| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
,2
* Department of Large Animal Sciences, Faculty of Life Sciences, University of Copenhagen, Groennegaardsvej 8, DK-1870 Frederiksberg C, Denmark
Department of Population Medicine and Diagnostic Sciences, College of Veterinary Medicine, Cornell University, Ithaca, NY
International EpiLab, National Veterinary Institute, Danish Technical University, Copenhagen, Denmark
1 Corresponding author: lrn{at}life.ku.dk
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Key Words: risk factor • Salmonella • surveillance program dairy cattle
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
In the surveillance program, all Danish cattle herds are classified into 3 infection categories. Dairy herds are classified based on bulk tank milk (BTM) samples that are tested using an ELISA based on an S. Dublin lipopolysaccharide antigen (O:9,12; Hoorfar et al., 1995). Other serotypes than S. Dublin may cross-react with this antigen because of O-antigens on the bacterial surface that are common to those carried by S. Dublin [e.g., O:12 in S. Typhimurium (Konrad et al., 1994; Hoorfar et al., 1995)]. Thus, the term Salmonella in this study implies S. Dublin or other serotypes that occur in cattle and can cross-react with the antigen used in the ELISA in the surveillance program. In Denmark, most of these would be S. Typhimurium (Anonymous, 2004).
For dairy herds, a Salmonella category is determined from the 4 most recent ELISA results. Samples are collected approximately every 3 mo, but extra samples at other intervals may also be requested by farmers. There has to be at least 3 wk and no more than 5 mo between each sample to be used for the Salmonella category determinations. The average proportion of background-corrected optical density value of the sample to a known positive control sample (ODC%) is calculated from these 4 measurements.
Herds not producing milk are classified based on ELISA measurements on blood samples that are collected routinely for evaluation of the herd status of bovine virus diarrhea or voluntarily submitted for analysis from animals above 8 mo of age (Nielsen et al., 2003).
Category (C) 1 is considered most likely free of Salmonella. If enough blood samples have been collected and all are below the cutoff of 50 ODC%, herds not producing milk are classified C1. Dairy herds are classified C1 if the average ODC% of the last 4 BTM measurements is below 25 and no increase of more than 20 ODC% is found when comparing the last measurement to the average of the 3 previous measurements. Until March 2006, C2 and C3 were divided into 2 sublevels. Sublevel C2a was considered likely to be infected because of antibody responses above the cutoff values. Sublevel C2b was not classifiable because of lack of data or because of contact to herds in C2a or C2b. Herds in C3a had clinical salmonellosis diagnosed by bacteriological culture (usually outbreak herds). Sublevel C3b had the bacteria detected by culture, but clinical salmonellosis had not been diagnosed or the herd had purchased cattle from a C3a or C3b herd. Laboratory results and classification categories are recorded in a database that is part of the Danish Cattle Database. The systematically collected data on Salmonella antibodies in all Danish dairy herds provides a unique opportunity to evaluate risk factors for the infection at herd level.
The use of BTM ELISA for herd classification of dairy herds was first suggested by Hoorfar et al. (1994) who recommended the use of this test for screening and certification programs. In a small study (n = 160 BTM samples) it was found that there was a good association between herd history of salmonellosis, herd location, and clinical status of the herd and the BTM ELISA response (Hoorfar et al., 1995). In the Danish surveillance program, the BTM ELISA measurements of 4 samples are used for the classification, and the validity of the program testing scheme in dairy herds was evaluated on a large scale using a simulation model based on field study data from dairy herds known to be infected with S. Dublin and S. Typhimurium (Warnick et al., 2006). This study found good validity of the classification (sensitivity = 95%, specificity = 96%, negative predictive value = 99%, and positive predictive value = 80%) at a national prevalence of 15% truly infected dairy herds. Infection was defined for that study as having at least one culture-positive fecal sample from cattle or a within-herd prevalence of at least 5% based on individual animal ELISA results.
Risk factors for Salmonella infection in cattle at the herd level can be evaluated from 2 perspectives: 1) the risk of introduction of the infection to a previously uninfected herd; and 2) the risk of currently being infected with Salmonella bacteria. The introduction has been shown to be influenced by management practices of the herd and location of the herd. Trade of living animals, grazing with cattle from other farms, and low level of biosecurity for professional visitors have been found as significant risk factors for introduction of infectious organisms to the herd (van Schaik et al., 2002). In a study of 1,429 Danish dairy herds, the risk of becoming infected with Salmonella (measured as a change from negative to positive BTM ELISA response) increased when nearby neighbors were infected. Also, the risk increased with the prevalence of seropositive herds in the region (Wedderkopp et al., 2001). Pastures recently contaminated with infected slurry are also an important risk for new infection (Taylor and Burrows, 1971; Taylor, 1973).
The aim of our study was to identify factors influencing the risk of dairy herds changing test status as a measure of Salmonella infection or recovery. Salmonella serotypes of primary interest were S. Dublin, S. Typhimurium or other serotypes that may cross-react with the S. Dublin-antigen used in the BTM ELISA testing scheme in the Danish surveillance program.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
National Surveillance Program Data
For the objective of this study, we used the ELISA ODC% results from the National Surveillance Program database to recalculate Salmonella test results for all dairy herds. The test program validity and the relationship between BTM ELISA measurements and individual cow antibodies were described elsewhere (Nielsen and Ersbøll, 2005; Warnick et al., 2006). In another study, it was shown that mainly serogroup B-serotypes such as S. Typhimurium may cross-react with a S. Dublin antigen because of common lipopolysaccharide O-antigens 1 and 12 on the cell surface (Konrad et al., 1994). The herds were test negative if the 4-BTM moving average was ODC% <25, and no increase of >20 ODC% was found when comparing the most recent measurement to the average of the 3 previous measurements. Test-positive dairy herds included those with BTM ELISA results that exceeded either of the ELISA test cut-off criteria described above. Thus, all dairy herds were denoted either test positive or test negative for each sampling date. The actual regulatory categories used in the surveillance program (C1, C2a, C2b, C3a, and C3b) were not used to define the response variables for these analyses, but C1 would be similar to test negative and C2a to test positive. Very few dairy herds were assigned C2b, C3a, and C3b in the surveillance program, and these herds were grouped according to their antibodies into test negative or positive.
For the analyses, all herds had a Salmonella test result assigned to each year-quarter (YQ) based on the last 4 consecutive BTM samples. If a herd was classified more than once during the same YQ, because more than one BTM were collected, the test result for that YQ was selected randomly. The full data set contained 70,871 data lines from 8,694 dairy herds from the period October 2001 to March 2004, and the herds had their Salmonella test result determined for between 1 and 11 YQ with a median of 9 and Q1 to Q3 of 8 to 10. Potential risk factors were constructed as either time dependent with measurements on a quarterly basis or as a onetime recording representing the entire sampling period. For several of the variables, however, data were only available for a limited period as described in Data Editing and Descriptive Statistics.
Herds not producing milk were grouped based on blood samples collected in the herds or at slaughter. The surveillance program in herds not producing milk had changed several times in the period 2002–2004 and was most likely of varying accuracy compared with the program for dairy herds. Therefore, data from these herds were only used to construct variables concerning characteristics of neighbor. They were not included as study herds as such. Overall, the apparent prevalence of Salmonella infection in cattle herds not producing milk was around 1.5%.
Study Herds
All dairy herds were included in the study regardless of whether they ceased operations during the period from which data were extracted (2001 to 2004). The BTM measurements were available starting in 2001 when data collection was initiated. Dairy herds were defined as herds that had weekly SCC measured from BTM as part of a compulsory milk quality control program because this definition gives the most updated information on which herds were truly milk-producing herds. Out of the 8,694 dairy herds with adequate samples, 1,007 (11.6%) were defined as Jersey herds and 7,682 (88.4%) were defined as large breed herds (Holstein Friesian, Red Danish Cattle, and mixed breeds; 5 herds had missing values for breed). The distinction between Jersey and large breeds was based on the geometric average percentage of fat in milk from weekly recordings in the Danish Cattle Database.
Data Editing and Descriptive Statistics
The following variables were extracted and constructed from the databases and used in the risk factor analyses. Even though some data (e.g., the Salmonella data) were available for the full period from 2001 to 2004, the final data sets used for the risk factor study had to be restricted to the 4 quarters of 2003 because of lack of data outside this period for most of the risk factors.
Year-Quarter.
The YQ were based on the months January to March, April to June, July to September, and October to December for each year 2001 to 2004. The distribution of test-positive herds and test-negative herds for each YQ in the study period is shown in Table 1
.
|
.
|
Organic.
Eight percent of the 8,694 dairy herds were recorded as organic in 2003. Information about which herds were organic was obtained from the Danish Plant Directorate and merged with data from the Danish Cattle Database. Although organic status was coded as a time-dependent variable, for practical purposes it can be considered as nonvarying because virtually all herds (>99%) were classified as organic or not organic for the entire study period.
Neighbors in 2-km Radius.
The number of neighbors included beef and dairy herds in a 2-km radius around each study herd. This number was calculated for all of 2003, so that if a herd were recorded active at any point in time during 2003 it would be counted as a neighbor and would be assigned neighbors as well. The number was used to represent the full study period because it was assumed not to change over time.
C2a or C3a Neighbors.
The number of cattle herds (beef and dairy) classified C2a or C3a in the surveillance program in a 2-km radius around each dairy herd in the previous YQ. Data were available from first YQ of 2003 to first YQ of 2004.
Close-Contact Neighbors.
The legally required area used to spread manure from each dairy and beef herd according to the regulations was calculated, and the radii were calculated for circles equivalent to this area centered around each herd. The area depended mainly on the herd size and breed. This variable showed how many neighboring dairy and beef herd circles in which each dairy herd was located during the previous YQ. Data were available from first YQ of 2003 to first YQ of 2004.
C2a or C3a Close-Contact Neighbor.
Similar to the previous variable, except this variable only counted the number of circles from herds that were classified in C2a or C3a in the surveillance program in the previous YQ.
Individual animal numbers, dates of birth, dates of entry to the holding, identification of the farm of origin, animal date, and reason for departure (e.g., breeding, slaughter, export, or death) were extracted from the Danish Cattle Database. These data were used to construct the following variables:
Calf Mortality.
Calf mortality in the previous YQ was calculated as the proportion of live-born calves that were ear-tagged and died within 1 to 90 d after birth. Data were available from fourth YQ of 2002 to third YQ of 2003.
Source Herds.
Total number of dairy and beef source herds for cattle purchased in the previous YQ. Data were available from second YQ of 2002 to first YQ of 2004.
Test-Positive Source Herds.
Total number of test-positive dairy source herds for cattle purchased in the previous YQ. Data were available from second YQ of 2002 to first YQ of 2004.
Purchased Cattle.
Total number of dairy and beef cattle purchased in the previous YQ. Data were available from second YQ of 2002 to first YQ of 2004.
Purchased Cattle from Test-Positive Herds.
Number of cattle purchased from test-positive dairy herds in the previous YQ. Data were available from second YQ of 2002 to first YQ of 2004.
Herd Size.
Total number of cattle (regardless of age and production type) on the premises counted per month and averaged for each YQ. Data were available from second YQ of 2001 to fourth YQ of 2003.
Statistical Method of Analysis
Two separate multivariable logistic regression models were used to analyze risk factors associated with a change of status between 2 YQ from test negative to positive and from test positive to negative, respectively. The factors analyzed in the model related to the YQ just before the change in test status (e.g., herd size, purchase of cattle, number of infected neighbors) or they could be factors that were considered fixed over the full study period (e.g., number of neighbor cattle herds in a 2-km radius). Continuous variables were checked visually for a linear relationship with the log odds of the response variable before inclusion in the model. Because all variables other than herd size did not show linear relationships, it was decided to categorize these (see Table 2 for details on the categories).
|
The models were constructed as multivariable logistic regression models controlling for the correlation of repeated measurements from the same herd by using generalized estimating equations (GEE). All main effects were included in the initial model together with all interactions with herd size. Backward selection was used to remove nonsignificant interaction terms and nonsignificant main effects in that order. Then 2-way interaction terms of the significant main effects were tested in the model. Significance level criteria for variables and interaction terms to remain in the final model were P < 0.05. For the variable selection all variables remained in the class statement to analyze on the same data set. After the final model with significant effects was determined, the model was tested on a data set including all data available for the class variables remaining in the model. This data set contained data from 4 YQ (first to fourth YQ of 2003). This restricted data set was a result of lack of data for several risk factors as illustrated in the section on Data Editing and Descriptive Statistics.
Odds ratios with 95% confidence limits were calculated as described by Hosmer and Lemeshow (2000). For the calculation of odds ratios for herd size and variables with significant interactions with herd size in the test-positive to test-negative model, 3 examples were selected for illustration (100, 200, and 300 cattle). It was necessary to choose specific numbers for the herd size as examples to illustrate the difference between combinations of risk factors because herd size was a continuous variable in the models. In Denmark, a total herd size of 100 would be considered a small dairy herd, 200 a medium-sized dairy herd, and 300 a large dairy herd.
Software
The SAS version 8.2 program was used for data editing and statistical analyses. The GENMOD procedure was used for logistic regression models with a repeated statement and an unstructured working correlation structure to account for the intraherd correlation of response measurements within the same herd. The "within-subject option" was used to allow for adjustment of correlation in the case of missing values for certain YQ.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
After the data set was reduced to including the relevant period with data for all important variables, the data set for test negative to positive changes had 21,007 observations, and changes occurred in 2% of these observations. The data set for test positive to negative changes had 6,168 observations, with changes occurring in 10% of these observations. Descriptive statistics including proportion of the 2 types of changes in herd status for each level of the categorical variables are given in Table 2
.
Model 1: Change from Test Negative to Positive Indicating Introduction of Infection
The parameter estimates and significance levels in the model for becoming test positive after being test negative, which is indicative of new infection, are given in Table 3. The higher the number of C2a or C3a neighbor herds was in the local area (2-km radius around the herd), the higher (P = 0.006) was the risk of changing from test negative to positive. The risk of changing was also higher (P < 0.0001) if a herd had purchased animals from test positive herds in the previous YQ than if it had only purchased animals from test-negative herds or not purchased animals at all. Because of high colinearity between number of purchased cattle and number of source herds, it was not possible to include both variables in the model simultaneously. When the number of test-positive source herds was tested in this model instead of number of purchased cattle from test-positive herds, this risk factor was also significant with increasing number of test-positive source herds leading to higher odds of changing from Salmonella test negative to positive (results not shown).
|
. Odds ratios with 95% confidence limits for each main effect and examples from the interaction terms are shown in Table 5. The odds of changing from test positive to negative were influenced by the local cattle herd density (measured as number of neighbor herd manure disposal areas that included the herd at risk). Herds that did not have such close-contact neighbors in the previous YQ had higher odds of becoming test negative than herds with one or more close-contact neighbors. This variable was highly dependent on the size of the neighbor herds because large herds would require larger areas for manure disposal than small herds. One other factor that described local cattle herd density was present in the final model (i.e., in the interaction between herd size and number of C2a or C3a neighbors in a 2-km radius). The calculation of the latter explanatory variable did not depend on the size of the neighboring herds. For medium and large herds, having 2 or more C2a or C3a herds in a 2-km radius lead to significantly smaller odds of becoming test negative in the current YQ after having been test positive in the previous YQ. For small herds there was no effect of having such neighbors.
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
The models for new infection and recovery of Salmonella infection in Danish cattle were based on herd test status determined by repeated bulk tank milk measurements. Thus, the outcome of interest was measured indirectly. There are no perfect diagnostic methods available to measure whether a herd is truly infected with Salmonella (Veling et al., 2002). With the method used in our study, the herd-positive predictive value in the program was not perfect (estimated to 80%), and this might have had an impact on the results (Warnick et al., 2006). Whereas Veling et al. (2002) found the sensitivity of BTM ELISA to be small (54%) when basing herd classification on one single BTM sample, combining the results from 4 repeated samples as is done in the Danish Salmonella surveillance program seemed to improve the sensitivity (95%; Warnick et al., 2006). Although detailed information on herds used for evaluation of the BTM ELISA (Nielsen and Ersbøll, 2005) raises confidence that risk factors identified in the study presented here are applicable to changes in herd infection status, potential effects of misclassification should be considered. For instance, a herd may become test positive after a rise in antibodies in the BTM when purchasing animals with antibodies from another herd without necessarily having introduced the infection into the herd. This would be more likely to occur if the source herd was test positive. If this phenomenon was common in a large number of herds, the effect of purchase on becoming infected could have been overestimated in our analysis. Other studies, however, support the finding that the risk of becoming infected increases with purchase from other herds, and biologically it makes sense that this risk is mainly increased if the source herd is infected and thus test positive (Vaessen et al., 1998; van Schaik et al., 2002). Based on clinical experience with the BTM ELISA test as well as bacteriological culture results and other observations of test program herds, we believe the analysis identified factors likely to be associated with changes in Salmonella infection status of the herd and not just with changes in herd test results. Consistency with published results from other field studies strengthens this conclusion for the effect of purchasing cattle from infected herds and a number of other risk factors identified.
One possible improvement of the models could be to include infection history from YQ earlier than the most recent YQ. If a test-negative herd had been infected with Salmonella (or had been test positive) within the last couple of years it was probably more likely to change to test positive again after reinfection from persistently infected animals or surviving bacteria in the environment (Wray et al., 1989; House et al., 1993). This should preferably be assessed using a data set containing data from a longer period than was available for this study. Another variable that could not be included in this study was the concurrent infection with other diseases, such as metabolic diseases, liver fluke infestation, or viral infections that may reduce the resistance to Salmonella in the herds (Aitken et al., 1981; Vaessen et al., 1998).
Model Results
The number of presumably infected (C2a or C3a) neighbor herds was a significant risk factor common for both models. The higher this number, the more likely change from test negative to positive was to occur and the less likely test-positive herds were to become test negative between 2 YQ. The results are supported by another study (Wedderkopp et al., 2001). In that study it was found that the risk for a dairy herd to change infection status was associated with the BTM status of the nearest neighbors and the prevalence of seropositive herds in the geographic area.
Herd size was also included in both models. The bigger the herd, the more likely it was to change from test negative to positive (model 1). Changing from test positive to negative, the effect of herd size was less clear. Large herds consisting of large breed or having C2a or C3a neighbors in a 2-km radius were less likely to recover, whereas the breed and neighbor factors were not found to be important for small herds. Herd size is often found to be a risk factor for infectious diseases (Vaessen et al., 1998; Warnick et al., 2001). Herd size, however, may be an indirect measure of management. In this study, we did not have access to data to investigate possible underlying reasons for the effect of herd size. In 2 studies in which there were more elaborate details of management available for the model, no association between Salmonella fecal shedding and herd size was detected (Fossler et al., 2005a,b).
The number of purchased cattle from test-positive herds was associated with becoming infected. Other studies have found purchase of live animals a significant risk factor for introduction of Salmonella infection to the herd (Wray et al., 1990, 1991; Vaessen et al., 1998; van Schaik et al., 2002). In the surveillance program, purchase of cattle from C2a dairy herds led to automatic classification in C2b for 3 mo from the day of purchase. This was a restriction implemented to attempt to minimize transmission between herds. The C2b was an indicator that the herd had been involved in high-risk activity (contact to assumed infected herds), and other herds were less likely to purchase cattle from the herds in this category. The fact that purchase from test-positive herds was found to be a significant risk factor for changing to test positive supports this control strategy. Whether the length of the automatic classification period based on trade could be improved was not investigated further in this study.
Organic production was only associated with persistence of test-positive results, not with becoming test positive. This may be because of regulations about management procedures in organic herds that allow for easy spread of infection from dams or calving environment to newborn calves and calving cows such as the practice of leaving the calf with the dam up to 3 to 4 d after birth, less aggressive antibiotic use, no preventive treatments allowed, and possible differences in feeding strategies (Fossler et al., 2005a,b).
In the final model there was no association between calf mortality in the previous YQ and a change in test results. Data quality for the variable of calf mortality was, however, not ideal. The variable was constructed as the number of dead calves out of all calves born. The calf mortality percentage varied dramatically because of the small number of calves born per YQ in some herds. It is likely that increased calf mortality more often would follow rather than precede the change of Salmonella status. For future studies, it should be investigated how to construct a reliable calf mortality or calf morbidity variable and to consider how to include it in a model as it is likely to be associated with many outbreaks and reinfections with S. Dublin and other types of Salmonella in dairy herds, and it may be useful as an early indicator of new or reinfection with Salmonella.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
The association between calf mortality (or calf morbidity) and Salmonella infection should be investigated using higher quality data than was available in this project. This would help determine if calf mortality, calf morbidity, or both are useful as an early warning of new infection in herds that have not yet had an increase in BTM ELISA.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
Received for publication May 24, 2006. Accepted for publication January 31, 2007.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
