Herd characteristics and management practices associated with seroprevalence of Mycobacterium avium subsp paratuberculosis infection in dairy herds

Saraya Tavornpanich Department of Medicine and Epidemiology, School of Veterinary Medicine, University of California, Davis, CA 95616.

Search for other papers by Saraya Tavornpanich in
Current site
Google Scholar
PubMed
Close
 DVM, MPVM, PhD
,
Wesley O. Johnson Department of Statistics, Donald Bren School of Information and Computer Sciences, University of California, Irvine, CA 92697.

Search for other papers by Wesley O. Johnson in
Current site
Google Scholar
PubMed
Close
 PhD
,
Randall J. Anderson Animal Health Branch, California Department of Food Agriculture, Sacramento, CA 95814.

Search for other papers by Randall J. Anderson in
Current site
Google Scholar
PubMed
Close
 DVM, MPVM
, and
Ian A. Gardner Department of Medicine and Epidemiology, School of Veterinary Medicine, University of California, Davis, CA 95616.

Search for other papers by Ian A. Gardner in
Current site
Google Scholar
PubMed
Close
 BVSc, MPVM, PhD

Abstract

Objective—To investigate herd characteristics and management practices associated with a high seroprevalence of Mycobacterium avium subsp paratuberculosis (MAP) in dairy herds in central California.

Sample Population—60 randomly selected cows from each of 21 dairy herds.

Procedures—Sera of selected cows were tested for antibodies against MAP by use of an ELISA test kit. Cows with a test sample-to-positive control sample (S:P) ratio of ≥ 0.25 were considered seropositive, and herds with ≥ 4% seropositive cows were considered high-seroprevalence herds. Data on herd characteristics and management practices were collected via interviews with owners. Bayesian logistic regression was used to model the predictive probability of a herd having a high seroprevalence on the basis of various herd characteristics and management practices.

Results—9 of 21 (43%) herds were classified as high-seroprevalence herds. Five variables (history of previous signs of paratuberculosis in the herd, herd size, exposing cattle to water from manure storage lagoons, feeding unsalable milk to calves, and exposing heifers ≤ 6 months old to manure of adult cows) were included in the predictive model on the basis of statistical and biological considerations. In large herds, the predictive probability of a high seroprevalence of MAP infection decreased from 0.74 to 0.39 when management changed from poor to good practices. In small herds, a similar decrease from 0.64 to 0.29 was predicted.

Conclusions and Clinical Relevance—The seroprevalence of MAP infection in California dairies may be reduced by improvements in herd management practices.

Abstract

Objective—To investigate herd characteristics and management practices associated with a high seroprevalence of Mycobacterium avium subsp paratuberculosis (MAP) in dairy herds in central California.

Sample Population—60 randomly selected cows from each of 21 dairy herds.

Procedures—Sera of selected cows were tested for antibodies against MAP by use of an ELISA test kit. Cows with a test sample-to-positive control sample (S:P) ratio of ≥ 0.25 were considered seropositive, and herds with ≥ 4% seropositive cows were considered high-seroprevalence herds. Data on herd characteristics and management practices were collected via interviews with owners. Bayesian logistic regression was used to model the predictive probability of a herd having a high seroprevalence on the basis of various herd characteristics and management practices.

Results—9 of 21 (43%) herds were classified as high-seroprevalence herds. Five variables (history of previous signs of paratuberculosis in the herd, herd size, exposing cattle to water from manure storage lagoons, feeding unsalable milk to calves, and exposing heifers ≤ 6 months old to manure of adult cows) were included in the predictive model on the basis of statistical and biological considerations. In large herds, the predictive probability of a high seroprevalence of MAP infection decreased from 0.74 to 0.39 when management changed from poor to good practices. In small herds, a similar decrease from 0.64 to 0.29 was predicted.

Conclusions and Clinical Relevance—The seroprevalence of MAP infection in California dairies may be reduced by improvements in herd management practices.

Paratuberculosis, caused by infection with MAP, is widespread throughout the United States and is estimated to affect approximately 70% of US dairy herds and 3% of dairy cattle.1,2 The disease is transmitted to susceptible young calves mainly via the fecal-oral route, through exposure to environments or substances (eg, food, water, and equipment) contaminated by feces of infected cows.3 Findings based on simulation studies4,5 and studies conducted on farms6 indicate that successful programs for reduction of within-herd prevalence of MAP infection depend on implementation of management practices that minimize the risk of MAP infection.

Many studies2,7–12 have been conducted to investigate associations between management factors and infection status of herds on the basis of results from ELISAs that detect antibodies against MAP in serum; however, with the exception of 1 study,8 such information is not available for large western US dairy herds. In those studies,7–12 various criteria were used to categorize herds as seropositive, but in most studies researchers detection of ≥ 1 seropositive cow in a herd was used to designate a seropositive herd, regardless of the number of cows tested. In other studies,2,9 additional information, such as a history of ≥ 5% of cows with clinical signs consistent with paratuberculosis, was used to minimize misclassification bias attributable to imperfect sensitivity and specificity of ELISAs. The detection of ≥ 4% seropositive cows in a herd by use of ELISA is considered indicative of MAP infection because there is consensus that serum and milk ELISAs are typically > 96% specific.13

Significant risk factors associated with herd seropositivity include maintaining a large herd size,2,8,10,12 allowing group housing of preweaned calves and periparturient cows,2 housing postweaned calves in calf barns and hutches rather than pens in cow barns,7 feeding calves unpasteurized whole milk,12 allowing use of exercise lots for lactating cows,9 purchasing replacement females that were born in other herds,8 and exposing calves ≤ 6 weeks old to feces of adult cows.1 Herd management practices associated with herd seronegativity for MAP antibody (ie, no MAP-infected cows identified by use of ELISA) include cleaning maternity pens after each use,9 removing a calf from its dam immediately after birth,2,10 feeding milk replacer,10 allowing cows to graze on paddocks free of MAP-contaminated manure, and applying lime to pasture.9

Results of studies of management practices associated with an increased probability of MAP infection have been inconsistent. For example, separating newborn calves from cows < 12 hours after birth and allowing females < 1 year old to graze in areas free of adult cattle were not identified as protective against paratuberculosis in 1 study.12 Such apparent differences may be attributable to variation of MAP prevalence and management practices among farms as well as dissimilarity of designs and sample sizes used in various studies.

In California, dairy management practices differ from those in other states and also vary within the state. In northern California, herds are small (mean size, approx 207 lactating cows), and many herds include cattle on pasture. In the central region, herds are larger (mean size, 806 lactating cows), and cows are housed mostly in drylots and freestalls with no access to pasture. Management practices commonly used in herds in central California include flushing walkways with lagoon water and housing sick cows and postpartum cows together in the hospital area. In southern California, herd sizes are larger than in northern California, and management practices are similar to those of central California dairies, with the exception that most dairies use drylots for cows.14,15 In 1 study,15 investigators estimated that the seroprevalence of MAP infection in dairy herds was 3.7%, 5.2%, and 6.9% in the central, southern, and northern regions of California, respectively.

Information on herd management practices associated with MAP infection in California dairies is not available, but such information would be useful to provide evidence in support of recommendations to reduce transmission of infection within herds. The objective of the study reported here was to investigate associations between herd characteristics and management practices and a high seroprevalence of MAP infection in dairy herds in central California.

Materials and Methods

Sample population—Twenty-one dairy herds were selected from among the clientele of a veterinary practice in central California. Herd sizes and management practices were considered typical of those for dairy herds in that geographic area. All herds had been confirmed as positive for infection with MAP on the basis of bacteriologic culture results of individual or pooled fecal samples or environmental slurry samples.16,17 The number of herds tested was chosen on the basis of funding constraints and availability of data obtained via questionnaire in 2001.

Within each herd, approximately 60 lactating cows were selected for participation by use of systematic random sampling, and the selection was performed for 2 time periods (August through November 2001 and January through June 2004). This sample size was sufficient for detecting ≥ 1 MAP-seropositive cow in a herd with assumptions that the pretest seroprevalence of MAP infection within a herd was ≥ 4% and that a 95% confidence interval for estimates of posttest seroprevalence was desired; it also ensured the error margin for within-herd posttest seroprevalence would be approximately 4% to 5%. Owners were asked to exclude firstlactation cows from the sampling in accordance with current testing recommendations for voluntary control of paratuberculosis.18

Herd characteristics and management data—Data were collected via interviews with herd owners or managers in August through November 2001. Information obtained for each herd included general herd characteristics (size of herd, breed of cow, percentage of firstlactation cows among adult cows, and type of housing); disease history (including previous detection of clinical signs consistent with paratuberculosis [diarrhea and persistent weight loss]); purchase policy for replacement heifers; management practices for colostrum, milk, and manure handling; and location and management of the calving area.

ELISA for serum antibodies against MAP—Two sets of blood samples were obtained for detection of antibodies against MAP. The first samples were collected in 2001 at the time interviews were conducted, and the second samples were collected in January through June 2004, which was approximately 2.5 years later. At both time periods, blood samples were collected from the coccygeal vein into evacuated tubes, placed on cold packs, and transported on the day of sample collection to the laboratory for analysis.a Blood samples were centrifuged; sera were harvested, divided into aliquots, and stored at 5°C until tested. Each serum sample was subsequently tested by use of an ELISA test kit,b as described elsewhere.16 Serum samples were assayed with positive- and negative-control samples, and results were reported as S:P ratios. Cows with ELISA S:P ratios of < 0.25 and ≥ 0.25 were classified as seronegative and seropositive for MAP, respectively, as recommended by the manufacturer.b Seroprevalence was calculated as the percentage of MAP-seropositive cows within each herd. Results for first-lactation cows were excluded from calculations when those cows were included in the testing. A herd was classified as a low- to zero-seroprevalence herd when it had ≤ 2 seropositive cows (< 4% seroprevalence) and a high-seroprevalence herd when ≥ 3 cows (≥ 4% seroprevalence) were seropositive.

Statistical analysis—Serologic results from the 2 sample collection periods were compared, but only results from the second collection period were used in subsequent analyses because those results most likely reflected the management practices used in the herds in 2001 (ie, the time when cows that were seropositive in 2004 were likely to have become infected with MAP). Herd characteristics and management factors were categorized on the basis of biological and analytic considerations. Small herds were defined as those with 285 to 675 lactating cows; large herds were defined as those with > 675 lactating cows. Separation of newborn calves from the dam was considered early when accomplished between 1 and 6 hours after birth. Associations between categorized herd variables and high seroprevalence status were evaluated by use of a χ2 test. Variables with a value of P < 0.20 in the univariable analyses or variables that were considered biologically important were included in a multivariable Bayesian logistic regression model to estimate the probability of a high–MAP seroprevalence herd on the basis of various herd characteristics and management practices.19,20

The informative Bayesian approach may improve model estimates by incorporating prior information on herd management factors, but analysis based solely on the sampled data or with the use of a noninformative prior distribution may be insufficient to yield reliable estimates of some predictive probabilities.21 Briefly, Bayesian analysis involves 3 key components: the likelihood function or L(y | θ), the prior distribution or p(θ), and the posterior distribution or p(θ | y). The likelihood function used was based on the binary test results from each of the 21 herds. In classic logistic regression, maximum likelihood–based methods are used to obtain estimates of variables of interest solely on the basis of current sample data. In Bayesian logistic regression, the estimate is based on current sample data and prior information in the form of a probability distribution. The posterior distributions used in our study combined information from the sampled data and the prior distribution. Statistically, these 3 components are connected through Bayes' rule: p(θ | y) α L(y | θ) × g(θ), where α indicates proportionality.

Estimation of the variables of interest was performed by use of the Gibbs sampler, which is an iterative algorithm that constructs a Markov chain and allows empiric approximation of the posterior distribution. The Bayesian analysis was performed by use of computer software.c Model convergence was assessed via visual examination of trace plots, assessment of Monte Carlo error, and running of multiple chains from dispersed starting values.22 Three parallel runs of 50,000 iterations were generated. The first 10,000 iterations were discarded, and posterior inferences were based on summaries of the final 40,000 iterations.

Two types of prior information were used for analyses. First, noninformative (diffuse) prior distributions were used in which the prior distributions for regression coefficients were modeled by use of a normal distribution (mean, 0; precision [1/variance], 2). This distribution was chosen because it resulted in noninformative prior distributions for the probabilities of high MAP seroprevalence for a range of covariate combinations. Second, partially informative prior distributions were used to incorporate scientific information about the regression coefficients of herd management variables. The partially informative prior distributions constrained the regression coefficients of variables believed to be biologically associated with increased or decreased risk of MAP infection to be positive or negative, respectively. Otherwise, they were noninformative as well.

The predictive probability of being a high-seroprevalence herd for various combinations of herdlevel risk factors was modeledd as logit (p) = β0 + (x × β), where p was the probability of the outcome, β0 was an intercept, × was a (row) vector of the covariates (ie, risk factors), and β was a (column) vector of the regression coefficients for the corresponding risk factors. In Bayesian logistic regression, the entire posterior distribution of each regression coefficient is obtained, rather than the point estimate and SE obtained in classic logistic regression. The Bayesian technique is useful to assess the posterior probability that a regression coefficient is positive (or negative) by calculating the proportion of Monte Carlo iterates from the posterior distribution that are positive (or negative) for each regression coefficient.e Regression coefficients with a posterior probability ≥ 0.8 were considered positively associated with a risk of high MAP seroprevalence, and regression coefficients with a posterior probability ≤ 0.2 were considered negatively associated with a risk of high seroprevalence.23 Predictive probabilities of being classified as a high-seroprevalence herd for various combinations of herd-level risk factors were also estimated from the final model.

Results

Description of sample population—Sizes of the 21 participating herds ranged from 285 to 1,778 lactating cows (median, 675 cows). Mean herd size (790 cows) was slightly lower than the value of 806 for the mean number of dairy cows per dairy in central California.14 Percentage of heifers in each herd ranged from 15% to 40%. Sixteen of 21 (76%) herds had only Holstein cattle; 2 (10%) had only Jersey cows, and 3 (14%) had both breeds. Fifteen (71%) herds housed their lactating cows in freestalls, 1 (5%) herd housed cows completely in drylots, and 5 (24%) herds used a combination of both housing styles. Nine (43%) herds were closed to introductions of cows from other herds.

Interval to separation of newborn calves from their dams ranged from 1 to > 24 hours after birth, but in most herds (14/21 [67%]), calves were separated within 6 hours. In most herds, the calving area was separated from housing areas for sick and lactating cows. Various areas were used for calving, including areas in which cows were penned individually and areas in which a small number of cows were penned together. Changing bedding between each calving was not a common practice, and in most herds (14/21 [67%]), bedding was changed after every 5 calvings.

Colostrum was sometimes pooled before it was fed to newborn calves in 19 of 21 (90%) herds, and unsalable milk was regularly fed to calves in 11 (52%) herds. Manure-removal equipment was used to handle feed in 13 of 21 (62%) herds. Lagoon water was used as flushing water in 10 (48%) herds, and heifers ≤ 6 months old were commonly exposed to manure of adult cows in 12 (57%) herds.

Agreement between estimates of seroprevalence—The percentage of seropositive cows ranged from 0% to 21% (median, 3%) for the first testing period and from 0% to 10% (median, 3%) for the second testing period. At both sampling periods, 7 and 4 herds were classified as low- and high-seroprevalence herds, respectively, and the status of 10 herds changed.

Univariable analysis of risk factors—On the basis of results of ELISAs performed on serum from blood samples obtained in 2004, MAP-seropositive cows were detected in 17 of 21 (81%) herds, and 9 (43%) herds had ≥ 4% seropositive cows. Median herd seroprevalence was 3.3% (range, 0% to 10%).

Univariable analyses revealed 4 variables that were qualified for inclusion in the Bayesian logistic regression model on the basis of the statistical criterion (ie, P < 0.20): herd size, use of manure-handling equipment to handle feed, exposure of cattle to lagoon water, and exposure of heifers ≤ 6 months old to manure of adult cows (Tables 1 and 2). History of previous signs of paratuberculosis in the herd and feeding of unsalable milk to calves were considered biologically important factors; therefore, their association with high seroprevalence status was also investigated by use of Bayesian logistic regression analysis. Although pooling colostrum was considered as a biologically important variable, most (19/21 [90%]) herds contained calves that were fed pooled colostrum, and hence, the association between pooling colostrum and high seroprevalence of MAP was not evaluated.

Table 1—

Univariable associations between herd characteristics and high seroprevalence of MAP infection (≥ 4% seropositive cows) in 21 dairy farms in central California.

CharacteristicNo. of herdsNo. (%) affectedP value*
No. of lactating cows0.13
 285 to 675113 (27)
 676 to 1778106 (60)
First-lactation heifer (%)0.53
 15 to 30105 (50)
 31 to 40114 (36)
Breed0.33
 Holstein167 (44)
 Jersey20 (0)
 Both32 (67)
Housing0.49
 Drylot10 (0)
 Freestall156 (40)
 Both53 (60)
Proportion of cows in herd with signs of paratuberculosis (%)0.24
 051 (20)
 0.1 to 1.4168 (50)
Proportion of females born on farm (%)0.45
 75 to 99126 (50)
 10093 (33)

Associations were evaluated by use of a χ2 test; variables with values of P < 0.20 were considered significant and included in subsequent multivariable analyses.

Table 2—

Univariable associations between management practices and high seroprevalence of MAP infection (≥ 4% seropositive cows) in 21 dairy farms in central California.

Management practiceNo. of herdsNo. (%) affectedP value*
Interval from birth until separation of newborn calves from their dam (h)0.35
 1 to 6147 (50)
 > 672 (29)
Calving area separated from sick cows area0.37
 No53 (60)
 Yes166 (38)
Calving area separated from housing for lactating cows0.58
 No62 (33)
 Yes157 (47)
Frequency of bedding changes in calving area0.59
 2 to 5 calvings42 (50)
 > 5 calvings145 (36)
 No bedding used32 (67)
Area used for calving0.65
 Penned separately42 (50)
 Penned as a group105 (50)
 Other72 (29)
Colostrum pooled0.35
 Never10 (0)
 Sometimes11 (100)
 Always198 (42)
Unsalable milk fed to calves0.34
 Never93 (33)
 Always116 (55)
Manure-handling equipment used to handle feed0.19
 Never82 (25)
 Sometimes or regularly137 (54)
Cattle exposed to lagoon water0.13
 No113 (27)
 Yes106 (60)
Heifers ≤ 6 mo old exposed to manure of adult cows0.09
 No92 (22)
 Yes127 (58) 

Data were unavailable for 1 herd.

See Table 1 for remainder of key.

Bayesian logistic regression—For each estimate of a variable, Monte Carlo error was small, and autocorrelation values indicated that iterates were not highly correlated with subsequent values. Trace plots of the last 40,000 iterations of the 3 independently generated Markov chains indicated adequate mixing for all variables. The final model was as follows:

article image

where p represents the probability of a high-seroprevalence herd for a given covariate combination, β0 is the intercept, β1 to β6 are the coefficients for the respective covariates, history represents previous signs of paratuberculosis in the herd, size represents herd size, lagoon represents exposure of cattle to lagoon water, unsalable represents feeding of unsalable milk to calves, equipment represents use of manure-handling equipment to handle feed, and manure represents exposure of heifers ≤ 6 months old to manure of adult cows.

The posterior distribution (median and 95% PI) of regression coefficients (βs) and posterior probability of the respective value of β being positive were obtained for models by use of noninformative and partially informative prior distributions (Table 3). When non informative prior distributions were used, all model variables were positively associated with an increased risk of high MAP seroprevalence. However, none of the factors was significant, and their 95% PIs included negative values. Values for the posterior probability of the regression coefficient being positive ranged from 0.52 for a history of previous signs of paratuberculosis in the herd to 0.71 for feeding of unsalable milk to calves. Use of manure-handling equipment to handle feed had the lowest probability among the factors related to management practices.

Table 3—

Posterior inferences for regression coefficients for herd management factors associated with high herd seroprevalence of MAP infection based on a Bayesian logistic regression model* with noninformative and partially informative prior distributions

Management factorNoninformative prior distributionsPartially informative prior distributions
Median95% PIPosterior probabilityMedian95% PIPosterior probability
Intercept−0.700−1.810 to 0.400NA−1.254−2.026 to −0.194NA
History of previous signs of paratuberculosis
 NoNANA0.480NANA0
 Yes0.026−1.052 to 1.1030.5200.3120.013 to 1.1051
Herd size
 SmallNANA0.320NANA0
 Large0.264−0.848 to 1.3660.6800.4190.020 to 1.3291
Cattle exposed to lagoon water
 NoNANA0.332NANA0
 Yes0.256−0.865 to 1.3800.6680.4100.018 to 1.3081
Calves fed unsalable milk
 NoNANA0.288NANA0
 Yes0.287−0.727 to 1.3210.7120.4320.022 to 1.3051
Manure-handling equipment used to handle feed
 NoNANA0.338NANA0
 Yes0.227−0.818 to 1.2600.6620.4200.021 to 1.2971
Heifers ≤ 6 mo old exposed to manure of adult cows
 NoNANA0.319NANA0
 Yes0.262−0.845 to 1.3710.6810.3940.019 to 1.2761

50,000 iterations with 3 Markov chains were performed; the first 10,000 iterations were discarded.

Values represent posterior probabilities that corresponding regression coeffcients are positive.

Small herd size was defined as 285 to 675 lactating cows; large herd size was > 675 cows. NA = Not applicable.

Use of partially informative prior distributions slightly affected the posterior distributions of most model variables but highly influenced the value for model intercept (β0) and the association of high seroprevalence of MAP infection with history of previous signs of paratuberculosis in the herd (Table 3). For that model, our assumption for constructing the prior distribution was that a history of previous signs of paratuberculosis in the herd should increase the risk of MAP transmission. Therefore, the posterior median of the regression coefficient for history of previous signs of paratuberculosis in the herd changed noticeably (from 0.026 to 0.312), and the 95% PI was necessarily confined to positive values (which was certain given the prior constraint). The changes in posterior distributions of the other 5 variables in the model were not substantial. The posterior medians remained as positive values and shifted slightly, with 95% PIs that were confined to positive values. This suggested that the constraints imposed by the prior distribution did not substantially modify inferences for herd size, exposure of cattle to lagoon water, feeding of unsalable milk to calves, use of manure-handling equipment to handle feed, and exposure of heifers ≤ 6 months old to manure of adult cows.

Predictive probability of high MAP seroprevalence—Predictive probabilities of high MAP seroprevalence within a herd for various combinations of herd-level risk factors were estimated (Tables 4 and 5). The use of partially informative prior distributions influenced the posterior distributions of model variables and affected the predictive probability of high-seroprevalence herds. When the noninformative prior distribution was used, the highest median for the predictive probability of high MAP seroprevalence (0.62) corresponded to large herds in which cattle were exposed to lagoon water, calves were fed unsalable milk, and heifers ≤ 6 months old were exposed to manure of adult cows. With the partially informative prior distribution, the highest median for the predictive probability of high MAP seroprevalence was 0.74 for the same combination of the risk factors.

In contrast, the smallest medians for predictive probability of high MAP seroprevalence for noninformative and partially informative prior distributions (0.35 and 0.29, respectively) corresponded to small herds with no exposure of cattle to lagoon water, no feeding of unsalable milk to calves, and no exposure of heifers ≤ 6 months old to manure of adult cows. Not surprisingly, the only factor that contributed to the differences in estimates obtained with the 2 prior distributions was the constraint imposed by the restriction of model variables to positive values. The reduction of the predictive probability of high MAP seroprevalence attributable to good management practices was 55% in small and 47% in large herds when estimated by use of the partially informative prior distribution.

In herds with a history of previous signs of paratuberculosis, preventing exposure of heifers ≤ 6 months old to manure of adult cows and refraining from feeding unsalable milk to calves reduced the predictive probability of high MAP seroprevalence by 49% in small herds and by 32% in large herds (Table 5). In herds without a history of previous signs of paratuberculosis, preventing exposure of heifers ≤ 6 months old to manure of adult cows and not feeding unsalable milk to calves reduced the predictive probability of high MAP seroprevalence by 46% in small herds and by 40% in large herds.

Table 4—

Medians and 95% PIs for predictive probabilities of high MAP seroprevalence in herds with various combinations of herd management factors estimated by use of Bayesian logistic regression with noninformative and partially informative prior distributions.

Management factorNoninformative prior distributionsPartially informative prior distributions
Herd size*Cattle Exposed to manure lagoonClaves fed unsalable milkHeifers exposed to manure of adult cowsMedian95% PIMedian95% PI
SmallNoNoNo0.3510.168–0.5820.2880.141–0.490
NoNoYes0.4180.168–0.7110.3930.198–0.637
YesNoNo0.4140.156–0.7250.3950.194–0.648
NoYesNo0.4230.179–0.7050.4010.201–0.645
YesNoYes0.4830.202–0.7740.5130.277–0.756
NoYesYes0.4910.196–0.7920.5190.281–0.763
YesYesNo0.4890.174–0.8130.5200.273–0.775
YesYesYes0.5580.233–0.8410.6370.378–0.849
LargeNoNoNo0.4120.158–0.7200.3930.194–0.647
NoNoYes0.4810.178–0.7990.5110.276–0.761
YesNoNo0.4780.178–0.7970.5140.270–0.766
NoYesNo0.4880.194–0.7910.5190.277–0.769
YesNoYes0.5480.246–0.8220.6300.375–0.844
NoYesYes0.5570.229–0.8420.6360.378–0.849
YesYesNo0.5530.215–0.8520.6380.371–0.857
YesYesYes0.6200.305–0.8650.7400.493–0.906

Small herds contained 285 to 675 lactating cows; large herds contained > 675 lactating cows.

Includes heifers ≤ 6 months old.

Table 5—

Medians and 95% PIs for predictive probabilities of high MAP seroprevalence in herds with various combinations of herd management factors estimated by use of Bayesian logistic regression with noninformative and partially informative prior distributions.

Management factorNoninformative prior distributionsPartially informative prior distributions
History of previous signs of paratuberculosisHerd size*Claves fed unsalable milkHeifers exposed to manure of adult cowsMedian95% PIMedian95% PI
NoSmallNoNo0.3490.157–0.6050.2700.125–0.474
SmallYesNo0.4180.159–0.7320.3720.176–0.626
LargeNoNo0.4200.159–0.7360.3770.181–0.639
SmallNoYes0.4300.177–0.7260.3830.186–0.632
LargeYesNo0.4920.180–0.8120.4970.251–0.757
SmallYesYes0.5010.192–0.8110.4990.257–0.753
LargeNoYes0.5060.206–0.8030.5070.266–0.759
LargeYesYes0.5780.240–0.8560.6240.359–0.846
YesSmallNoNo0.3750.161–0.6420.3620.183–0.588
SmallYesNo0.4460.183–0.7400.4770.253–0.720
LargeNoNo0.4470.171–0.7560.4850.256–0.731
SmallNoYes0.4570.194–0.7460.4880.264–0.726
LargeYesNo0.5200.216–0.8110.6020.350–0.824
SmallYesYes0.5300.230–0.8110.6050.357–0.822
LargeNoYes0.5320.233–0.8110.6130.365–0.828
LargeYesYes0.6020.297–0.8490.7180.477–0.891

See Table 4 for key.

Discussion

In the study reported here, we evaluated associations between herd management practices and high MAP seroprevalence in dairy herds in central California. All herds that participated were considered infected on the basis of positive bacteriologic culture results, and management practices used in herds classified as having a high seroprevalence of MAP infection (≥ 4% of cows seropositive) were compared with practices used in herds with low or zero seroprevalence of MAP infection (< 4% of cows seropositive). Herds in the lowto zero-seroprevalence classification were managed with practices that kept the prevalence of MAP infection low, even though the herds were likely to have been infected on the basis of results of previous bacteriologic cultures. Therefore, the study targeted management practices associated with dissemination of MAP within herds once they were already infected with MAP. Herds that were not confirmed as infected with MAP were excluded from statistical analyses because inclusion of those herds would have complicated differentiation of factors associated with a lower risk of introduction of MAP infection from factors associated with a lower risk of transmission of infection.

The ELISA test kit was chosen to estimate seroprevalence because it has been used for detection of antibodies against MAP in most studies2,7–10 of risk factors for MAP infection, thereby facilitating comparison of results. To account for the long incubation period of paratuberculosis, only the results of the second serum ELISA (on blood samples obtained approx 2.5 years after data on management practices were collected) were used in the statistical analyses. The agreement between seroprevalence estimates at the 2 sample collection times was moderate. Although the differences detected may have been real, they were more likely attributable to sampling variation and possible clustering of the seroprevalence of MAP infection by production group. In California, cows are commonly grouped by age, stage of lactation, milk yield (relative value compared with lactation-matched herdmates), and milk salability. These factors are associated with MAP seropositivity.24

Results of the study reported here may be applicable to the larger population of California dairies. Although the herds were not randomly selected from all dairy herds in central California, certain characteristics (distributions of herd sizes and seroprevalences of MAP infections within herds) correspond with those of a larger seroprevalence study15 conducted in central California. The larger study involved random selection of herds to estimate seroprevalence of paratuberculosis; therefore, the correspondence between results of that study15 and our study provides strong evidence that our results were representative of the region as a whole.

In our study, Bayesian logistic regression was used with a noninformative prior distribution and a partially informative prior distribution to incorporate minimal information about herd management risk factors. The partially informative prior distribution was used to improve model estimates by constraining 95% PIs for all regression coefficients to positive values. Because all variables in the model were presumed to be biologically associated with an increased risk of MAP transmission, it is reasonable to incorporate that belief into the model using the partially informative prior distribution.

When noninformative prior distributions were used in the predictive model, the posterior probabilities of positive regression coefficients of most variables were > 0.6 but < 0.8, which indicated positive associations between high seroprevalence of MAP infection and the variables investigated; however, the associations were not significant (ie, P < 0.8).23 In the Bayesian logistic regression model, each regression coefficient was represented as a distribution and not as a single value. Hence, the posterior probability of a positive or negative regression coefficient was obtained by calculating the proportion of Monte Carlo iterates from the posterior distribution for that regression coefficient. This proportion ranged from 0 to 1, where 0 represented 100% certainty of a negative regression coefficient or negative association with a herd having high seroprevalence and 1 represented 100% certainty of a positive regression coefficient or positive association with a herd having high seroprevalence. In the present study, the cutoff value for determining a significantly positive association between a predictor and high seroprevalence was set at 0.8, and the cutoff for a significantly negative association was set at 0.2. These cutoff values will vary depending on the requirements of the investigators.

Management practices related to milk and manure handling likely influenced the high seroprevalence of MAP infection in the herds in our study. Herds in which management practices involved feeding of unsalable milk to calves, flushing of cow walkways with recycled lagoon water, handling of feed with manurehandling equipment, and exposing heifers ≤ 6 months old to manure of adult cows had an increased risk of a high seroprevalence of MAP infection. Indirect or direct exposure to lagoon water is a risk factor for calves becoming infected with MAP,24 and most herds in the study reported here were exposed to water from MAP-positive lagoons.17,25

A significant relationship between herd size and high seroprevalence of MAP infection was also detected. Large size of dairy herds was associated with a higher seroprevalence of MAP infection in the Netherlands10 and in some areas in Wisconsin.7 A biological explanation for this relationship is that higher cattle-stocking densities may result in higher burdens of MAP in the environment and thus a greater risk of infection, compared with the burdens of MAP and risk of infection in smaller herds.

Other management practices regarding calving area (ie, separation of calving areas from housing for lactating cows, frequency of bedding changes in calving area, and interval from parturition until separation of newborn calves from their dams) were not significantly associated with the seroprevalence of MAP infection in the herds in our study. The lack of significance may have been attributable to the small number of herds included, which may have limited the power to detect significant risk factors associated with high seroprevalence of MAP infection in herds. However, another study11 of 54 dairy herds also revealed that the interval from birth until separation of newborn calves from their dams was not associated with the probability of infection with MAP.

Associations detected in the study reported here were biologically plausible and consistent with those detected in other studies.2,8,10,11 Our predictive model indicated that herd size influenced the probability of a herd having a high seroprevalence of MAP infection and that good management practices related to use of unsalable milk and handling of manure were also important. Incorporation of prior information regarding herd management factors was useful in evaluating associations with a high seroprevalence of MAP infection in the selected herds. We chose to evaluate herd-level management factors as opposed to cow-level factors because herd-level management factors can be modified by the herd owner. Moreover, cow-level factors are more difficult to change, and although there is some evidence of a genetic predisposition to paratuberculosis,26 information regarding genealogy was not available for the herds in the present study. When cow-level factors are also of interest, then a hierarchic regression model that combines cow- and herd-level data and includes a random effect should be used to evaluate cow and herd factors.27

ABBREVIATIONS

MAP

Mycobacterium avium subsp paratuberculosis

S:P ratio

Test sample-to-positive control sample ratio

PI

Probability interval

a.

California Animal Health and Food Safety Laboratory, University of California, Davis, Calif.

b.

HerdChek, IDEXX Laboratories Inc, Westbrook, Me.

c.

WinBUGS, version 1.4, MRC Biostatistics Unit, Cambridge, England. Available at: www.mrc-bsu.cam.ac.uk/bugs. Accessed April 1, 2008.

d.

Logit function, WinBUGS, version 1.4, MRC Biostatistics Unit, Cambridge, England.

e.

Step function, WinBUGS, version 1.4, MRC Biostatistics Unit, Cambridge, England.

References

  • 1.

    Lombard JE, Wagner BA, Smith RL, et al. Evaluation of environmental sampling and culture to determine Mycobacterium avium subspecies paratuberculosis distribution and herd infection status on US dairy operations. J Dairy Sci 2006;89:41634171.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 2.

    Wells SJ, Wagner BA. Herd-level risk factors for infection with Mycobacterium paratuberculosis in US dairies and association between familiarity of the herd manager with the disease or prior diagnosis of the disease in that herd and use of preventive measures. J Am Vet Med Assoc 2000;216:14501457.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 3.

    Sweeney RW. Transmission of paratuberculosis. Vet Clin North Am Food Anim Pract 1996;12:305312.

  • 4.

    Groenendaal H, Nielen M, Jalvingh AW, et al. A simulation of Johne's disease control. Prev Vet Med 2002;54:225245.

  • 5.

    Groenendaal H, Galligan DT. Economic consequences of control programs for paratuberculosis in midsize dairy farms in the United States. J Am Vet Med Assoc 2003;223:17571763.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 6.

    Jubb TF, Galvin JW. Effect of a test and control program for bovine Johne's disease in Victorian dairy herds 1992–2002. Aust Vet J 2004;82:228232.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 7.

    Collins MT, Sockett DC, Goodger WJ, et al. Herd prevalence and geographic distribution of, and risk factors for, bovine paratuberculosis in Wisconsin. J Am Vet Med Assoc 1994;204:636641.

    • Search Google Scholar
    • Export Citation
  • 8.

    Hirst HL, Garry FB, Morley PS, et al. Seroprevalence of Mycobacterium avium subsp paratuberculosis infection among dairy cows in Colorado and herd-level risk factors for seropositivity. J Am Vet Med Assoc 2004;225:97101.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 9.

    Johnson-Ifearulundu YJ, Kaneene JB. Management-related risk factors for M paratuberculosis infection in Michigan, USA, dairy herds. Prev Vet Med 1998;37:4154.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 10.

    Muskens J, Elbers A, Van Weering H, et al. Herd management practices associated with paratuberculosis seroprevalence in Dutch dairy herds. J Vet Med B Infect Dis Vet Public Health 2003;50:372377.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 11.

    Obasanjo IO, Grohn YT, Mohammed HO. Farm factors associated with the presence of Mycobacterium paratuberculosis infection in dairy herds on the New York State Paratuberculosis Control Program. Prev Vet Med 1997;32:243251.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 12.

    Ridge SE, Baker IM, Hannah M. Effect of compliance with recommended calf-rearing practices on control of bovine Johne's disease. Aust Vet J 2005;83:8590.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 13.

    Collins MT, Gardner IA, Garry FB, et al. Consensus recommendations on diagnostic testing for the detection of paratuberculosis in cattle in the United States. J Am Vet Med Assoc 2006;229:19121919.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 14.

    National Agriculture Statistics Services. California livestock and dairies: milk cows, dairies, cows per dairy and milk production by county and region, 2002–2003. Available at: www.nass.usda.gov/Statistics_by_State/California/index.asp. Accessed Apr 1, 2008.

    • Search Google Scholar
    • Export Citation
  • 15.

    Adaska JM, Anderson RJ. Seroprevalence of Johne's-disease infection in dairy cattle in California, USA. Prev Vet Med 2003;60:255261.

  • 16.

    Tavornpanich S, Gardner IA, Anderson RJ, et al. Evaluation of microbial culture of pooled fecal samples for the detection of Mycobacterium avium subsp paratuberculosis in large dairy herds. Am J Vet Res 2004;65:10611070.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 17.

    Berghaus RD, Farver TB, Anderson RJ, et al. Environmental sampling for detection of Mycobacterium avium ssp paratuberculosis on large California dairies. J Dairy Sci 2006;89:963970.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 18.

    USDA APHIS. Uniform program standards for the Voluntary Bovine Johne's Disease Control Program, effective June 1, 2006. USDA, 2006. Available at: www.aphis.usda.gov/animal_health/animal_diseases/johnes/downloads/johnes-umr.pdf. Accessed Apr 1, 2008.

    • Search Google Scholar
    • Export Citation
  • 19.

    Chen MH, Shao QM, Ibrahim JG. Variable selection for logistic regression models. In: Monte Carlo methods in Bayesian computations. New York: Springer, 2000;268275.

    • Search Google Scholar
    • Export Citation
  • 20.

    Mila AL, Yang XB, Carriquiry AL. Bayesian logistic regression of soybean stem rot prevalence in the US North-Central region: accounting for uncertainty in parameter estimation. Phytopathology 2003;93:758764.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 21.

    Carlin PB, Louis TA. The Bayes approach. In: Bayes and empirical Bayes methods for data analysis. New York: Chapman & Hall, 2000;20.

  • 22.

    Gelman A, Rubin DB. Inference from iterative simulation using multiple sequences. Stat Sci 1992;7:457472.

  • 23.

    Thurmond MC, Branscum AJ, Johnson WO, et al. Predicting the probability of abortion in dairy cows: a hierarchical Bayesian logistic-survival model using sequential pregnancy data. Prev Vet Med 2005;68:223239.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 24.

    Tavornpanich S, Gardner IA, Carpenter TE, et al. Evaluation of cost-effectiveness of targeted sampling methods for detection of Mycobacterium avium subsp paratuberculosis infection in dairy herds. Am J Vet Res 2006;67:821828.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 25.

    Aly SS, Thurmond MC. Evaluation of Mycobacterium avium subsp paratuberculosis infection of dairy cows attributable to infection status of the dam. J Am Vet Med Assoc 2005;227:450454.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 26.

    Koets AP, Adugna G, Janss LL, et al. Genetic variation of susceptibility to Mycobacterium avium subsp. paratuberculosis infection in dairy cattle. J Dairy Sci 2000;83:27022708.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 27.

    Jackson C, Best N, Richardson S. Improving ecological inference using individual-level data. Stat Med 2006;25:21362159.

  • 1.

    Lombard JE, Wagner BA, Smith RL, et al. Evaluation of environmental sampling and culture to determine Mycobacterium avium subspecies paratuberculosis distribution and herd infection status on US dairy operations. J Dairy Sci 2006;89:41634171.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 2.

    Wells SJ, Wagner BA. Herd-level risk factors for infection with Mycobacterium paratuberculosis in US dairies and association between familiarity of the herd manager with the disease or prior diagnosis of the disease in that herd and use of preventive measures. J Am Vet Med Assoc 2000;216:14501457.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 3.

    Sweeney RW. Transmission of paratuberculosis. Vet Clin North Am Food Anim Pract 1996;12:305312.

  • 4.

    Groenendaal H, Nielen M, Jalvingh AW, et al. A simulation of Johne's disease control. Prev Vet Med 2002;54:225245.

  • 5.

    Groenendaal H, Galligan DT. Economic consequences of control programs for paratuberculosis in midsize dairy farms in the United States. J Am Vet Med Assoc 2003;223:17571763.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 6.

    Jubb TF, Galvin JW. Effect of a test and control program for bovine Johne's disease in Victorian dairy herds 1992–2002. Aust Vet J 2004;82:228232.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 7.

    Collins MT, Sockett DC, Goodger WJ, et al. Herd prevalence and geographic distribution of, and risk factors for, bovine paratuberculosis in Wisconsin. J Am Vet Med Assoc 1994;204:636641.

    • Search Google Scholar
    • Export Citation
  • 8.

    Hirst HL, Garry FB, Morley PS, et al. Seroprevalence of Mycobacterium avium subsp paratuberculosis infection among dairy cows in Colorado and herd-level risk factors for seropositivity. J Am Vet Med Assoc 2004;225:97101.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 9.

    Johnson-Ifearulundu YJ, Kaneene JB. Management-related risk factors for M paratuberculosis infection in Michigan, USA, dairy herds. Prev Vet Med 1998;37:4154.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 10.

    Muskens J, Elbers A, Van Weering H, et al. Herd management practices associated with paratuberculosis seroprevalence in Dutch dairy herds. J Vet Med B Infect Dis Vet Public Health 2003;50:372377.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 11.

    Obasanjo IO, Grohn YT, Mohammed HO. Farm factors associated with the presence of Mycobacterium paratuberculosis infection in dairy herds on the New York State Paratuberculosis Control Program. Prev Vet Med 1997;32:243251.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 12.

    Ridge SE, Baker IM, Hannah M. Effect of compliance with recommended calf-rearing practices on control of bovine Johne's disease. Aust Vet J 2005;83:8590.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 13.

    Collins MT, Gardner IA, Garry FB, et al. Consensus recommendations on diagnostic testing for the detection of paratuberculosis in cattle in the United States. J Am Vet Med Assoc 2006;229:19121919.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 14.

    National Agriculture Statistics Services. California livestock and dairies: milk cows, dairies, cows per dairy and milk production by county and region, 2002–2003. Available at: www.nass.usda.gov/Statistics_by_State/California/index.asp. Accessed Apr 1, 2008.

    • Search Google Scholar
    • Export Citation
  • 15.

    Adaska JM, Anderson RJ. Seroprevalence of Johne's-disease infection in dairy cattle in California, USA. Prev Vet Med 2003;60:255261.

  • 16.

    Tavornpanich S, Gardner IA, Anderson RJ, et al. Evaluation of microbial culture of pooled fecal samples for the detection of Mycobacterium avium subsp paratuberculosis in large dairy herds. Am J Vet Res 2004;65:10611070.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 17.

    Berghaus RD, Farver TB, Anderson RJ, et al. Environmental sampling for detection of Mycobacterium avium ssp paratuberculosis on large California dairies. J Dairy Sci 2006;89:963970.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 18.

    USDA APHIS. Uniform program standards for the Voluntary Bovine Johne's Disease Control Program, effective June 1, 2006. USDA, 2006. Available at: www.aphis.usda.gov/animal_health/animal_diseases/johnes/downloads/johnes-umr.pdf. Accessed Apr 1, 2008.

    • Search Google Scholar
    • Export Citation
  • 19.

    Chen MH, Shao QM, Ibrahim JG. Variable selection for logistic regression models. In: Monte Carlo methods in Bayesian computations. New York: Springer, 2000;268275.

    • Search Google Scholar
    • Export Citation
  • 20.

    Mila AL, Yang XB, Carriquiry AL. Bayesian logistic regression of soybean stem rot prevalence in the US North-Central region: accounting for uncertainty in parameter estimation. Phytopathology 2003;93:758764.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 21.

    Carlin PB, Louis TA. The Bayes approach. In: Bayes and empirical Bayes methods for data analysis. New York: Chapman & Hall, 2000;20.

  • 22.

    Gelman A, Rubin DB. Inference from iterative simulation using multiple sequences. Stat Sci 1992;7:457472.

  • 23.

    Thurmond MC, Branscum AJ, Johnson WO, et al. Predicting the probability of abortion in dairy cows: a hierarchical Bayesian logistic-survival model using sequential pregnancy data. Prev Vet Med 2005;68:223239.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 24.

    Tavornpanich S, Gardner IA, Carpenter TE, et al. Evaluation of cost-effectiveness of targeted sampling methods for detection of Mycobacterium avium subsp paratuberculosis infection in dairy herds. Am J Vet Res 2006;67:821828.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 25.

    Aly SS, Thurmond MC. Evaluation of Mycobacterium avium subsp paratuberculosis infection of dairy cows attributable to infection status of the dam. J Am Vet Med Assoc 2005;227:450454.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 26.

    Koets AP, Adugna G, Janss LL, et al. Genetic variation of susceptibility to Mycobacterium avium subsp. paratuberculosis infection in dairy cattle. J Dairy Sci 2000;83:27022708.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 27.

    Jackson C, Best N, Richardson S. Improving ecological inference using individual-level data. Stat Med 2006;25:21362159.

Advertisement