- Email: [email protected]
Bovine tuberculosis and udder health in Irish dairy herds
The Veterinary Journal 192 (2012) 71–74
Contents lists available at ScienceDirect
The Veterinary Journal journal homepage: www.elsevier.com/locate/tvjl
Bovine tuberculosis and udder health in Irish dairy herds F. Boland a,⇑, G.E. Kelly a, M. Good b, S.J. More c a
UCD School of Mathematical Sciences, University College Dublin, Belfield, Dublin 4, Ireland Department of Agriculture, Fisheries and Food, Kildare St., Dublin 2, Ireland c Centre for Veterinary Epidemiology and Risk Analysis, UCD School of Agriculture, Food Science and Veterinary Medicine, University College Dublin, Belfield, Dublin 4, Ireland b
a r t i c l e
i n f o
Article history: Accepted 1 April 2011
Keywords: Bovine tuberculosis (TB) Lactations Somatic cell counts Random effects h-likelihood Ireland
a b s t r a c t The association between bovine tuberculosis (TB) infection status based on results from the single intradermal comparative tuberculin test (SICTT) and milk production has been described in dairy cows in TBinfected herds in Ireland. The biological basis was uncertain, but could be related to increased TB susceptibility among lower producing dairy cows. In this study, the relationship between somatic cell count (as an objective measure of udder health) and SICTT reactivity (as a proxy for TB infection status) was investigated. Somatic cell counts of TB infected cows, both during and prior to the lactation of diagnosis of TB infection, were examined and compared to non-infected cows. All Irish dairy herds restricted from trading between June 2004 and May 2005 as a result of two or more TB reactors (test positive) to the SICTT were considered for study. Data were collected on 4340 cows from 419 herds. Previous lactation data for the cows were taken into consideration and all lactations on a cow were analysed together with the years of lactations. There was an inherent hierarchical structure in the data, with lactations nested within cows and cows within herds and so a linear mixed model with two random effects was used to describe the data. Milk production (305-day milk yield) was also included in the model as a fixed effect. The results of the study showed that for all lactations and years under investigation, somatic cell counts for SICTT reactor cows when compared to the non-reactor cows were not significantly different. In this study population, TB infection status was not associated with udder health. Ó 2011 Elsevier Ltd. All rights reserved.
Introduction In earlier work, we described an association between bovine tuberculosis (TB) infection status caused by infection with Mycobacterium bovis based on the result of a single intradermal comparative tuberculin test (SICTT) and milk production in dairy cows in TB-infected herds in Ireland (Boland et al., 2010). SICTT reactors had significantly lower milk yields (120–573 kg per lactation) compared to non-reactors, both during the year of TB detection and in each preceding lactation. In other words, low milk production preceded and was an important risk factor for SICTT reactivity (as a proxy for TB infection) in this population. Based on data from the UK, Brotherstone et al. (2010) recently noted that animals of higher genetic merit for milk yield were also less likely to be susceptible to TB. The biological basis for these results is uncertain. However, based on available knowledge and through a process of elimination, Boland et al. (2010) suggested that the results most plausibly reflected an increased TB susceptibility among lower producing dairy cows. Many factors are associated with TB infection risk in people, including poverty, stress, malnutrition, and concurrent disease ⇑ Corresponding author. Tel.: +353 1 716 7151. E-mail address: fi[email protected] (F. Boland). 1090-0233/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tvjl.2011.04.004
(in particular, acquired immunodeficiency disease or AIDS), and similar factors are likely to be important with cattle. Host genetics is also an important contributor to TB susceptibility, both in humans (see, for example, Bellamy et al., 2000) and cattle (Bermingham et al., 2009; Brotherstone et al., 2010). As a prelude to further investigations, we considered it would be useful to gain an insight into other phenotypic differences between SICTT reactors and non-reactor animals, both during the lactation of diagnosis and previously. This information may help to understand better the mechanisms underpinning the association between milk production and TB infection status. The aim of the present study was to investigate the relationship between somatic cell count (as an objective measure of udder health) and SICTT reactivity (as a proxy for TB infection status) in Irish dairy cows, both during and prior to the lactation of diagnosis. Materials and methods The data The data consisted of cows in all dairy herds restricted from trading between 1 June 2004 and 31 May 2005 following disclosure of two or more TB reactors at testing. Typically herds had between two and seven SICTT reactors, and there were approximately 10 times more non-reactor cows than TB reactors. The data set
72
F. Boland et al. / The Veterinary Journal 192 (2012) 71–74
was extremely large and would require an extensive amount of computational force for even elementary calculations (i.e., long periods of time and huge storage). For this reason a systematic random sample of five non-reactors from each herd was selected thus reducing the size of the data set for modelling purposes. This was carried out using the SURVEYSELECT procedure in SAS (SAS Institute Inc., Cary, NC, USA) in which each non-reactor was numbered from 1 to n, the first cow was selected at random and then every n/5 thereafter. This systematic method of sampling was used because of the way the data were recorded, the lowest cow numbers corresponded to the oldest cows and the highest to the youngest cows. Therefore systematic sampling was used to prevent the possibility of all younger or older cows being selected from a herd and to provide a uniform coverage of the entire age range. In total there were 419 herds and 4340 cows (2342 TB reactors and 1998 nonreactors). For each of the first five lactations for each cow in the study, the data included a record of the year of the lactation, the lactation number (i.e., 1–5), the average somatic cell counts (103 cells/mL) over the entire lactation, the 305-day milk yield and a record of the TB status of a cow (0 = non-reactor, 1 = TB reactor). Obtained somatic cell count (SCC) values were transformed into natural logarithms (ln SCC) for normality and homogeneity of variances and all analysis was carried out on the transformed data. Descriptive statistics Initially, summary statistics were computed. The data were divided into TB reactors and non-reactors and the TB reactor cows were further divided into five groups depending on the lactation in which TB was disclosed (last lactation). The ln SCC for the lactation in which TB was disclosed and all previous lactations was compared to the equivalent lactations for the non-reactor cows. The Bonferroni method was used to control the overall Type I error rate in making multiple comparisons of means. P values <0.05 were deemed significant. Linear mixed models analysis Linear mixed models using the estimation methods of h-likelihood (Lee et al., 2006) were used to analyse the data and to take into account the possible correlations between lactations within a cow or possible correlation between cows in the same herd. Two random effects were included in the model, herd and cow within herd, to model these correlations. The random effects, herd (w) and cow within herd (a), were assumed to have normal distributions with means 0 and variances r2w and r2a , respectively. Together with the random effects, year, lactation number, TB status, all interactions between these three variables and 305-day milk yield were included in the model as fixed effects. Thus the model was as follows:
yijkl ¼ l þ sðmilkÞ þ a þ bi þ abi þ cj þ acj þ ðbcÞij þ ðabcÞij þ wk þ akl þ eijkl
Linear mixed models analysis Both random effects, herd (P < 0.0001) and cow within herd (P < 0.0001), were significant in the model. The fixed term, TB year lactation, was significant in the model (P = 0.0408) and hence, by the hierarchical principle, all lower order terms involving these variables (TB, lactation, year, TB lactation, TB year and lactation year) were included in the model. Additionally, 305-day milk yield was significant in the model (P < 0.0001). The estimated differences in ln SCC between the TB reactor animals and the non-reactor animals from the model for all years and all lactations were calculated. The estimated differences were then back transformed and are presented in Table 2. However, as a result of analysing the data on the log scale, the estimates presented are a ratio of the mean somatic cell count of the TB reactors to the non-reactors. Therefore, a ratio of 1 indicates that the TB reactors and non-reactors had the same mean SCC. If the ratio is <1, the TB reactors had a higher mean SCC than the non-reactors and if the ratio is >1, the non-reactors had a higher mean SCC than the TB reactors. The further the ratio is away from 1, the bigger the difference in mean SCC. The ratio values varied randomly between 0.027 and 2.137. However, only two values, at the 5% significance level, indicated a significant difference between TB reactors and non-reactors (1998 lactation 4 (P = 0.0105), 1999 lactation 5 (P = 0.0035)). Both of these significant values were <1 indicating that the non-reactor animals had a higher mean SCC than the TB reactor animals (Table 2). Again, to control for multiple comparisons, P values were adjusted using the Bonferroni method. Diagnostic plots of the residuals showed no evidence of model inadequacy. While the model had a few large standardised residuals, this can generally be expected for a large data set. When an adjustment was made for multiple comparisons none of the standardised residuals were significantly different from zero. Discussion
ð1Þ
where yijkl is the ln SCC value for lactation i in year j of cow l from herd k; l is the overall mean effect; s is the coefficient of 305-day milk yield (milk); a is the effect of TB; bi is the ith lactation effect (i = 1, . . . , 5); cj is the jth year effect (j = 1, . . . , 7); wk is the effect of herd k, wk N 0; r2w ; akl is the effect of cow l in herd k, akl N 0; r2a ; eijkl is the random residual effect of each observation, eijkl N 0; r2e . The data were analysed using the MIXED procedure (SAS, 1999) in SAS version 9.1 (SAS Institute Inc., Cary, NC, USA) for linear mixed models. The residuals from the final model were assessed for normality using normal probability plots. The residuals were also plotted against the fitted values and tested for outliers with the significance level adjusted using the Bonferroni method.
Results Descriptive statistics A comparison of mean ln SCC between TB reactor and non-reactor cows is presented in Table 1. For TB reactors, ln SCC is further categorised by the lactation number of TB detection (last lactation). The majority of comparisons of TB reactors with the non-reactor cows, matched by lactation number were not significant (i.e. P > 0.05). However, there were three exceptions. There was a significant difference in a TB reactor’s last lactation, compared to the equivalent lactation in non-reactors, if the last lactation was lactation 3 (P = 0.01) or 5 (P = 0.0234). In addition, there was a significant difference between TB reactors and non-reactors for lactation 2 (P = 0.0475), if a TB reactor’s last lactation was lactation 5. These significant differences were however negative indicating that the non-reactor animals had a higher mean ln SCC than the TB reactor animals. Again, the Bonferroni method was used to control for multiple comparisons.
There is a measurable increase in TB infection risk, and likely TB genetic susceptibility, among lower producing dairy cows, both in Ireland (Boland et al., 2010) and the UK (Brotherstone et al., 2010). The current study was conducted against the backdrop of these earlier findings, with the aim to better understand the nature of TB susceptibility among lower producing dairy cows in Ireland. Specifically, we query whether TB infection risk is also associated with increased risk of mastitis, being one example of another infectious disease where detailed data are available. Note that any observed association between TB and mastitis would not imply causality, given that tuberculous mastitis is now a rare event in countries such as Ireland with an ongoing programme of national TB control (Doran et al., 2009). In this study, there was no evidence of an association between TB infection status and SCC, either during or prior to the lactation of diagnosis. As highlighted in Table 2, no significant differences in SCC between TB reactors and non-reactors were seen in almost all years and lactations, with the exceptions of lactation 4 in 1998 and lactation 5 in 1999. A large dataset was used in this analysis, and it is likely that any biologically meaningful association would have been identified. We are aware of the potential confounding effect of milk production, noting the well-recognised association between increased somatic cell counts and reduced milk yield. Estimates of milk loss from high SCC range from 0.3 to 1.8 L/cow/ day, depending on the stage of lactation and SCC level (Hortet and Seegers, 1998; Green et al., 2006), although a slightly lower reduction in yield was measured after accounting for the effect of dilution on SCC among high-yielding dairy cows (Green et al., 2006). For this reason, 305-day milk yield was included in the model to control for this potential confounding effect.
73
F. Boland et al. / The Veterinary Journal 192 (2012) 71–74
Table 1 A comparison of mean log somatic cell count (SCC) among single intradermal comparative tuberculin test (SICTT) reactor cows and non-reactor cows, in bovine tuberculosis (TB) infected Irish dairy herds, restricted from trading between 1 June 2004 and 31 May 2005. For reactor cows, lactation log SCC is further categorised by the lactation of TB detection (last lactation). The Bonferroni method was used to account for multiple comparisons. Lactation
1 1 1 1 1 2 2 2 2 3 3 3 4 4 5 a b c
SICTT reactors
Non-reactors a
Last lactation
n
Mean
n
Mean
1 2 3 4 5 2 3 4 5 3 4 5 4 5 5
581 447 311 242 404 522 353 255 436 402 264 467 292 499 544
11.40 11.41 11.39 11.36 11.29 11.43 11.42 11.59 11.41 11.54 11.74 11.60 11.82 11.76 11.93
1913 1913 1913 1913 1913 1232 1232 1232 1232 741 741 741 425 425 213
11.36 11.36 11.36 11.36 11.36 11.50 11.50 11.50 11.50 11.69 11.69 11.69 11.86 11.86 12.11
SEc of difference
P value
0.04 0.04 0.03 0.00 0.07 0.07 0.08 0.09 0.09 0.15 0.05 0.09 0.05 0.10 0.18
0.037 0.041 0.048 0.053 0.042 0.044 0.050 0.058 0.045 0.058 0.067 0.054 0.076 0.061 0.079
0.2840 0.2235 0.5303 0.9999 0.0934 0.1145 0.1109 0.1233 0.0475 0.0100 0.4542 0.0941 0.6001 0.1028 0.0234
ln SCC. Mean ln SCC of SICTT reactors minus non-reactor animals. Standard error.
Table 2 The estimated ratio of differences in mean somatic cell count (single intradermal comparative tuberculin test (SICTT) reactor cows to non-reactor cows), the two sample t tests of differences and the sample sizes for all lactations and all years in bovine tuberculosis (TB) infected Irish dairy herds restricted from trading between 1 June 2004 and 31 May 2005. The Bonferroni method was used to account for multiple comparisons.
a
Mean differenceb a
Lactation
Year
Estimated ratiosa
P value
Sample size (n)
1 1 1 1 1 1 1 2 2 2 2 2 2 2 3 3 3 3 3 3 3 4 4 4 4 4 4 4 5 5 5 5 5 5
1998 1999 2000 2001 2002 2003 2004 1998 1999 2000 2001 2002 2003 2004 1998 1999 2000 2001 2002 2003 2004 1998 1999 2000 2001 2002 2003 2004 1998 1999 2001 2002 2003 2004
0.754 1.643 0.985 1.063 0.944 1.104 0.945 1.128 0.741 0.713 1.564 1.133 0.952 0.732 0.391 0.849 0.795 2.137 1.095 0.819 0.675 0.044 0.603 0.987 0.507 1.142 1.111 0.971 0.841 0.027 0.464 1.550 1.029 1.522
0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9345 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.8365 0.0105 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.0035 0.9999 0.9999 0.9999 0.9999
127 172 259 380 479 797 935 101 126 167 232 376 521 700 52 104 119 148 228 394 463 35 51 94 110 145 236 345 11 34 76 103 153 177
Ratio of the mean SCC of the TB reactors to the non-reactors.
This study focused on SICTT reactivity as a proxy for TB infection status, and must be interpreted with caution, noting that some misclassification is likely. Due to both the nature of the infection and current diagnostic capacity, false negative animals (SICTT neg-
ative but TB infected) are much more likely than false positive animals (SICTT positive but not TB infected). The operating characteristics of the SICTT are well-described (de la Rua-Domenech et al., 2006), with the most recent Irish data suggesting a sensitivity and specificity of 52.9–60.8% and 99.2–99.8%, respectively (Clegg et al., 2011). In the current study, false positive misclassification was further minimised by limiting positive infection status to those SICTT-positive reactors in herds restricted from trading following disclosure of two or more such animals. Greiner and Gardner (2000) outline the issues associated with misclassification by diagnostic tests when conducting risk factor studies. A range of methodological issues were faced during this work, as discussed in detail previously (Boland et al., 2010). The study focuses on milk-recording herds, where data on individual animal milk yield and SCC are available. In Ireland, farmers pay for milk recording, with the generated information most commonly being used to assist with on-farm decision-making. During 2010, milk recording was conducted on 5924 Irish dairy herds (approximately 33% of dairy herds but 50% of the national dairy cow population). In these herds, milk recording was conducted most frequently on 4 (32.3% of milk-recording herds in 2010), 5 (11.5%), 6 (10.9%) or 7 (16.5%) occasions during the year. SCC is known to be lower in milk-recorded herds compared to herds where milk recording is not conducted (Kelly et al., 2009). We also highlight the two significant P values, after Bonferroni correction, in Table 2 (1998 lactation 4 (P = 0.0105) and 1999 lactation 5 (P = 0.0035)). These two small P values appear extremely unusual in comparison to all the other values in Table 2 (P > 0.8365 in all cases). These values were examined further and no unusual observations were found. For both of the years and lactations in question the majority of the SCC values for the TB infected cows were smaller than the values for the non-infected cows. However, there were no apparent influential observations leading to such small P values for 1998 lactation 4 and 1999 lactation 5. In summary, in this study in a sample of 419 herds and 4340 cows, there was insufficient evidence of a difference in ln SCC between TB reactors and non-reactors.
Conclusions In this study, there was no evidence of an association between TB infection status and SCC (as an objective measure of udder
74
F. Boland et al. / The Veterinary Journal 192 (2012) 71–74
health), and thus mastitis in Irish dairy herds, either during or prior to the lactation of diagnosis. No significant differences in SCC between TB reactors and non-reactors were seen in almost all years and lactations. Conflict of interest statement None of the authors of this paper has a financial or personal relationship with other people or organisations that could inappropriately influence or bias the content of the paper. Acknowledgements The authors wish to thank the Department of Agriculture, Fisheries and Food, firstly for providing funding for the first author and secondly for providing the TB testing data through CVERA, University College Dublin. We also wish to thank the Irish Cattle Breeding Federation (ICBF) for providing the milk production data. References Bellamy, R., Beyers, N., McAdam, K.P.W.J., Ruwende, C., Gie, R., Samaai, P., Bester, D., Meyer, M., Corrah, T., Collin, M., Camidge, D.R., Wilkinson, D., Hoal-van Helden, E., Whittle, H.C., Amos, W., van Helden, P., Hill, A.V.S., 2000. Genetic susceptibility to tuberculosis in Africans: A genome-wide scan. Proceedings of the National Academy of Science USA 14, 8005–8009. Bermingham, M.L., More, S.J., Good, M., Cromie, A.R., Higgins, I.M., Brotherstone, S., Berry, D.P., 2009. Genetics of tuberculosis in Irish Holstein-Friesian dairy herds. Journal of Dairy Science 92, 3447–3456. Boland, F., Kelly, G., Good, M., More, S.J., 2010. Bovine tuberculosis and milk production in infected dairy herds in Ireland. Preventive Veterinary Medicine 92, 153–161.
Brotherstone, S., White, I.M.S., Coffey, M., Downs, S.H., Mitchell, A.H., Clifton-Hadley, R.S., More, S.J., Good, M., Woolliams, J.A., 2010. Evidence of genetic resistance of cattle to infection with Mycobacterium bovis. Journal of Dairy Science 93, 1234– 1242. Clegg, T.A., Duignan, A., Whelan, C., Gormley, E., Good, M., Clarke, J., Toft, N., More, S.J., 2011. Using latent class analysis to estimate the test characteristics of the cinterferon test, the single intradermal comparative tuberculin test and a multiplex ELISA under Irish conditions. Veterinary Microbiology. doi:10.1016/ j.vetmic.2011.02.027. de la Rua-Domenech, R., Goodchild, A.T., Vordermeier, H.M., Hewinson, R.G., Christiansen, K.H., Clifton-Hadley, R.S., 2006. Ante mortem diagnosis of tuberculosis in cattle: A review of the tuberculin tests, c-interferon assay and other ancillary diagnostic techniques. Research in Veterinary Science 81, 190– 210. Doran, P., Carson, J., Costello, E., More, S.J., 2009. An outbreak of tuberculosis affecting cattle and people on an Irish dairy farm in 2005, following the consumption of raw milk from a cow with tuberculous mastitis. Irish Veterinary Journal 62, 390–397. Green, L.E., Schukken, Y.H., Green, M.J., 2006. On distinguishing cause and consequence: Do high somatic cell counts lead to lower milk yield or does high milk yield lead to lower somatic cell count? Preventive Veterinary Medicine 76, 74–89. Greiner, M., Gardner, I.A., 2000. Application of diagnostic tests in veterinary epidemiologic studies. Preventive Veterinary Medicine 45, 43–59. Hortet, P., Seegers, H., 1998. Calculated milk production losses associated with elevated somatic cell counts in dairy cows: Review and critical discussion. Veterinary Research 29, 497–510. Kelly, P.T., O’Sullivan, K., Berry, D.P., More, S.J., Meaney, W.J., O’Callaghan, E.J., O’Brien, B., 2009. Farm management factors associated with bulk tank somatic cell count in Irish dairy herds. Irish Veterinary Journal, Suppl. 62, 45–51. Lee, Y., Nelder, J.A., Pawitan, Y., 2006. Generalised Linear Models with Random Effects Unified Analysis via H-likelihood. Chapman and Hall/CRC Press, Boca Raton, FL. SAS, 1999. SAS/STAT User’s Guide, Version 8. The MIXED procedure, SAS Institute Inc., Cary, NC, USA (Chapter 41).
Contents lists available at ScienceDirect
The Veterinary Journal journal homepage: www.elsevier.com/locate/tvjl
Bovine tuberculosis and udder health in Irish dairy herds F. Boland a,⇑, G.E. Kelly a, M. Good b, S.J. More c a
UCD School of Mathematical Sciences, University College Dublin, Belfield, Dublin 4, Ireland Department of Agriculture, Fisheries and Food, Kildare St., Dublin 2, Ireland c Centre for Veterinary Epidemiology and Risk Analysis, UCD School of Agriculture, Food Science and Veterinary Medicine, University College Dublin, Belfield, Dublin 4, Ireland b
a r t i c l e
i n f o
Article history: Accepted 1 April 2011
Keywords: Bovine tuberculosis (TB) Lactations Somatic cell counts Random effects h-likelihood Ireland
a b s t r a c t The association between bovine tuberculosis (TB) infection status based on results from the single intradermal comparative tuberculin test (SICTT) and milk production has been described in dairy cows in TBinfected herds in Ireland. The biological basis was uncertain, but could be related to increased TB susceptibility among lower producing dairy cows. In this study, the relationship between somatic cell count (as an objective measure of udder health) and SICTT reactivity (as a proxy for TB infection status) was investigated. Somatic cell counts of TB infected cows, both during and prior to the lactation of diagnosis of TB infection, were examined and compared to non-infected cows. All Irish dairy herds restricted from trading between June 2004 and May 2005 as a result of two or more TB reactors (test positive) to the SICTT were considered for study. Data were collected on 4340 cows from 419 herds. Previous lactation data for the cows were taken into consideration and all lactations on a cow were analysed together with the years of lactations. There was an inherent hierarchical structure in the data, with lactations nested within cows and cows within herds and so a linear mixed model with two random effects was used to describe the data. Milk production (305-day milk yield) was also included in the model as a fixed effect. The results of the study showed that for all lactations and years under investigation, somatic cell counts for SICTT reactor cows when compared to the non-reactor cows were not significantly different. In this study population, TB infection status was not associated with udder health. Ó 2011 Elsevier Ltd. All rights reserved.
Introduction In earlier work, we described an association between bovine tuberculosis (TB) infection status caused by infection with Mycobacterium bovis based on the result of a single intradermal comparative tuberculin test (SICTT) and milk production in dairy cows in TB-infected herds in Ireland (Boland et al., 2010). SICTT reactors had significantly lower milk yields (120–573 kg per lactation) compared to non-reactors, both during the year of TB detection and in each preceding lactation. In other words, low milk production preceded and was an important risk factor for SICTT reactivity (as a proxy for TB infection) in this population. Based on data from the UK, Brotherstone et al. (2010) recently noted that animals of higher genetic merit for milk yield were also less likely to be susceptible to TB. The biological basis for these results is uncertain. However, based on available knowledge and through a process of elimination, Boland et al. (2010) suggested that the results most plausibly reflected an increased TB susceptibility among lower producing dairy cows. Many factors are associated with TB infection risk in people, including poverty, stress, malnutrition, and concurrent disease ⇑ Corresponding author. Tel.: +353 1 716 7151. E-mail address: fi[email protected] (F. Boland). 1090-0233/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tvjl.2011.04.004
(in particular, acquired immunodeficiency disease or AIDS), and similar factors are likely to be important with cattle. Host genetics is also an important contributor to TB susceptibility, both in humans (see, for example, Bellamy et al., 2000) and cattle (Bermingham et al., 2009; Brotherstone et al., 2010). As a prelude to further investigations, we considered it would be useful to gain an insight into other phenotypic differences between SICTT reactors and non-reactor animals, both during the lactation of diagnosis and previously. This information may help to understand better the mechanisms underpinning the association between milk production and TB infection status. The aim of the present study was to investigate the relationship between somatic cell count (as an objective measure of udder health) and SICTT reactivity (as a proxy for TB infection status) in Irish dairy cows, both during and prior to the lactation of diagnosis. Materials and methods The data The data consisted of cows in all dairy herds restricted from trading between 1 June 2004 and 31 May 2005 following disclosure of two or more TB reactors at testing. Typically herds had between two and seven SICTT reactors, and there were approximately 10 times more non-reactor cows than TB reactors. The data set
72
F. Boland et al. / The Veterinary Journal 192 (2012) 71–74
was extremely large and would require an extensive amount of computational force for even elementary calculations (i.e., long periods of time and huge storage). For this reason a systematic random sample of five non-reactors from each herd was selected thus reducing the size of the data set for modelling purposes. This was carried out using the SURVEYSELECT procedure in SAS (SAS Institute Inc., Cary, NC, USA) in which each non-reactor was numbered from 1 to n, the first cow was selected at random and then every n/5 thereafter. This systematic method of sampling was used because of the way the data were recorded, the lowest cow numbers corresponded to the oldest cows and the highest to the youngest cows. Therefore systematic sampling was used to prevent the possibility of all younger or older cows being selected from a herd and to provide a uniform coverage of the entire age range. In total there were 419 herds and 4340 cows (2342 TB reactors and 1998 nonreactors). For each of the first five lactations for each cow in the study, the data included a record of the year of the lactation, the lactation number (i.e., 1–5), the average somatic cell counts (103 cells/mL) over the entire lactation, the 305-day milk yield and a record of the TB status of a cow (0 = non-reactor, 1 = TB reactor). Obtained somatic cell count (SCC) values were transformed into natural logarithms (ln SCC) for normality and homogeneity of variances and all analysis was carried out on the transformed data. Descriptive statistics Initially, summary statistics were computed. The data were divided into TB reactors and non-reactors and the TB reactor cows were further divided into five groups depending on the lactation in which TB was disclosed (last lactation). The ln SCC for the lactation in which TB was disclosed and all previous lactations was compared to the equivalent lactations for the non-reactor cows. The Bonferroni method was used to control the overall Type I error rate in making multiple comparisons of means. P values <0.05 were deemed significant. Linear mixed models analysis Linear mixed models using the estimation methods of h-likelihood (Lee et al., 2006) were used to analyse the data and to take into account the possible correlations between lactations within a cow or possible correlation between cows in the same herd. Two random effects were included in the model, herd and cow within herd, to model these correlations. The random effects, herd (w) and cow within herd (a), were assumed to have normal distributions with means 0 and variances r2w and r2a , respectively. Together with the random effects, year, lactation number, TB status, all interactions between these three variables and 305-day milk yield were included in the model as fixed effects. Thus the model was as follows:
yijkl ¼ l þ sðmilkÞ þ a þ bi þ abi þ cj þ acj þ ðbcÞij þ ðabcÞij þ wk þ akl þ eijkl
Linear mixed models analysis Both random effects, herd (P < 0.0001) and cow within herd (P < 0.0001), were significant in the model. The fixed term, TB year lactation, was significant in the model (P = 0.0408) and hence, by the hierarchical principle, all lower order terms involving these variables (TB, lactation, year, TB lactation, TB year and lactation year) were included in the model. Additionally, 305-day milk yield was significant in the model (P < 0.0001). The estimated differences in ln SCC between the TB reactor animals and the non-reactor animals from the model for all years and all lactations were calculated. The estimated differences were then back transformed and are presented in Table 2. However, as a result of analysing the data on the log scale, the estimates presented are a ratio of the mean somatic cell count of the TB reactors to the non-reactors. Therefore, a ratio of 1 indicates that the TB reactors and non-reactors had the same mean SCC. If the ratio is <1, the TB reactors had a higher mean SCC than the non-reactors and if the ratio is >1, the non-reactors had a higher mean SCC than the TB reactors. The further the ratio is away from 1, the bigger the difference in mean SCC. The ratio values varied randomly between 0.027 and 2.137. However, only two values, at the 5% significance level, indicated a significant difference between TB reactors and non-reactors (1998 lactation 4 (P = 0.0105), 1999 lactation 5 (P = 0.0035)). Both of these significant values were <1 indicating that the non-reactor animals had a higher mean SCC than the TB reactor animals (Table 2). Again, to control for multiple comparisons, P values were adjusted using the Bonferroni method. Diagnostic plots of the residuals showed no evidence of model inadequacy. While the model had a few large standardised residuals, this can generally be expected for a large data set. When an adjustment was made for multiple comparisons none of the standardised residuals were significantly different from zero. Discussion
ð1Þ
where yijkl is the ln SCC value for lactation i in year j of cow l from herd k; l is the overall mean effect; s is the coefficient of 305-day milk yield (milk); a is the effect of TB; bi is the ith lactation effect (i = 1, . . . , 5); cj is the jth year effect (j = 1, . . . , 7); wk is the effect of herd k, wk N 0; r2w ; akl is the effect of cow l in herd k, akl N 0; r2a ; eijkl is the random residual effect of each observation, eijkl N 0; r2e . The data were analysed using the MIXED procedure (SAS, 1999) in SAS version 9.1 (SAS Institute Inc., Cary, NC, USA) for linear mixed models. The residuals from the final model were assessed for normality using normal probability plots. The residuals were also plotted against the fitted values and tested for outliers with the significance level adjusted using the Bonferroni method.
Results Descriptive statistics A comparison of mean ln SCC between TB reactor and non-reactor cows is presented in Table 1. For TB reactors, ln SCC is further categorised by the lactation number of TB detection (last lactation). The majority of comparisons of TB reactors with the non-reactor cows, matched by lactation number were not significant (i.e. P > 0.05). However, there were three exceptions. There was a significant difference in a TB reactor’s last lactation, compared to the equivalent lactation in non-reactors, if the last lactation was lactation 3 (P = 0.01) or 5 (P = 0.0234). In addition, there was a significant difference between TB reactors and non-reactors for lactation 2 (P = 0.0475), if a TB reactor’s last lactation was lactation 5. These significant differences were however negative indicating that the non-reactor animals had a higher mean ln SCC than the TB reactor animals. Again, the Bonferroni method was used to control for multiple comparisons.
There is a measurable increase in TB infection risk, and likely TB genetic susceptibility, among lower producing dairy cows, both in Ireland (Boland et al., 2010) and the UK (Brotherstone et al., 2010). The current study was conducted against the backdrop of these earlier findings, with the aim to better understand the nature of TB susceptibility among lower producing dairy cows in Ireland. Specifically, we query whether TB infection risk is also associated with increased risk of mastitis, being one example of another infectious disease where detailed data are available. Note that any observed association between TB and mastitis would not imply causality, given that tuberculous mastitis is now a rare event in countries such as Ireland with an ongoing programme of national TB control (Doran et al., 2009). In this study, there was no evidence of an association between TB infection status and SCC, either during or prior to the lactation of diagnosis. As highlighted in Table 2, no significant differences in SCC between TB reactors and non-reactors were seen in almost all years and lactations, with the exceptions of lactation 4 in 1998 and lactation 5 in 1999. A large dataset was used in this analysis, and it is likely that any biologically meaningful association would have been identified. We are aware of the potential confounding effect of milk production, noting the well-recognised association between increased somatic cell counts and reduced milk yield. Estimates of milk loss from high SCC range from 0.3 to 1.8 L/cow/ day, depending on the stage of lactation and SCC level (Hortet and Seegers, 1998; Green et al., 2006), although a slightly lower reduction in yield was measured after accounting for the effect of dilution on SCC among high-yielding dairy cows (Green et al., 2006). For this reason, 305-day milk yield was included in the model to control for this potential confounding effect.
73
F. Boland et al. / The Veterinary Journal 192 (2012) 71–74
Table 1 A comparison of mean log somatic cell count (SCC) among single intradermal comparative tuberculin test (SICTT) reactor cows and non-reactor cows, in bovine tuberculosis (TB) infected Irish dairy herds, restricted from trading between 1 June 2004 and 31 May 2005. For reactor cows, lactation log SCC is further categorised by the lactation of TB detection (last lactation). The Bonferroni method was used to account for multiple comparisons. Lactation
1 1 1 1 1 2 2 2 2 3 3 3 4 4 5 a b c
SICTT reactors
Non-reactors a
Last lactation
n
Mean
n
Mean
1 2 3 4 5 2 3 4 5 3 4 5 4 5 5
581 447 311 242 404 522 353 255 436 402 264 467 292 499 544
11.40 11.41 11.39 11.36 11.29 11.43 11.42 11.59 11.41 11.54 11.74 11.60 11.82 11.76 11.93
1913 1913 1913 1913 1913 1232 1232 1232 1232 741 741 741 425 425 213
11.36 11.36 11.36 11.36 11.36 11.50 11.50 11.50 11.50 11.69 11.69 11.69 11.86 11.86 12.11
SEc of difference
P value
0.04 0.04 0.03 0.00 0.07 0.07 0.08 0.09 0.09 0.15 0.05 0.09 0.05 0.10 0.18
0.037 0.041 0.048 0.053 0.042 0.044 0.050 0.058 0.045 0.058 0.067 0.054 0.076 0.061 0.079
0.2840 0.2235 0.5303 0.9999 0.0934 0.1145 0.1109 0.1233 0.0475 0.0100 0.4542 0.0941 0.6001 0.1028 0.0234
ln SCC. Mean ln SCC of SICTT reactors minus non-reactor animals. Standard error.
Table 2 The estimated ratio of differences in mean somatic cell count (single intradermal comparative tuberculin test (SICTT) reactor cows to non-reactor cows), the two sample t tests of differences and the sample sizes for all lactations and all years in bovine tuberculosis (TB) infected Irish dairy herds restricted from trading between 1 June 2004 and 31 May 2005. The Bonferroni method was used to account for multiple comparisons.
a
Mean differenceb a
Lactation
Year
Estimated ratiosa
P value
Sample size (n)
1 1 1 1 1 1 1 2 2 2 2 2 2 2 3 3 3 3 3 3 3 4 4 4 4 4 4 4 5 5 5 5 5 5
1998 1999 2000 2001 2002 2003 2004 1998 1999 2000 2001 2002 2003 2004 1998 1999 2000 2001 2002 2003 2004 1998 1999 2000 2001 2002 2003 2004 1998 1999 2001 2002 2003 2004
0.754 1.643 0.985 1.063 0.944 1.104 0.945 1.128 0.741 0.713 1.564 1.133 0.952 0.732 0.391 0.849 0.795 2.137 1.095 0.819 0.675 0.044 0.603 0.987 0.507 1.142 1.111 0.971 0.841 0.027 0.464 1.550 1.029 1.522
0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9345 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.8365 0.0105 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.0035 0.9999 0.9999 0.9999 0.9999
127 172 259 380 479 797 935 101 126 167 232 376 521 700 52 104 119 148 228 394 463 35 51 94 110 145 236 345 11 34 76 103 153 177
Ratio of the mean SCC of the TB reactors to the non-reactors.
This study focused on SICTT reactivity as a proxy for TB infection status, and must be interpreted with caution, noting that some misclassification is likely. Due to both the nature of the infection and current diagnostic capacity, false negative animals (SICTT neg-
ative but TB infected) are much more likely than false positive animals (SICTT positive but not TB infected). The operating characteristics of the SICTT are well-described (de la Rua-Domenech et al., 2006), with the most recent Irish data suggesting a sensitivity and specificity of 52.9–60.8% and 99.2–99.8%, respectively (Clegg et al., 2011). In the current study, false positive misclassification was further minimised by limiting positive infection status to those SICTT-positive reactors in herds restricted from trading following disclosure of two or more such animals. Greiner and Gardner (2000) outline the issues associated with misclassification by diagnostic tests when conducting risk factor studies. A range of methodological issues were faced during this work, as discussed in detail previously (Boland et al., 2010). The study focuses on milk-recording herds, where data on individual animal milk yield and SCC are available. In Ireland, farmers pay for milk recording, with the generated information most commonly being used to assist with on-farm decision-making. During 2010, milk recording was conducted on 5924 Irish dairy herds (approximately 33% of dairy herds but 50% of the national dairy cow population). In these herds, milk recording was conducted most frequently on 4 (32.3% of milk-recording herds in 2010), 5 (11.5%), 6 (10.9%) or 7 (16.5%) occasions during the year. SCC is known to be lower in milk-recorded herds compared to herds where milk recording is not conducted (Kelly et al., 2009). We also highlight the two significant P values, after Bonferroni correction, in Table 2 (1998 lactation 4 (P = 0.0105) and 1999 lactation 5 (P = 0.0035)). These two small P values appear extremely unusual in comparison to all the other values in Table 2 (P > 0.8365 in all cases). These values were examined further and no unusual observations were found. For both of the years and lactations in question the majority of the SCC values for the TB infected cows were smaller than the values for the non-infected cows. However, there were no apparent influential observations leading to such small P values for 1998 lactation 4 and 1999 lactation 5. In summary, in this study in a sample of 419 herds and 4340 cows, there was insufficient evidence of a difference in ln SCC between TB reactors and non-reactors.
Conclusions In this study, there was no evidence of an association between TB infection status and SCC (as an objective measure of udder
74
F. Boland et al. / The Veterinary Journal 192 (2012) 71–74
health), and thus mastitis in Irish dairy herds, either during or prior to the lactation of diagnosis. No significant differences in SCC between TB reactors and non-reactors were seen in almost all years and lactations. Conflict of interest statement None of the authors of this paper has a financial or personal relationship with other people or organisations that could inappropriately influence or bias the content of the paper. Acknowledgements The authors wish to thank the Department of Agriculture, Fisheries and Food, firstly for providing funding for the first author and secondly for providing the TB testing data through CVERA, University College Dublin. We also wish to thank the Irish Cattle Breeding Federation (ICBF) for providing the milk production data. References Bellamy, R., Beyers, N., McAdam, K.P.W.J., Ruwende, C., Gie, R., Samaai, P., Bester, D., Meyer, M., Corrah, T., Collin, M., Camidge, D.R., Wilkinson, D., Hoal-van Helden, E., Whittle, H.C., Amos, W., van Helden, P., Hill, A.V.S., 2000. Genetic susceptibility to tuberculosis in Africans: A genome-wide scan. Proceedings of the National Academy of Science USA 14, 8005–8009. Bermingham, M.L., More, S.J., Good, M., Cromie, A.R., Higgins, I.M., Brotherstone, S., Berry, D.P., 2009. Genetics of tuberculosis in Irish Holstein-Friesian dairy herds. Journal of Dairy Science 92, 3447–3456. Boland, F., Kelly, G., Good, M., More, S.J., 2010. Bovine tuberculosis and milk production in infected dairy herds in Ireland. Preventive Veterinary Medicine 92, 153–161.
Brotherstone, S., White, I.M.S., Coffey, M., Downs, S.H., Mitchell, A.H., Clifton-Hadley, R.S., More, S.J., Good, M., Woolliams, J.A., 2010. Evidence of genetic resistance of cattle to infection with Mycobacterium bovis. Journal of Dairy Science 93, 1234– 1242. Clegg, T.A., Duignan, A., Whelan, C., Gormley, E., Good, M., Clarke, J., Toft, N., More, S.J., 2011. Using latent class analysis to estimate the test characteristics of the cinterferon test, the single intradermal comparative tuberculin test and a multiplex ELISA under Irish conditions. Veterinary Microbiology. doi:10.1016/ j.vetmic.2011.02.027. de la Rua-Domenech, R., Goodchild, A.T., Vordermeier, H.M., Hewinson, R.G., Christiansen, K.H., Clifton-Hadley, R.S., 2006. Ante mortem diagnosis of tuberculosis in cattle: A review of the tuberculin tests, c-interferon assay and other ancillary diagnostic techniques. Research in Veterinary Science 81, 190– 210. Doran, P., Carson, J., Costello, E., More, S.J., 2009. An outbreak of tuberculosis affecting cattle and people on an Irish dairy farm in 2005, following the consumption of raw milk from a cow with tuberculous mastitis. Irish Veterinary Journal 62, 390–397. Green, L.E., Schukken, Y.H., Green, M.J., 2006. On distinguishing cause and consequence: Do high somatic cell counts lead to lower milk yield or does high milk yield lead to lower somatic cell count? Preventive Veterinary Medicine 76, 74–89. Greiner, M., Gardner, I.A., 2000. Application of diagnostic tests in veterinary epidemiologic studies. Preventive Veterinary Medicine 45, 43–59. Hortet, P., Seegers, H., 1998. Calculated milk production losses associated with elevated somatic cell counts in dairy cows: Review and critical discussion. Veterinary Research 29, 497–510. Kelly, P.T., O’Sullivan, K., Berry, D.P., More, S.J., Meaney, W.J., O’Callaghan, E.J., O’Brien, B., 2009. Farm management factors associated with bulk tank somatic cell count in Irish dairy herds. Irish Veterinary Journal, Suppl. 62, 45–51. Lee, Y., Nelder, J.A., Pawitan, Y., 2006. Generalised Linear Models with Random Effects Unified Analysis via H-likelihood. Chapman and Hall/CRC Press, Boca Raton, FL. SAS, 1999. SAS/STAT User’s Guide, Version 8. The MIXED procedure, SAS Institute Inc., Cary, NC, USA (Chapter 41).