somatic cell count in dairy herds

SOMATIC CELL COUNT IN DAIRY HERDS

J.J. Lievaart Utrecht 2010

Lievaart, Jan Johannes Somatic cell count in dairy herds 2010 Dissertation Utrecht University, Faculty of Veterinary Medicine -met een samenvatting in het Nederlands ISBN: Lay-out: Harry Otter, division Multimedia. Cover design: Udo Prinsen, Studio Carambolas Printed by: Offset drukkerij Ridderprint, Ridderkerk

SOMATIC CELL COUNT IN DAIRY HERDS Celgetal op melkveebedrijven (met een samenvatting in het Nederlands)

Proefschrift ter verkrijging van de graad van doctor aan de Universiteit Utrecht op gezag van de rector magnificus, prof. dr. J.C. Stoof, ingevolge het besluit van het college voor promoties in het openbaar te verdedigen op vrijdag 3 september 2010 des ochtends te 10.30 uur

door

Jan Johannes Lievaart geboren op 15 september 1970 te Giessen

Promotoren:

Prof. dr. J.A.P. Heesterbeek Prof. dr. W.D.J. Kremer Prof. dr. H.W. Barkema

Voor Karianne, Jurre, Taeke en Sylke

Paranimfen:

drs. B. van de Griend ir. L.W.C. Rosengarten

Contents Chapter 1

General Introduction

1

Chapter 2

Reliability of the bulk milk somatic cell count as an indication of average herd somatic cell count

11

Chapter 3

Comparison of bulk milk, yield-corrected, and average somatic cell counts as parameters to summarize the subclinical mastitis situation in a dairy herd

27

Chapter 4

Influence of the sampling interval on the accuracy of bulk milk somatic cell count data

35

Chapter 5

Effect of herd characteristics, management practices, and season on different categories of the herd somatic cell count

47

Chapter 6

Subclinical mastitis and associated risk factors on New South Wales dairy farms

63

Chapter 7

Prediction of the herd somatic cell count of the following month using a linear mixed effect model

79

Chapter 8

Summarising Discussion

97

Summary

113

Samenvatting

119

Dankwoord – Acknowledgements

127

Curriculum Vitae

133

List of publications

134

Chapter 1

GENERAL INTRODUCTION

J.J. Lievaart

Department of Farm Animal Health, Ruminant Health Unit Faculty of Veterinary Medicine, Utrecht University

Yalelaan 7, 3584 CL, Utrecht, The Netherlands

General introduction

2

Chapter 1

Introduction The number of somatic cells in milk, expressed as the somatic cell count (SCC), is one of the most important parameters available to monitor subclinical mastitis in the individual cow (defined here as SCC > 200,000 cells/ml) and at herd level. The somatic cells mainly consist of leukocytes or white blood cells including macrophages, lymphocytes, and polymorphonuclear neutrophils (PMN) (Kelly et al., 2000). When pathogens invade the mammary gland the primary phagocytic cells, PMN and macrophages, comprise the first line of defence and cause an increase in the SCC (Harmon, 1994; Paape et al., 2002). In the most common method to determine the SCC, a DNA-specific dye (ethidium bromide) is used and each cell produces an electrical pulse, which is amplified and recorded by an automated fluorescence microscope (International Dairy Federation, 1995). The SCC can be measured as an individual quarter sample, as a composite sample of four quarters per cow (Bradley and Green, 2005; Breen et al., 2009), or determined in the bulk tank as the bulk milk SCC (BMSCC) representing the average of the herd (Dohoo and Meek, 1982; Schukken et al., 2003; Jayarao et al., 2004). Guidelines to monitor udder health have been established, but depend on the goals used to monitor the SCC. The individual SCC (ISCC) data are used by farmers and their advisors as a tool, together with milk cultures, to determine individual treatments during lactation, at drying off (Torres et al., 2008), or to determine when to cull a cow (Bascom and Young, 1998). At the herd level, the BMSCC, the proportion of cows with subclinical mastitis or the arithmetic mean of ISCC will help a farmer to support management decisions regarding udder health (Sargeant et al., 1998; Jayarao et al., 2004). Since Neave et al. (1969) first published recommendations on control of contagious mastitis, there have been numerous control strategies outlining mechanisms for the prevention of intramammary infections (IMI). A large number of these studies outlined risk factors associated with management practices (Bartlett et al., 1992; Barkema et al., 1998; Barnouin et al., 2004; Wenz et al., 2007; Kelly et al., 2009), herd characteristics (Laevens et al., 1997; Berry et al., 2007; Green et al., 2008), or seasonal variation (Berry et al., 2006; Green et al., 2006, Olde Riekerink et al., 2007). Management styles and practices have a considerable impact and therefore have formed the basis of many control strategies; however, the exact contribution of each management practice on either the cow or seasonal component of the average herd SCC (HSCC) is unknown. Furthermore, the effect of management practises on udder health is frequently examined for European and North American husbandry systems without exact knowledge as to whether these management practices are applicable for other husbandry systems worldwide (Hutton et al., 1990; Barkema et al., 1999; Peeler et al., 2000; Wenz et al., 2007).

3


In assessing the effect of these management practices, herd characteristics or seasonal variations on the prevalence of subclinical mastitis, many studies have used BMSCC as an indicator representing the herd average (Barkema et al., 1998a,b; Barnouin et al., 2004; Green et al., 2006; Wenz et al., 2007). It has been hypothesised that there is often a significant difference between BMSCC and the average SCC of all individual cow milk samples, as milk from cows with a high SCC is often excluded. Supporting assertions are studies that suggest that the arithmetic mean SCC of all cows would be a more appropriate parameter to summarise a herd subclinical mastitis situation (Schukken et al., 2003; Bradley and Green, 2005; Valde et al., 2005). Monitoring of both BMSCC and ISCC is based on retrospective data and on long sampling intervals ranging from 2 to 6 weeks. In udder health management there is a need for a more prospective tool, which enables the farmer to act before a problem arises. A prediction of the parameters used, such as the BMSCC or HSCC, would therefore be a valuable supplementary tool to assess if changes in management should be implemented when an increase in average herd SCC is expected within a short time period. For close monitoring at herd level, the sampling interval of the SCC parameters used is an important issue. When describing a series of SCC data, sampling intervals should probably be as short as possible so that an upward trend in the SCC data will be detected early and, if necessary, a timely intervention can be implemented. At present Dairy Herd Improvement sampling and BMSCC sampling intervals vary per country and the effect of the interval period between samples should be examined more closely. Aim and Scope of this Thesis The overall aim of this thesis was to critically review how useful current SCC parameters are for monitoring udder health on herd level by producers and processors, and if necessary to suggest improvements and recommendations for the use of these parameters. More specifically the aims of this thesis were: 1. To evaluate the usefulness of BMSCC as an indicator of the average herd ISCC; 2. To determine the association between the udder health parameters BMSCC, the average of all individual cows (ISCC), and the weighted average ISCC, and the proportion of subclinically infected cows within a herd;

4

Chapter 1

3. To quantify the contribution of management, herd and seasonal variables to the average HSCC for a European husbandry system; 4. To examine if current management practices in Europe regarding udder health management apply to the husbandry system in Australia; 5. To develop a predictive model for monthly average HSCC; 6. To determine the accuracy of describing series of BMSCC data with various sampling intervals, in order to support decision making in udder health management on farms without individual SCC data available. Outline of the Thesis Before any assessment of the udder health situation can be made, the SCC parameters currently used have to be evaluated to assess if they reflect the real situation within a herd. Therefore, in Chapter 2, BMSCC data were compared with the average herd SCC (based on individual herd recordings) to find out if and to what extent a difference existed. Subsequently, in Chapter 3, the parameters BMSCC, the weighted average SCC and the average ISCC, were compared to determine which parameter had the highest association with the prevalence of subclinical mastitis. The results in Chapter 4 illustrate the effect of various sampling intervals on the accuracy of describing series of BMSCC data. As stated in the introduction, many studies have described an association between the average herd SCC and management practices, but to date the contribution of any management practice to the average herd SCC (HSCC) has not been quantified; either the quantitative contribution of seasonal effects or herd characteristics. In Chapter 5 a linear mixed effect model was used to determine not only which management practices, herd characteristics or seasonal variables contributed to the average HSCC, but also the magnitude of their contribution. Because all management practices described in Chapter 5 are based on a system of cows housed indoors during winter and partly during summer and in a European climate, some key management practices were also tested under different climate and husbandry systems in Australia. In Chapter 6, the results of a survey regarding these management practices conducted under Australian conditions are described. Besides the monitoring of udder health based on retrospective data an additional useful tool would be a prediction of the average herd SCC of the subsequent period. This outcome could predict if penalty thresholds will be exceeded or if interventions in management are needed to maintain or improve the current situation. In Chapter 7, the outcomes of the linear mixed effect model from Chapter 5 were used to predict the average HSCC of the subsequent period.

5


Table 1. Explanation of used abbreviations in this thesis AIC

Akaike information criterion

ANN

artificial neural network

ANOVA

analysis of variance

BDCT

blanket dry cow therapy

BMSCC

bulk milk somatic cell count

CHSCC

yield-corrected herd somatic cell count

CV

coefficient of variation

DCT

dry cow treatment

DHI

dairy herd improvement

DIM

days in milk

GLM

general linear model

HSCC

average herd somatic cell count

IMM

intramammary infection

ISCC

individual somatic cell count

LGDIFF

large difference

LME

linear mixed effect

NSW

New South Wales

OR

odds ratio

SCC

somatic cell count

SCM

subclinical mastitis

SD

standard deviation

SDCT

selective dry cow therapy

SMDIFF

small difference

6

Chapter 1

References Barkema, H. W., Y. H. Schukken, T. J. G. M. Lam, M. L. Beiboer, G. Benedictus, and A. Brand. 1998a. Management practices associated with low, medium, and high somatic cell counts in bulk milk. J. Dairy Sci. 81:1917–1927. Barkema, H. W., Y. H. Schukken, T. J. G. M. Lam, M. L. Beiboer, H. Wilmink, G.Benedictus, and A. Brand. 1998b. Incidence of clinical mastitis in dairy herds grouped in three categories by bulk milk somatic cell counts. J. Dairy Sci. 81:411-419. Barkema, H. W., J. D. van der Ploeg, Y. H. Schukken, T. J. G. M. Lam, G. Benedictus, and A. Brand. 1999. Management style and its association with bulk milk somatic cell count and incidence rate of clinical mastitis. J. Dairy Sci. 82:1655-1663. Barnouin, J., M. Chassagne, S. Bazin, and D. Boichard. 2004. Management practices from questionnaire surveys in herds with very low somatic cell score through a national mastitis program in France. J. Dairy Sci. 87:3989-3999. Bartlett, P. C., G. Y. Miller, S. E. Lance, and L. E. Heider. 1992. Environmental and managerial determinants of somatic cell counts and clinical mastitis incidence in Ohio dairy herds. Prev. Vet. Med. 14:195-207. Bascom, S. S., and A. J. Young. 1998. A summary of the reasons why farmers cull cows. J. Dairy Sci. 81: 2299-2305. Berry, D. P., J. M. Lee, K. A. Macdonald, K. Stafford, L. Matthews, and J. R. Roche. 2007. Associations among body condition score, body weight, somatic cell count, and clinical mastitis in seasonally calving dairy cattle. J. Dairy Sci. 90:637-648. Berry, D. P., B. O’Brien, E. J. O’Callaghan, K. O. Sullivan, and W. J. Meaney. 2006. Temporal trends in bulk tank somatic cell count and total bacterial count in Irish dairy herds during the past decade. J. Dairy Sci. 89:4083-4093. Bradley, A., and M. Green. 2005. Use and interpretation of somatic cell count data in dairy cows. In Pract. 27:310-315. Breen, J.E., A.J. Bradley, and M.J. Green. 2009. Quarter and cow risk factors associated with a somatic cell count greater than 199,000 cells per milliliter in United Kingdom dairy cows. J. Dairy Sci. 92:3106-3115. Dohoo, I.R., and A. H. Meek. 1982. AH. Somatic cell counts in Bovine milk. Can. Vet. J. 23:119-125. Green, M. J., A. J. Bradley, G. F. Medley, and W. J. Browne. 2008. Cow, farm, and herd management factors in the dry period associated with raised somatic cell counts in early lactation. J. Dairy Sci. 91:1403-1415. Green, M. J., A. J. Bradley, H. Newton, and W.J. Browne. 2006. Seasonal variation of bulk milk somatic cell counts in UK dairy herds: Investigations of the summer rise. Prev. Vet. Med. 74:293-308.

7

General introduction Harmon, R. Physiology of mastitis and factors affecting somatic cell counts. 1994. J. Dairy Sci. 77:21032112. Hutton, C. T., L. K. Fox, and D. D. Hancock. 1990. Mastitis control practices: differences between herds with high and low milk somatic cell counts. J. Dairy Sci. 73:1135-1143. International Dairy Federation. 1995. Enumeration of somatic cells. FIL-IDF Standard no. 148A. IDF, Brussels, Belgium. Jayarao, B. M., S. R. Pillai, A. A. Sawan, D. R. Wolfgang, and N. V. Hegde. 2004. Guidelines for monitoring bulk tank milk somatic cell and bacterial counts. J. Dairy Sci. 87:3561-3573. Kelly, A. L., D. Tiernan, C. O'Sullivan, and P. Joyce. 2000. Correlation between bovine milk somatic cell count and polymorphonuclear leukocyte level for samples of bulk milk and milk from individual cows. J. Dairy Sci. 83:300-304. Kelly, P. T., K. O'Sullivan, D. P. Berry, S. J. More, W. J. Meaney, E. J. O'Callaghan, and B. O'Brien. 2009. Farm management factors associated with bulk milk somatic cell count in Irish dairy herds. Irish Vet. J. 62:45-51. Laevens, H., H. Deluyker, Y. H. Schukken, L. De Meulemeester, R. Vandermeersch, E. De Mulenaere, and A. De Kruif. 1997. Influence of parity and stage of lactation on the somatic cell count in bacteriologically negative dairy cows. J. Dairy Sci. 80:3219-3226. Neave, F. K., F. H. Dodd, R. G. Kingwill, and D. R. Westgarth. 1969. Symposium: Mastitis control: Control of mastitis in the dairy herd by hygiene and management. J. Dairy Sci. 52:696-707. Olde Riekerink, R. G., H. W. Barkema, and H. Stryhn. 2007. The effect of season on somatic cell count and the incidence of clinical mastitis. J. Dairy Sci. 90:1704-1715. Paape, M., J. Mehrzad, X. Zhao, J. Detilleux, and C. Burvenich. 2002. Defence of the bovine mammary gland by polymorphonuclear neutrophil leukocytes. J. Mammary Gland Biol. Neoplasia 7:109122. Peeler, E. J., M. J. Green, J. L. Fitzpatrick, K. L. Morgan, and L. E. Green. 2000. Risk factors associated with clinical mastitis in low somatic cell count British dairy herds. J. Dairy Sci. 83:2464-2472. Sargeant, J. M., Y. H. Schukken, and K. E. Leslie. 1998. Ontario bulk milk somatic cell count reduction program: progress and outlook. J. Dairy Sci. 81:1545-1554. Schukken, Y. H., D. J. Wilson, F. Welcome, L. Garrison-Tikofsky, and R. N. Gonzalez. 2003. Monitoring udder health and milk quality using somatic cell counts. Vet. Res. 34:579-596. Sordillo, L. M., K. Shafer-Weaver, and D. DeRosa. 1997. Immunobiology of the mammary gland. J. Dairy Sci. 80:1851-1865. Torres, A. H., P. J. Rajala-Schultz, F. J. DeGraves, and K. H. Hoblet. 2008. Using dairy herd improvement records and clinical mastitis history to identify subclinical mastitis infections at dry-off. J. Dairy Res. 75:240-247.

8

Chapter 1 Valde, J. P., O. Osteras, and E. Simensen. 2005. Description of herd level criteria for good and poor udder health in Norwegian dairy cows. J. Dairy Sci. 88:86–92. Wenz, J., S. Jensen, J. Lombard, B. Wagner, and R. Dinsmore. 2007. Herd management practices and their association with bulk tank somatic cell count on United States dairy operations. J. Dairy Sci.90:3652-3659.

9

Chapter 2


J.J. Lievaart, H.W. Barkema, H. Hogeveen and W.D.J. Kremer

Journal of Dairy Research 76 (2009), 1–7


Abstract Bulk milk somatic cell count (SCC) is a frequently used parameter to estimate the subclinical mastitis situation in a dairy herd, but it often differs considerably from the average SCC of all individual cows. In this study, first the sampling variation was determined on 53 dairy farms with a bulk milk SCC (BMSCC) ranging from 56,000 to 441,000 cells/mL by collecting five samples on each farm of the same bulk tank. The average absolute sampling variation ranged from 1,800 to 19,800 cells/mL. To what extent BMSCC represents all lactating cows was evaluated in another 246 farms by comparing BMSCC to the average herd SCC corrected for milk yield (CHSCC), after the difference was corrected for the sampling variation of BMSCC. On average BMSCC was 49,000 cells/mL lower than CHSCC, ranging from -10,000 cells/mL to 182,000 cells/mL, while the difference increased with an increasing BMSCC. Subsequently, management practices associated with existing differences were identified. Farms with a small (250,000 cells/mL) BMSCC class and corresponded to the study of Barkema et al. (1998a, b) in which the data were collected of the second group of 300 farms used in this study to quantify the difference between BMSCC and CHSCC and associated management practices. To determine the sampling variation of a BMSCC sample on each of the 53 farms, five samples were collected consecutively from the same bulk tank. The samples were collected during a routine milk delivery of every two or three days conform a standard protocol which instructed sample collectors letting the bulk tank stir for at least 5 min before sample collection and check if the temperature was below 4o Celsius. All samples were analysed on the same Fossomatic machine (FossomatictmFC). The second group of 300 farms housed lactating cows in a free-stall barn during the winter, participated in a 3 or 4-weekly milk recording system, had annual production quota between 300,000 and 900,000 kg, and had cows of the Holstein-Friesian or Dutch Friesian breeds. Three questionnaires were conducted to collect information on mastitis prevention and control management practices (Barkema et al. 1998a, b)(Table 1). Table 1. Descriptive results of management practices per category of difference between bulk milk SCC (BMSCC) and average herd SCC (CHSCC) SMDIFF (< 20%)† 121 farms

LGDIFF (≥ 20%)‡ 125 farms

Management Practices Size of the property (hectares)

44.8

50.1

Cows present, no.

57.3

61.6

Cows inside in summer (%)

3.7

3.7 b

Cows indoor during night (in the summer) (%)

31.4

37.6

Clipping of hair of cows in winter (%)

90.9

88.0

Dry cow treatment of all cows (%)

87.6

82.4

Clinical mastitis history included (%) High SCC history included (%) Production history included (%)

14

b

5.7

9.6

32.2a,b

9.6

a,b

11.0

17.4

Chapter 2 Disinfection of teats before administering dry cow treatment (%) Dry cows checked daily for clinical mastitis (%)

56.2a,b

42.4

79.3a,b

52.8

a,b

Heifers checked daily for clinical mastitis (%)

63.6

40.0

Calving area also used as sick cow pen (%)

63.6

70.4

Post-milking teat disinfection in summer (%)

36.3

40.0

Cows restrained in head gates after milking: in summer (%) in winter (%)

26.4 33.0

18.4 23.2

Recording clinical mastitis cases (%)

27.2

28.0

Always treat clinical mastitis: intramammary with antibiotics (%) intramuscular with antibiotics (%)

63.6 18.2a,b

56.0 9.8

Dry udder preparation (%)

77.7

72.0

24.7a,b

74.4

34.7

41.6

Feeding calves: high SCC milk or milk with antibiotic residues (%) milk replacer (%) fresh milk (unselected) (%)

52.1

56.8

Average dry period (days) (%)

55.4

55.2

1.7

1.7

2.3a,b

1.9

a,b

3.2

Minimal days of antibiotic treatment of clinical mastitis Number of treatments per clinical mastitis case

Minimum DIM milk excluded from bulk tank 4.1 after calving † Farms with an average difference between BMSCC and CHSCC < 20% ‡ Farms with an average difference between BMSCC and CHSCC ≥20% a Different (P < 0.05) from LGDIFF herds b Different (P < 0.25) from LGDIFF herds

The Dutch Breeding Organization (NRS, Arnhem, The Netherlands) provided the monthly test-day SCC and BMSCC data per farm for a period of 3 consecutive years. In a previous study was found that the interval between sampling of BMSCC and test-day SCC did not influence the association between BMSCC and individual SCC if BMSCC and test-day SCC were sampled within two days of each other (Lievaart et al. 2007b). Therefore, only data points of farms were included when BMSCC and individual test-day SCC data were determined within two days of each other (Lievaart et al. 2007b). Additionally, only farms that had at least three data points available were included. In total 246 out of the 300 herds fitted both criteria. To determine associated management practices, only those variables that are known from

15


a previous study to influence CHSCC (Lievaart et al. 2007a) were selected from the questionnaires. Data Handling and Statistical Analyses The five bulk tank SCC samples from each of the 53 farms were used to calculate the average absolute sampling variation and the coefficient of variation (CV) as a normalized measure of dispersion. A linear regression model was used to examine if the average absolute sampling variation or CV were associated with the level of BMSCC. The absolute variation or CV was set as a continuous dependent variable and BMSCC as a continuous independent variable. Statistical analyses were performed using R (http://www.r-project.org/version 2.2.0; last assessed May 2010) and EXCEL (Microsoft, 2002). The R2 was used to describe the proportion of variation that was explained by the model. The regression equation of the average absolute sampling variation was used to calculate the sampling variation for each individual BMSCC value of the remaining 246 farms. Thereafter, the absolute sampling variation was subtracted from the difference between BMSCC and CHSCC before the difference was assessed. The CHSCC was calculated by multiplying the test-day SCC with the test-day milk yield for each cow and averaging this figure over the total kg of milk produced by the herd on that test-day. For each farm an average difference between BMSCC and CHSCC was calculated from the available data points. The difference was calculated as an absolute and percentage difference value with BMSCC as the baseline; and again a linear regression model were used to determine a possible relationship between the differences, absolute and percentage, and the level of BMSCC. To determine whether management practices were associated with a small or a large difference between BMSCC and CHSCC all farms were divided into two categories: 1) farms with 250,000 cells/mL is an excellent parameter to summarize the subclinical mastitis situation, average herd SCC may be another useful complementary parameter. Bulk milk SCC is mainly used by dairy processors as a quality parameter, but it is also frequently used in research on udder health management and control strategies (e.g., Goodger et al., 1993; Faye et al., 1994; Barkema et al., 1998; Barkema et al., 1999; Tadich et al., 2003; De Vliegher et al., 2004). It is questionable, however, whether BMSCC is the most appropriate parameter, because for a variety of reasons, milk is withheld from the bulk tank. Instead of using BMSCC, another option would be to use individual cow DHI SCC data and determine a yield-corrected test-day SCC (CHSCC) or the arithmetic average test-day SCC of the lactating herd (HSCC). These options were previously mentioned in other studies but have not been validated (Schukken et al., 2003; Bradley and Green, 2005; Valde et al., 2005). It seems logical that parameters calculated by using the SCC of all lactating animals in the herd would more accurately reflect the subclinical mastitis situation in a dairy herd, because the milk of some cows is withheld from the bulk tank. The goal of this study was therefore to determine whether CHSCC or HSCC would be a more appropriate parameter than BMSCC to summarize the subclinical mastitis situation in a dairy herd. Materials and Methods Most of the data used in this study were presented earlier (Barkema et al., 1998). In short, milk-recording and BMSCC data were collected on 300 dairy farms in the northern part of the Netherlands in a study on bulk mastitis and management practices. Mean HSCC of a DHI test day was calculated as the arithmetic mean SCC of all lactating animals. Additionally, a herd SCC corrected for milk yield (CHSCC) was calculated by multiplying the daily test-day milk yield of each cow with its test-day SCC, and averaging this figure over all lactating cows in the herd. The prevalence of high SCC at a test day was defined as the percentage of lactating cows with SCC of >250,000 cells/mL. This cut-off value is also used by the Dutch DHI organization in its monthly test results. Bulk milk SCC and milk-recording data were included in the analysis if the BMSCC sample was collected within 2 d of the individual milkrecording data and if the BMSCC did not exceed 400,000 cells/mL, which is currently used as the penalty limit in the Netherlands. To compare BMSCC, CHSCC, and HSCC in the same range, this cut off value of 400,000 cells/mL was used for all

29

Comparison of Bulk Milk, Yield-Corrected, and Average Somatic Cell Counts Parameters

3parameters. In total, 786 test days and BMSCC of 246 farms were included in the analyses. The correlations between the percentage of high-SCC cows and the 3 parameters, BMSCC, CHSCC, and HSCC, were examined by using a partial correlation function to correct for the number of data per farm (SPSS version 12.0, SPSS Inc., Chicago, IL). The heteroskedasticity of the data was corrected by using the weighted variances instead of the normal variances. To determine the effect of the interval between BMSCC sampling and DHI test day, separate models were run for 0-, 1-, and 2-d intervals. The proportion of variance in the outcome variable that was explained by a dependent variable was calculated as the correlation coefficient, R2. To test the difference in R2 between the dependent variables, and between the 3 BMSCC models with different intervals with DHI test day, the model was also tested by using the statistical program R (http://www.r-project.org/ version 2.2.0). This program provided an Akaike information criterion (AIC) value for each model that represented the goodness of fit (Akaike, 1973). A model was considered as having a better fit if the AIC value of the model increased with more than the absolute number of 2 compared with another model. The model with the lowest AIC value was considered to be the best linear regression model (Akaike, 1973). Somatic cell counts have a log-normal distribution. Therefore, in most studies of individual and herd-level SCC, a logarithmic transformation is used. In our opinion, this was not necessary in this study because the contribution of individual cows to the bulk tank needed to be calculated. Results and Discussion Bulk milk SCC, CHSCC, and HSCC had R2 with a prevalence of high SCC of 0.64 [95% confidence interval (CI): 0.63 to 0.65; P < 0.001; Figure 1), 0.74 (95% CI: 0.71 to 0.77; P < 0.001; Figure 2), and 0.89 (95% CI: 0.87 to 0.91; P < 0.001; Figure 3), respectively.

30

Chapter 3

Figure 1. Correlation (R2) between bulk milk SCC (BMSCC) and the percentage of cows with a test-day SCC of >250,000 cells/mL

Figure 2. Correlation (R2) between yield-corrected SCC (CHSCC) and the percentage of cows with a test-day SCC of >250,000 cells/mL

31

Comparison of Bulk Milk, Yield-Corrected, and Average Somatic Cell Counts Parameters

Figure 3. Correlation (R2) between the average SCC (HSCC) and the percentage of cows with a test-day SCC of >250,000 cells/mL

The models with BMSCC, CHSCC, and HSCC had AIC values of 4,875, 4,109, and 4,035, respectively. The R2of the prevalence of high SCC and BMSCC with an interval of 0, 1, and 2 d were 0.62 (95% CI: 0.60 to 0.64; P < 0.001), 0.63 (95% CI: 0.60 to 0.66; P < 0.001), and 0.61 (95% CI: 0.58 to 0.64; P < 0.001), respectively. Conclusions We conclude that, based on the highest correlation with the percentage of cows with subclinical mastitis, HSCC is a more appropriate parameter to summarize the subclinical mastitis situation in a dairy herd than the frequently used BMSCC, and that the use of HSCC is preferable if DHI data are available. The correlation of BMSCC with the prevalence of high SCC is moderate, and there is no difference in the correlation of BMSCC and the percentage of high-SCC cows if the BMSCC sample was collected 0, 1, or 2 d after milk recording. Apart from the influence of individual cow milk yield, a possible reason for the merely moderate correlation between 32

Chapter 3

BMSCC and the percentage of high-SCC cows is the withholding of milk of highSCC cows. Acknowledgment We thank the Dutch Royal Herd Service for providing the test day data. References Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267– 281 in Second Int. Symp. on Information Theory. Budapest, Hungary. Barkema, H. W., Y. H. Schukken, T. J. G. M. Lam, M. L. Beiboer, G. Benedictus, and A. Brand. 1998. Management practices associated with low, medium, and high somatic cell counts in bulk milk. J. Dairy Sci. 81:1917–1927. Barkema, H. W., J. D. Van der Ploeg, Y. H. Schukken, T. J. G. M. Lam, G. Benedictus, and A. Brand. 1999. Management style and its association with bulk milk somatic cell count and incidence rate of clinical mastitis. J. Dairy Sci. 82:1655–1663. Bradley, A., and M. Green. 2005. Use and interpretation of somatic cell count data in dairy cows. In Pract. 27:310–315. De Vliegher, S., H. Laevens, H.W. Barkema, I. R. Dohoo, H. Stryhn, G. Opsomer, and A. de Kruif. 2004. Management practices and heifer characteristics associated with early lactation somatic cell count of Belgian dairy heifers. J. Dairy Sci. 87:937–947. Faye, B., N. Dorr, F. Lescourret, J. Barnouin, and M. Chassagne. 1994. Farming practices associated with the udder infection complex. Vet. Res. 25:213–218. Goodger, W. J., T. Farver, J. Pelletier, P. Johnson, G. de Snayer, and J. Galland. 1993. The association of milking management practices with bulk tank somatic cell counts. Prev. Vet. Med.15:235–251. Schukken, Y. H., D. J. Wilson, F. Welcome, L. Garrison-Tikofsky, and R. N. Gonzalez. 2003. Monitoring udder health and milk quality using somatic cell counts. Vet. Res. 34:579–596. Tadich, N., J. Kruze, G. Locher, and L. E. Green. 2003. Risk factors associated with BMSCC greater than 200,000 cells/ml in dairy herds in southern Chile. Prev. Vet. Med. 58:15–24. Valde, J. P., O. Osteras, and E. Simensen. 2005. Description of herd level criteria for good and poor udder health in Norwegian dairy cows. J. Dairy Sci. 88:86–92.

33

Chapter 4

Influence of the Sampling Interval on the Accuracy of Bulk Milk Somatic Cell Count Data

J.J. Lievaart, W.D.J. Kremer, J.K. Reneau, and H.W. Barkema

Submitted for publication


Abstract Although the bulk milk somatic cell count (SCC) is, in most instances, not a good proxy for the actual average herd SCC, for a large number of farms worldwide bulk milk SCC (BMSCC) is the SCC parameter available to monitor trends in udder health. The frequency of sampling BMSCC varies considerably between countries, and it is unknown to what extent the sampling interval of BMSCC or variation in BMSCC data itself influences the accuracy. The aim of this study was to assess the effect of sampling interval and variation of the BMSCC data on the accuracy to describe measured detailed series of BMSCC data. Because BMSCC is measured with regular time intervals, an Artificial Neural Network (ANN) was used to determine both the effect of sampling interval and variation of the BMSCC data. The intervals examined in this study ranged from 4 up to 14 d and were compared to the baseline of a standard 2 d sampling interval. The BMSCC data were collected every other day for a 24 mo period on 949 farms and all series were created by exclusion of BMSCC data in between the original 2 d sampling interval series. The effect of variation of BMSCC was determined by comparing the error of the ANN model in two subsets of farms, those with the lowest SD (n=239) and those with a high SD of BMSCC data (n=236). No significant differences were found in any of the sampling intervals between the two cohorts of low and high SD in BMSCC. Overall, compared to the two-day sampling interval, on average the error of the ANN model was 32,6000 cells/mL for all farms included, ranging from 15,000 cells/mL (4 d interval) to 41,000 cells/mL for the 14 d sampling interval. Keywords: bulk milk somatic cell count, sampling frequency, variation, Artificial Neural Network

36

Chapter 4

Introduction The average of all individual SCC data is the most accurate parameter to summarize the subclinical mastitis situation in a dairy herd (Lievaart et al., 2007). However, a large proportion of dairy herds does not participate in DHI milk recording (Wiggans, 2009), and therefore bulk milk SCC (BMSCC) is in these herds the only available parameter to assess trends in actual average herd SCC and the prevalence of subclinical mastitis at herd level (Jayarao et al., 2004). Compared to assessments from individual-level data, we have shown that BMSCC, when sampled at the same frequency, can lead to an additional inaccuracy of tens of thousands of cells (Lievaart et al., 2009). In the present paper we investigate how the sampling frequency itself further adds to this inaccuracy when using BMSCC as a proxy to gauge the herd SCC status. The frequency of sampling BMSCC varies between countries, and even between dairy processors within countries, and ranges from sampling every milk collection to one sample every 14 d; as a result not every bulk tank is monitored for BMSCC (Goodrigde et al., 2004). To date, it is unknown what the influence of the sampling interval is on the accuracy of BMSCC data, and whether this influence is likely to be different for farms with a low and a high variability in BMSCC (Anderson et al., 1989). The relatively high sampling frequency of BMSCC data in some countries provides an opportunity to address these questions. For our study we use data from farms sampled every two d. Artificial neural networks are frequently used to analyze time series with a high correlation between data points, and these networks can also handle data with an irregular pattern (Gurney, 1997). We use this technique to produce 2-d model predictions from data sampled at increasingly longer intervals to quantify how the error increases with interval length. Materials and Methods Data Collection and Preparation The original dataset included BMSCC data that were collected daily or every other day for at least 2 mo of the 24-mo period (January 2003 until December 2004) from 1,501 (6.4% of the total population) US Upper Midwest (North and South Dakota, Wisconsin, and Minnesota) dairy farms (Lukas et al., 2008a,b). For the current study, only farms of the original dataset were included that had a complete BMSCC data set for all months of the 24-mo period and a sampling interval of BMSCC of 2 d (baseline data set). In total 949 of the 1,501 herds fitted these criteria. Six additional datasets with a sampling interval of 4, 6, 8, 12 and 14 d, respectively, were created through ‘resampling’ of the baseline data, by exclusion of BMSCC data in the baseline at 37


appropriate times depending on the interval. The influence of the sampling interval on accuracy may be influenced by the variation of the data (Anderson et al., 1989). To test this, we divided the selected 949 farms with a complete dataset into a low and a high BMSCC variation cohort. To create these two cohorts, first BMSCC was transformed on a natural logarithmic scale to gain a normal distribution of the BMSCC data. The standard deviation (SD) of BMSCC was determined over this twoyear period and the farms with the low or high variation of BMSCC data were defined as the farms belonging to the lowest and highest 25% SD of BMSCC, respectively. Data Analysis Using an Artificial Neural Network Model Artificial neural networks (ANNs) are frequently used in human and veterinary medicine and agriculture to predict the outcome of a disease process (e.g. BartoschHärlid et al., 2008), as diagnostic support (e.g. Heald et al., 2000), or simulation of nonlinear data series (e.g. Njubi et al., 2010). These ANNs can be seen as an organized network of artificial neurons or data points that process these data with a constant feedback to constantly optimize the end result (Gurney, 1997; Hassan, 2007). For this study, the option of a time series analysis of the Alyuda NeuroIntelligence 2.2 software program (2005) was used. For each BMSCC dataset with the sampling intervals ranging from 4 to 14 d the 3 mo of data preceding the predicted BMSCC value were used to train, validate and test the ANN model. For all series we used an input layer with all data points of the preceding 3 mo (which now becomes the number of neurons), one hidden layer with half the number of neurons, and one output neuron (the described data point) (see example neural network Fig. 1).

38

Chapter 4

Input Layer Input LN BMSCC data

Hidden Layer

Output Layer Described LN BMSCC data point

Figure 1. Schematic overview of an artificial neural network with an input, hidden and output layer

The program used 60% of the data points to train the model, 20% to validate the model and 20% to test the chosen model in how well it described the baseline BMSCC series. For this study the algorithm QuickProp (Alyuda NeuroIntelligence 2.2, Los Altos, CA, United States) was used with 5000 iterations per model to find the best combination of weights for each neuron (or in this case BMSCC data point) to include for the next layer of neurons. To assess different sampling intervals and if variation of the BMSCC data (low versus high SD cohort) influenced the sampling interval the error of the ANN model that fitted the BMSCC data series was used. The 2-d sampling interval was used as the baseline to assess the other sampling intervals ranging from 4 to 14 d. We used the ANN to predict a 2-d time series for all re-sampled data sets, and then calculated the error between this prediction and the actual 2-d baseline values. A General Linear Model (GLM) was used to examine whether sampling interval or SD cohort did have an effect on the error and if there was an interaction between those two variables. The significant variables (P < 0.05) from the GLM models were further investigated using an ANOVA with a post-hoc Bonferroni test to examine which sampling interval was different to the others, while a Student’s t-test was used whether a difference existed between SD cohorts (low versus high SD of BMSCC). All data were back-transformed from the natural logarithmic scale for presentation purposes.

39


Results The 949 farms had a geometric average BMSCC of 267,000 cells/mL (range 111,000665,000 cells/mL) with an average SD of 108,000 cells/mL. The cohort of low SD farms included 239 farms that had an average BMSCC of 186,000 (range 111,000 – 424,000 cells/mL) and a SD of 75,000 cells/mL. The cohort of high SD farms included 241 farms with an average BMSCC of 478,000 (range 151,000 - 665,000 cells/mL) and SD of 149,000 cells/mL. Outcomes of the GLM Model and Optimal Sampling Interval Sampling interval was the only significant variable (P < 0.01) in the GLM model to determine the influence of sampling interval or cohort (high or low SD) of BMSCC. No significant interaction was found between the variables of cohort (high and low SD) and sampling interval. The average error of the ANN model in describing BMSCC series for all sampling intervals was 32,600 cells/mL. The sampling interval of 4 d had the lowest error of 15,300 cells/mL and was significantly lower (P=0.02) compared to all other intervals. The errors from the other sampling intervals ranged from 27,200 cells/mL for the 6 d interval to 40,400 cells/mL for the 14 d sampling interval (Fig. 2).

40

Chapter 4

45 40

BMSCC * 1,000cells/mL

35 30 25 20 15 10 5 0 d4

d6

d8

d10

d12

d14

Sampling Interval

Figure 2. The average error of bulk milk somatic cell count series (BMSCC; *1,000 cells/mL) for 6 sampling intervals (d) using artificial neural network (ANN) models

Because the prediction error of the ANN models were not influenced by the low or high SD of BMSCC no separate data were presented.

Discussion In general, the accuracy of describing a time series variable depends on the sampling interval (Anderson et al., 1989). For BMSCC data, the sampling interval various worldwide from every bulk tank, typically containing 2 d of milk production, to once every 14 d. Information on the possible error to describe BMSCC series due to the sampling interval is therefore interesting. Although BMSCC is an important udder health parameter, which is in most cases readily available with a relatively short sampling interval, it is not always the most reliable tool to monitor udder health on dairy farms (Lievaart et al., 2007; Lievaart et al., 2009). In a previous study the correlation between the proportion of cows with a SCC >250,000 cells/mL and BMSCC was 64% (Lievaart et al., 2007). This moderate correlation could partly be a result of withholding high SCC cows from the bulk tank, a practice commonly used by farmers to avoid penalties for exceeding 41


the SCC threshold for milk quality (Lievaart et al., 2009). A study which quantified the difference between BMSCC and average herd SCC corrected for milk yield (CHSCC) found a difference between the BMSCC and CHSCC parameters 49,000 cells/mL on average, ranging from -10,000 cells/mL to 182,000 cells/mL, while the difference increased with an increasing BMSCC (Lievaart et al., 2009). The current study shows that possibly an additional deviation of between 15,000 and 40,000 cells/mL is caused by the choice of sampling interval. To monitor udder health at herd level using BMSCC data would therefore only be suitable to identify rough trends, and seems of limited use to underpin decisions in short term udder health management. Given these limitations still a large number of farmers depend on the BMSCC as the only indicator of udder health, or more specifically to monitor subclinical infections in their herd. The proportion of dairy herds that participates in a DHI program ranges from 21% in Poland to 94% in Denmark (International Committee for Animal Recording, 2008). In January 2009, 48% of the US dairy cow population was not enrolled in a DHI program, while SCC was determined on 94% of the enrolled cows (Wiggans, 2009). However, when farmers do have access to individual SCC recordings, these are sometimes of limited use for short term monitoring of udder health as a result of a relative long sampling interval of DHI programs ranging from three up to six weeks. Dairy processors would also benefit from a shortest sampling interval possible because it will result in more precise monitoring of the on-farm situation regarding milk quality or food safety. For example, shelf life of milk is an important issue for dairy processors, and it has a strong negative correlation with the level of SCC of milk (Ma et al., 2000). A higher sampling frequency will increase the odds of detecting an increase in BMSCC before it exceeds the farms’ udder health objectives or BMSCC penalty limits, and it will encourage farmers to take timely action to decrease the proportion of high SCC cows and lower BMSCC. This could also have an impact on another important milk quality issue, antibiotic residues. The level of BMSCC is positively associated with an increased likelihood of antibiotic residues and pathogens in milk (Ruegg and Tabone, 2000; Van Schaik et al., 2002; Olde Riekerink et al., 2006). Although processors do check for residues frequently, preventing the need to use antibiotics by improved on-farm monitoring as a result of frequent available BMSCC data would be the preferred strategy. In this study we have not quantified the absolute error of the BMSCC as a proxy to measure the actual herd SCC. The latter is unknown and we should therefore make do with a relative assessment. For this we have assumed that the accuracy of BMSCC is constant throughout the year for a given farm and sampling strategy. We further assume that the most frequent sampling gives the most accurate description of the actual situation, even though we do not quantify how accurate that is. Our relative measure only refers to additional uncertainty introduced by less frequent sampling. In addition, because in this study the sampling intervals were artificially created by 42

Chapter 4

deleting data points within a single baseline time series, there is dependency in our procedure. One also has to keep in mind that farms used in this dataset were monitored every other day and therefore could be aware of an increase in BMSCC within 48 h, and react on this information. Farms with an actual sampling interval of 14 d would automatically have a delay to a response in BMSCC, which could result in an increased variation in BMSCC data, which we have not taken into account. Therefore, our quantitative results should be seen as indicative and not as absolute predictions of error. The results should be tested on different datasets from different countries with different sampling intervals. In the current study we only included the variation of BMSCC to examine its impact on the sampling interval. Other factors could contribute to further inaccuracy. Days in milk and average parity are both associated with an increasing prevalence of subclinical mastitis and increased SCC (Breen et al., 2009; Laevens et al., 1997). Therefore, in future studies other variables such as herd size, or average parity that potentially could influence variation in BMSCC should be examined more closely. Another interesting variable to be studied is seasonality, which has been documented in the literature to be influential risk factor for subclinical mastitis and increases in SCC (Green et al., 2006; Olde Riekerink et al., 2007). The type of pathogen causing the subclinical infections within a herd is another variable known to cause variation in SCC. For example an Escherichia coli IMI has a different SCC pattern over time compared to a Staphylococcus aureus IMI (Sol et al., 2000; De Haas et al., 2004; Shum et al., 2009). It would therefore be interesting including information on the pathogens that are most frequently found in bulk milk or clinical mastitis cases in the herds included in a study. In this study no influence was found of variability, measured by the SD, of BMSCC on the accuracy of BMSCC for the different sampling intervals. The reason may lie in the ANN technique used and the ability of this technique to deal with data demonstrating an irregular pattern (Gurney, 1997). As an alternative to the modelling approach we have chosen, a possibility would be to assess the sensitivity of different sampling intervals to quantify the number of times within a BMSCC series from different sampling intervals that the value exceeds a predefined limit such as the penalty limit of dairy processors. Conclusions and Implications In addition to the known restrictions of the BMSCC parameter to assess average SCC at herd level, an increased sampling interval of BMSCC does cause an additional inaccuracy. Compared to the standard 2 d sampling interval used in this study, the error in describing BMSCC data increased from 15,000 to 41,000 cells/mL for the 4 43


and 14 d sampling interval, respectively. Therefore, to the length of the sampling interval greatly influences the usefulness of BMSCC in gauging whether milk quality is likely to violate penalty limits. Future research in determining which other factors influence the accuracy of BMSCC data should include herd size, average parity and type of pathogen causing subclinical mastitis. References Anderson, S. M., I. L. Mao, and J. L. Gill 1989. Effect of frequency and spacing of sampling on accuracy and precision of estimating total lactation milk yield and characteristics of the lactation curve. J. Dairy Sci. 72:2387-2394. Alyuda Research, 2005. Alyuda NeuroIntelligence 2.2. Los Altos, CA. Bartosch-Härlid, A., B. Andersson, U. Aho J. Nilsson, and R. Andersson. 2008. Artificial neural networks in pancreatic disease. Br. J. Surg. 95:817-826. Breen, J. E., M. J. Green, and A. J. Bradley. 2009. Quarter and cow risk factors associated with the occurrence of clinical mastitis in dairy cows in the United Kingdom. J. Dairy Sci. 92:25512561. De Haas, Y., R. F. Veerkamp, H. W. Barkema, Y. T. Grohn, and Y. H. Schukken. 2004. Associations between pathogen-specific cases of clinical mastitis and somatic cell count patterns. J. Dairy Sci. 87:95–105. Goodridge, L., A. R. Hill, and R. W. Lencki. 2004. A review of international standards and the scientific literature on farm milk bulk-tank sampling protocols. J. Dairy Sci. 87:3099-3104. Green, M. J., A. J. Bradley, H. Newton, and W. J. Browne. 2006. Seasonal variation of bulk milk somatic cell counts in UK dairy herds: Investigations of the summer rise. Prev. Vet. Med. 74:293-308. Gurney, K. 1997. An introduction to neural networks. UCL Press. London. Hassan, K. J. 2007. Application of Artificial Neural Networks for understanding and diagnosing the state of mastitis in dairy cattle. MS Thesis. Lincoln Univ., Canterbury, New Zealand. Heald, C. W., T. Kim, W. M. Sischo, J. B. Cooper, and D. R. Wolfgang. 2000. A computerized mastitis decision aid using farm-based records: an artificial neural network approach. J. Dairy Sci. 83:711-720. Jayarao, B. M., S. R. Pillai, A. A. Sawant, D. R. Wolfgang, and N. V. Hegde. 2004. Guidelines for monitoring bulk tank milk somatic cell and bacterial counts. J. Dairy Sci. 87:3561-3573. Laevens, H., H. Deluyker, Y. H. Schukken, R. Vandermeersch, E. De Meulenaere, and A. De Kruif. 1997. Influence of parity and stage of lactation on the somatic cell count in bacteriologically negative dairy cows. J. Dairy Sci. 80:3219-3226.

44

Chapter 4 Lievaart J. J., W. D. J. Kremer, and H. W. Barkema. 2007. Comparison of bulk milk, yield-corrected, and average somatic cell counts as parameters to summarize the subclinical mastitis situation in a dairy herd. J. Dairy Sci. 90:4145-4148. Lievaart, J. J., H. W. Barkema, H. Hogeveen, and W. D. J. Kremer. 2009. Reliability of the bulk milk somatic cell count as an indication of average herd somatic cell count. J. Dairy Res. 76: 490496. Lukas, J. M., J. K. Reneau, and M. L. Kinsel. 2008a. Predicting somatic cell count standard violations based on herd's bulk tank somatic cell count. Part I: Analyzing variation. J. Dairy Sci. 91:427432. Lukas, J. M., J. K. Reneau, C. Munoz-Zanzi, and M. L. Kinsel. 2008b. Predicting somatic cell count standard violations based on herd's bulk tank somatic cell count. Part II: Consistency index. J. Dairy Sci. 91:433-441. Ma, Y., C. Ryan, D. M. Barbano, D. M. Galton, M. A. Rudan, and K. J. Boor. 2000. Effects of somatic cell count on quality and shelf-life of pasteurized fluid milk. J. Dairy Sci. 83:264-274. Njubi, D. M., J. W. Wakhungu, and M. S. Badamana. 2010. Use of test-day records to predict first lactation 305-day milk yield using artificial neural network in Kenyan Holstein-Friesian dairy cows. Trop. Anim. Hlth. Prod. 42:639-644. Shum, L.W., C. S. McConnel, A. A. Gunn, and J. K. House. 2009. Environmental mastitis in intensive high-producing dairy herds in New South Wales. Aust. Vet J. 87:469-475. Sol, J., O. C. Sampimon, H. W. Barkema, and Y. H. Schukken. 2000. Factors associated with cure after therapy of clinical mastitis caused by Staphylococcus aureus. J. Dairy Sci. 83:278-284. Olde Riekerink, R. G. M., H. W. Barkema, S. Veenstra, D. E. Poole, R. T. Dingwell, and G. P. Keefe. 2006. Prevalence of contagious mastitis pathogens in bulk tank milk in Prince Edward Island. Can. Vet. J. 47:567-572. Olde Riekerink, R.G.M., H. W. Barkema, and H. Stryhn. 2007. The effect of season on somatic cell count and the incidence of clinical mastitis. J. Dairy Sci. 90:1704-1715. Van Schaik, G., M. Lotem, and Y. H. Schukken. 2002. Trends in somatic cell counts, bacterial counts, and antibiotic residue violations in New York State during 1999-2000. J. Dairy Sci. 85:782-789. Wiggans,

G. R. 2009. USDA Summary of DHI Participation (DHI Report http://aipl.arsusda.gov/publish/dhi/current/partall.html. (Last accessed May 2010)

K-1).

45

Chapter 5

Effect of Herd Characteristics, Management Practices, and Season on Different Categories of the Herd Somatic Cell Count

J. J. Lievaart, H. W. Barkema, W. D. J. Kremer, J. van den Broek, J. H. M. Verheijden, and J. A. P. Heesterbeek

Journal of Dairy Science 90 (2007), 4137–4144

Effect of Herd Characteristics, Management Practices and Season

Abstract In this study, the contribution of management practices, herd characteristics, and seasonal variables to the herd somatic cell count (SCC) was quantified in herds with low (200,000 cells/mL) herd SCC (HSCC). Selection of the variables was performed using a linear mixed effect model; HSCC was calculated as the arithmetic mean of the individual cow’s SCC. The data concerning management practices were derived from 3 questionnaires on mastitis prevention and management practices on 246 Dutch dairy farms. The monthly Dairy Herd Improvement test data of these 246 farms were used to calculate the herd characteristics and seasonal effects. None of the management practices were associated with HSCC in all 3 HSCC categories. Some variables only had a significant association with HSCC in one HSCC category, such as dry premilking treatment (−9,100 cells/mL in the low HSCC category) or feeding calves with high SCC milk (11,100 cells/ mL in the medium HSCC category). Others had an opposite effect on HSCC in different HSCC categories, such as average parity (−6,400 and 11,000 cells/mL in the low and medium HSCC category, respectively) and feeding calves with fresh milk (10,300 and −9,700 cells/mL in the low and high HSCC category, respectively). We conclude that, given the individual Dairy Herd Improvement data and information on management practices of an individual farm, it is possible to provide quantitative insight into the contribution of these different variables to the HSCC of an individual farm. Being able to provide such insight is a prerequisite for interpretation, prediction, and control of HSCC on individual dairy farms. Keywords: herd somatic cell count, season, herd characteristic, management practice, quantification

48

Chapter 5

Introduction The herd-level SCC is a result of multifactorial cow factors, management practices, and seasonal fluctuations. Cow factors that influence herd SCC are composed of herd size, average parity, DIM, production category, and breed (De Vliegher et al., 2004; Rodrigues et al., 2005; Sewalem et al., 2006). The pathogen distribution among the herd also influences the level of herd SCC (HSCC); herds that are Staphylococcus aureus-positive in the bulk milk have higher bulk milk SCC (BMSCC) than S. aureusnegative herds (Pitkälä et al., 2004; Olde Riekerink et al., 2006). Dry cow treatment, milking technique, postmilking teat disinfection, and antibiotic treatment of clinical mastitis are examples of management factors with a significant effect on BMSCC (Igono et al., 1988; Wilson et al., 1997; Barkema et al., 1998a; Barnouin et al., 2004). Seasonal fluctuations such as the summer peak can also have a major effect but do not occur in all herds (Igono et al., 1988; Green et al., 2006). Ideally one would want to continuously monitor and interpret SCC on the herd level and to detect an increase or decrease in the trend over time (Barkema et al., 1999; Jayarao et al., 2004). Especially when bonus programs are installed such as in many countries and states of the United States, which apply a cut-off value varying from 150,000 to 250,000 cells/mL (Sampson, 2006). It is important to know in farms that may exceed a cutoff value what is needed to bring it below the bonus program requirements. Most of the research done on the association between management practices and herd-level SCC uses BMSCC as the outcome variable (Barkema et al., 1998a). Bulk milk SCC does, however, not always provide a good summary of the SCC situation in the lactating herd (Valde et al., 2005). The average HSCC of all lactating cows is in that respect a better parameter management practices is studied most often a linear association is assumed (Goodger et al., 1993; Barkema et al., 1998a), but it is likely that because of a different pathogen distribution management practices have a different effect in herds with a different BMSCC (De Haas et al., 2004). The interpretation and judgment of the HSCC is essential, but there is still little knowledge about the precise quantitative contributions of the significant variables and possible difference of these quantitative effects on different HSCC categories. Therefore, in this study the contribution of herd characteristics, management practices, and seasonal effects was quantified on a low, medium, and high long-term average HSCC by means of a linear mixed effect (LME) model.

49


Materials and Methods Data Collection Farms were selected that housed lactating cows in a free-stall barn during the winter, participated in a 3 or 4-weekly milk recording system, had annual production quota between 300,000 and 900,000 kg, and had cows of the Holstein-Friesian or Dutch Friesian breeds. Three questionnaires were conducted to collect information on mastitis prevention and control management practices (Barkema et al., 1998a,b). The Dutch Breeding Organization (NRS, Arnhem, The Netherlands) provided the monthly DHI data per farm for a period of 2.5 consecutive years. The HSCC was calculated as the arithmetic mean of the individual cows’ SCC for each milk recording date (Lievaart et al., 2007). Per farm, the mean from the initial 6 mo was used to divide the farms into 3 HSCC categories: low (200,000 cells/mL) HSCC. The reason to divide the farms into 3 categories is derived from the used data set of Barkema et al. (1998a,b), which described the difference in management variables related to different categories of BMSCC. In total, 246 farms out of the original 300 farms had a complete data set of milk recording data and questionnaires distributed over 81, 86, and 79 farms with a low, medium, and high mean HSCC, respectively. Because of missing individual SCC data, 54 farms had to be excluded from the study. The remaining 2 yr of HSCC data were used to examine the effect of season, herd characteristics, and management practices on HSCC for each category. From all variables for which information was collected, only those variables that are known from literature to influence SCC were selected (Table 1). Additionally, to examine the influence of the previous HSCC and the withinherd distribution of previous individual SCC data, both variables were included in the category of herd characteristics. To characterize this within-herd distribution, the individual SCC were divided into 4 categories, with the precise cut-off value given in Table 1. We then specified the distribution at a given recording date as the percentage of individual cows in each category at that date. Data Handling and Statistical Analysis Statistical analyses were performed using R (http:// www.r-project.org/ version 2.2.0). A LME model was designed with HSCC as the dependent variable to assess the contribution of the explanatory variables on HSCC (Pinheiro and Bates, 2000). The explanatory variables in the group’s seasonal effects, herd characteristics, and management practices were evaluated. The following LME model represents the basis to test the fit of all variables per HSCC category: 50

Chapter 5

HSCC = intercept + β1 × seasonal effects + β2 × management practices + β3 × herd characteristics + random effect “herd” + ε. Table 1. Summary of the variables; seasonal factors, herd characteristics, and management practices Component

Variables

Seasonal factors

Month and year

Herd characteristics

Herd size, average DIM, average parity, previous herd SCC, percentage of cows with a SCC 0 to 50, 51 to 150, 151 to 250, and 251 to 500 within the previous recording date

Management practices

Size of the property, clipping hair of cows, zero grazing, locked in head gates after milking, dry off procedure, clinical mastitis checks during dry period, planned length of dry period, presence of calving pen, udder preparation, postmilking teat disinfection, dry cow treatment, method of treatment of clinical mastitis, minimal number of antibiotic treatments per clinical mastitis case, registration of clinical mastitis cases, time after calving milk is added to bulk tank, feeding milk with high SCC, antibiotic residues, or milk replacer to the young calves

This model was fitted for each of the 3 HSCC categories (low, medium, and high) separately. The Akaike information criterion (AIC) was used to select the best fitting LME model (Akaike, 1973) using a backwards-stepwise elimination procedure. This selection criterion was defined as follows: AIC = −2(loglikelihood) + 2 k, where k is the number of explanatory variables (+ intercept) included in the model. The AIC values were then used to compare a series of LME models, and the model with the lowest AIC was considered to be the best LME model (Akaike, 1973). The elimination of variables stopped when the AIC did not change with more than the absolute number of 2 and had the lowest number of variables included. Therefore, a variable was declared significant if the AIC value of the LME model did not change with more than the absolute number of 2. Even though variables for which β values were significant in the best LME model, there was variation between significant variables in how pronounced their effect was. Finally, we decided to include variables into the final model if the standard error was less than one-half of the β value and if they had a clear effect of at least 1% of the average HSCC in the relevant category. Clear effects had a wide range of importance of up to almost 30,000 cells/mL, and they could be positive and negative. A random farm effect was used as well as a first 51


order autoregressive (AR1) correlation structure to model dependence in time. Inclusion of the AR1 structure did lower the AIC value, and therefore it was included in the model. The variances in the model were allowed to differ for each category. The linear part of the model was used to quantify the significance and the contribution on HSCC per variable. Outliers of the data did not influence the outcome of the models, and the residuals of the models were normally distributed. Results The AIC values of the starting models, including all variables, and the final models with the remaining variables are 8,260.9 vs. 8,236.1 (low HSCC category), 3,529.1 vs. 3,513.9 (medium HSCC category), and 7,327.9 vs. 7,288.2 (high HSCC category), respectively. In each HSCC category, mean HSCC of all farms remained within the predefined limits of that category, except for 2 occasions in the low and medium HSCC category (Figure 1).

Figure 1. Mean herd SCC (HSCC) of herds in the low (200,000 cells/mL) HSCC category.

52

Chapter 5

In March, 5 farms exceeded the predefined limits in the medium category of HSCC, and in August 3 farms within the low category of HSCC, respectively (Figure 1). The highest HSCC value in the high category of HSCC was 593,000 cells/mL. Over the 2-yr period all HSCC categories had a similar pattern of fluctuation, and the standard deviation per category was roughly constant. The results of the LME model including all significant variables are presented in Tables 2 to 4. Table 2. Seasonal variables included in the final linear mixed effect model for 3 herd SCC (HSCC; x 1,000 cells/mL) categories Herd SCC 200 (n = 79)

SE

ß

SE

ß

SE

16.6

–20.8

42

47.7

39.9

1

Month 1

Ref

Ref

Ref

Month 2

–1.0

6.6

–2.8

4.7

11

6

Month 3

3.4

6

3.2

4.3

10.1

5.5

Month 4

9.9

6.2

10.8

4.4

20.4

5.6

Month 5

24.5

6.1

18.7

4.4

27.9

5.6

Month 6

8.1

6.1

13.7

4.4

18.5

5.7

Month 7

7.4

6.1

14

4.4

22

5.6

Month 8

21

6.1

16.6

4.4

19.4

5.6

Month 9

6.3

6.1

12.7

4.3

12.2

5.6

Month 10

13.9

6.2

7.9

4.4

13

5.6

Month 11

1.1

6

4

4.3

6.1

5.4

Month 12

4.1

6.6

6.4

4.7

11.7

6

Year 1

—2

Year 2

—

Ref —

–6.9

Ref 1.2

–7.2

1.6

1

Ref = reference category Deleted from the model via backward elimination based on the lowest Akaike information criterion value

2

53


Figure 2. Contribution of the variable month to the mean herd SCC (HSCC) in the low (200,000 cells/mL) HSCC category

In the low HSCC category, the monthly contribution had the largest value during May (24,500 cells/mL) and August (21,000 cells/mL; Table 2; Figure 2). The same held for the medium HSCC category. The months May (27,900 cells/mL) and July (22,000 cells/mL) demonstrated the largest contribution for the high HSCC category (Table 2; Figure 2). The year effect was not significant in the low HSCC category and indicated a significant decrease of 6,900 cells/mL and 7,200 cells/mL in the second year for the medium and high HSCC category, respectively. The significant herd characteristics are presented in Table 3. The HSCC increased with increasing average parity in the medium (11,300 cells/mL) and high (8,400cells/mL) HSCC category, but an opposite association (6,400 cells/mL) was found in the low HSCC category. Average milk yield and size of the lactating herd were associated in only 2 HSCC categories: HSCC decreased with 900 and 1,200 cells/mL per kg increase in daily milk yield in the medium and high HSCC category, respectively. In the medium and high HSCC categories, HSCC increased with increasing lactating herd size: 200 and 100 cells/mL per head, respectively. Average DIM was not associated with HSCC in any of the 3 HSCC categories. Herd SCC at the previous milk recording date had the largest contribution of all variables to the current HSCC in the LME model. This variable had a contribution of 0.60, 0.61, or 0.73 times the previous HSCC in the low, medium, and 54

Chapter 5

high HSCC categories, respectively. The percentage of cows with an individual SCC from 51,000 to 150,000 cells/mL, and those with SCC 151,000 to 250,000 cells/mL at the previous test day had a significant contribution in the low HSCC category, but the latter was not a clear effect (Table 3). In the medium and high HSCC category of HSCC, all categories of SCC had a significant association (Table 3). Table 3. Herd characteristics in the final linear mixed effect model based on the lowest Akaike information criterion value for 3 herd SCC (HSCC; x 1,000 cells/mL) categories Herd SCC 200 (n = 79)

SE

ß

SE

ß

SE

3.4

11.3

2.9

8.4

3.1

—

–0.90

0.3

1.2

0.4

0.1

0

—

—

Average daily production (kg)

—

Herd size (lactating cows)

0.2

0.01

Previous HSCC

0.6

0.03

0.61

0.03

0.73

0.03

% cows SCC 0 to 50 previous HSCC

—

—

0.79

0.36

1.32

0.33


0.59

0.09

0.99

0.38

1.66

0.37


0.46

0.24

0.84

0.36

1.82

0.36


—

—

1.15

0.33

1.52

0.33

1

Deleted from the model via backward elimination based on the lowest Akaike information criterion value.

The significant management variables are presented in Table 4. None of the variables were significant in all 3 HSCC categories, whereas some variables had an opposite effect in different HSCC categories. Checking the dry cows visually for clinical mastitis (daily 27,700 cells/mL or weekly 24,600 cells/mL), time after calving that the milk is added to the bulk tank (2,700 cells/mL per day), and locking cows in the head gates after milking during winter and summer period (13,500 cells/mL vs. 18,800 cells/mL) were only associated with HSCC in the low HSCC category. Feeding calves with milk of high SCC cows (11,100 cells/mL) was only manifest in the medium HSCC category. In the high HSCC category, registration of clinical mastitis cases (10,000 cells/mL), minimal days of treatment (5,500 cells/mL per day),

55


and feeding calves with milk replacer (7,000 cells/mL) demonstrated significant contributions. Table 4. Management practices in the final linear mixed effect model based on the lowest Akaike information criterion (AIC) value for 3 herd SCC (HSCC; x 1,000 cells/mL) categories HSCC

Variable

200 (n = 79) ß SE

–8.8

—

–8.4

4

—

—

—

—

8

—

—

—

—

8

—

—

—

— —

Heifers not visually checked for mastitis

—

Ref

—

Ref

Heifers visually checked for mastitis every day

—

—

–6.3

4

–11.5

3

Heifers visually checked for mastitis every week

—

—

–8.3

4

–7.0

3

Wet premilking treatment

Ref

—

—

—

—

Dry premilking treatment

–9.1

3

—

—

—

—

Time after calving milk is added to bulk tank (d)

–2.7

1

—

—

—

—

Registration of clinical mastitis cases

—

—

—

—

–10.0

3

Minimal days of treatment of clinical mastitis

—

—

—

—

–5.5

2

Postmilking teat disinfection in summer

4.3

2

–5.8

3

—

—

Calves fed milk with high SCC

—

—

11.1

3

—

—

10.3

2

—

—

–9.7

3

—

—

—

—

–7.0

3

—

—

—

—

Calves fed with fresh milk Calves fed with milk replacer Cows not fed and not locked in head gates after milking in the winter season Cows fed and not locked in head gates after milking in winter season Cows fed and locked in head gates after milking in winter season Cows not fed and not locked in head gates after milking in the summer season Cows fed and not locked in head gates after milking in summer season Cows fed and locked in head gates after milking in summer season 1

Ref 14.2

5

—

—

—

—

13.5

4

—

—

—

—

—

—

—

—

Ref –13.1

5

—

—

—

—

–18.8

4

—

—

—

—

Deleted from the model via backward elimination based on the lowest AIC value. Ref = referent category

2

56

Chapter 5

Checking the heifers for clinical mastitis weekly or daily was associated with a decreased HSCC in the medium and high HSCC category of 6,300 and 11,500 cells/mL for the everyday checking and 8,300 and 7,000 cells/ mL for every week checking, respectively. Opposite correlations were found between the low and medium category of HSCC for postmilking teat disinfection in the summer (4,300 vs. −5,800 cells/mL), and between the low and high category of HSCC for feeding calves fresh milk (10,300 vs. −9,700 cells/mL). To assess the contribution of continues variables in the LME model, Table 4 presents the mean value and contribution per significant variable in each category of HSCC. Discussion The initial focus of this study was to quantify the contribution to HSCC of a large number of variables to support the interpretation of the current HSCC. Subsequently, the knowledge on fixed quantitative effects of management variables and seasonal effects together with the present variable individual SCC data can be used in modelling the HSCC of the next period for individual farms. We studied influence of variables in 3categories of average HSCC: low, medium, and high. Surprisingly, during this study we detected that none of the variables were significantly associated with HSCC in all of the predefined categories, and some even had an opposite effect in 2 HSCC categories. This was not mentioned earlier in relevant literature, and therefore this study provides new insight into the way different variables influence SCC on the herd category. There are 3 possible explanations for the observed differences in association with HSCC among HSCC categories: 1) a difference in awareness of farmers in different HSCC categories, 2) a different approach or overestimation of the effects of management practices in the different HSCC categories, and 3) different pathogen distributions in the different HSCC categories. All farms were visited 3 times to conduct a questionnaire; every year of the study there was a meeting for all participating farmers, a whole herd quarter milk sample collection was conducted, and the milking machine was evaluated during milking. Additional contact with the farmers occurred every 6 to 8 wk when clinical mastitis samples were collected. All these contacts may have increased the awareness of the effect of udder health management practices resulting in a decreasing HSCC in the medium and high HSCC category in the second year of the study (Figure 1). The different approach of management practices among farms with a different BMSCC category is also described in other studies on udder health management (Barkema et al., 1999; Barnouin et al., 2004; De Vliegher et al., 2004; Rodrigues et al., 2005) described in a study on management style and its association with BMSCC and incidence rate of clinical mastitis, 2 clusters of farmers, a “quick and dirty” cluster and 57


a “clean and accurate” cluster. Farmers with a low BMSCC were more precise than fast, whereas farmers with a high BMSCC showed the opposite attitude. Tarabla and Dodd (1996 ) carried out a survey on associations between farmers’ personal characteristics, management practices, and farm performances (milk yield and quality) and reported that the variables related to farmers’ attitude and socio-demographic profile explained a similar or greater amount of the farm performance than the group of management variables. These studies confirmed that the influence of the farmers’ attitude is underestimated, and the effect of management practices is perhaps overestimated. The third possible explanation of different pathogen distributions is based on studies that found different pathogens on different levels of individual or HSCC (Schukken et al.,1990; De Haas et al., 2004). These pathogen distributions are not included in the model, but determination of the pathogen distribution is very important when giving recommendations on farm-specific udder health management. Also, the large monthly fluctuations could not be explained throughout this study. Two possible explanations for these fluctuations could again be the pathogen distribution and the influence of temperature. Apart from those 3 possible explanations, awareness, approach, and pathogen distributions, logical pathways for the absence of effects or opposite directions for several management practices are not always easy to provide. Also, a remarkable outcome was found for locking the cows in head gates after milking. This management practice had different effects during different seasons (winter or summer). So far no biological reason for the outcome can be provided. The authors suggest that more research regarding the influences on different levels of HSCC including pathogen distribution is needed, especially because all outcomes were compared with studies that used the BMSCC as a dependent variable instead of the HSCC parameter. The parameter used in this study, HSCC, was different from the frequently used BMSCC. Originally, BMSCC was introduced as a quality parameter by dairy processors. As a consequence of not including milk of all cows in the bulk milk tank the BMSCC will not always be a reliable reflection of the SCC of all cows in a herd. On that account, in this study only the individual SCC were used to calculate the HSCC. In comparison with the original study of Barkema et al. (1998a) on management practices, which uses the BMSCC parameter instead of the HSCC parameter, some inexplicable differences in significant management practices were found. The most important management practices between herds with a low BMSCC of the original study and the other herds were postmilking teat disinfection (only a difference between the low and medium HSCC category in the current study), duration of treatment of clinical mastitis cases (only an effect in the high HSCC category of the current study), and drying after wet premilking treatment (no effect in the current study). Not finding an association between HSCC and a management practice within a certain HSCC category, however, does not automatically imply that 58

Chapter 5

this management practice would not have an effect on cow- and herd-category SCC. If the proportion of herds that has adopted this management practice is low or high, a larger number of herds would need to be included in the study or a different study design would be necessary to increase the power of the study and find a significant association with HSCC. The low number of significant management practices and in some cases their small contribution to the total HSCC imply the need for a profound study on these particular management practices. The low number of significant management practices and in some cases their small contribution to the total HSCC implies the need for a profound study on these particular management practices. The design of this study, particularly the classification of all herds within a defined category, could prevent finding important significant management practices. Therefore, a subsequent study should include farms with a less constant level of HSCC and a more intensive monitoring of the management practices. Regarding the quantification of the significant management variables, this study provides a range for each HSCC category based on the accepted practices on the individual farm. The range between the sum of increasing or decreasing significant management practices was 99,300, 34,000, and 43,700 cells/mL in the low, medium, and high HSCC category, respectively. In total this range is rather large for the low HSCC category and small for the medium and high HSCC category. A general explanation for the difference in range could be the accuracy of the farmers to carry out the necessary management practices in herd within the high HSCC category. Conclusions None of the management variables was significantly associated with HSCC in all 3 HSCC categories, whereas some variables had an opposite effect between HSCC categories. This suggests that care must be taken in drawing generalized conclusions for the possible effectiveness of, for example, management changes aimed at reducing HSCC for a particular farm. This also suggests the need for a reinvestigation of various management practices factors and their effects on HSCC. Acknowledgment The authors would like to thank the Dutch Breeding Organization for providing the individual milk production recording data used in this study.

59


References Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267– 281 in Second Int. Symp. Inf. Theory, Budapest, Hungary. Barkema, H. W., Y. H. Schukken, T. J. G. M. Lam, M. L. Beiboer, G. Benedictus, and A. Brand. 1998a. Management practices associated with low, medium, and high somatic cell counts in bulk milk. J. Dairy Sci. 81:1917–1927. Barkema, H. W., Y. H. Schukken, T. J. G. M. Lam, M. L. Beiboer, G. Benedictus, and A. Brand. 1999. Management practices associated with the incidence rate of clinical mastitis. J. Dairy Sci. 82:1643–1654. Barkema, H. W., Y. H. Schukken, T. J. G. M. Lam, M. L. Beiboer, H. Wilmink, G. Benedictus, and A. Brand. 1998b. Incidence of clinical mastitis in dairy herds grouped in three categories by bulk milk somatic cell counts. J. Dairy Sci. 81:411–419. Barnouin, J., M. Chassagne, S. Bazin, and D. Boichard. 2004. Management practices from questionnaire surveys in herds with very low somatic cell score through a national mastitis program in France. J. Dairy Sci. 87:3989–3999. De Haas, Y., R. F. Veerkamp, H. W. Barkema, Y. T. Gro¨hn, and Y. H. Schukken. 2004. Associations between pathogen-specific cases of clinical mastitis and somatic cell count patterns. J. Dairy Sci. 87:95–105. De Vliegher, S., H. Laevens, H. W. Barkema, I. R. Dohoo, H. Stryhn, G. Opsomer, and A. de Kruif. 2004. Management practices and heifer characteristics associated with early lactation somatic cell count of Belgian dairy heifers. J. Dairy Sci. 87:937–947. Goodger, W. J., T. Farver, J. Pelletier, P. Johnson, G. de Snayer, and J. Galland. 1993. The association of milking management practices with bulk tank somatic cell counts. Prev. Vet. Med. 15:235–251. Green, M. J., A. J. Bradley, H. Newton, and W. J. Browne. 2006. Seasonal variation of bulk milk somatic cell counts in UK dairy herds: Investigations of the summer rise. Prev. Vet. Med. 74:293–308. Igono, M. O., H. D. Johnson, B. J. Steevens, W. A. Hainen, and M. D. Shanklin. 1988. Effect of season on milk temperature, milk growth-hormone, prolactin, and somatic-cell counts of lactating cattle. Int. J. Biometeorol. 32:194–200. Jayarao, B. M., S. R. Pillai, A. A. Sawant, D. R. Wolfgang, and N. V. Hegde. 2004. Guidelines for monitoring bulk tank milk somatic cell and bacterial counts. J. Dairy Sci. 87:3561–3573. Lievaart, J. J., W. D. J. Kremer, and H. W. Barkema. 2007. Comparison of bulk milk somatic cell count yield corrected and average somatic cell count as parameters to summarize the subclinical mastitis situation in a dairy herd. J Dairy Sci. 90:4145-4148. Olde Riekerink, R. G., H. W. Barkema, S. Veenstra, D. E. Poole, R. T. Dingwell, and G. P. Keefe. 2006. Prevalence of contagious mastitis pathogens in bulk tank milk in Prince Edward Island. Can. Vet. J. 47:567–572.

60

Chapter 5 Pinheiro, J. C., and D. M. Bates. 2000. Mixed effects model in S and S-Plus. Springer, New York, NY. Pitkälä , A., M. Haveri, S. Pyörälä , V. Myllys, and T. Honkanen- Buzalski. 2004. Bovine mastitis in Finland 2001. Prevalence, distribution of bacteria, and antimicrobial resistance. J. Dairy Sci. 87:2433–2441. R: A language and environment for statistical computing, R Development Core Team, R Foundation for Statistical Computing, Vienna, Austria, 2005. http://www.R-project.org (last accessed May 2010). Rodrigues, A. C. O., D. Z. Caraviello, and P. L. Ruegg. 2005. Management of Wisconsin dairy herds enrolled in milk quality teams. J. Dairy Sci. 88:2660–2671. Sampson, R. 2006. Dairy farmer’s response to the quality bonus program. Proc. NMC Summer meeting, August 9–10, 2006, Charlottetown, Prince Edward Island, Canada. Schukken, Y. H., J. Buurman, A. Brand, D. van der Geer, and F. J. Grommers. 1990. Population dynamics of bulk milk somatic cell counts. J. Dairy Sci. 73:1343–1350. Sewalem, A., G. J. Kistemaker, F. Miglior, and B. J. Van Doormaal. 2006. Analysis of inbreeding and its relationship with functional longevity in Canadian dairy cattle. J. Dairy Sci. 89:2210–2216. Valde, J. P., O. Osteras, and E. Simensen. 2005. Description of herd level criteria for good and poor udder health in Norwegian dairy cows. J. Dairy Sci. 88:86–92. Wilson, D. J., H. H. Das, R. N. Gonzalez, and P. M. Sears. 1997. Association between management practices, dairy herd characteristics, and somatic cell count of bulk tank milk. J. Am. Vet. Med. Assoc. 210:1499–1502.

61

Chapter 6


K. Plozza, J.J. Lievaart, G. Potts, and H.W. Barkema

Accepted for publication


Abstract To determine the current prevalence of subclinical mastitis and associated risk factors on New South Wales dairy farms a survey was sent to 382 New South Wales dairy farmers to acquire information on relevant risk factors associated with subclinical mastitis. The average herd prevalence of the 189 respondents (response rate 49.5%) of subclinical mastitis was 29%. Farmers that had herds with a low prevalence of subclinical mastitis (200,000 cells/ml) wore gloves during milking more frequently when compared to farmers that had herds with a high prevalence of subclinical mastitis (>30% cows with ISCC >200,000 cells/ml) (62 versus 26%), used individual paper towels for udder preparation (62 versus 16%), feeding cows directly post-milking (87 versus 47%) and more frequently treated high somatic cell count cows (80 versus 69%). Farmers that had herds with a high prevalence of subclinical mastitis (>30% cows with ISCC >200,000 cells/ml) more often used selective dry cow therapy (52 versus 24%) compared to low prevalence herds. The prevalence of subclinical mastitis in this cross sectional study is comparable or lower than other studies in North America and the European Union. The outcome of the survey will provide a benchmark for the current New South Wales dairy industry to focus in the near future on the management practices associated with the low prevalence of subclinical mastitis such as wearing gloves, using paper towels and feeding cows directly after milking. Keywords: management practices, prevalence, subclinical mastitis

64

Chapter 6

Introduction Subclinical mastitis (SCM) is of great importance to dairy farmers which results in reductions in milk yield, undesired compositional changes within the milk Brightling et al., 1998; Seegers et al., 2003; Halasa et al., 2009) as well as costs associated with control strategies that are implemented. Previous studies outlining recommended control strategies for SCM commonly focus on risk factors associated with management (Barkema et al., 1998; Lievaart et al., 2007; Halasa et al., 2009; O’Brien et al., 2009), cow (Harmon et al., 1994; Laevens et al., 1997), and environment (Green et al., 2006, Olde Riekerink et al., 2007; Green et al., 2008; Lievaart et al., 2009). However, the majority of these studies are conducted overseas, utilise housed animals, and are often done in climates very different to those encountered in Australia. This is of importance as SCM has the greatest effect on bulk milk somatic cell count (SCC) (Dohoo and Meek, 1982; Green et al., 2006; Olde Riekerink et al., 2007; Lievaart et al., 2009) and often penalty systems are enforced at the dairy processor level for high bulk milk SCC (BMSCC) milk. In Australia, the 2008 national average BMSCC was 217,000 cells/ml (Australian Diary Herd Improvement report, 2008) with some dairy processors currently deeming milk to be premium when 600,000 cells/mL. Levels of acceptable BMSCC are likely to be continually assessed in future due to international marketing bodies continually decreasing their allowable BMSCC range. The European Union (EU) for example currently regards milk or milk products >400,000 cells/mL to be unfit for human consumption (Dairy Australia. Milk Quality, 2009) which places pressure on Australia as an exporting nation to continually decrease the acceptable BMSCC. The management of clinical and subclinical mastitis largely focuses on decreasing the presence and spread of contagious pathogens on farm and is commonly monitored by the use of herd recording systems which allow the individual cow SCC (ISCC) to be recorded regularly (Shum et al., 2009). As there are little published data from within Australia regarding the current prevalence of SCM and related risk factors, this study aims to increase the understanding and awareness of the current New South Wales situation by first, determining the current prevalence of SCM within NSW, and secondly, outlining associated risk factors. Materials and Methods Data Collection Questionnaires were distributed to 382 farms across NSW that participated in dairy herd improvement production monitoring with a return address envelope to encourage

65


participation. Each farm was marked with an 8 digit code which was used instead of farmer/property names throughout the study. The questionnaire included 26 questions relating to management practices and current protocols for high ISCCs such as milking procedures, dry cow therapy, heifer calving management, nutrition and mastitis management. All questions asked are reported in the Table 1-3. The survey is available on request by emailing the corresponding author of this study. Dichotomous questions were used in this study for their simplicity to answer (Scholl et al., 1994) and the responses were coded for statistical analysis. Each returned questionnaire was consigned to a farm number using the farm address. Farmers who did not wish to be identified were able to do so by removing their address from the returned questionnaire. These farmers were then removed from the risk factor study. In addition, ISCC data from lactating cows from January 2006 to June 2009 were provided by the herd recording company “Dairy Express” (Armidale, NSW, Australia). Data Handling and Statistical Analyses Individual farm data sets were presented in Excel (Microsoft 2007) format and calculations for the percentage of cows with an ISCC >200,000 cells/mL were completed on an individual farm basis for each month that herd recording was conducted on farm. Subclinical mastitis was defined as ISCC >200,000 cells/ml which is regarded as the cut off value for cows with an intramammary infection (Brolund, 1985; Schepers et al., 1997). At the farm level, a high prevalence of SCM was defined as having on average >30% of cows with SCM during the last 42 months of herd recordings, while low prevalence was defined as on average 200,000 cells/mL

40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0

May-09

Mar-09

Jan-09

Nov-08

Sep-08

Jul-08

May-08

Mar-08

Jan-08

Nov-07

Sep-07

Jul-07

May-07

Mar-07

Jan-07

Nov-06

Sep-06

Jul-06

May-06

Mar-06

Jan-06

0.0

Figure 1. Prevalence of subclinical mastitis for all farms and farms with a low and high proportion of cows with subclinical mastitis

In total, 52 and 62 farms fitted in the low (on average 30% of cows) SCM prevalence categories, respectively (Figure 1). The mean herd size for farms grouped into the low and high SCM categories was 172 and 154 cows, respectively. The prevalence of SCM in the farms in the low SCM category

67


was on average 16.1% (95% C.I. 15.1 to 17.0%), while for the farms in the high category the average prevalence of SCM was 36.5% (95% C.I. 36.2 to 39.8%).

Average proportion of cows with and individual SCC >200,000 cells/mL

45

40

35

30

25

20

15

10

5

0 1

11

21

31

41

51

61

71

81

91

101

111

121

131

141

151

161

171

181

Farm number

Figure 2. Average proportion of cows with an individual somatic cell count > 200,000 cells/mL over the 42 month period, calculated for all 189 farms included in the study

Management Related Risk Factors The association of the management categorical explanatory variables at P 200,000 cells/mL c P 30% of cows with ISCC > 200,000 cells/mL c P 200,000 cells/mL c P