Risk Factors for Salmonella Infection in Fattening Pigs – An Evaluation of Blood and Meat Juice Samples
S. Hotes. Institute of Animal Breeding and Husbandry, Christian-Albrechts-University, Olshausenstraße 40, D-24098 Kiel, Germany.
Tel.: +49 4318 802 572;
Fax: +49 4318 802 588;
The main objective of this study was to analyse potential herd-level factors associated with the detection of Salmonella antibodies in fattening pigs. Two independent datasets, consisting of blood and meat juice samples respectively, were used. Additional information about husbandry, management and hygiene conditions was collected by questionnaire for both datasets. The serological analysis showed that 13.8% of the blood samples and 15.7% of the meat juice samples had to be classified as Salmonella-positive. Logistic-regression models were used to assess statistically significant risk factors associated with a positive sample result. The results of the statistical blood sample analysis showed that the application of antibiotics increased the odds ratio (OR) by a factor of 5.21 (P < 0.001) compared to untreated pigs. A fully slatted floor decreased the prevalence of Salmonella as well as the use of protective clothing or the cleaning of the feed tube (ORs 0.35–0.54, P < 0.001). It was shown that a distance of less than 2 km to other swine herds increased the chance of a positive Salmonella result (OR = 3.76, P < 0.001). The statistical analysis of the meat juice samples revealed the importance of feed aspects. The chance of obtaining a positive meat juice sample increased by a factor of 3.52 (P < 0.001) by using granulated feed instead of flour. It also became clear that liquid feeding should be preferred to dry feeding (OR = 0.33, P < 0.001). A comparison of the blood sample analysis to the meat juice model revealed that the latter was less powerful because data structure was less detailed. The expansion of data acquisition might solve these problems and improve the suitability of QS monitoring data for risk factor analyses.
Infections with Salmonella are one of the most important sources of food-borne diseases.
With 40 000–50 000 reported human illnesses every year (Robert Koch Institute, 2007, 2008, 2009) Salmonella still represents a major public health problem in Germany. Pork is one of the most frequent sources of infection. About 20% of human salmonellosis are associated with contaminated pork products (Steinbach and Kroell, 1999). The European Food Safety Authority (2008) reported that the prevalence of slaughtered pigs infected with Salmonella is 10.9% for Germany, a little higher than the average for the European Union. The European regulation for the control of Salmonella and other specified food-borne zoonotic agents (EC No 2160/2003) from 2003 states that ‘The protection of human health against diseases and infections transmissible directly or indirectly between animals and humans (zoonoses) is of paramount importance’. Furthermore, it regulates that national control programmes have to be established, which provide the detection of zoonoses and specify control measures. Such a control programme has existed in Germany on a voluntary basis for some time. The QS Qualität und Sicherheit GmbH was founded in 2001 to create a voluntary basis for a system of proven quality assurance (QS Qualität und Sicherheit GmbH, 2009). This so-called QS system or QS monitoring was developed for meat and meat products and met all statutory requirements. The system distinguishes between three risk categories of fattening farms:
- I. Farms with a herd prevalence of less than 20%;
- II. Farms with a herd prevalence of between 20% and 40% and
- III. Farms with herd prevalence of more than 40%.
The classification is determined by a certain number of samples. The required number of samples per year depends on the quantity of delivered market hogs. Farms producing more than 200 hogs per year have to examine 60 pigs or carcasses. In the majority of cases, the sampling takes place at the slaughterhouse in the form of meat juice samples. The monitoring system is based on an enzyme-linked immunosorbent assay (ELISA) for the detection of Salmonella antibodies. Results are given as optical density % (OD%). A sample is regarded as Salmonella-positive if the cut-off of OD% 40 is exceeded. The data on sample taking and the results have to be collected in a special Salmonella database. Fatteners belonging to Category III are compelled to identify the sources of Salmonella and to establish targeted measures to reduce the prevalence. Thus, the knowledge of risk factors associated with exposure to Salmonella is essential.
The aim of the study was to identify risk factors associated with the detection of Salmonella antibodies in fattening pigs. Two datasets, blood samples and meat juice samples, were analysed by a logistic regression. A comparison of the analyses should emphasise the strengths and weaknesses of the respective type of data. Recommendations were made for further analysis and data collection.
Materials and Methods
Collection of blood samples
The data for the blood sample dataset came from a previous project, within which more than 4 200 sows and fattening pigs were sampled (Meyer, 2004). The data collection took place between March 2001 and April 2002 and was supported by the ZNVG (Vermarktungsgemeinschaft für Zucht- und Nutzvieh). The ZNVG is one of the largest production associations of Schleswig-Holstein with a market share of 40%. All sampled farms are members.
In the current study, data from 32 conventional fattening farms including 59 fattening barns was used. Statistical analysis was carried out at barn level considering the proportion of positive samples per barn. The investigated farms were located throughout Schleswig-Holstein. Farms were chosen using a stratified random sample based on spatial distribution. Within the strata, the farms were selected by a lottery system. The median number of fattening pigs was 965, with a minimum of 272 animals and a maximum of 2810 animals. The numbers of examined pigs differed between farms.
The calculation of Noordhuizen et al. (1997) was used to ensure that the sampling enabled the estimation of Salmonella prevalence at herd level. The expected prevalence was set to 20%, the absolute accuracy to 10%, and the probability of error to 5% (Meyer, 2004). The calculated number of blood samples varied between 50 and 65 per farm. Pigs of all barns were sampled. Several compartments were considered within each barn. The pig sampling itself was randomised across pens and each animal was tested only once. The producers had no prior information about the prevalence of Salmonella in fattening pigs. QS categorisation had already become an issue but was not started until 2003. Therefore, the farmers had great interest in the project and no-one refused participation. Thousand eight hundred and thirty-six fattening pigs were sampled in total.
Analysis of blood samples
The blood samples were analysed for antibodies against Salmonella O-antigens with the SALMOTYPE® meat juice ELISA. This ELISA is based on the detection of Salmonella O-antigens 1, 4, 5, 6, 7 and 12. Samples with an OD% higher than 40 were considered positive (Meyer, 2004).
Questionnaire for the blood sample dataset
A survey was carried out to determine possible relations between the herd-level prevalence and husbandry practices at fattening. The veterinarian interviewed the farmer during the visit and completed the questionnaire for each fattening barn. The questionnaire included 56 questions on farm size, pig purchase and housing system, housing conditions, manure storage, drinking and feeding practice, pest control, cleaning and disinfection procedures, environmental circumstances as well as hygiene and health aspects (questionnaire is available from the corresponding author). Most of the questions were designed as multiple-choice to obtain answers as concise as possible.
Collection of meat juice samples
The information on the meat juice samples came from the QS database of the ZNVG. The selection of the farms was determined by the participation in the meat juice survey and a sufficient reply to the questions. All QS monitoring samples, taken between July 2007 and December 2008, were considered and the proportion of positive samples was calculated for every farm. The final meat juice dataset contained 4 204 sample records from 37 fattening and farrow-to-finishing farms from Schleswig-Holstein.
Questionnaire for the meat juice sample dataset
The data on the management, husbandry and hygiene aspects were obtained by the same questions as described above. Unlike the blood sample survey, each farm participated in only one questionnaire; regardless of whether they had more fattening barns. The analysis had to be accomplished at farm level because a sample, taken at the slaughterhouse, could not be traced back to the barn, just to the farm. The producers were to state the most common practice. The questionnaire was sent by post dispatched by the ZNVG. Due to the fact that only 14 of 88 farm managers answered, the remaining farms were called to complete the questionnaire with the producer or an appropriate person. At the end of the survey, 67 farmers had participated in the study, but only 37 could be considered for analysis. Fourteen farms were excluded because the producers did not complete the questionnaire sufficiently. Additionally, 13 farms were not taken into account because they worked with a continuous-flow system. These farms were neglected due to inconsistent answers throughout the survey. The producers stated that they worked with a continuous-flow system but used the given answers for all-in all-out production without satisfying specification. This problem did not appear for the blood sample dataset when the questionnaire was carried out on-site. One farm was neglected because the regular sample size of 60 samples was not achieved. Furthermore, two farms were excluded because they used straw as bedding material.
Due to the high number of excluded farms, the farrow-to-finishing farms were left in the dataset. Finally, the meat juice dataset considered 26 fattening farms and 11 farrow-to-finishing farms. The number of fattening pigs varied between 400 and 2500 per farm with a median of 950 animals.
The same statistical methods were applied to both datasets. Following Nyman et al. (2007), continuous covariates were categorised using the quartiles or biologically important values as cut-points. The distribution of the original categorical covariates was also checked. Categories with too few observations were pooled when the new classification made biological or logical sense. Otherwise, this covariate could not be considered in further analysis (Nyman et al., 2007). Finally, 19 potential risk factors for the blood sample dataset and 17 for the meat juice dataset remained (Table 1). The blood samples were analysed at barn level. Compartments within the same barn but with heterogeneous equipment or different feeding systems were considered as autonomous barns. Analyses of the blood and meat juice samples were carried out with the proportion of positive samples as outcome variable.
Table 1. Potential risk factors of the blood sample dataset and meat juice sample dataset
|Common potential risk factors||Number of fattening pigs||≤350||Number of fattening pigs||≤750|
|> 1000|| ||>1500|
|Floor||Fully slatted floor||Floor||Fully slatted floor|
|Partly slatted floor|| ||Partly slatted floor|
|Changing room||Yes||Changing room||Yes|
|Pest occurrence||Few rodents and no birds||Pest occurrence||Few rodents and no birds|
|Increased rodents and birds|| ||Increased rodents and birds|
|Feeding system||Trough feeding||Feeding system||Dry feeding|
|Liquid feeding|| ||Liquid feeding|
|Mash feeding|| ||Mash feeding|
| || ||‘Mix’ of several systems|
|Feed structure||Granulate||Feed structure||Granulate|
| || ||‘Mix’ of several structures|
|Acidification of feed or water||Yes||Acidification of feed or water||Yes|
|Application of antibiotics||Yes||Application of antibiotics||Yes|
|Housing of diseased animals||Special compartment/barn||Housing of diseased animals||Special compartment/barn|
|Special pen|| ||Special pen|
|No special housing|| || |
|Carcass disposal||Disposal from yard||Carcass disposal||Disposal from yard|
|Disposal from road|| ||Disposal from road|
|Varying potential risk factors||Pigs suppliers||Established pigs suppliers||Farm type||Finishing farm|
|Varying pigs suppliers|| ||Farrow-to-finishing farm|
|Pen partition||Latticed pen partition||Number of fattening barns||1 fattening unit|
|Closed pen partition|| ||2 fattening units|
|Protective clothing||Yes|| ||3 fattening units|
|No|| ||4 fattening units|
|Pets in barn||Yes||Number of persons with access||1 person|
|No||to barn(s) (vet excluded)||2 persons|
|Cleaning feed tube||Regularly|| ||>2 persons|
|Sometimes||Care of foreign livestock||Yes|
|Cleaning walls||Regularly||Cleaning ventilation||Regularly|
|Cleaning boots||Regularly/sometimes||Water origin||Well water|
|Never|| ||Municipal water|
|Feed origin||Solely bought||Proximity to sewage||Yes|
|Solely/mainly own crop || ||No|
|Proximity to other swine||Closer than 2 km|| || |
|Herds||Further away than 2 km|| || |
A random farm or barn effect was assessed neither for the blood sample model nor for the meat juice model. Both are observational studies; without an orthogonal study design risk factors and random effect would be confounded. Hence, fitting the random effect would have involved the risk of leaving significant covariates undetected. To account for the ubiquitousness of Salmonella and its importance in public health, the risk of mistaking a factor for significant seemed to be justified.
Collinearity between all risk factors was analysed pair-wise using a χ²-test of independence. In cases where the expected cell frequencies were so small that the χ²-test may not have been valid, Fisher’s exact test was additionally calculated to ensure the result. The effect size was assessed by calculation of the Phi Coefficient or Cramer’s V, depending on whether both variables had only two categories or at least one had more. The cut-off for a remarkable collinearity was set to a value higher than 0.80. No variable combination showed such an effect size – neither in the blood sample variables, nor the meat juice data.
Logistic regression models were fitted using automated stepwise selection procedure (sas, proc logistic). The selection adds or removes a covariate to the model based on the χ² score considering a significance level of P < 0.05. The selection process terminates if no further covariate can be added to the model or the covariate admitted is the only one excluded in the subsequent step (SAS Institute Inc., 2004). This procedure was additionally applied to several, slightly changed sets of potential risk factors to assess the importance of the selected covariates for the final model.
The goodness of fit of the final models was evaluated by Nagelkerke’s R² and by visual examination of the standardised deviance residuals plotted against the linear predictor (Collett, 2003).
The serological analysis showed that 13.8% of the blood samples and 15.7% of the meat juice samples had to be classified as Salmonella-positive. The QS classification, based on the respective herd prevalence, is shown in Table 2. The relative occupancy of the risk categories is very similar, comparing the frequencies for the blood and meat juice data.
Table 2. QS categorisation based on the serological results
|Category I||≤ 20%||23||71.88||25||67.57|
|Category II||> 20% and ≤ 40%||7||21.88||9||24.32|
|Category III||> 40%||2||6.25||3||8.11|
For the blood sample dataset, the model selection detected the covariates ‘Floor’, ‘Pest occurrence’, ‘Application of antibiotics’, ‘Pen partition’, ‘Protective clothing’, ‘Cleaning feed tube’ and ‘Proximity to other swine herds’ as significant independent from changes in the set of potential risk factors. The residual plot of the blood sample estimation did not show any trend. Nagelkerke’s R² reached a value of 19.7, indicating that the model was able to explain about one fifth of the variance in Salmonella prevalence.
For the meat juice model the covariates ‘Feeding system’, ‘Feed structure’ and ‘Acidification of feed or water’ were significant independent from changes in the set of potential risk factors. Furthermore, the covariates ‘Application of antibiotics’; ‘Number of fattening barns’ and ‘Cleaning ventilation’ were significant except the risk factor ‘Housing of diseased animals’ was excluded from the set of potential risk factors. On the one hand this might be due to the high amount of missing values in the covariate ‘Housing of diseased animals’ or on the other hand to multicolliniarity among the four covariates. Consequently, the estimates of the covariates ‘Application of antibiotics’; ‘Number of fattening barns’ and ‘Cleaning ventilation’ were biased. The residuals of the meat juice model were distributed randomly but the explanatory power of the model was worse than for the blood sample estimation. Nagelkerke’s R² reached a value of 12.0%.
Risk factor analyses
The results of the logistic regression models are shown in Table 3. Due to missing answers for the risk factors considered, the estimations were based on 55 barns (out of 59 barns) for the blood sample analysis and 29 farms (out of 37 farms) for the meat juice analysis.
Table 3. Most important risk factors associated with sero-positivity for Salmonella
|Blood sample dataset||Floor||Fully slatted floor||0.51||0.35, 0.74||0.0005|
|Partly slatted floor||1||–|
|Pest occurrence in barn(s)||Few rodents and no birds||3.04||2.02, 4.56||<0.0001|
|Increased rodents and birds||1||–|
|Application of antibiotics||Yes||5.21||3.51, 7.73||<0.0001|
|Pen partition||Latticed pen partition||0.56||0.40, 0.79||0.0010|
|Closed pen partition||1||–|
|Protective clothing||Yes||0.54||0.38, 0.77||0.0007|
|Cleaning feed tube||Regularly||0.40a||0.27, 0.60||< 0.0001|
|Proximity to other swine herds||Closer than 2 km||3.76||2.47, 5.73||< 0.0001|
|Further away than 2 km||1||–|
|Meat juice sample dataset||Feeding system||Liquid feeding||0.33a||0.21, 0.53||< 0.0001|
|Mash feeding||1.21b||0.86, 1.69|
|‘Mix’ of several systems||2.28c||1.37, 3.79|
|Feed structure||Granulate||3.52a||2.23, 5.55||< 0.0001|
|‘Mix’ of several structures||3.19a||1.85, 5.50|
|Acidification of feed or water||Yes||1.80||1.30, 2.49||0.0004|
|Application of antibiotics||Yes||0.72||0.56, 0.92||0.0080|
|Number of fattening barns||1 fattening unit||2.20a||1.49, 3.24||< 0.0001|
|2 fattening units||1.15b||0.73, 1.81|
|3 fattening units||2.21a||1.41, 3.47|
|4 fattening units||1||–|
|Cleaning ventilation||Regularly||0.99a||0.74, 1.32||0.0001|
Results for the blood sample dataset showed that the Odds Ratio (OR) for the application of antibiotics was 5.21 (see Table 3). Accordingly, pigs treated with antibiotics had a five times higher chance of being tested positive than untreated pigs. Furthermore, the proximity to other swine herds increased the chance of obtaining a positive blood sample. However, a fully slatted floor, protective clothing and the – even irregular – cleaning of the feed tube reduced the chance of a positively tested pig. The results suggest that a latticed partition between pens and the increasing number of rodents and birds in the barn may decrease the chance of a positive result, too.
Estimated ORs for the meat juice model revealed that liquid feeding should be preferred to a dry feeding system (Table 3). Moreover, the feeding of flour decreased the chance of a positive meat juice sample compared to a granulated feed structure. Estimations for the ‘mix’ groups cannot be interpreted. They comprised all farms with more than one feeding system or different feed structures. The acidification of feed or water showed a significant impact on the sample results. As described above, the remaining effects seemed to be biased.
The blood and meat juice herd prevalence investigated was 13.8% and 15.7% respectively. Tenhagen et al. (2009) reported a Salmonella prevalence of 13.5% of lymph nodes for German pigs. Studies for Schleswig-Holstein are rare. The Federal Institute of Risk Assessment (Bundesinstitut für Risikobewertung, 2008) quantified meat juice prevalence in Schleswig-Holstein at 23.6%. This value is much higher than detected in the present study, but could be explained by the use of a cut-off of OD% 20 instead of OD% 40.
The comparability of blood and meat juice results, as proved by Szabó et al. (2008) or Nielsen et al. (1998), was an important assumption for the present study. Against this background, Table 2 suggests that Salmonella herd prevalence did not obviously change between 2001/2002 and 2007/2008. Even more important seems to be the detection of risk factors associated with Salmonella in fattening pigs.
The number of farms available for the blood sample analysis and the meat juice analysis were 32 and 37, respectively. The farm selection for the blood sample dataset was randomised with regard to spatial distribution. Sampling was carried out in the year 2000/2001. Due to the fact that the introduction of obligatory Salmonella monitoring was imminent, fatteners were interested in their herd prevalence and all contacted farmers participated. Six years later, when the meat juice survey started, only 14 of 88 producers answered at the first onset.
The present meat juice survey revealed the problem of obtaining satisfying and complete answers about the current farm situation. Without any doubt, the questionnaire performed as a telephone survey or sent by post was unfavourable in obtaining precise answers to all questions. A personal interview in the form of a farm visit might have improved the willingness of the farmers to participate but would have been expensive and time-consuming. Regrettably, it was not possible for the advisors of the ZNVG to carry out the survey during their regular farm visits. Due to the fact that the production of QS-distinguished pork requires regular compliance audits, a simultaneous collection of farm data by the QS delegate might be a possibility to receive a more detailed and convincing database. Such a continuous acquisition enables long-term analysis as well. However, the question of how to deal with mixed systems at a particular farm still remains difficult. Data collection at barn level, including a certain traceability of sample results from slaughterhouse to barn, would be hard to implement. But the results of the meat juice model emphasised the difficulties of mixed groups. A meaningful interpretation of these groups is thus not possible.
Both blood and meat juice models were fitted without a random barn or farm effect because data came from an observational study. Consequently, selection of farms was not based on an orthogonal study design. Hence, random effect and risk factors might be confounded. Hosmer and Lemeshow (2000) referred to a confounded relationship between risk factor and outcome variable if a covariate is associated with both the outcome variable and an independent variable or risk factor. Brill and Barbone (2004) pointed out that confounding is a major threat to validity. For the present study, the fitting of the random effect had involved the risk of leaving significant covariates undetected. On the other hand, some risk factors might be mistaken as significant. Preliminary analysis considering a random effect in the model confirmed the presented approach: ORs and confidence limits were almost identical and risk factors remained significant.
Risk factor analyses
In accordance with other studies (Davies et al., 1997; Nollet et al., 2004; Vonnahme et al., 2008) the results showed that a fully slatted floor is associated with a decreased risk of seropositivity for Salmonella: contaminated faeces flow away much faster and have a minor chance of infecting susceptible pigs in the pen. Regular cleaning of the feed tube prevents the settlement and growth of Salmonella and decreases the risk of seropositivity as well. A preventive effect was also revealed for protective clothing and a great distance to other swine herds. These effects reduce the possibilities of Salmonella entrance and thereby decrease Salmonella prevalence. Pest control is supposed to work in a similar manner (Funk and Gebreyes, 2004) but the estimated OR did not show this. An increasing number of pests in the barn was associated with Salmonella decrease. Farzan et al. (2006) was also unable to prove a Salmonella-increasing effect caused by rodents. The difficulties might have been due to the subjective assessment. The awareness and sensitivity towards rodents or birds in the barn is certainly different between farm managers and consequently their answers did not underlie the same magnitude of impartiality. In contrast to Lo Fo Wong et al. (2004), the analysis could not point out that pigs which were able to have snout contact with neighboured pigs because of a latticed or low pen separation had a higher chance of being tested sero-positive. Conversely, we observed a protective effect of more open pen partitions. This result is doubtful because more contacts increase the chance of an infectious contact. However it has to be considered that nose-nose transmission is less frequently the reason for infection than faecal-oral transmission (Schwartz, 1999).
In line with van der Wolf et al. (2001), the application of antibiotics was associated with positive testing. The estimation showed a more than five times higher chance of a Salmonella-positive blood sample. This effect can be explained by the disturbance of the endogenous flora resulting in decreased colonisation resistance which in turn reduces the minimal infectious or colonisation doses (Nurmi et al., 1992; van den Bogaard and Stobberingh, 1999).
Several studies reported a Salmonella-preventive effect of liquid feeding or a prevalence-increasing impact of dry feeding systems respectively (van der Wolf et al., 2001; Bahnson et al., 2006; Farzan et al., 2006; Benschop et al., 2008). These relations also became clear in the meat juice analysis executed. Furthermore, it could be shown that flour decreased the amount of positive samples compared to granulated feed but not by contrast with pellets. The insignificant difference between pellets and flour is in line with Jørgensen et al. (2001) but in contrast with Hansen et al. (2001), who was able to prove a significant difference between pellets and meal feed. The positive effect of meal was explained by the more coherent structure and a lesser tendency towards phase separation. These properties may have stimulated lactobacilli and caused a high concentration of organic acids in the gastric content. The resulting decrease in pH deteriorates the conditions for Salmonella. Accordingly, Jørgensen et al. (2001) reported a relative risk of 2.7 for Salmonella in pens without acidification of feed compared to pens in which acid was used. This result seems to be in contrast with present examinations but it has to be considered that the use of acid is a common recommendation for Salmonella prevention in Germany. In the years 2001/2002, acid was used on 11 finishing farms of the blood sample dataset. All farms were low-risk farms (prevalence smaller than 20%), but one was classed as Category II with a herd prevalence of 20.7%. In comparison, six years later, 17 farms of the meat juice dataset used acid in feed or water. Six of them were Category II farms and one had a high risk for Salmonella infection with a herd prevalence of 61.2%. It seems that especially fatteners with high herd prevalence use supplemental acid to achieve Salmonella reduction. Another explanation for the association between acid additives and high Salmonella prevalence might be the generation of acid tolerance. Foster (1995) described the adaptation of Salmonella typhimurium to acidity below pH 4.0 if the organisms were first adapted to a moderate acid pH. For the present study, time series data on the supplementation of acid would be necessary to ascertain whether the use of additives was a reaction to an increasing prevalence and whether Salmonella decrease was actually achieved.
Concluding the present study, important risk factors associated with Salmonella in fattening pigs were detected. The analysis of the blood and meat juice samples demonstrated especially the importance of hygienic principles and feeding aspects, respectively. Finally, capabilities to improve data acquisition for risk factors analyses were shown. The obligatory Salmonella monitoring system offers enormous information on Salmonella prevalences in German fattening pigs. If this information could be connected with farm data more precisely, risk factor analyses would be more comprehensive and convincing.