, Lisa O’Connor, Rob A.A. van Oosterom, Jacek Osek, José Luis Paramio Lucas, Antonio
Petrini, Christodoulos Pipis, Saara Raulo, Tatjana Ribakova, Antonia Ricci, Petr Šatrán, Snieguole
Sceponaviciene, Joseph Schon, Jelena Sõgel, Ana María Troncoso González, Kilian Unger,
François Veillet, Luc Vanholme, and Dimitris Vourvidis.
Acknowledgements
The Task Force on Zoonoses Data Collection wishes to acknowledge the contribution of the
Working Group that prepared this report: Thomas Blaha, Kristen Barfod, Alex Cook, Pedro Rubio
Nistal, Micheál O’Mahony, Arjen W. van de Giessen, Kris De Smet, Billy Amzal, Didier Verloo,
Pierre-Alexandre Belœil and Frank Boelaert. The Task Force on Zoonoses Data Collection also
wishes to acknowledge the contribution of statistical analysis by the personnel of Hasselt
University, Center for Statistics: Marc Aerts, José Cortiñas, Christel Faes, Saskia Litière and Kaatje
Bollaerts; the Danish Zoonosis Centre: Mariann Chriel and Tine Hald; Local Health Unit Reggio
Emilia: Stefano Guazzetti.
The implementation of the baseline survey by the Competent Authorities of the MSs and Norway
is gratefully acknowledged.
66
Abbreviations
CI
CRL
ECDC
EFSA
EU
MS(s)
NRL
OR
RDNC
RR
TESSy
VIF
Confidence Interval
Community Reference Laboratory
European Centre for Disease Prevention and Control
European Food Safety Authority
European Union
Member State(s)
National Reference Laboratory
Odds Ratio
‘Reacts but Does Not Conform’
Relative Risk
European Surveillance System
Variance Inflation Factor
67
List of Tables
Table 1.
The final random effect logistic model for factors associated with Salmonella
infection in lymph nodes of slaughter pigs, in the EU, 2006-2007. ...........................28
Table 2.
Random effect logistic models for factors associated with Salmonella infection in
slaughter pigs in participating countries .....................................................................30
Table 3.
The final random effect logistic model for factors associated with the Salmonella
surface contamination of carcasses of slaughter pigs, in a group of 13 MSs, 20062007.............................................................................................................................44
Table 4.
Random effect logistic models for factors associated with Salmonella surface
contamination of carcasses of slaughter pigs for the 13 participating MSs................46
Table 5.
Most likely clusters of Salmonella spp., S. Typhimurium, S. Derby, S. Enteritidis,
S. Infantis, and S. Rissen in MSs ................................................................................47
Table 6.
Number of isolations of Salmonella serovars from slaughter pigs (baseline survey
2006-2007), pig feed (2006), humans (ECDC), broiler flocks (baseline survey 20052006), laying hen holdings (baseline survey 2004-2005), and turkey flocks (baseline
survey 2006-2007). .....................................................................................................52
Table 7.
Distribution of the S. Enteritidis phage types in slaughter pigs in four reporting MSs,
EU slaughter pigs baseline survey, 2006-2007...........................................................53
Table 8.
Distribution of the S. Typhimurium phage types in slaughter pigs in four reporting
MSs, EU slaughter pigs baseline survey, 2006-2007. ................................................54
Table 9.
Distribution of the S. Typhimurium phage types in slaughter pigs in the Netherlands,
2006-2007. ..................................................................................................................54
Table 10.
Comparison of S. Enteritidis phage types isolated from humans and slaughter pigs
(ileo-caecal lymph nodes and carcass swabs).............................................................55
Table 11.
Comparison of S. Typhimurium phage types isolated from humans and slaughter pigs
(ileo-caecal lymph nodes and carcass swabs).............................................................56
Table 12.
Comparison of S. Typhimurium phage types isolated from humans and slaughter pigs
(ileo-caecal lymph nodes and carcass swabs) in the Netherlands...............................57
68
List of Figures
Figure 1.
Bar plot of the weight of the lymph node samples tested, by country and for the EU, and
by Salmonella status ....................................................................................................... 16
Figure 2.
Box plot of the number of lymph nodes (LN) per sample, per country.......................... 17
Figure 3.
Box plot of the number of lymph nodes per sample by Salmonella status of sample .... 18
Figure 4.
Frequency distribution of the weighted number of days between sampling and testing of
lymph node samples, by country and for the EU, and by Salmonella status.................. 19
Figure 5.
Weighted Salmonella prevalence by number of days between sampling and testing, with
95% confidence intervals, in the EU............................................................................... 20
Figure 6.
Bar plot of the weighted number of lymph node samples collected by month and
country, and for the EU, and by Salmonella status......................................................... 21
Figure 7.
Weighted Salmonella prevalence by the month of sampling, with 95% confidence
intervals, in the EU ......................................................................................................... 22
Figure 8.
Bar plot of the weighted number of tested lymph node samples, by sampling quarter and
country, and for the EU, and by Salmonella status......................................................... 23
Figure 9.
Weighted Salmonella lymph node prevalence by sampling quarter, with 95% confidence
intervals, in the EU ......................................................................................................... 24
Figure 10.
Bar plot of the weighted number of lymph node samples collected, by hour of sampling
and country, and for the EU, by Salmonella status......................................................... 25
Figure 11.
Weighted Salmonella prevalence by hour of sampling, with 95% confidence intervals,
in the EU ......................................................................................................................... 26
Figure 12.
Box plot of carcass weights for slaughter pigs sampled for lymph nodes, per country . 27
Figure 13.
Box plot of carcass weight for slaughter pigs sampled for lymph nodes, by Salmonella
status ............................................................................................................................... 27
Figure 14.
Frequency distribution of the time (days) between sampling and testing of carcass
swabs, by MS and for the 13-MS group, and by Salmonella status ............................... 34
Figure 15.
Weighted Salmonella prevalence of carcass contamination by number of days between
sampling and testing, with 95% confidence intervals, in the 13-MS group ................... 35
Figure 16.
Weighted prevalence of Salmonella contaminated carcasses by Salmonella infection
status of the slaughter pig, with 95% confidence intervals, in the 13-MS group a ......... 35
Figure 17.
Bland-Altman graph of the proportion of Salmonella positive lymph nodes against the
differences between proportions of positive carcasses and lymph nodes per
slaughterhouse................................................................................................................. 36
69
Figure 18.
Box plot of the proportion of Salmonella contaminated carcasses by the proportion of
Salmonella positive pigs in 146 slaughterhouses............................................................ 37
Figure 19.
Bar plot of the weighted number of carcass swabs collected by month and MS, and for
the 13-MS group, by Salmonella status .......................................................................... 38
Figure 20.
Weighted Salmonella prevalence of carcass contamination by month of sampling, with
95% confidence interval, in the 13-MS group ................................................................ 38
Figure 21.
Bar plot of the weighted number of carcass swabs collected by quarter and MS, and for
the 13-MS group, and by Salmonella status ................................................................... 39
Figure 22.
Weighted Salmonella prevalence of carcass contamination by sampling quarter, with
95% confidence intervals, in the 13-MS group .............................................................. 40
Figure 23.
Bar plot of the weighted number of carcass swabs collected by hour of sampling and
MS, and for the 13-MS group, and by Salmonella status ............................................... 41
Figure 24.
Weighted Salmonella prevalence carcass contamination by hour of sampling, with 95%
confidence interval, in the 13 MS-group ........................................................................ 42
Figure 25.
Box plot of the weight of the carcasses sampled with carcass swabs per MS................ 42
Figure 26.
Box plot of the weight of carcasses sampled with carcass swabs, by Salmonella status,
in the 13-MS group ......................................................................................................... 43
Figure 27.
Most likely clusters of Salmonella spp., S. Typhimurium, S. Derby, S. Enteritidis, S.
Infantis, and S. Rissen in slaughter pigs (ileo-caecal lymph nodes)............................... 48
Figure 28.
Relative frequency distribution of S. Typhimurium, S. Derby, S. Enteritidis, S. Anatum,
S. Infantis, S. Agona and other Salmonella serovars isolated in MSs from ileo-caecal
lymph nodes from slaughter pigs, 2006-2007................................................................. 49
Figure 29.
Comparison of the Salmonella serovar distribution in ileo-caecal lymph nodes from
slaughter pigs and humans (TESSy, 2006) in MSs and Norway.................................... 50
Figure 30.
Comparison of the Salmonella serovar distribution in carcass swabs from slaughter pigs
and humans (TESSy, 2006) in seven MSs...................................................................... 51
70
References
Berends B.R., Van Knapen F., Mossel D.A., Burt S.A., Snijders J.M. 1998. Impact on human health
of Salmonella on pork in The Netherlands and the anticipated effects of some currently proposed
control strategies. Int. J. Food Microbiol., 44, 219-29.
Champagne M.J., Ravel A., Daignault D. 2005. A comparison of sample weight and culture methods
for the detection of Salmonella in pig feces. J. Food Prot., 68, 1073-6.
EC (European Commission). 2003. Regulation No 2160/2003 of the European Parliament and of the
Council of 17 November 2003 on the control of Salmonella and other specified food-borne zoonotic
agents. OJ L 325, 12.12.2003, p. 1.
<http://eur-lex.europa.eu/LexUriServ/site/en/oj/2003/l_325/l_32520031212en00010015.pdf>
EC (European Commission). 2006. Commission Decision of 29 September 2006 concerning a
financial contribution from the Community towards a baseline survey on the prevalence of
Salmonella in slaughter pigs to be carried out in the Member States. OJ L 275, 6.10.2006, p. 51.
<http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2006:275:0051:0061:EN:PDF>
EFSA (European Food Safety Authority). 2006a. The Community Summary Report on Trends and
Sources of Zoonoses, Zoonotic Agents, Antimicrobial Resistance and Foodborne Outbreaks in the
European Union in 2006, The EFSA Journal (2006), 130.
<http://www.efsa.europa.eu/EFSA/efsa_locale-1178620753812_1178671312912.htm>
EFSA (European Food Safety Authority). 2006b.Opinion of the Scientific Panel on Biological
Hazards on “Risk assessment and mitigation options of Salmonella in pig production”, The EFSA
Journal (2006), 341, 1-131.
EC (European Commission). 2007. Commission Decision of 30 March 2007 concerning a financial
contribution from the Community towards a baseline survey on the prevalence of Salmonella in
slaughter pigs to be carried out in Bulgaria and Romania. OJ L 95, 5.4.2007, p. 41.
<http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2007:095:0041:0049:EN:PDF>
EFSA (European Food Safety Authority). 2007a. Report of the Task Force on Zoonoses Data
Collection on the Analysis of the baseline study on the prevalence of Salmonella in holdings of laying
hen flocks of Gallus gallus, The EFSA Journal (2007) 97. <http://www.efsa.europa.eu/EFSA/efsa_locale1178620753812_1178620761896.htm>
EFSA (European Food Safety Authority). 2007b. Report of the Task Force on Zoonoses Data
Collection on the Analysis of the baseline survey on the prevalence of Salmonella in broiler flocks of
Gallus gallus, in the EU, 2005-2006 [1] - Part B: factors related to Salmonella flock prevalence,
distribution of Salmonella serovars, and antimicrobial resistance patterns. The EFSA Journal (2007)
101, 1-86.
71
EFSA (European Food Safety Authority). 2008a. Report of the Task Force on Zoonoses Data
Collection on the analysis of the baseline survey on the prevalence of Salmonella in slaughter pigs,
Part A, Salmonella prevalence estimates. The EFSA Journal 135, 1-111.
EFSA (European Food Safety Authority). 2008b. Report of the Task Force on Zoonoses Data
Collection on the Analysis of the baseline survey on the prevalence of Salmonella in turkey flocks, in
the EU, 2006-2007: Part B: factors related to Salmonella flock prevalence and distribution of
Salmonella serovars. The EFSA Journal (2008) 198, 1-224.
Funk J.A., Davies P.R., Nichols M.A. 2000. The effect of fecal sample weight on detection of
Salmonella enterica in swine feces. J.Vet. Diagn. Invest., 12, 412-418.
Hald T., Wingstrand D., Swanenburg M., von Altrock A., Thorberg B.M. 2003. The occurence and
epidemiology of Salmonella in European pig slaughterhouses. Epidemiol. Infect., 131, 1187-1203.
Hald T., Vose D., Wegener H.C., Koupeev T. 2004. A Bayesian approach to quantify the contribution
of animal-food sources to human salmonellosis. Risk Anal., 24, 255-69.
McDowell S.W.J., Porter R., Madden R., Cooper B., Neill S.D. 2007. Salmonella in slaughter pigs in
Northern Ireland: Prevalence and use of statistical modelling to investigate sample and abattoir
effects. Int. J. Food Microb., 118, 116-125.
Poisson D.M., Lanotte P., Perola V. 1993. Storage of stools specimens and of Mueller-Kauffmann
enrichment broths during week-ends does not hinder the isolation of Salmonella strains. Pathol. Biol.
(Paris), 41, 906-8.
Sorensen L.L., Alban L. Nielsen B., Dahl J. 2004. The correlation between Salmonella serology and
isolation of Salmonella in Danish pigs at slaughter. Vet. Microbiol., 101, 131-141.
72
List of Annexes
Annex I.
Statistical methodology used in the analysis of the Report B produced for the EUwide baseline survey on the prevalence of Salmonella in slaughter pigs…….........76
Annex II.
Descriptive analysis of factors potentially associated with Salmonella positivity in
lymph nodes collected from slaughter pigs…………………………………...…...82
Annex III.
Descriptive analysis of factors potentially associated with Salmonella surface
contamination of pig carcasses…………………………………………...………..93
Annex IV.
Model building……………………………………..…………………...………...100
73
SCIENTIFIC REPORT
Analysis of the baseline survey on the prevalence of
Salmonella in slaughter pigs, in the EU, 2006-2007
Part B: factors associated with Salmonella infection
in lymph nodes, Salmonella surface contamination of
carcasses, and distribution of Salmonella serovars1
Annexes
Report of the Task Force on Zoonoses Data Collection
(Question N° EFSA-Q-2006-042B)
Adopted on 14 November 2008
1
For citation purposes: Report of the Task Force on Zoonoses Data Collection on the Analysis of the baseline survey on the
prevalence of Salmonella in slaughter pigs, Part B, The EFSA Journal (2008) 206, 1-111
Table of contents
ANNEX I
STATISTICAL METHODOLOGY USED IN THE ANALYSIS OF THE
REPORT B PRODUCED FOR THE EU-WIDE BASELINE SURVEY ON THE
PREVALENCE OF SALMONELLA IN SLAUGHTER PIGS. ........................................ 76
I.1
Data import and management............................................................................................................... 76
I.2
Methodology and tools for descriptive analysis................................................................................... 76
I.3
Methodology and tools for the regression analysis ............................................................................. 77
I.3.1 Multicollinearity analysis among risk factors ................................................................................. 77
I.3.2 Multivariable modelling ................................................................................................................... 78
ANNEX II DESCRIPTIVE
ANALYSIS
OF
FACTORS
POTENTIALLY
ASSOCIATED WITH SALMONELLA POSITIVITY IN LYMPH NODES
COLLECTED FROM SLAUGHTER PIGS. ..................................................................... 82
II.1
Factors related to the sensitivity of the sampling and testing process.............................................. 82
II.2
Factors related to the lymph node infection ........................................................................................ 88
ANNEX III DESCRIPTIVE
ANALYSIS
OF
FACTORS
POTENTIALLY
ASSOCIATED WITH SALMONELLA SURFACE CONTAMINATION OF PIG
CARCASSES. ........................................................................................................................ 93
III.1 Factors related to the sensitivity of the sampling process .................................................................. 93
III.2 Factors related to the surface contamination of carcasses................................................................. 95
ANNEX IV
MODEL BUILDING................................................................................... 100
IV.1 Definition of new independent variables for “month of sampling” and “hour of sampling”...... 100
IV.2 Analysis of multicollinearity among potential factors ...................................................................... 103
IV.2.1 Lymph nodes ............................................................................................................................... 103
IV.2.2 Carcass swabs .............................................................................................................................. 105
IV.3 Complementary information on multivariable models .................................................................... 106
IV.3.1 Model on lymph nodes infection with Salmonella..................................................................... 106
IV.3.2 Model on carcass surface contamination with Salmonella ........................................................ 107
LIST OF TABLES............................................................................................................................................... 109
LIST OF FIGURES............................................................................................................................................. 111
75
Annex I
Statistical methodology used in the analysis of
the Report B produced for the EU-wide baseline survey on
the prevalence of Salmonella in slaughter pigs.
I.1 Data import and management
All data management and statistical analysis in this report were performed using SAS,
whereas figures were constructed using R.
The data provided by EFSA contain information at the level of the samples within a slaughter
pig. For simplification purposes, a variable “cTestType” is created which reflects whether the
sample was obtained from a bacteriological analysis of the ileocaecal lymph nodes (LN) or
from the bacteriological analysis of the carcass swabs (CS). Additionally, a variable
“TestResult” is created which equals 1 when the test for the presence of Salmonella was
positive, and 0 when the test was negative.
I.2 Methodology and tools for descriptive analysis
The descriptive section presents a thorough description of the samples by all independent
variables. This descriptive analysis is based on boxplots, frequency tables, simple chi-squared
or trend tests and simple weighted logistic regression models. Note that these results should
be interpreted only within the context of an exploratory analysis. Further analysis using
appropriate modelling techniques should be used to validate these results in their proper
context.
In what follows, some detailed discussion is provided on the tests used to study association
between Salmonella prevalence and the independent variables. Note that this association is
studied using the data at the community level.
•
Chi-square tests
Consider two categorical variables X and Y, X having I levels and Y having J levels. The IJ
possible combinations of outcomes can be displayed in a contingency table having I rows for
the categories of X, and J columns for the categories of Y. For instance, Table I.1 represents
the notation used in such a 3 X 13 table.
76
Table I.1
Table notation for the cross classification of Salmonella prevalence in
slaughter pigs by month of sampling.
Salmonella
Month of sampling
Oct/06
Nov/06
…
Oct/07
Total
No
Yes
Inconclusive
Total
The null hypothesis H0 of independence is equivalent to cell probabilities satisfying
. For a sample of size n with cell counts
, the values
represent
when H0 is true. The
the expected frequencies, i.e, the values of the expectations
sample cell counts can then be compared to the expected frequencies to judge whether the
data contradict H0. If the null hypothesis is true, then
should be close to
in each cell.
, the stronger the evidence against H0. In practice,
can
The larger the differences
.
be estimated by
A test statistic which uses this property is the Pearson chi-squared statistic, given by
,
which is asymptotically distributed according to a chi-square with
freedom.
degrees of
It is difficult to evaluate whether the available sample size is large enough for these
asymptotic results to be valid. A general rule of thumbs is given by
. When the
sample size is small, one can resort to inferences using exact distributions rather than largesample approximations. (Agresti, 1996; Agresti, 2002)
I.3 Methodology and tools for the regression analysis
I.3.1 Multicollinearity analysis among risk factors
A formal method to detect multicollinearity is given by the variance inflation factor or VIF.
The VIF measures how much the variances of the estimated regression coefficients are
inflated as compared to when the predictor variables are not linearly related. Essentially, each
risk factor
is regressed on the other risk factors
in the model. The corresponding
coefficient of multiple determination
is then used to calculate the VIF:
Note that the VIF is equal to 1 when
, i.e., when
is not linearly related to the other
, then the VIF will be larger than 1, indicating an inflated variance
risk factors. When
77
for the estimated regression coefficients due to correlations among the risk factors. A
maximum VIF exceeding 10 is frequently interpreted as an indication of multicollinearity.
For the categorical covariates, the VIF can be calculated in a similar way using
,
with
and
representing the maximized log-likelihoods for the fitted model and
the “null” model, containing only the intercept, and n referring to the sample size.
(Neter et al., 1996 ; Agresti A., 1996)
I.3.2 Multivariable modelling
The hierarchical structure in the data can essentially be expressed as follows: slaughter pigs
within a slaughterhouse, and slaughterhouse within a country. Interest goes to the pig-level
prevalence. Therefore, let be the probability for a pig to be positive, let
be the number of
pigs in slaughterhouse j from country i. The starting point for inference on the ‘pig-level
prevalence’ of the different outcome variables is the binomial distribution for the number of
positive pigs in slaughterhouse j from country i:
.
(1)
In a fully random sample these numbers
could be combined in a straightforward way to
estimate the prevalence for country i. The main complications here are
1. the assumptions on the binomial distribution are violated
2. the sample is not drawn at random (but essentially stratified)
Indeed,
– violation of independence: outcomes from the same slaughterhouse are expected to be
more alike (correlated) as compared to outcomes from a different slaughterhouse
(hierarchical correlation structure),
– violation of constant probability: samples, even from the same slaughterhouse might
have different probabilities to be infected (heterogeneity of probability).
•
Clustering
To account for the possibility of samples from the same slaughterhouse being more alike than
from different slaughterhouses, there exist, broadly, three approaches:
– Ignore the correlation. While this typically leaves the consistency of point estimation
intact, the same is not true for measures of precision. In case of a “positive”
correlation (i.e., samples within a slaughterhouse are more alike than between
slaughterhouses), then ignoring this aspect of the data, just as ignoring
overdispersion, overestimates precision and hence underestimates standard errors and
widths of confidence intervals.
– Account for correlation. The existence of correlation is recognized but considered as
a nuisance characteristic. A crude way of correcting for clustering is done by
computing a so-called design effect. Roughly, the design effect is a factor comparing
78
the precision under simple random sampling with the precision of the actual design.
Standard errors, computed as if the design had been simple random sampling, can
then be artificially inflated using the design effect.
– Model correlation. In contrast to the previous view-point, one can have a genuine
scientific interest in the correlation itself. The intra-class correlation should be
addressed in order to obtain valid statistical inferences, and specialized methods
which model the correlation should be used.
Obviously the third method is much broader. Hence analysis strategies consistent with an
interest in the intra-cluster dependence can be applied. There exist two important families of
models which can be used for this purpose: random-effects models and marginal models.
In a marginal or population-averaged model, marginal distributions are used to describe the
outcome vector , given a set
of predictor variables. A marginal model can be used to
evaluate the overall (or population-averaged) trend as a function of covariates. Alternatively,
in a random-effects model, also called cluster-specific models or multilevel models, the
predictor variables
are supplemented with a vector of random effects (specific to the
cluster/slaughterhouse), conditional upon which the components of are usually assumed to
be independent. Thus, cluster-specific models are differentiated from population-averaged
models by the inclusion of parameters which are specific to the cluster/slaughterhouse. In
random-effects models, the intracluster correlation is assumed to arise from natural
heterogeneity in the parameters across clusters (slaughterhouse).
There are two routes to introduce randomness into the model parameters. The first approach
introduces the random effects on the probability scale, such as the beta-binomial model
(Skellam, 1948). The second approach introduces the random effects in the linear predictor,
yielding the classical mixed-effects models (Stiratelli et al., 1984). A random effects logistic
regression model is an example of the second approach, where it is assumed that the number
of positive samples in slaughterhouse j in country i follow a binomial distribution:
,
(1)
with mean modeled through a linear predictor containing fixed regression parameters
slaughterhouse-specific parameters :
and
.
It is assumed that the slaughterhouse-specific effects are normally distributed with mean zero
and some variance , i.e.,
. The above model can be interpreted as a logistic
regression model for each slaughterhouse, where some of the regression parameters are
express
specific (random effects), while others are not (fixed effects). The random effects
how unit-specific trends deviate from the population-averaged trends. In case of repeated
samples, the above model can be generalized by inclusion of a general time trend (fixed
effect) and slaughterhouse-specific time trends (random effect). This is often called a randomslopes model.
Unlike for correlated Gaussian outcomes, the parameters of the cluster-specific and
population-averaged models for correlated binary data describe different types of effects of
the covariates on the response probabilities (Neuhaus, 1992). The choice between populationaveraged (i.e. marginal models) and cluster-specific (i.e. mixed models) strategies may
heavily depend on the scientific goals. Population-averaged models evaluate the overall risk
79
as a function of covariates. With the cluster-specific approach, the response rates are modelled
as a function of covariates and parameters, specific to a slaughterhouse. In such models, the
interpretation of fixed-effect parameters is conditional on a constant level of the
slaughterhouse-specific parameter (e.g. random effect). Diggle et al. (1994) and Diggle et al.
(2002) recommended the random-effect model for inferences about individual responses and
the marginal model for inferences about margins, that is, the objectives (or the types of
inferences) in a study should determine which suitable statistical model to use. For more
details, see e.g. Aerts et al. (2002) and Molenberghs and Verbeke (2005).
•
Weighting
Most statistical procedures analyze the data as if they were collected as a simple random
sample. As a result, these procedures may lead to biased estimates and may underestimate the
variability present in the data, when the data actually arise from complex surveys. Assigning
weights to the observations is one possible approach to correct for the differences between the
complex survey design and simple random sampling. In general, by using weights, we try to
‘reconstruct the total population’, in order to avoid that certain strata or subpopulations are
over- or underrepresented.
First note that the target population of the survey is defined as the population of slaughter pigs
in the slaughterhouses representing 80% of the national throughput. For most countries with
important pig sectors, the sampling rate is close to the target of 80%. However, some
sampling rates were different from the target. Therefore, we need to take this disproportionate
sampling into account by means of weights. In this report, Proxy - slaughterhouse weights, as
represented in Table I.2 were used for the risk factor analysis.
Table I.2
Weights to account for the disproportionate sampling in the baseline
survey on slaughter pigs.
Level
Weight
EU
WY1*WY2
Country
WY1 = 80 % of national throughput / sum of throughput of sampled slaughterhouses
Slaughterhouse
WY2 = total # pigs / sampled # of pigs (in a slaughterhouse)
Ideally, the country-level weight would be expressed in terms of the total and sampled
number of slaughterhouses within a country (similar to the approach in report A for the BST).
However, there is no information available on the number of non-sampled slaughterhouses.
Therefore, an approximation like WY1 could be used to reflect the disproportionate sampling.
Observe that now the information of the sampled slaughterhouse throughput is used in both
WY1 as well as in WY2. Additionally, the definition of WY1 assumes that the number of pigs
processed by each slaughterhouse in a country is roughly the same. It would be of interest to
study how good an approximation WY1 really is to the “ideal” country-level weight,
however, this is outside the scope of the current report.
80
Note that the total number of pigs in a slaughterhouse is provided by the variable
V051_Throughput, whereas the sampled number of pigs can be calculated from the data. The
weights to be used when studying prevalence are therefore,
• on the level of the European Community: W_EU = WY1*WY2
• on the level of the MS and Norway: W_MS = WY2
The proxy weights as defined in WY1 will therefore not impact the country-specific
prevalence estimates.
Finally, observe that the sum of these weights gives an indication of the total number of
slaughter pigs N in the EU, or within each Member State. To avoid overemphasizing the
importance of the pigs used in the sample, we therefore need to standardize the calculated
weights so that they sum to Ns, i.e., the sample size. In general, this implies that, for pig k, in
slaughterhouse j, in country i:
If
then
Therefore, we will use the standardized weights
.
.
On the level of each member state separately, a similar standardization procedure can be
, where
applied. We will then use the following standardized weights
( ) denotes the sample size (population size) within Country c.
81
Annex II
Descriptive analysis of factors potentially
associated with Salmonella positivity in lymph nodes
collected from slaughter pigs.
II.1 Factors related to the sensitivity of the sampling and
testing process
•
Table II.1
Distribution of the number and percentage of lymph node samples by the
weight of the sample, per country and in the EU.
15 - 24 g
Lymph Node Weight
25 - 34 g
35 - 44 g
Austria
236 38%
231 38%
Belgium
137 23%
464 77%
Country
Bulgaria
3 2%
155 88%
Cyprus
249 69%
110 31%
12 2%
303 46%
Czech Republic
Denmark
95 15%
≥ 45 g
52 8%
Total
614 a
601
18 10%
176
359
146 22%
192 30%
653
204 20%
794 80%
Estonia
29 7%
121 29%
88 21%
182 43%
420
Finland
94 22%
185 44%
61 15%
79 19%
419
France
b
Germany
Greece
Hungary
Ireland
Italy
690 59%
435 37%
27 2%
11 1%
1,163
1,235 48%
1,239 48%
42 2%
51 2%
2,567
74 21%
246 71%
21 6%
4 1%
345
213 32%
205 31%
78 12%
162 25%
658
72 17%
136 32%
118 28%
96 23%
422
15 2%
6 1%
374 53%
313 44%
Latvia
10 3%
382 97%
Lithuania
35 8%
236 51%
177 15%
963 82%
Luxembourg
Poland
998
708
392
130 28%
60 13%
22 2%
14 1%
312 100%
461
313
1,176
Portugal
655 100%
3 0%
Slovakia
119 31%
178 46%
59 15%
29 8%
385
Slovenia
193 45%
217 51%
16 4%
3 1%
429
Spain
658
1,841 70%
778 30%
Sweden
188 48%
200 51%
5 1%
Netherlands
168 16%
917 84%
1 <1%
United Kingdom b
EU
Norway
202 34%
267 45%
7,212 39%
9,392 50%
3 <1%
229 56%
2,619
125 21%
1,067 6%
74 18%
1 <1%
394
1,086
5 <1%
948 5%
102 25%
599
18,614
408
The weight of 3 lymph node samples were missing.
b
France and United Kingdom reported additional data on lymph node weight after publication of report part A.
a
82
Table II.2
Weighted frequencies of Salmonella positive/negative lymph node samples
by lymph node sample weight at the EU level (Norway included), and
corresponding Chi-square statistic.
Frequency
Row Pct
Col Pct
Negative
Positive
2
Salmonella
Lymph node weight
2
15 - 24 g
25 - 34 g
35 - 44 g
≥ 45 g
Total
6,631
8,487
666
454
16,238
41%
52%
4%
3%
100%
82%
87%
89%
89%
85%
1,435
1,217
80
55
2,788
100%
51%
44%
3%
2%
18%
13%
11%
11%
15%
Total
8,067
9,703
746
Chi-square statistic: 113.2 (P-value <0.0001)
509
19,025
Each cell in this contingency table of weighted Salmonella by the weight of the LN represents:
on the first line (‘Frequency’): the weighted number of samples by infection outcome and by the weight of the LN.
Note that these are not integer values, therefore rounding errors can lead to some discrepancies with margin totals.
Some difference between the weighted number total and the total number of samples (cf. Table II.1) may occur.
on the second line (‘Row Pct’): the distribution of the samples by the weight of the LN, for each infection outcome
separately
- on the third line (‘Col Pct’): the distribution of the samples by infection outcome, for each class of LN weigths
separately
83
•
Number of lymph nodes in the sample
Table II.3
Central tendency (mean, median) and dispersion (standard deviation and
quartiles) measures of the number of lymph nodes in the sample, by
country and at the EU level.
Number of lymph nodes per sample
Country
a
a
Median
Q3
12
16
a
Mean
StD
13.6
6.1
Austria
10
Belgium
5
5
5
5.4
2.2
Bulgaria
8
10
12
10.6
3.4
Cyprus
12
15
16
13.7
3.6
Czech Republic
12
17
27
21.4
13.4
Denmark
10
10
14
12.7
7.6
Estonia
18.5
23
29
24.0
8.7
Finland
10
16
24
18.1
9.7
France
.
.
.
.
.
Germany
5
9
13
10.1
5.8
Greece
8
11
15
12.0
5.7
Hungary
5
5
5
5.0
0.0
Ireland
16
20
24
21.2
7.2
Italy
5
5
6
6.4
3.0
Latvia
9
11
13
11.1
2.8
Lithuania
12
16
19
16.4
6.2
Luxembourg
5
5
6
5.4
0.6
Poland
5
5
9
7.2
3.3
Portugal
13
16
21
17.1
6.4
Slovakia
15
19
25
20.8
8.5
Slovenia
13
17
22
18.2
6.8
Spain
.
.
.
.
.
Sweden
9
11
14
11.7
4.1
The Netherlands
7
9
12
10.3
5.4
United Kingdom
10
15
19
15.3
6.5
EU
6
10
16
12.5
8.0
Norway
15
15
20
17.8
6.3
a
Table II.4
Q1
Q1: 25% quantile, Q3: 75% quantile, StD: standard deviation
Odds ratio estimate and corresponding 95% confidence interval obtained
from a weighted logistic regression, modelling the probability of observing
a positive sample using an intercept and (separately) the number of lymph
nodes per sample at the global level.
Outcome of interest
OR Estimate
Salmonella
1.002
Lower
Bound
0.994
Upper
Bound
1.010
84
•
Table II.5
number of days between the sampling date and the starting date of testing,
per country and in the EU.
Number of days delay between sampling date and the starting date of testing lymph node samples
Country
0
1
2
3
4
Austria
89 14%
342
55%
155
25%
23
4%
8 1%
Belgium
55 9%
187
31%
216
36%
80
13%
30 5%
105
60%
42
24%
27
15%
1 1%
16
4%
38
11%
10 3%
9
1%
Bulgaria
Cyprus
179 50%
103
29%
Czech Republic
415 64%
204
31%
20
3%
4 <1%
576
58%
197
20%
380
90%
14
3%
241
58%
163
39%
Denmark
Estonia
25 6%
Finland
1 <1%
France
18 2%
155
16%
10
2%
5
7
2%
9
1%
469
40%
417
36%
91 8%
5
485
19%
219
9%
164 6%
79
Greece
87 25%
200
58%
42
12%
8
2%
2 1%
Hungary
53 8%
400
61%
141
21%
45
7%
6 1%
Ireland
52 12%
256
61%
62
15%
12
3%
31 7%
304 43%
327
46%
57
8%
13
2%
3 <1%
275
70%
107
27%
7
2%
3 1%
Lithuania
50 11%
350
76%
61
13%
Luxembourg
92 29%
109
35%
63
20%
39
13%
7 2%
Poland
115 10%
869
74%
187
16%
2
<1%
25
Portugal
397 60%
211
32%
25
4%
86 22%
200
52%
98
25%
Slovenia
43 10%
168
39%
207
48%
10
EU
55%
417
16%
330
13%
174 7%
297
75%
25
6%
64
16%
3 1%
13 1%
1,070
99%
3
Norway
16 4%
Table II.6
434
72%
50
54%
3,322
280
69%
35
359
1
<1%
998
176
2
<1%
2
<1%
653
1,163
<1%
3%
33
1%
5
1%
6
1%
2
<1%
3
<1%
1
<1%
8
<1%
6
2%
345
8
1%
658
1
<1%
18%
9%
1
<1%
1
<1%
312
2
<1%
1
<1%
1,176
1
<1%
37
2
1%
<1%
2,619
394
1,086
103
1,636
59
599
17%
12 2%
9%
608 3%
168
14%
14 3%
2
1%
<1%
52
<1%
2
<1%
43
<1%
Outcome of interest
OR Estimate
Lower Bound
Upper Bound
Salmonella positivity
1.151
1.113
1.190
385
429
18,617
408
Odds ratio estimate and corresponding 95% confidence interval from a
weighted logistic regression, modelling the probability of observing a
positive sample using an intercept and (separately) the number of days
between sampling and testing at the global level.
Time between sampling
and testing
422
392
<1%
8%
2,567
708
1 <1%
2%
1,452
10,018
2%
658
5 1%
2,770 15%
6
4%
207 8%
United Kingdom
601
461
Slovakia
Netherlands
1%
1%
420
14%
Spain
1
8
419
43%
Sweden
1%
4 1%
163
Latvia
6
1 <1%
1,099
Italy
3%
3 <1%
54 5%
Total
7
617
19
480 19%
Germany
6
85
Table II.7
number of days delay between the sampling date and the starting date of
testing, per country and in the EU.
Country
Austria
Belgium
Bulgaria
Cyprus
Czech Republic
Denmark
Estonia
Finland
France
Germany
Greece
Hungary
Ireland
Italy
Latvia
Lithuania
Luxembourg
Poland
Portugal
Slovakia
Slovenia
Spain
Sweden
Netherlands
United Kingdom
EU
Norway
Number of days delay between the sampling date
and the starting date of testing lymph node sample
0 – 2 days
586 95%
458 76%
147 84%
298 83%
639 98%
777 78%
419 100%
405 97%
650 56%
2,064 80%
329 95%
594 90%
370 88%
688 97%
382 97%
461 100%
264 85%
1,171 100%
633 96%
384 100%
418 97%
2,076 79%
327 83%
1,086 100%
484 81%
16,110 87%
331 81%
3 – 4 days
31 5%
110 18%
28 16%
48 13%
12 2%
209 21%
1 <1%
14 3%
508 44%
383 15%
10 3%
51 8%
43 10%
16 2%
10 3%
46 15%
2 <1%
25 4%
33
1
13
2
12
5%
1%
4%
<1%
1%
5
120
6
13
9
4
<1%
5%
2%
2%
2%
1%
2 1%
3 <1%
1 <1%
11 3%
504 19%
67 17%
115 19%
2,244 12%
73 18%
Total
5 – 7 days
39 1%
263 1%
4 1%
617
601
176
359
653
998
420
419
1,163
2,567
345
658
422
708
392
461
312
1,176
658
385
429
2,619
394
1,086
599
18,617
408
86
Table II.8
Weighted frequencies of Salmonella positive/negative lymph node samples
by the number of days delay between the sampling date and the starting
date of testing for lymph node samples at the EU level (Norway included),
and corresponding Chi-square statistic.
Lymph node samples
Frequency
Row Pct
Col Pct
Negative
Positive
3
Salmonella
Time to testing
0 – 2 days
3 – 4 days
5 – 7 days
Total
13,712
2,235
294
16,240
84%
14%
2%
100%
86%
79%
85%
85%
2,153
582
53
2,788
77%
21%
2%
100%
14%
21%
15%
Total
15,865
2,817
346
Chi-square statistic: 129.6 (P-value<0.0001)
15%
19,028
3
Each cell in this contingency table of weighted Salmonella by the number of days delay between the sampling date and the
starting date of detection testing for LN represents:
on the first line (‘Frequency’): the weighted number of samples by infection outcome and by the number of days
delay between the sampling date and the starting date of detection testing for LN. Note that these are not integer
values, therefore rounding errors can lead to some discrepancies with margin totals. Some difference between the
weighted number total and the total number of samples (cf. Table II.7) may occur.
on the second line (‘Row Pct’): the distribution of the samples by the number of days delay between the sampling
date and the starting date of detection testing for LN, for each infection outcome separately.
on the third line (‘Col Pct’): the distribution of the samples by infection outcome, for each class of days of delay
between the sampling date and the starting date of detection testing for LN, separately
87
II.2
•
Month of sampling
Table II.9
Country
Austria
Factors related to the lymph node infection
Distribution of the number and percentage of pigs by month of sampling.
Month of sampling
Oct.06
Nov.06
Dec.06
Jan.07
Feb.07
Mar.07
Apr.07
May07
Jun.07
Jul.07
Aug.07
Sep.07
45
7%
55
9%
82
14%
40
6%
63
10%
59
10%
46
8%
44
7%
51
8%
60
10%
45
7%
70
11%
50
8%
7
2%
56
9%
76
8%
31
7%
30
7%
44
12%
53
8%
85
9%
38
9%
35
8%
8
1%
225
9%
24
7%
74
11%
35
8%
1
0%
28
8%
54
8%
85
9%
34
8%
35
8%
84
7%
212
8%
15
4%
6
1%
33
8%
2
0%
35
10%
53
8%
93
9%
35
8%
35
8%
104
9%
223
9%
40
12%
85
13%
33
8%
83
12%
51
11%
24
8%
120
10%
35
10%
54
8%
89
9%
35
8%
35
8%
132
11%
221
9%
47
14%
32
5%
33
8%
103
15%
43
11%
72
16%
21
7%
98
8%
76
12%
30
8%
38
9%
220
8%
36
9%
94
9%
52
9%
1691
9%
34
8%
32
9%
55
8%
82
8%
34
8%
35
8%
127
11%
223
9%
18
5%
69
10%
37
9%
103
15%
65
17%
46
10%
27
9%
101
9%
104
16%
34
9%
38
9%
221
8%
31
8%
91
8%
49
8%
1728
9%
34
8%
31
9%
54
8%
70
7%
36
9%
30
7%
116
10%
216
8%
23
7%
36
5%
38
9%
78
11%
51
13%
54
12%
25
8%
73
6%
81
12%
29
8%
37
9%
214
8%
32
8%
95
9%
53
9%
1593
9%
34
8%
54
9%
53
9%
33
19%
29
8%
57
9%
76
8%
35
8%
38
9%
101
9%
218
8%
34
10%
5
1%
36
9%
104
15%
46
12%
55
12%
25
8%
89
8%
87
13%
36
9%
40
9%
232
9%
34
9%
87
8%
42
7%
1646
9%
34
8%
43
7%
56
9%
41
23%
29
8%
55
8%
81
8%
35
8%
38
9%
112
10%
224
9%
36
10%
57
9%
35
8%
57
8%
49
13%
57
12%
23
7%
76
6%
75
11%
29
8%
37
9%
230
9%
39
10%
93
9%
46
8%
1653
9%
31
8%
57
9%
50
8%
38
22%
31
9%
54
8%
78
8%
35
8%
34
8%
113
10%
213
8%
28
8%
17
3%
37
9%
54
8%
55
14%
42
9%
23
7%
134
11%
76
12%
41
11%
35
8%
242
9%
40
10%
85
8%
51
9%
1663
9%
33
8%
54
9%
55
9%
37
21%
36
10%
54
8%
81
8%
35
8%
35
8%
117
10%
225
9%
25
7%
12
2%
34
8%
49
7%
39
10%
42
9%
34
11%
103
9%
78
12%
31
8%
37
9%
242
9%
37
9%
99
9%
52
9%
1644
9%
34
8%
36
6%
50
8%
27
15%
22
6%
54
8%
102
10%
37
9%
39
9%
149
13%
206
8%
52
15%
265
40%
36
9%
74
10%
44
11%
42
9%
38
12%
122
10%
81
12%
31
8%
39
9%
315
12%
32
8%
93
9%
53
9%
2039
11%
37
9%
Belgium
Bulgaria
Cyprus
Czech Republic
Denmark
Estonia
Finland
France
Germany
Greece
161
6%
3
1%
Hungary
Ireland
35
8%
Italy
Latvia
Lithuania
Luxembourg
Poland
17
5%
84
7%
29
9%
96
8%
26
8%
80
7%
19
5%
31
7%
40
10%
33
8%
47
2%
29
7%
88
8%
47
8%
1168
6%
36
9%
29
8%
32
8%
417
16%
32
8%
81
7%
50
8%
1440
8%
31
8%
Portugal
Slovakia
Slovenia
Spain
Sweden
Netherlands
United Kingdom
European Union
Norway
20
5%
88
8%
52
9%
755
4%
34
8%
36
32
8%
239
9%
32
8%
92
8%
52
9%
1603
9%
36
9%
Total
617
100%
601
100%
176
100%
359
100%
653
100%
998
100%
420
100%
419
100%
1163
100%
2567
100%
345
100%
658
100%
422
100%
708
100%
392
100%
461
100%
312
100%
1176
100%
658
100%
385
91%
429
100%
2619
100%
394
100%
1086
100%
599
100%
18617
100%
408
100%
88
Table II.10
Weighted frequencies of Salmonella positive/negative lymph node samples by the month of sampling at the EU level
(Norway included), and corresponding Chi-square statistic.
Frequency4
Row Pct
Col Pct
Negative
Positive
Month of sampling
Oct. 06
Nov. 06
Dec. 06
Jan. 07
Feb. 07
Mar. 07
Apr. 07
May 07
Jun. 07
Jul. 07
Aug. 07
Sep. 07
Total
692
1,097
1,292
1,496
1,564
1,489
1,372
1,374
1,330
1,411
1,345
1,778
16,240
4%
7%
8%
9%
10%
9%
8%
8%
8%
9%
8%
11%
100%
90%
91%
82%
86%
87%
86%
86%
85%
83%
85%
82%
84%
85%
79
103
278
245
228
250
223
243
267
242
286
343
2,788
3%
4%
10%
9%
8%
9%
8%
9%
10%
9%
10%
12%
100%
10%
9%
18%
14%
13%
14%
14%
15%
17%
15%
18%
16%
15%
1,792
1,739
1,595
1,618
1,598
1,653
1,631
2,121
19,028
Total
771
1,200
1,570
1,741
4
Each cell in this contingency table of Salmonella by the month of sampling represents
on the first line (‘Frequency’): the weighted number of samples by infection outcome and by month of sampling. Note that these are not integer values, therefore rounding errors
can lead to some discrepancies with margin totals. Some difference between the weighted number total and the total number of samples (cf. Table II.9) may occur.
on the second line (‘Row Pct’): the distribution of the samples by month of sampling, for each infection outcome separately
on the third line (‘Col Pct’): the distribution of the samples by infection outcome, for each month of sampling separately
•
Sampling quarter
Table II.11
Country
Lymph node samples - Distribution of the number and percentage of lymph
node samples by sampling quarter, per country and at the EU level.
Sampling quartera
Country
Sampling quatera
3
4
Latvia
108
28%
146
37%
138
35%
392
100%
601
100%
Lithuania
169
37%
166
36%
126
27%
461
100%
102
58%
176
100%
Luxembourg
72
23%
72
23%
73
23%
95
30%
312
100%
2
3
4
Austria
140
23%
163
26%
167
27%
147
24%
617
100%
Belgium
145
24%
142
24%
159
26%
155
26%
74
42%
1
Total
2
1
Bulgaria
a
Total
Cyprus
79
22%
102
28%
89
25%
89
25%
359
100%
Poland
260
22%
319
27%
238
20%
359
31%
1176
100%
Czech Republic
163
25%
162
25%
166
25%
162
25%
653
100%
Portugal
0
0%
180
27%
243
37%
235
36%
658
100%
Denmark
246
25%
264
26%
227
23%
261
26%
998
100%
Slovakia
88
23%
100
17%
94
24%
103
27%
385
91%
Estonia
103
25%
104
25%
106
25%
107
25%
420
100%
Slovenia
96
22%
108
25%
114
27%
111
26%
429
100%
Finland
100
24%
105
25%
106
25%
108
26%
419
100%
Spain
464
18%
680
26%
676
26%
799
31%
2,619
100%
France
92
8%
363
31%
329
28%
379
33%
1,163
100%
Sweden
81
21%
99
25%
105
27%
109
28%
394
100%
Germany
598
23%
667
26%
658
26%
644
25%
2,567
100%
Netherlands
257
24%
277
26%
275
25%
277
26%
1,086
100%
Greece
42
12%
105
30%
93
27%
105
30%
345
100%
United Kingdom
149
25%
153
26%
141
24%
156
26%
599
100%
Hungary
80
12%
186
28%
98
15%
294
45%
658
100%
European Union
3,361
18%
5,020
27%
4,891
26%
5,345
29%
18,617
100%
Ireland
103
24%
103
24%
109
26%
107
25%
422
100%
Norway
101
25%
104
25%
99
24%
104
25%
408
100%
Italy
3
0%
289
41%
239
34%
177
25%
708
100%
Sampling quarter: 1: Oct.-Dec.2006; 2: Jan.-Mar.2007; 3: Apr.-Jun.2007; 4: Jul.-Sept.2007
Table II.12
Weighted frequencies of Salmonella positive/negative lymph node samples by
quarter of sampling at the EU level (Norway included), and corresponding
Chi-square statistic.
Frequency5
Row Pct
Col Pct
Sampling quarter
1
2
Negative
3,081
4,549
19%
28%
87%
86%
Positive
460
722
17%
26%
13%
14%
Total
3,541
5,272
•
3
4
4,077
25%
85%
734
26%
15%
4,811
4,534
28%
84%
871
31%
16%
5,405
Total
16,241
100%
85%
2,788
100%
15%
19,028
Hour of sampling
Table II.13
Lymph node samples - Chi-square statistic of the weighted Salmonella
prevalence by hour of sampling in the EU.
Chi-square statistic (P-value)
LN
129.6 (<0.0001)
5
Each cell in this contingency table of weighted Salmonella by the number of days delay between the sampling date and
the starting date of detection testing for LN represents:
on the first line (‘Frequency’): the weighted number of samples by infection outcome and by quarter of sampling.
Note that these are not integer values, therefore rounding errors can lead to some discrepancies with margin
totals. Some difference between the weighted number total and the total number of samples (cf. Table II.11) may occur.
on the second line (‘Row Pct’): the distribution of the samples by quarter of sampling, for each infection outcome
separately.
on the third line (‘Col Pct’): the distribution of the samples by infection outcome, for each quarter of sampling,
separately
91
•
Weight of carcasses
Table II.14
Lymph node samples - Central tendency (mean, median) and dispersion
measures (standard deviation and quartiles) of the weight of the carcass, at
the country and EU level.
Lymph node samples: Carcass weight
Country
a
Median
Q3
Mean
a
StD
Austria
90
95
100
94.4
8.3
Belgium
85
90
97
92.5
8.9
Bulgaria
80
85
89
84.5
5.3
Cyprus
63
70
70
66.3
5.1
Czech Republic
77
80
82
79.6
2.9
Denmark
78
83
88
83.3
7.7
Estonia
71
72
76
73.9
5.2
Finland
80
85
90
85.2
8.6
France
87
92
98
92.3
8.5
Germany
90
95
99
94.3
9.0
Greece
60
70
80
72.9
15.0
Hungary
108
110
117
109.9
11.0
Ireland
70
82
90
82.0
10.3
Italy
127
132
136
129.1
10.2
Latvia
68
72
79
74.4
9.9
Lithuania
78
87
100
87.4
12.9
Luxembourg
79
83
87
83.0
6.0
Poland
79
85
100
89.1
14.5
Portugal
74
80
85
80.2
9.0
Slovakia
86
91
99
92.7
12.0
Slovenia
77
82
90
83.5
10.5
Spain
71
71
71
73.9
10.2
Sweden
83
88
92
87.5
7.9
Netherlands
86
89
93
89.8
6.6
United Kingdom
74
79
84
79.2
7.9
European Union
74
85
95
87.0
15.5
Norway
71
75
81
75.6
8.9
a
Table II.15
Q1a
Q1: 25% quantile, Q3: 75% quantile, StD: standard deviation
Lymph node samples - Odds ratio estimate and corresponding 95%
confidence interval obtained from a weighted logistic regression, modelling
the probability of observing a positive sample using an intercept and
(separately) the weight of the carcass.
Weight of the carcass
Outcome of interest
OR
Estimate
Lower
Bound
Upper
Bound
Salmonella
0.987
0.984
0.990
92
Annex III Descriptive analysis of factors potentially
associated with Salmonella surface contamination of pig
carcasses.
III.1
•
Factors related to the sensitivity of the sampling process
Table III.1
Distribution of the number and percentage of carcass swabs by the number of
days delay between the sampling date and the starting date of testing, per
Member State and in the 13-MS group.
Number of days delay between sampling date and starting date of detection testing for CS
Country
0
1
2
3
4
5
Austria
Belgium
Cyprus
Czech Republic
Denmark
France
Ireland
Latvia
Lithuania
Poland
Slovenia
Sweden
United Kingdom
89
27
185
264
1
7
52
3
50
59
84
6
14%
7%
52%
63%
<1%
2%
12%
1%
11%
13%
19%
1%
342
134
99
132
198
63
256
274
351
335
192
303
467
55%
35%
28%
32%
58%
15%
61%
70%
76%
75%
44%
75%
73%
155
127
15
13
71
154
62
104
60
51
148
25
51
25%
33%
4%
3%
21%
37%
15%
27%
13%
11%
34%
6%
8%
23
49
38
5
53
152
12
7
4%
13%
11%
1%
15%
37%
3%
2%
8
19
10
2
17
35
31
3
1%
5%
3%
<1%
5%
8%
7%
1%
10
64
111
2%
16%
17%
4
4
12
1%
1%
2%
13 - MS Group
827
14%
3146
55%
1036
18%
524
9%
145
3%
Table III.2
15
7
6
1%
<1%
1%
1
3
<1%
1%
37
6
2%
1
<1%
2
<1%
1
<1%
10
<1%
4%
2%
4
1
6
1%
7
Tot
4
5
1%
1%
1
1
<1%
617
381
359
417
344
413
422
391
461
447
441
402
641
11
<1%
5736
Carcass swabs - Odds ratio estimates and corresponding 95% confidence
intervals from a weighted logistic regression, modelling the probability of
observing a positive sample using an intercept and (separately) the number of
days between sampling and testing in 13-MS group.
Number of days
between sampling
and testing
Outcome of interest
OR
Estimate
Lower
Bound
Upper
Bound
Salmonella
1.224
1.134
1.322
93
Table III.3
Carcass swabs - Distribution of the number and percentage of samples by the
detection testing for carcass swabs, per Member State and in the 13-MS
group.
Number of days delay between sampling and testing of carcass swabs
Country
0 day
1 day
2 days
3-7 days
Total
Austria
89
14%
342
55%
155
25%
31
5%
617
Belgium
27
7%
134
35%
127
33%
93
24%
381
Cyprus
185
52%
99
28%
15
4%
60
17%
359
Czech Republic
264
63%
132
32%
13
3%
8
2%
417
Denmark
1
<1%
198
58%
71
21%
74
22%
344
France
7
2%
63
15%
154
37%
189
46%
413
Ireland
59
14%
253
60%
59
14%
51
12%
422
Latvia
3
1%
274
70%
104
27%
10
3%
391
Lithuania
50
11%
351
76%
60
13%
-
461
Poland
59
13%
335
75%
51
11%
2
<1%
447
Slovenia
84
19%
192
44%
148
34%
17
4%
441
Sweden
6
1%
303
75%
25
6%
68
17%
402
467
7%3
51
8%
123
19%
641
3,143
55%
1,033
18%
726
13%
5,736
United Kingdom
13 MS group
Table III.4
834
15%
Weighted frequencies of Salmonella positive/negative carcass swabs by the
testing, per Member State and in the 13-MS group, with Chi-square statistic.
Time to testing
Frequency6
Row Pct
Col Pct
Negative
Positive
Total
0 day
1 day
2 days
3-7
days
Total
455
9%
96%
20
4%
4%
2,575
50%
92%
225
42%
8%
1,153
22%
88%
164
30%
12%
1,011
19%
88%
133
24%
12
5,194
100%
91%
542
100%
9%
475
2,730
1,317
1,144
5,736
Chi-square statistic: 41 (P-value < 0.0001)
6
Each cell in this contingency table of weighted Salmonella by the number of days delay between the sampling date and the
starting date of detection testing for carcass swabs represents:
on the first line (‘Frequency’): the weighted number of samples by infection outcome and by the number of days delay
between the sampling date and the starting date of detection testing for CS. Note that these are not integer values,
therefore rounding errors can lead to some discrepancies with margin totals.
on the second line (‘Row Pct’): the distribution of the samples by the number of days delay between the sampling date
and the starting date of detection testing for CS, for each infection outcome separately.
- on the third line (‘Col Pct’): the distribution of the samples by infection outcome, for each class of days of delay
between the sampling date and the starting date of detection testing for CS, separately
94
III.2 Factors related to the surface contamination of carcasses
•
Month of sampling
Table III.5
Country
Austria
Carcass swabs - Distribution of the number and percentage of pigs by month
of sampling, per Member State and 13-MS group.
Month of sampling
Total
Oct.06
Nov.06
Dec.06
Jan.07
Feb.07
Mar.07
Apr.07
May07
Jun.07
Jul.07
Aug.07
Sep.07
45
7%
55
9%
40
6%
59
10%
44
7%
60
10%
70
11%
54
9%
43
7%
57
9%
54
9%
36
6%
617
100%
57
15%
36
9%
33
9%
32
8%
29
8%
31
8%
32
8%
34
9%
32
8%
32
8%
33
9%
381
100%
Belgium
Cyprus
7
2%
44
12%
28
8%
35
10%
35
10%
32
9%
31
9%
29
8%
29
8%
31
9%
36
10%
22
6%
359
100%
Czech Republic
37
9%
37
9%
34
8%
28
7%
36
9%
36
9%
31
7%
32
8%
34
8%
35
8%
39
9%
38
9%
417
100%
Denmark
24
7%
28
8%
27
8%
27
8%
31
9%
27
8%
22
6%
23
7%
25
7%
23
7%
37
11%
50
15%
344
100%
3
1%
25
6%
39
9%
47
11%
35
8%
38
9%
34
8%
51
12%
38
9%
39
9%
64
15%
413
100%
35
8%
33
8%
33
8%
33
8%
37
9%
38
9%
36
9%
35
8%
37
9%
34
8%
36
9%
422
100%
43
11%
65
17%
51
13%
46
12%
48
12%
55
14%
39
10%
44
11%
391
100%
51
11%
73
16%
46
10%
54
12%
55
12%
57
12%
42
9%
42
9%
41
9%
461
100%
France
Ireland
35
8%
Latvia
Lithuania
Poland
27
6%
38
9%
26
6%
53
12%
34
8%
42
9%
25
6%
36
8%
33
7%
48
11%
35
8%
50
11%
447
100%
Slovenia
31
7%
34
8%
34
8%
37
8%
38
9%
39
9%
37
8%
41
9%
37
8%
37
8%
37
8%
39
9%
441
100%
Sweden
27
7%
30
7%
32
8%
32
8%
36
9%
31
8%
32
8%
34
8%
39
10%
40
10%
37
9%
32
8%
402
100%
United Kingdom
55
9%
54
8%
55
9%
53
8%
54
8%
51
8%
55
9%
52
8%
51
8%
54
8%
53
8%
54
8%
641
100%
13 - MS Group
288
5%
415
7%
370
6%
480
8%
536
9%
530
9%
515
9%
504
9%
516
9%
529
9%
514
9%
539
9%
5,736
100%
95
Table III.6
Carcass swabs - Weighted Salmonella prevalence by month of sampling in the 13-MS group with chi-square statistic.
Frequency7
Row Pct
Col Pct
Negative
Positive
Total
Month of sampling
Oct/06
Nov/06
Dec/06
Jan/07
Feb/07
Mar/07
Apr/07
May/07
Jun/07
Jul/07
Aug/07
Sep/07
Total
257
385
380
511
499
465
369
399
417
430
449
632
5,194
5%
7%
7%
10%
10%
9%
7%
8%
8%
8%
9%
12%
100%
98%
95%
93%
94%
89%
93%
90%
91%
86%
88%
88%
88%
91%
6
22
27
31
59
33
43
42
69
57
64
90
542
1%
4%
5%
6%
11%
6%
8%
8%
13%
10%
12%
17%
100%
2%
5%
7%
6%
11%
7%
10%
9%
14%
12%
12%
12%
9%
5,736
263
407
407
542
558
498
412
440
486
487
513
722
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
7
on the first line (‘Frequency’): the weighted number of samples by infection outcome and by month of sampling. Note that these are not integer values, therefore rounding errors
can lead to some discrepancies with margin totals.
on the second line (‘Row Pct’): the distribution of the samples by month of sampling, for each infection outcome separately
on the third line (‘Col Pct’): the distribution of the samples by infection outcome, for each month of sampling separately
•
Sampling quarter
Table III.7
Distribution of the number and percentage of carcass swabs by sampling
quarter, per Member State and 13-MS group.
Country
Sampling quarter
Total
1
2
3
4
Austria
140
23%
163
26%
167
27%
147
24%
617
100%
Belgium
93
24%
94
25%
97
25%
97
25%
381
100%
Cyprus
79
22%
102
28%
89
25%
89
25%
359
100%
Czech Republic
108
26%
100
24%
97
23%
112
27%
417
100%
Denmark
79
23%
85
25%
70
20%
110
32%
344
100%
France
28
7%
121
29%
123
30%
141
34%
413
100%
Ireland
103
24%
103
24%
109
26%
107
25%
422
100%
Latvia
0
0%
108
28%
145
37%
138
35%
391
100%
Lithuania
0
0%
170
37%
166
36%
125
27%
461
100%
Poland
91
20%
129
29%
94
21%
133
30%
447
100%
Slovenia
99
22%
114
26%
115
26%
113
26%
441
100%
Sweden
89
22%
99
25%
105
26%
109
27%
402
100%
United Kingdom
164
26%
158
25%
158
25%
161
25%
641
100%
13 - MS Group
1,073
19%
1,546
27%
1,535
27%
1,582
28%
5,736
100%
97
Table III.8
- Weighted frequencies of Salmonella positive/negative carcass swabs by
sampling quarter in the 13-MS group, and corresponding Chi-square
statistic.
Carcass swabs samples
Frequency8
Row Pct
1
Col Pct
Negative
1,022
Positive
Salmonella
Sampling quarter
2
3
4
Total
5,194
1,476
1,185
1,511
6%
9%
7%
9%
32%
95%
92%
89%
88%
91%
55
122
154
211
542
2%
4%
6%
8%
19%
5%
8%
11%
12%
9%
1,722
5,736
Total
1,077
1,598
1,339
8
on the first line (‘Frequency’): the weighted number of samples by infection outcome and by quarter of sampling. Note
that these are not integer values, therefore rounding errors can lead to some discrepancies with margin totals.
on the second line (‘Row Pct’): the distribution of the samples by quarter of sampling, for each infection outcome
separately
on the third line (‘Col Pct’): the distribution of the samples by infection outcome, for each quarter of sampling
separately
98
•
Hour of sampling
Table III.9
Carcass swabs – Chi-square statistic of the weighted Salmonella prevalence
by hour of sampling in the 13 – MS group.
Chi-square statistic (P-value)
97.1 (<0.001)
•
Weight of carcasses
Table III.10 Carcass swabs - Central tendency (mean, median) and dispersion measures
(standard deviation and quartiles) of the weight of the carcass, by outcome,
MS and the 13-MS group.
Carcass swabs: Carcass weight
Country
a
a
Q1
Median
Q3a
Mean
StDa
Austria
90
95
100
94.4
8.3
Belgium
86
91
100
93.5
9.3
Cyprus
63
70
70
66.3
5.1
Czech Republic
77
79
82
79.3
2.9
Denmark
78
82
88
83.0
7.6
France
87
92
99
92.8
8.5
Ireland
70
82
90
82.0
10.3
Latvia
68
72
79
74.3
9.9
Lithuania
78
87
100
87.4
12.8
Poland
80
85
100
89.3
14.1
Slovenia
77
82
90
83.5
10.5
Sweden
82
88
92
87.4
8.0
United Kingdom
74
79
84
79.1
7.9
13 - MS Group
76
84
92
84.4
12.0
Q1: 25% quantile; Q3: 75% quantile; StD: standard deviation
Table III.11 Carcass swabs - Odds ratio estimate and corresponding 95% confidence
interval obtained from a weighted logistic regression, modelling the probability of
observing a positive sample using an intercept and (separately) the weight of the carcass.
Weight of the carcass
Outcome of interest
OR
Estimate
Lower
Bound
Upper
Bound
Salmonella
1.018
1.010
1.027
99
Annex IV
Model building
IV.1 Definition of new independent variables for “month of
sampling” and “hour of sampling”
Considering month of sampling as a continuous variable would imply a linear trend, however, a
seasonal trend is expected to occur. In some periods of the year the samples could be more
affected by Salmonella than in others. On the other hand, including all months as categories of a
class variable may imply an overparameterisation of the model, especially when countries are
considered separately.
To study the effect of month of sampling in more detail, we have estimated (using communitylevel weights and considering all data over all participating countries) the prevalence of
Salmonella within each sample using a simple random-effects model including only a fixed
intercept and a random intercept for each slaughterhouse. These prevalence estimates were then
plotted by month of sampling and a loess smoothing technique was used to obtain a mean profile
over time (see Figure IV.1). When the information on all the MS are pooled there does not seem to
be any specific trend observed by month of sampling, not for LN samples nor for CS samples.
Note that these results are obtained by pooling all information, without taking into account the
effect of other possibly important covariates. A more extensive model building exercise is
necessary to confirm this result.
Figure IV.1
Month of sampling by pig-specific prevalence of Salmonella from (a) lymph node
samples and (b) carcass swab samples. Loess smoother was used to obtain average
over the pig-level prevalence estimates.
(a) Lymph Node samples
(b) Carcass Swab samples
As mentioned before, including month of sampling in the country-specific models as a
categorical variable results in many extra parameters and will complicate the model fit.
Therefore, in the country-specific model building exercise a new variable, labeled “quarter”, has
been created such that it equals 1 when the pig was sampled in the period Oct.-Dec., 2 when
sampled in the period Jan.-Mar., 3 in the period Apr.-Jun., and 4 in the period Jul.-Sept.
100
Similarly, treating the hour of sampling as a categorical variable implies a nominal variable with
24 categories, and will result in an additional 23 parameters. The effect of this variable is again
studied using a plot of pig-level prevalence estimates by the hour of sampling (see Figure IV.2).
The graph suggests a lower prevalence during the day time for LN samples (Figure IV.2a). Note
however that few observations were sampled before 5h and after 18h. The increase in the loess
smoother in this period could be explained by these possibly influential prevalence estimates. The
reverse average evolution is observed for the CS samples. In Figure IV.2(b) it is shown that
daytime samples seem to be slightly more affected by Salmonella.
Figure IV.2
Hour of sampling by pig-specific prevalence of Salmonella from (a) lymph node
samples and (b) carcass swab samples. Loess smoother was used to obtain average
over the pig-level prevalence estimates.
Another interesting evolution is shown in Figure IV.3. In this plot we have displayed the pig-level
Salmonella prevalence by the hour of sampling during a period of one week time.
Figure IV.3
Hour of sampling over a one week period (starting on Sunday, 0 am) by pig-specific
prevalence of Salmonella from (a) lymph node samples and (b) carcass swab
samples. Loess smoother was used to obtain average over the pig-level prevalence
estimates.
101
The graph starts at midnight (0 am) on Sunday and finishes at 23h (11pm) on Saturday. A sine
type of evolution can be seen in Figure IV.3(a), with peaks during the night time. A similar trend,
although less pronounced, is also observed for CS samples in Figure IV.3 (b). This trend could be
taken into account in the analysis in two ways: one can consider a sine transformation of the hour
of sampling or this effect can be modelled by considering both the weekday and the hour of
sampling as a categorical variable. These two approaches will be compared using a simple
analysis performed in NLMIXED and the corresponding AIC. Note that the AIC is an informal
selection tool based on the likelihood, while penalizing for the number of parameters in the
model. The general guideline using the AIC is “the smaller the better”.
For the first approach, the hour of sampling (on a week basis) will be transformed using the
function
where refers to the period, t refers to the time of sampling, and refers to
makes sure that the peaks of the fitted function are repeated
the phase angle. Basically,
every day, whereas the phase angle shifts the curve to fit the peaks in the observed data. We have
included this new variable into a simple random-intercept model, where the prevalence of
Salmonella is modelled using a fixed intercept and this sine-variable. Note that we have included
as an additional parameter in the model, to obtain a correct estimate for further use. Note that
this is just an exploratory analysis to help in deciding which of the two approaches could be used
in the final analysis. At this point no weights and no other important covariates are taken into
account. For the lymph nodes samples, this analysis has resulted in an AIC equal to 13784 and an
estimate for equal to 1.36. For the carcass swabs, we obtained an AIC equal to 2479.9 and an
estimate for equal to 1.72.
In the second approach, we considered another NLMIXED model, including a fixed intercept and
both the hour of sampling during the day and weekday as categorical variables in the linear
predictor. Note that to reduce the 24 categories of hour of sampling, we have created a new
variable with 4 categories: one category contains the samples taken at night, such that it contains
around 20% of the data. The remaining hours are divided over another three categories. This
categorization was different for the data based on the lymph node samples and the carcass swabs.
This analysis has resulted in an AIC equal to 13794 for the lymph node data and 2481.4 for the
carcass swabs. Although the difference between the two models is small, the AIC favours the
sine approach both for the lymph node samples and the carcass swabs, as it requires much less
additional model parameters. Therefore, in the model building we will proceed with the new
, where for the lymph node samples corresponds to 1.36 and for the
variable
carcass swabs equals 1.72. Note that a significant effect of this variable would imply that there
exists a clear seasonal trend such that day and night time results differ considerably.
102
IV.2 Analysis of multicollinearity among potential factors
IV.2.1
Lymph nodes
The VIF values resulting from the analysis studying multicollinearity among the risk factors in all
countries performing the lymph node analysis is shown in Table IV.1. Note that the countryspecific intercept was not included in this analysis to avoid separation problems. From this table
it seems that multicollinearity will not be an issue for the global models.
A similar table was also constructed for each of the countries separately. The results are shown in
Table IV.2. Note that some elements in this matrix are left empty. This is the result of the fact
that (i) no information for the variable was available, (ii) quasi-complete separation issues
occurred. These covariates will be excluded from the model. For the other variables we do not
observe any issues with multicollinearity.
Table IV.1
Lymph nodes - Variance Inflation Factors (Community-level)
Factor
VIF
4.31
Weight of LN
2.10
Number of LN
1.30
Weight of carcass
1.15
Time of sampling (sine)
1.04
Quarter
1.03
103
1.03
1.05
1.71
1.05
2.74
1.34
Belgium
1.01
1.19
1.42
1.11
2.41
1.06
Number of LN
Weight of LN
Austria
Country
Weight of carcass
Quarter
and testing
Lymph nodes - Variance Inflation Factors (Country-level)
Table IV.2
Bulgaria
1.10
1.56
5.46
1.33
3.74
1.52
Cyprus
1.11
1.31
1.26
1.15
3.61
1.17
Czech Republic
1.02
1.09
2.32
1.01
7.46
1.39
Denmark
1.02
1.05
1.48
1.03
2.89
1.01
Estonia
1.24
1.33
2.09
1.24
3.97
1.22
Finland
1.06
1.43
2.23
1.03
3.08
1.58
France
1.01
1.27
3.26
1.01
2.20
N/A
Germany
1.02
1.07
3.12
1.01
2.70
1.07
Greece
1.08
1.41
4.44
1.04
6.43
1.48
Hungary
1.05
1.95
1.52
1.08
5.08
1.00
Ireland
1.07
1.14
1.22
1.14
4.00
1.20
Italy
1.03
1.98
4.06
1.14
7.15
1.33
Latvia
1.05
1.18
3.68
2.15
3.49
2.01
Lithuania
1.09
1.18
1.93
1.12
4.00
1.14
Luxembourg
1.07
1.19
4.00
1.09
3.86
1.10
Poland
1.04
1.12
5.64
1.05
8.58
1.09
Portugal
1.02
1.08
3.83
1.01
3.00
1.01
Slovakia
1.07
1.53
2.17
1.01
3.91
1.83
Slovenia
1.01
1.70
4.37
1.03
3.25
2.80
Spain
1.12
1.17
1.27
1.22
3.23
N/A
Sweden
1.02
1.12
3.86
1.04
1.69
1.09
Netherlands
1.01
1.03
3.78
1.02
4.00
1.01
United Kingdom
1.01
1.05
1.91
1.02
1.53
1.05
Norway
1.04
1.50
3.15
1.06
3.35
1.24
104
IV.2.2
Factor
VIF
3.91
2.18
Quarter
1.04
Weight of carcass
1.01
1.01
and testing
Weight of carcass
Country
Carcass swabs – Variance Inflation Factors (Country-level).
Quarter
Table IV.4
Carcass swabs - Variance Inflation Factors (EU-level).
Time of sampling
Table IV.3
Carcass swabs
Austria
1.01
1.03
3.27
1.01
2.76
Belgium
1.03
1.21
1.92
1.13
2.36
Cyprus
1.09
1.22
1.92
1.06
3.58
Czech Republic
1.02
1.02
2.70
1.01
7.58
Denmark
1.06
1.11
2.35
1.04
3.03
France
1.02
1.31
1.55
1.02
2.18
Ireland
1.11
1.08
1.72
1.10
4.12
Latvia
1.08
1.11
2.73
1.08
3.27
Lithuania
1.03
1.03
3.34
1.03
4.00
Poland
1.02
1.06
2.45
1.03
3.84
Slovenia
1.01
1.05
2.50
1.02
6.49
Sweden
1.02
1.04
3.54
1.03
1.61
United Kingdom
1.02
1.03
1.42
1.01
1.52
105
IV.3 Complementary information on multivariable models
IV.3.1
Table IV.5
Model on lymph nodes infection with Salmonella
Lymph node infection model for Salmonella – Global model. Parameter
estimates for fixed and country-specific intercept.
Effect
Table IV.6
Estimate
SE
Intercept
-6.35
0.95
Austria
2.25
0.99
Belgium
4.20
0.96
Cyprus
4.39
0.96
Czech Republic
3.20
0.97
Denmark
3.52
0.97
Estonia
3.07
1.00
France
4.60
0.95
Germany
4.00
0.95
Greece
4.93
0.96
Hungary
3.91
0.97
Ireland
4.29
0.95
Italy
4.63
0.95
Latvia
3.23
0.98
Lithuania
2.05
1.03
Luxembourg
4.61
1.00
Poland
3.22
0.96
Portugal
5.14
0.96
Slovakia
3.45
0.99
Slovenia
3.48
0.97
Spain
5.34
0.96
Sweden
2.18
1.00
Netherlands
3.74
0.95
United Kingdom
4.74
0.95
Norway
0.00
.
Lymph node infection model for Salmonella - Estimate and standard error of
the variance of the random intercepts for slaughterhouse.
Estimate
0.313
SE
0.051
106
IV.3.2
Table IV.7
Model on Carcass surface contamination with Salmonella
Carcass contamination model for Salmonella – Global model. Parameter
estimates for fixed and country-specific intercept.
Effect
Table IV.8
Estimate
SE
Intercept
-1.99
0.32
Austria
-2.57
0.68
Belgium
0.56
0.41
Cyprus
-1.48
1.30
Czech Republic
-1.05
0.54
Denmark
-1.87
0.52
France
0.36
0.40
Ireland
0.64
0.62
Latvia
-1.90
1.26
Lithuania
-2.33
1.05
Poland
-2.69
0.46
United Kingdom
0.00
.
Carcass contamination model for Salmonella – Estimates and standard errors
of the variance of the random intercept and slope.
Variance of the
Random intercepta
a
b
Variance of the
Random slopeb
Estimate
SE
Estimate
SE
1.05
0.25
1.23
0.63
Significance random intercept: P-value ≤ 0.0001
Significance of random slope: P-value = 0.0008
107
References
Aerts, M., Geys, H., Molenberghs, G. and Ryan, L. 2002. Topics in Modelling of Clustered Data.
Chapman & Hall, London, United Kingdom.
Agresti, A. 1996. An Introduction to Categorical Data Analysis. John Wiley and Sons, New
York, USA.
Agresti, A. 2002. Categorical Data Analysis. John Wiley and Sons, Hoboken, New Jersey, USA.
Diggle, P.J., Liang, K.Y., and Zeger, S.L. 1994. Analysis of Longitudinal data. Oxford
University Press, New York, USA.
Diggle, P.J., Heagerty, P.J., Liang, K.-Y., and Zeger, S.L. 2002. Analysis of Longitudinal data
(2nd ed.). Oxford University Press, New York, USA.
Molenberghs, G., and Verbeke, G. 2005. Models for Discrete Longitudinal Data. Springer, New
York, USA.
Neter, J., Kutner, M.H., Nachtsheim, C.J., and Wasserman, W. 1996. Applied Linear Statistical
Models (4th ed.). McGraw-Hill, Blacklick, Ohio, USA.
Neuhaus, J.M 1992. Statistical methods for longitudinal and clustered designs with binary
responses. Statistical Methods in Medical Research 1, 249-273
Skellam, J.G. 1948. A probability distribution derived from the binomial distribution by
regarding the probability of success as variable between the sets of trials. Journal of the
Royal Statistical Society, series B 10, 257-261.
Stiratelli, R., Laird, N., and Ware, J. 1984. Random effects models for serial observations with
dichotomous responses. Biometrics 40,961-972
108
List of tables
Table I.1 Table notation for the cross classification of Salmonella prevalence in slaughter pigs by
month of sampling..........................................................................................................................77
Table I.2 Proxy-slaughterhouse weights to account for the disproportionate sampling in the
baseline survey on slaughter pigs...................................................................................................80
Table II.1 Lymph node samples - Distribution of the number and percentage of samples by the
weight of the lymph nodes, per Member State and in the EU. ......................................................82
Table II.2 Weighted Salmonella prevalence by the weight of the lymph nodes in the community
and corresponding chi-square statistic. ..........................................................................................83
Table II.3 Lymph node samples - Central tendency (mean, median) and dispersion measures
(standard deviation and quantiles) of the weight of the carcass, by country and the community..84
Table II.4 Lymph node samples - Odds ratio estimates and corresponding 95% confidence
intervals obtained from a weighted logistic regression, modelling the probability of observing a
positive sample using an intercept and (separately) the number of lymph nodes..........................84
number of days delay between the sampling date and the starting date of detection testing for
lymph nodes, per Member State and in the EU..............................................................................85
Table II.6 Odds ratio estimates and corresponding 95% confidence intervals from a weighted
logistic regression, modelling the probability of observing a positive sample using an intercept
and (separately) the number of days between sampling and testing in EU....................................85
number of days delay between the sampling date and the starting date of detection testing for
lymph nodes, per Member State and in the EU..............................................................................86
Table II.8 Weighted Salmonella prevalence by the number of days delay between the sampling
date and the starting date of detection testing for lymph nodes, per Member State and in the EU –
with chi-square statistic..................................................................................................................87
Table II.9 Lymph node samples - Distribution of the number and percentage of pigs by month of
sampling. ........................................................................................................................................88
Table II.10 Lymph node samples - Weighted Salmonella prevalence by month of sampling in the
EU with corresponding chi-square statistic....................................................................................89
Table II.11 Lymph node samples - Distribution of the number and percentage of pigs by quarter
of sampling.....................................................................................................................................90
Table II.12 Lymph node samples - Weighted Salmonella prevalence by quarter of sampling in the
EU with chi-square statistic............................................................................................................91
Table II.13 Lymph node samples - Chi-square statistic of the weighted Salmonella prevalence by
hour of sampling in the EU. ...........................................................................................................91
Table II.14 Lymph node samples - Central tendency (mean, median) and dispersion measures
(standard deviation and quantiles) of the weight of the carcass, country and the community.......92
109
Table II.15 Lymph node samples - Odds ratio estimates and corresponding 95% confidence
intervals obtained from a weighted logistic regression, modelling the probability of observing a
positive sample using an intercept and (separately) the weight of the carcass. .............................92
Table III.1 Carcass swabs - Distribution of the number and percentage of samples by the number
of days delay between the sampling date and the starting date of detection testing for carcass
swabs, per Member State and in the 13 – MS group......................................................................93
Table III.2 Carcass swabs - Odds ratio estimates and corresponding 95% confidence intervals
from a weighted logistic regression, modelling the probability of observing a positive sample
using an intercept and (separately) the number of days between sampling and testing in 13 – MS
group...............................................................................................................................................93
Table III.3 Carcass swabs - Distribution of the number and percentage of samples by the number
of days delay between the sampling date and the starting date of detection testing for carcass
swabs, per Member State and in the 13 – MS group......................................................................94
Table III.4 Weighted Salmonella prevalence by the number of days delay between the sampling
date and the starting date of detection testing for carcass swabs, per Member State and in the 13 –
MS group, with chi-square statistic................................................................................................94
Table III.5 Carcass swabs - Distribution of the number and percentage of pigs by month of
sampling, per Member State and 13 – MS group...........................................................................95
Table III.6 Carcass swabs - Weighted Salmonella prevalence by month of sampling in the 13 –
MS group with chi-square statistic.................................................................................................96
Table III.7 Carcass swabs - Distribution of the number and percentage of pigs by quarter of
sampling, per Member State and 13 – MS group...........................................................................97
Table III.8 Carcass swabs - Weighted Salmonella prevalence by quarter of sampling in the 13 –
MS group........................................................................................................................................98
Table III.9 Carcass swabs – Chi-square statistic of the weighted Salmonella prevalence by hour
of sampling in the 13 – MS group..................................................................................................99
Table III.10 Carcass swabs - Central tendency (mean, median) and dispersion measures (standard
deviation and quantiles) of the weight of the carcass, by outcome, MS and the 13 – MS group. .99
Table III.11 Carcass swabs - Odds ratio estimates and corresponding 95% confidence intervals
obtained from a weighted logistic regression, modelling the probability of observing a positive
sample using an intercept and (separately) the weight of the carcass............................................99
Table IV.1 Lymph nodes - Variance Inflation Factors (Community-level) ................................103
Table IV.2 Lymph nodes - Variance Inflation Factors (Country-level) ......................................104
Table IV.3 Carcass swabs - Variance Inflation Factors (EU-level). ............................................105
Table IV.4 Carcass swabs – Variance Inflation Factors (Country-level). ...................................105
Table IV.5 Lymph node samples – Global model. Parameter estimates for fixed and countryspecific intercept. .........................................................................................................................106
Table IV.6 Lymph node samples - Estimates and standard errors of the variance of the random
intercepts for slaughterhouse........................................................................................................106
Table IV.7 Carcass swab samples – Global model. Parameter estimates for fixed and countryspecific intercept. .........................................................................................................................107
110
Table IV.8 Carcass swab samples – Estimates and standard errors of the variance of the random
intercepts. .....................................................................................................................................107
List of Figures
Figure IV.1 Month of sampling by pig-specific prevalence of Salmonella from (a) lymph node
samples and (b) carcass swab samples. Loess smoother was used to obtain average over the piglevel prevalence estimates............................................................................................................100
Figure IV.2 Hour of sampling by pig-specific prevalence of Salmonella from (a) lymph node
samples and (b) carcass swab samples. Loess smoother was used to obtain average over the piglevel prevalence estimates............................................................................................................101
Figure IV.3 Hour of sampling over a one week period (starting on Sunday, 0 am) by pig-specific
prevalence of Salmonella from (a) lymph node samples and (b) carcass swab samples. Loess
smoother was used to obtain average over the pig-level prevalence estimates. ..........................101
111

Analysis of the baseline survey on the prevalence of Salmonella in

Transcription

Similar documents

Salmonella - European Food Safety Authority

Diagnostic Laboratory Updates — Pennsylvania Animal Diagnostic

6734-2 CS6000 Leaflet

Introduction Legislation Training Inspection

Chromosomal rearrangements in Salmonella spp. s2-2

Is it Just a Swollen Node?

South Towne Expo Center - The Great Salt Lake Business Conference

South Towne Expo Center and Salt Palace

4. Open up the stomach cavity. 5. Remove the stomach and

Carcass handling - Wild Boar Best Practice Guides website