Accurate classification of herd infection status is a critically important component of a successful control program for paratuberculosis. However, a scientifically sound and cost-effective method of ascertaining with high confidence the correct classification of herds as infected or noninfected has yet to be determined. In 1998, the VJDHSP was developed by the US Animal Health Association's National Johne's Working Group to provide guidelines for identification of herds with a low risk of paratuberculosis.1 The screening method used at the first level of VJDHSP testing, in which 30 randomly selected cows in the second or higher lactation were tested by use of a combination of ELISA results for serum antibodies and microbial culture of feces in seropositive cows, misclassified 67% of dairy herds that had a low within-herd prevalence of MAP infection; those misclassifications were false negatives because the specificity of fecal microbial culture as a confirmatory test was 100%.2 To increase the probability of detecting paratuberculosis in herds with low within-herd prevalence, more cows must be tested. This could result in higher costs and lower herd-level specificity if the specificity of the diagnostic test used was imperfect. In addition, increased costs could decrease owner participation in the voluntary control program.
If early detection of MAP-infected herds is the primary goal of the VJDHSP program, the detection probability could be increased by targeted sampling of subgroups of cows that, if infected, would have an increased probability of having a detectable serologic response or fecal shedding of the organism. Targeted sampling has successfully been used for detecting subclinical Salmonella spp infection in Danish swine herds and for screening cattle for BSE in Switzerland.3,4 A study4 of BSE revealed that targeted screening of highrisk cattle (such as cows culled because of health problems) was 49 to 58 times as likely to detect cows with BSE as the passive surveillance measures that are used in most countries.4 Earlier studies5–7 of paratuberculosis revealed that cows infected with MAP at a certain age, parity, and stage of lactation develop higher serum concentrations of antibody against the bacterium and shed high numbers of the organism in feces. Results of other studies8–11 also indicated there were associations with low milk production, a high proportion of cows with mastitis, and a high premature culling rate in infected cows. However, to our knowledge, the effect of sampling targeted groups of cows on detection of MAP-infected herds has not been investigated.
The present study evaluated the epidemiologic and financial impacts of sampling that was targeted on the basis of lactation number, stage of lactation, and milk production on classification of herd infection status. Sampling without replacement was used to estimate the probability of detection of MAP-infected herds for different sampling methods and sample sizes. Marginal cost-effectiveness analysis12 was used to measure the changes in cost per unit of increase in detection probability and to compare the difference in costs among sampling methods. Objectives of the study reported here were to compare the probability of detecting an MAP-infected herd on the basis of sampling from targeted groups of cows with that resulting from random sampling of cows in lactation 2 or higher (the criterion for level 1 testing in the VJDHSP) and to determine the cost-effectiveness of sampling methods for detection of paratuberculosis in large dairy herds with low-to-moderate within-herd prevalence.
Materials and Methods
Study populations—Data were obtained from 5 dairy herds in the San Joaquin Valley of California. Data were collected from 4 dairy herds from January to March 2003 and from a fifth herd during January 2004. Four herds were MAPinfected herds, and 1 herd was likely noninfected (determined on the basis of results of a previous study13 in which the prevalences of infection were estimated by use of subsets of cows in the herds). In each herd, all cows in lactation 1 or higher that were available on the days of sample collection were included in the study and used as the sampling frame for the simulation studies. Blood samples were collected from all cows, and fecal samples were collected from cows that were seropositive for antibodies against MAP. Cow-level information, including lactation number, days in lactation, and milk production, was extracted from a dairy management information systema at the time of sample collection.
Testing methods and classification of MAP-infection status in individual cows—The infection status of each cow was classified on the basis of results of serologic testing and fecal bacterial culture as performed at the California Animal Health and Food Safety Laboratories in Davis and Tulare. Sera were tested for antibodies against MAP by use of a commercially available ELISA kit,b and results were reported as the ratio of sample values to positive control values (S:P). Fecal samples from all cows with ELISA S:P ratios ≥ 0.25 were cultured by use of a double-incubation and centrifugation technique on an egg-yolk medium.c Samples with positive culture results were confirmed on the basis of mycobactin dependency, acid-fast staining, and evidence of colony and microscopic morphologic characteristics consistent with MAP. If cow infection status was determined on the basis of ELISA results alone, an S:P ratio ≥ 0.25 was used to classify cows as infected with MAP. If the ELISA-FC was used to classify cow infection status, cows with positive results for both ELISA and fecal culture were considered infected.
Evaluation of herd sampling methods—Thirteen sampling scenarios, with sample sizes ranging from 30 to 150 cows, were used to select cows from each herd. Test results from selected cows were used to calculate the number of cows with positive test results from the total number of tested cows. For each sampling scenario, the number of samples collected began with 30 cows, which is the sample size used for the initial testing in the VJDHSP; the number of samples collected was increased in 30-cow increments thereafter until a ≥ 95% detection probability was obtained.
The scenarios included sampling the selected number of cows from 1) all cows, 2) cows in lactation 1, and 3) cows in lactation 2 or higher; cows in lactation 2 or higher were subcategorized as 4) cows that were due to calve in 60 days or designated as not lactating in the computerized records,c 5) cows < 91 days into lactation or designated as newly calved cows in records, 6) cows 91 to 200 days into lactation or designated as being in midlactation in records, 7) cows > 200 days into lactation or designated as being in late lactation in records, and 8) cows in lactation 3 or higher. Cows in lactation 3 or higher were subcategorized as 9) nonlactating cows, 10) newly calved cows, 11) cows in midlactation, and 12) cows in late lactation; the last group was 13) cows with low milk production. For each cow, the classification of low milk production was made on the basis of the value of the 305-day mature equivalent milk yield, which is a predicted value for milk production that is adjusted for age and stage of lactation in each cow relative to the mean mature equivalent milk yield of each herd. On the basis of each farm's computerized records, the mean value for mature equivalent milk yield of the herd was set at 100; a cow with a relative value < 100 was classified as a low-production cow, and cows with a relative value > 100 were classified as high-production cows. For each herd, a low-to-high milk-production ratio was calculated from the number of cows with low milk production divided by the number of cows with high milk production and stratified by stage of lactation. Scenarios 3 to 13 constituted the targeted sampling methods used in the present study.
Simulation method—The Monte Carlo simulation method was used to obtain the probability of detection of an MAP-infected herd for each sampling method. Briefly, for each of the 4 herds, all cows were used to create a sampling frame. A sample of randomly selected cows was taken from the sampling frame without replacement via 1 of the 13 sampling methods, and the number of cows with positive test results (x) from n sampled cows was determined. The herd was classified as infected if × ≥ 1. Each sampling scenario was simulated 1,000 times to obtain an estimate of detection probability, which was equal to the proportion of events in which × ≥ 1 out of 1,000 times for a given sampling scenario and sample size. This is equivalent to the probability that the infected herd was accurately classified as infected. The term herd-level sensitivity was not applicable to this study because the simulation output was derived from a single herd. The simulation outcome derived by use of the sampling frame from the noninfected herd was termed the misclassification probability, or the probability that the noninfected herd was misclassified as infected. All simulations were performed with statistical software.d
Cost-effectiveness analysis—Cost-effectiveness of each sampling method was defined as the ratio of cost to detection probability. Cost was estimated on the basis of current prices for services at California Animal Health and Food Safety laboratory facilities, which were $6 for the ELISA and $16 for fecal culture. The detection probability was obtained from the simulation model as the measure of effectiveness of each sampling method. To be consistent with the level 1 (ie, initial) testing method used by the VJDHSP, the detection probability for each sampling method was made on the basis of results of ELISA-FC. If 2 sampling methods had the same detection probability, the method that yielded the lower cost was defined as the more cost-effective sampling method. For the same sampling method, cost and detection probability increased with larger sample sizes. Marginal cost-effectiveness was used to measure how much the cost would increase relative to the increase in detection probability.12
where C1 is the cost of method 1, C2 the cost of method 2, DP1 is the detection probability of method 1, and DP2 is the detection probability of method 2.
Results
Descriptive analysis—Herd sizes (numbers included all lactating and nonlactating cows) for the 5 participating herds designated as A, B, C, D, and E were 329, 1,023, 1,301, 2,474, and 1,154 cows, respectively. Frequency distributions of cows by lactation number, stage of lactation, and milk production were summarized (Table 1). Most cows in the study were in lactation 2 or higher, and cows in lactation 1 comprised 5% to 10% of the herds except in herds B and D, in which 33% of cows were in lactation 1; the proportion of cows in lactation 1 in herds A, C, and E was small because some cows in lactation 1 were not available for sampling on the day of sample collection. Therefore, the effect of stage of lactation on detection probability was not investigated for cows in lactation 1. The frequency distribution by stage of lactation and milk production was calculated for cows in each herd. Cows had a dissimilar distribution of stage of lactation across dairies, although within a given dairy, most cows were lactating, and overall, 13% to 18% of cows were in the dry period (not lactating). In all herds except herd B, in which data for milk production were not available, there was a higher percentage of cows with low milk production than cows with high milk production. The difference was small (4%) in herd A but was large (18% to 20%) in herds C and D. The herd-specific low-to-high milk-production ratio was highest in early-lactation cows (range, 2.7 in herd A to 39 in herd D) and lowest in late-lactation cows (range, 0.35 in herd D to 0.81 in herd A).
Frequency (%) distribution of cows in 5 California dairy herds classified by lactation number, stage of lactation, and milk production. Herds A, B, C, and D were known to contain cows infected with MAP, whereas Herd E was a noninfected herd.
| Subgroup | Herd A | Herd B | Herd C | Herd D | Herd E |
|---|---|---|---|---|---|
| All cows | 329 | 1,023 | 1,301 | 2,474 | 1,154 |
| Lactation No.* | |||||
| 1 | 32 (10) | 339 (33) | 65 (5) | 811 (33) | 112 (10) |
| 2 | 125 (38) | 217 (21) | 543 (42) | 619 (25) | 439 (38) |
| 3 | 73 (22) | 196 (19) | 362 (28) | 451 (18) | 285 (25) |
| 4 | 45 (14) | 144 (14) | 220 (17) | 312 (13) | 185 (15) |
| 5 or higher | 54 (16) | 127 (13) | 105 (8) | 281 (11) | 133 (12) |
| Stage of lactation† | |||||
| Not lactating | 49 (15) | 130 (13) | 229 (18) | 394 (16) | 176 (15) |
| < 91 days into lactation | 61 (19) | 354 (35) | 325 (25) | 782 (32) | 250 (22) |
| 91–200 days into lactation | 96 (29) | 294 (29) | 371 (29) | 627 (26) | 298 (26) |
| > 200 days into lactation | 123 (37) | 237 (23) | 353 (28) | 647 (26) | 416 (37) |
| Milk production | |||||
| Low | 170 (52) | NA | 784 (60) | 1450 (59) | 643 (56) |
| High | 159 (48) | NA | 517 (40) | 1024 (41) | 511 (44) |
Six cows in herd C had a missing value for lactation number.
Eight, 23, 24, and 14 cows had missing values for stage of lactation in herds B, C, D, and E, respectively. The percentage was calculated on the basis of available data.
NA = Data not available.
Test results for ELISA alone and ELISA-FC for each herd were summarized (Tables 2 and 3, respectively). The prevalence of cows with positive results of ELISA testing alone varied from 1.2% (14/1,154 cows) in herd E to 8.1% (201/2,474 cows) in herd D, and the overall prevalence in all infected herds (herds A to D) was 6.1%. The prevalence of cows with positive results of combined ELISA-FC testing ranged from 0% in herd E to 2.8% (n = 71) in herd D, and the overall prevalence in all infected herds (herds A to D) was 2.3%. There was substantial variation in the proportion of cows with positive ELISA results among the subgroups, and within-herd prevalence was positively correlated with the degree of heterogeneity. Interestingly, there was little variation in the proportion of cows with positive ELISA results among the subgroups in the noninfected herd (herd E). Generally, the proportions of cows with positive ELISA results and cows with positive ELISA-FC–results increased as the number of lactations increased. In 3 of the 4 infected herds, the proportion of cows with positive ELISA-FC results was highest in cows in late lactation (ie, > 200 days into lactation), and the same was true for the subset of cows in lactation 2 or higher and the subset of cows in lactation 3 or higher. Stratification of the data on the basis of lactation number and stage of lactation (ie, nesting of the stage of lactation within the lactation number) revealed evidence of interaction between those 2 variables (Figure 1). Therefore, the effect of stage of lactation on the detection probability was evaluated separately on the basis of the subset of cows in lactation 2 or higher and cows in lactation 3 or higher. In general, cows with low milk production were more likely to have positive ELISA and ELISAFC results. The differences in prevalence as indicated by ELISA-FC results between cows with low milk production and cows with high milk production ranged from 0.2% to 2.3%.
Frequency and prevalence (%) of cows with positive ELISA results for serum antibodies against MAP in the same herds as in Table 1, classified by lactation number, stage of lactation, and milk production.
| Subgroup | Herd A | Herd B | Herd C | Herd D | Herd A to D | Herd E |
|---|---|---|---|---|---|---|
| All cows | 10 (3.0) | 73 (7.1) | 31 (2.4) | 201 (8.1) | 315 (6.1) | 14 (1.2) |
| Lactation No. | ||||||
| 1 | 0 (0) | 17 (5.1) | 1 (1.5) | 14 (1.7) | 32 (2.6) | 1 (0.9) |
| 2 | 0 (0) | 11 (5.0) | 4 (0.7) | 32 (0.8) | 47 (3.1) | 5 (1.1) |
| 3 | 1 (1.4) | 6 (3.1) | 11 (3.0) | 50 (11) | 68 (6.2) | 4 (1.4) |
| 4 | 3 (7.0) | 18 (12.5) | 10 (4.5) | 50 (16) | 81 (11.2) | 3 (1.6) |
| 5 or higher | 6 (11.1) | 21 (16.5) | 5 (4.8) | 55 (20) | 87 (15.3) | 1 (0.8) |
| Subset of cows in lactation 2 or higher* | ||||||
| Not lactating | 1 (3.4) | 10 (9.1) | 5 (3.0) | 32 (13.4) | 48 (9.6) | 0 (0) |
| < 91 days into lactation | 3 (5.0) | 17 (7.6) | 4 (1.2) | 53 (10.0) | 77 (7.3) | 5 (2.1) |
| 91–200 days into lactation | 3 (3.3) | 11 (7.2) | 8 (2.1) | 42 (9.5) | 64 (6.6) | 4 (1.4) |
| > 200 days into lactation | 3 (2.6) | 18 (9.3) | 13 (3.7) | 59 (13.5) | 93 (8.3) | 4 (1.0) |
| Subset of cows in lactation 3 or higher† | ||||||
| Not lactating | 1 (6.3) | 9 (13.2) | 5 (4.9) | 27 (17.9) | 42 (13.3) | 0 (0) |
| < 91 days into lactation | 3 (7.9) | 14 (8.8) | 2 (1.2) | 43 (13.3) | 62 (9.4) | 3 (2.1) |
| 91–200 days into lactation | 3 (5.6) | 8 (8.3) | 8 (3.6) | 32 (11.8) | 51 (8.6) | 3 (1.8) |
| 200 days into lactation | 3 (4.7) | 14 (9.9) | 11 (5.9) | 52 (18.1) | 80 (11.5) | 2 (1.0) |
| Milk production | ||||||
| Low | 8 (4.7) | NA | 21 (2.6) | 114 (7.8) | 143 (5.9) | 8 (1.2) |
| High | 2 (1.3) | NA | 10 (1.9) | 87 (8.5) | 99 (5.8) | 6 (1.2) |
Calculation made on the basis of a subset of cows in lactation 2 or higher.
Calculation made on the basis of a subset of cows in lactation 3 or higher.
NA = Data not available.
Frequency and prevalence (%) of cows with positive results of both serum ELISA testing and fecal culture for MAP in the same herds as in Tables 1 and 2, classified by lactation number, stage of lactation, and milk production.
| Subgroup | Herd A | Herd B | Herd C | Herd D | Herd A to D | Herd E |
|---|---|---|---|---|---|---|
| All cows | 6(1.8) | 28.(2.8) | 11(0.8) | 71(2.8) | 116(2.3) | 0(0) |
| Lactation No. | ||||||
| 1 | 0 (0) | 8 (2.4) | 0 (0) | 4 (0.5) | 12 (1.0) | 0 (0) |
| 2 | 0 (0) | 4 (1.8) | 1 (0.2) | 9 (1.5) | 14 (0.9) | 0 (0) |
| 3 | 1 (1.4) | 2 (1.0) | 5 (1.4) | 15 (3.3) | 23 (2.1) | 0 (0) |
| 4 | 1 (2.2) | 8 (5.6) | 2 (0.9) | 18 (5.8) | 29 (4.0) | 0 (0) |
| 5 or higher | 4 (7.4) | 6 (4.8) | 3 (2.9) | 25 (8.9) | 38 (6.7) | 0 (0) |
| Subset of cows in lactation 2 or higher* | ||||||
| Not lactating | 0 (0) | 1 (0.9) | 1 (0.6) | 13 (5.4) | 15 (3.0) | 0 (0) |
| < 91 days into lactation | 3 (5.0) | 6 (2.7) | 1 (0.3) | 10 (1.9) | 20 (1.9) | 0 (0) |
| 91–200 days into lactation | 2 (2.2) | 5 (2.5) | 4 (1.1) | 16 (3.6) | 27 (2.7) | 0 (0) |
| > 200 days into lactation | 1 (0.9) | 8 (5.3) | 5 (1.4) | 27 (6.2) | 41 (3.8) | 0 (0) |
| Subset of cows in lactation 3 or higher† | ||||||
| Not lactating | 0 (0) | 0 (0) | 1 (1.0) | 11 (7.3) | 12 (3.8) | 0 (0) |
| < 91 days into lactation | 3 (7.9) | 6 (3.8) | 0 (0) | 8 (2.5) | 17 (2.5) | 0 (0) |
| 91–200 days into lactation | 2 (3.7) | 4 (2.8) | 4 (1.8) | 11 (4.1) | 21 (3.3) | 0 (0) |
| > 200 days into lactation | 1 (1.6) | 6 (6.1) | 5 (2.7) | 27 (9.4) | 39 (6.0) | 0 (0) |
| Milk production | ||||||
| Low | 5 (2.9) | NA | 6 (0.7) | 46 (3.1) | 57 (2.4) | 0 (0) |
| High | 1 (0.6) | NA | 5 (0.9) | 25 (2.4) | 31 (1.8) | 0 (0) |
See Table 2 for key.
Assuming values of 25% and 35% for sensitivity of detection by use of ELISA and fecal culture, respectively, and a value for specificity of 100% for fecal culture,13 the sensitivity and specificity of ELISA-FC were estimated to be 9% and 100%, respectively, on the basis of serial interpretation of 2 independent tests.14 Therefore, true within-herd prevalence of the disease for each herd was estimated by use of the following formula: true prevalence = (apparent prevalence + specificity −1)/(sensitivity + specificity − 1),15 which resulted in prevalence estimates of 20%, 31%, 9%, and 31% for herds A, B, C, and D, respectively. Estimation of true prevalence in herd D, assuming a value of 25% for ELISA sensitivity and 98% for ELISA specificity,12 was consistent with the assumption that the herd was noninfected.
Distribution of proportion of cows with positive results of the ELISA-FC test combination in lactations 2, 3, 4, and 5 or higher for all cows and stratified by stage of lactation to nonlactating, newly calved, midlactation, and late-lactation cows.
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.821
Detection probability—Estimated detection probabilities for all sampling scenarios were dependent on the within-herd prevalence, number of samples, and target group of sampled cows (Tables 4 and 5). Results from the simulation revealed that sampling from cows in lactation 1 (scenario 2) typically yielded the lowest detection probability, which was not unexpected because of the long incubation period associated with MAP infection. For the ELISA-FC combination, sampling 30 cows in lactation 2 or higher (scenario 3) detected only 23% of infected cows in herd C (which had an estimated true within-herd prevalence of 9%), 47% in herd A (estimated true within-herd prevalence of 20%), 60% in herd B (estimated true within-herd prevalence of 31%), and 70% in herd D (estimated true within-herd prevalence of 31%). Sampling from cows in lactation 3 or higher, regardless of lactation stage (scenario 8), yielded a detection probability that was 1.5, 1.1, 1.6, and 1.2 times as high as that determined from sampling cows in lactation 2 or higher for herds A, B, C, and D, respectively.
Detection probability of 13 sampling scenarios designed on the basis of testing with ELISA results alone and with ELISA-FC results on samples from 30 cows/herd. Values given are percentages.
| Sampling scenario | ELISA | ELISA-FC | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Herd A | Herd B | Herd C | Herd D | Herd E | Herd A | Herd B | Herd C | Herd D | |
| 1 | 62 | 90 | 52 | 92 | 31 | 44 | 57 | 22 | 59 |
| 2 | 0 | 80 | 45 | 41 | 26 | 0 | 54 | 14 | 0 |
| 3* | 66 | 93 | 52 | 97 | 32 | 47 | 60 | 23 | 70 |
| 4 | NA | 97 | 64 | 99 | 0 | NA | 28 | 18 | 83 |
| 5 | 89 | 92 | 32 | 96 | 48 | 88 | 59 | 8 | 45 |
| 6 | 70 | 92 | 49 | 95 | 35 | 54 | 57 | 28 | 69 |
| 7 | 60 | 97 | 69 | 99 | 27 | 25 | 84 | 36 | 86 |
| 8 | 82 | 95 | 69 | 99 | 34 | 70 | 67 | 36 | 82 |
| 9 | NA | 99 | 83 | 99 | 0 | NA | 0 | 29 | 92 |
| 10 | 98 | 95 | 34 | 98 | 45 | 99† | 73 | 0 | 55 |
| 11 | 91 | 96 | 69 | 98 | 42 | 79 | 62 | 45 | 72 |
| 12 | 85 | 97 | 86 | 99 | 25 | 45 | 88† | 58† | 95† |
| 13 | 79 | NA | 56 | 91 | 31 | 62 | NA | 20 | 61 |
Indicates the VJDHSP screening method.
Indicates the sampling method in the ELISA-FC category that yielded the highest detection probability for each herd.
NA = Not applicable because the number of cows in subgroup was less than the sample size required (herd A), or data were not available (herd B).
Comparisons of detection probability percentage and cost-effectiveness value between a sampling scenario in which cows in lactation 2 or higher were randomly sampled (S1) and the sampling method that yielded the highest detection probability from simulation results (S2). The testing method was ELISA-FC and sample sizes ranged from 30 to 150 cows.
| Sampling scenario | Detection probability Sample size | Cost-effectiveness Sample size | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 30 | 60 | 90 | 120 | 150 | 30 | 60 | 90 | 120 | 150 | |
| Herd A | ||||||||||
| S1 | 47 | 74 | 88 | 95 | 98 | 3.9 | 5.1 | 6.5 | 8.0 | 9.7 |
| S2 | 99 | NA | NA | NA | NA | 2.2 | NA | NA | NA | NA |
| Herd B | ||||||||||
| S1 | 60 | 85 | 94 | 98 | 99 | 3.3 | 4.6 | 6.3 | 7.8 | 9.7 |
| S2 | 88 | 99 | 100 | NA | NA | 2.4 | 4.3 | 6.4 | NA | NA |
| Herd C | ||||||||||
| S1 | 23 | 43 | 56 | 67 | 76 | 8.0 | 8.6 | 9.9 | 11.0 | 12.1 |
| S2 | 58 | 85 | 96 | 99 | 100 | 3.6 | 4.9 | 6.5 | 8.4 | 10.4 |
| Herd D | ||||||||||
| S1 | 70 | 91 | 97 | 99 | 99 | 2.9 | 4.4 | 6.2 | 8.0 | 10.1 |
| S2 | 95 | 99 | 100 | 100 | 100 | 2.8 | 5.4 | 8.0 | 10.6 | 13.3 |
NA = Not applicable because the number of cows in the subgroup was less than the sample size required.
The detection probability was strongly influenced by stage of lactation. For 3 of the 4 infected herds (herds B, C, and D), sampling from cows in lactation 3 or higher that were in late lactation (scenario 12) yielded the highest detection probability for the ELISA-FC test. Interestingly, the detection probability of sampling from late-lactation cows in lactation 2 or higher (scenario 7) was higher than that derived by sampling all cows in lactation 3 or higher (scenario 8) for herds B and D. The ratios of detection probability for sampling scenario 12 to the detection probability for sampling scenario 3 were 1.5, 2.5, and 1.4 in herds B, C, and D, respectively. However, in herd A, sampling from cows in lactation 3 or higher that were in early lactation (scenario 10) yielded the highest detection probability, and the ratio of detection probability derived on the basis of that sampling method, compared with the detection probability derived on the basis of sampling scenario 3, was 2.1. There was substantial variation in detection probability among herds when sampling was performed on the basis of milk production. However, it was more likely that sampling from cows with low milk production yielded a higher detection probability than sampling from cows with high milk production.
The detection probability estimates derived by use of ELISA results alone were greater than the estimates derived by use of testing with ELISA-FC, regardless of sampling scheme or prevalence. If ELISA was used solely to test cows in noninfected herds, cows with falsepositive results would be classified as infected. Simulation results from the noninfected herd revealed that the misclassification probability increased as the number of cows tested increased. Probabilities of misclassifying the noninfected herd as infected were 31%, 53%, 69%, 79%, and 86% if 30, 60, 90, 120, and 150 cows in lactation 2 or higher were sampled, respectively.
Cost-effective sampling methods—The most costeffective sampling method does not necessarily have the highest detection probability. Cost-effective values and detection probabilities for sampling cows in lactation 2 or higher (reference method), compared with sampling cows in lactation 3 or higher that were in the early stage of lactation (herd A) or late stage of lactation (herds B, C, and D; alternative methods), were summarized (Table 4). Overall, the alternative methods were more cost-effective (ie, yielded lower cost-effective values) and had higher detection probabilities, compared with the reference sampling method, except when the detection probability approached 100%. The difference in cost-effective values between the reference and alternative sampling methods was greater when the withinherd prevalence was low and sample size was small. In herd A, for which within-herd prevalence of MAP infection was estimated to be 20%, the alternative sampling method yielded a higher detection probability and a cost-effective value that was 1.8 times that derived by use of the reference sampling method. The same was true for herd C, for which within-herd prevalence was estimated to be 9%. For herd C, the alternative sampling method yielded cost-effective values that were 1.2 to 2.2 times those derived from reference sampling and had higher detection probabilities for a range of 30 to 150 sampled cows. For herds B and D, which had the highest within-herd prevalence (31%), cost-effective values associated with the alternative method were lower than values associated with the reference method when sample size was small; however, when the sample size was large, cost-effective values with the alternative method were higher than those with the reference method. The higher cost-effective values yielded by the alternative method occurred because the cost-per-unit increase in the detection probability was higher when the detection probability approached 100% than when the detection probability was low. Moreover, for the same sample size, cost of the alternative method was usually higher than cost of the reference method because the alternative method detected more cows with positive results via ELISA-FC testing.
Marginal cost-effectiveness was used to measure the ratio of the change in cost to the change in detection probability. Values for mean cost-effectiveness of the alternative sampling method were calculated for each of 2 consecutive sample sizes (Table 5). Low values for mean cost-effectiveness were evident when the starting detection probabilities were small, and high values for mean cost-effectiveness were evident when starting detection probabilities were large. For example, in the herd with a 9% true prevalence (herd C), it would cost $19 per unit increase in detection probability from 85% to 96%, compared with $69 per unit increase in detection probability from 96% to 99%. The nonlinear relationship between the change in cost and the change in detection probability was summarized by comparing the alternative sampling method with the reference method in herds C and D (Figure 2). For herd D, which had an estimated within-herd prevalence of 31%, a dairy owner would spend approximately $250 if the alternative sampling method was used, compared with approximately $550 if the reference method was used, to achieve a detection probability of 95%. The difference in detection probability between the 2 methods decreased as the cost increased, and the detection probability approached 100%. For herd C, which had an estimated within-herd prevalence of 9%, testing costs to achieve a 95% detection probability by use of the alternative sampling method were approximately $600. In comparison, the reference method yielded only a 60% detection probability for the same testing cost.
Association between cost and detection probability (DP) between sampling cows in lactation 2 or higher (S1) and sampling cows in lactation 3 or higher (S2) that were in the late stage of lactation in herds C and D. Prevalence estimates for herds C and D were 9% and 31%, respectively.
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.821
Discussion
Results indicated that sampling a targeted group of cows with a higher probability of MAP infection could increase the detection probability or decrease costs associated with classification of herd status. Targeted sampling is a useful tool when detection of disease is the primary objective and there is prior knowledge of disease clustering by factors such as age or clinical signs. The technique has advantages over random sampling, in which every animal has the same probability of inclusion in the sample and which has been used in many epidemiologic studies. Targeted sampling may aid substantially in the detection of MAP infection at the herd level because in prior studies5–7,8–11 in cattle, associations were detected between age, stage of lactation, and milk production in infected cows and infection with MAP, as determined by serum antibody concentration and number of organisms shed in feces. To our knowledge, the present study is the first in which the epidemiologic and financial effects of sampling target subpopulations of cows on detection probability in cattle herds have been investigated.
In the present study, the prevalence of cows deemed infected by use of ELISA-FC results increased as lactation number increased. Sampling cows that were in lactation 3 or higher yielded a higher detection probability than did sampling cows in lactation 2 or higher. Although there were high proportions of cows with positive ELISA fecal-culture results in the subgroups of cows in lactations 4 and 5 or higher, we did not investigate sampling from cows in lactation 4 or higher because few cows remained in the 4 infected herds if cows in lactations 2 and 3 were excluded. These findings agree with those of earlier studies,5–7 in which it was found that cows in higher lactations were more likely to have positive test results because of the chronic nature of paratuberculosis. We further investigated the effect of stage of lactation on the detection probability in subgroups of cows in lactation 2 or higher and 3 and higher and found that sampling cows in late lactation yielded the highest detection probability in most instances. Although the highest detection probability in herd A was obtained by sampling cows in early lactation, the effect of stage of lactation on detection probability warrants investigation in a future study in which more herds are enrolled. It is likely that cows in peak lactation are under higher energy and physiologic demands and therefore would be more susceptible to the effects of stress, resulting in greater multiplication of the MAP organism and possibly increased shedding in feces.
Cows with clinical signs consistent with paratuberculosis and cows designated to be culled are also high-risk groups in which sampling could maximize the probability of detection in MAP-infected herds.8–11 Decreased milk production is an important early indicator of paratuberculosis9,10 and is a finding that often becomes evident before more specific clinical signs of infection occur. Therefore, the effect of sampling cows with low milk production on detection of MAP infection was investigated. We did not detect a substantial difference in detection probability associated with targeted sampling of cows with low milk production, compared with sampling all cows. This may have been because most infected cows in those herds were only subclinically infected and milk production had not been substantially affected. We were unable to investigate the effect of targeted sampling of cows designated to be culled on the probability of detection of infected herds because of the small number of cows infected with MAP that had clinical signs of paratuberculosis. This limitation may be explained by aggressive culling practices (annual rates of approx 30% to 40%), which are typical in California dairies. Nonetheless, sampling from targeted groups of cows might provide an economic and practical sampling strategy for testing in herds in which management changes to control transmission of MAP infection have not been implemented.
The major finding in our study was that detection of MAP-infected herds could be greatly improved, and therefore cost of testing could be substantially reduced, by sampling targeted groups of cows, such as cows in late lactation and in lactation 3 or higher. Assuming that a minimum detection probability of 95% was desired, costs of testing could be reduced by approximately $300 in a high-prevalence (31% infection) herd and by $800 or more in a low-prevalence (9%) herd by sampling late-lactation cows that were in lactation 3 or higher, compared with sampling cows in lactation 2 or higher. Use of targeted sampling methods could benefit participants in the VJDHSP, which mandates testing of 30 randomly selected cows in lactation 2 or higher as the screening method for a herd entering the program. We found that the VJDHSP screening method yielded 60% and 70% detection probabilities in herds B and D (with 31% within-herd prevalence), respectively, and detection probabilities of only 47% and 23%, respectively, when within-herd prevalence was 20% (herd A) and 9% (herd B).
If the VJDHSP screening method was used to detect MAP in the population of 1,000 MAP-infected dairy herds in which 70%, 20%, and 10% of the infected herds had 9%, 20%, and 31% within-herd prevalence of infection, respectively, the method would detect approximately 32% of the infected herds or misclassify 68% of them as noninfected during the first round of testing. If the same sample size of 30 cows was used but sampling was targeted to subgroups of cattle (eg, cows in lactation ≥ 3), the detection probability would be increased 1.1to 1.6-fold, compared with results of the VJDHSP screening method, and would detect approximately 47% of the infected herds. Therefore, 150 infected herds in this population that were misclassified as noninfected herds by the VJDHSP screening method would be correctly classified by the targeted sampling method. If the cost of herd testing was $1,800 for a herd progressing to level 2 of VJDHSP, $270,000 ($1,800 × 150) could be saved by sampling cows in lactation 3 or higher. Moreover, purchasing cows from those herds would be associated with an increased risk of introducing infected cows into noninfected herds. According to a prior study16 in which the risk of introducing MAP infection into a dairy herd from the purchase of replacement cows was assessed, the risk could be as high as 67% when ELISA testing alone was used prior to purchase and as high as 99% in instances in which no testing was conducted. Our study did not account for the probability or cost of purchasing infected cows from infected herds that were misclassified as noninfected by the VJDHSP screening method, compared with the targeted sampling method.
Findings were derived from results of simulation testing with data from 4 infected herds. Although the small number of herds investigated in this study limits generalization of our findings to larger populations, results were consistent with those of prior studies.2,17 The inadequate detection probability associated with the VJDHSP screening method that we detected was consistent with findings of a prior study,2 in which a herd sensitivity of 33% was yielded in lowprevalence herds (0.1% to 4.9% culture-positive cows), a herd sensitivity of 68% was yielded in moderate-prevalence herds (5% to 9.9% of cows with positive culture results), and a herd sensitivity of 84% was yielded in high-prevalence herds (≥ 10% of cows with positive culture results). The lower detection probability for the VJDHSP method found in the present study may be partly attributable to the larger size and variability in baseline prevalence among the herds studied (mean herd size in the earlier study,2 116 cows). The potential effect of false-positive results associated with sampling on the basis of ELISA results alone was also investigated in the present study. We found that the probability of misclassifying a noninfected herd as infected ranged from 32% (with a sample size of 30 cows) to 86% (with a sample size of 150 cows). In contrast, results of a study17 in which the specificity of the same ELISA used in our study was evaluated indicated that, on the basis of ELISA results alone, 6 of 7 noninfected herds were misclassified as infected when all cows in the herds were tested.
In the study reported here, 2 testing methods were used: ELISA and ELISA-FC. Although the ideal situation would have been to evaluate the true infection status of each cow, a definitive antemortem test or combination of tests for paratuberculosis does not exist. We therefore designed the study to investigate circumstances that typically apply to herd testing programs, in which ELISA with follow-up fecal culture is the most commonly used testing method. The limitation of the ELISA-FC test is low test sensitivity; to increase the detection probability by improving test sensitivity without substantial increases in cost, a combination of targeted sampling with alternative testing methods, such as pooled fecal cultures and culture of environmental samples, should be evaluated in future studies.
ABBREVIATIONS
| MAP | Mycobacterium avium subsp paratuberculosis |
| VJDHSP | Voluntary Johne's Disease Herd Status Program |
| ELISA-FC | Combination of ELISA results for serum antibodies and fecal culture results for growth of MAP |
| BSE | Bovine spongiform encephalopathy |
Dairy COMP 305, Valley Agricultural Software, Tulare, Calif.
IDEXX HerdCheck test kits, IDEXX Laboratories Inc, Westbrook, Maine.
Shin SJ. New methods for reduction in bacterial and fungal contamination from fecal-culture for Mycobacterium paratuberculosis in Proceedings. 93rd Annu Meet U S Anim Health Assoc 1989;381.
S-PLUS 6.1, Insightful Corp, Seattle, Wash.
References
- 1.↑
Bulaga LL, Gardner IA & Tavornpanich S, et al. U.S. Voluntary Johne's Disease Herd Status Program for cattle. Proc Annu Meet U S Anim Health Assoc 1998;102:402–423.
- 2.↑
Wells SJ, Whitlock RH & Wagner BA, et al. Sensitivity of test strategies used in the Voluntary Johne's Herd Status Program for detection of Mycobacterium paratuberculosis infection in dairy cattle herds. J Am Vet Med Assoc 2002;220:1053–1057.
- 3.
Christensen J, Gardner IA. Herd-level interpretation of test results for epidemiologic studies of animal diseases. Prev Vet Med 2000;45:83–106.
- 4.↑
Doherr MG, Hiem D & Fatzer R, et al. Targeted screening of high-risk cattle populations for BSE to augment mandatory reporting of clinical suspects. Prev Vet Med 2001;51:3–16.
- 5.
Nielsen SS, Enevoldsen CE, Gröhn YT. The Mycobacterium avium subsp. paratuberculosis ELISA response by parity and stage of lactation. Prev Vet Med 2002;54:1–10.
- 6.
Nielsen SS, Gröhn YT, Enevoldsen CE. Variation of the milk antibody response to paratuberculosis in naturally infected dairy cows. J Dairy Sci 2002;85:2795–2802.
- 7.
van Schaik G, Rossiter CR & Stehman SM, et al. Longitudinal study to investigate variation in results of repeated ELISA and culture of fecal samples for Mycobacterium avium subsp. paratuberculosis in commercial dairy herds. Am J Vet Res 2003;64:479–484.
- 8.
Benedictus G, Dijkhuizen AA, Stelwagen J. Economic losses due to paratuberculosis in dairy cattle. Vet Rec 1987;121:142–146.
- 9.
Buergelt CD, Duncan JR. Age and milk production data of cattle culled from a dairy herd with paratuberculosis. J Am Vet Med Assoc 1978;173:478–480.
- 10.
Nordlund KV, Goodger WJ & Pelletier J, et al. Associations between subclinical paratuberculosis and milk production, milk components, and somatic cell counts in dairy herds. J Am Vet Med Assoc 1996;208:1872–1876.
- 11.
Wilson DJ, Rossiter C & Han HR, et al. Association of Mycobacterium paratuberculosis infection with reduced mastitis, but with decreased milk production and increased cull rate in clinically normal diary cows. Am J Vet Res 1993;54:1851–1857.
- 12.↑
Weinstein WD, Stason WB. Foundations of cost-effectiveness analysis for health and medical practices. N Engl J Med 1977;296:716–721.
- 13.↑
Tavornpanich S, Gardner IA & Anderson R, et al. Evaluation of microbial culture of pooled fecal samples for detection of Mycobacterium avium subsp. paratuberculosis in large dairy herds. Am J Vet Res 2004;65:1061–1070.
- 14.↑
Gardner IA, Stryhn H & Lind P, et al. Conditional dependence between tests affects the diagnosis and surveillance of animal diseases. Prev Vet Med 2000;45:107–122.
- 15.↑
Rogan WJ, Gladen B. Estimating prevalence from the results of a screening test. Am J Epidemiol 1978;107:71–76.
- 16.↑
Carpenter TE, Gardner IA & Collins MT, et al. Effects of prevalence and testing by enzyme-linked immunosorbent assay and fecal culture on the risk of introduction of Mycobacterium avium subsp. paratuberculosis-infected cows into dairy herds. J Vet Diagn Invest 2004;16:31–38.
- 17.↑
Collins MT, Wells SJ & Petrini K, et al. Evaluation of five antibody detection tests for bovine paratuberculosis. Clin Diagn Lab Immunol 2005;12:685–692.
