Milk Production and Somatic Cell Count in Michigan Dairy Herds

Milk Production and Somatic Cell Count in Michigan Dairy Herds PAUL C. BARTLETT Department of Large Animal Clinical Sciences Michigan Stale University East lansing 48824 GAY Y. MILLER Ohio Stale University Columbus 43210 CLYDE R. ANDERSON Michigan Dairy Herd Improvement Association Box 23153

lansing 48909 JOHN H. KIRK Auburn University Auburn, AL 36849

ABSTRACT

The established association between milk production and sec in dairy cattle is increasingly used to estimate lost production due to mastitis. Such cost estimates are used to make decisions regarding cost effective mastitis prevention and control. It is therefore important to verify the relationship between sec and milk production using data from different areas of the country and by using different analytical methodology. Our study used the 1985 to 1986 Michigan DIDA data base and analyzed daily milk production records rather than lactation summary records as used in the past. One advantage to our approach was that it did not give equal weight to all lactations, regardless of their duration. Also, it enabled inclusion of cows that had incomplete lactations caused by culling, or had other reasons for removal from the herd. A statistical model was constructed to predict milk production OIl the basis of herd, cow within herd, stage in lactation, month of calving, lactation, and sec. The data base contained 397,172 milk

Received February I, 1989. Accepted May 14, 1990. 1990 ] Dairy Sci 73:2794-2800

test records obtained from Michigan DIDA from 504 Holstein herds in Michigan's lower peninsula. Our fmal model predicted 78% of the variation in milk production. Prediction of milk loss for each herd was highly correlated (r = .98) with the prediction model adopted by most DIDA organizations. Our model predicted that the mean herd lost a mean of 1.17 kg of milkIcow per d associated with sec. (Key words: mastitis, model, economics) INTRODUCTION

Michigan DHIA has adopted use of the sec linear score because of its nonnal distribution and relationship with lost milk production. The most widely recognized and accepted relationship between sec and lost milk production is based upon research at the University of Wisconsin (2, 8). Because important management decisions regarding cost effective prevention and control of mastitis are based on this relationship, it is prudent to confirm this relationship with different analytical techniques for different geographical areas and management systems. Several different approaches have been used to measure the causal relationship between sec and milk production. One approach uses the opposite quarter of the same cow for com2794

MILK PRODUcnON AND SOMATIC CELL COUNT

parison with an infected quarter (4, 7, 13). This approach has been criticized because of the possibility of compensatory increased milk production in healthy quarters in response to decreased milk. production in a locally infected quarter. Also, such studies are frequently unable to include cattle with more than one infected quarter. In addition, large dairy data bases, such as those found at the DHIA processing centers, do not have data identifying a quarter that is infected or quantities of milk produced for each quarter. Another approach has been to model production on a cow basis, with milk. production (dependent variable) as a function of see and several independent variables (5, 8, 11, 12). This is the approach used in this study. The objective of this study was to determine the relationship between milk production and see in Michigan and to compare our fmdings to those of others who have studied this relationship in other parts of the US.

2795

with subsequent categories increasing by 30 d increments until the 10th category, which included any milk. test obtained at >270 d postcalving. Lactation number was transformed to a dichotomous variable, with "I" for all first lactations and "2" for all subsequent lactations. The sec (in units of 100,000 cells/ml) obtained from DHIA was transformed by adding 1 to see then taking the natural logarithm of this sum. The estimated mean score (1.5) was then subtracted so that the new variable (LNSeC) would be approximately balanced around a mean of zero to reduce the amount of polynomial correlation between the linear, quadratic and cubic forms of the same variable (1). Actual daily milk. was used as the dependent variable. The model (Model A) was as follows:

=
+ Hi + Cjj + Sit + MI + Lm + bI(LNSeC) + ~(LNSeC)2 + ~(LNSCC)3 + &jIdmn

MATERIALS AND METHODS

where: Michigan DHIA tested 46% (2200 out of 4835) of Michigan dairy herds in 1986. A search of these computer records identified 504 herds in Michigan's lower peninsula that had used the individual cow see option for at least 8 of the 12 mo between March 1985 and March 1986. The monthly records from these herds were obtained on a tape. Somatic cell counts were detennined by the Michigan DHIA laboratory using a Fossomatic (Fossomatic Model 15600, AISN N. Foss Electric, llierod, Denmark) and were reported in integer units to the nearest 100,000 cells/mI. A data set of 480,043 records was edited to a data set of 397,172 records that were used for the analysis. The data set was edited to remove all dry cow records (69,841), all lactating cow records missing see (11,884), and all erroneous records (1146). Erroneous records included 1142 records with a test month greater than 12, or less than 1, and 4 records showing greater than 70 kg of milk produced by a single cow on the day of test. The data set was then analyzed with the general linear models procedure of the SAS statistical program (10). The number of days postcalving was transformed to a new variable having 10 different categories for days postcalving. The first category included 0 to 29 d,

Yijklmn
Hi Cjj Sit MI

Lm LNSee bi ~

b.3 ~jkImn

= =

test day milk. yield, = intercept, effect of herd i (504 classes), = effect of cow j within herd i (53,034 classes), effect of stage k of lactation (10 classes), = effect of month I of calving (12 classes), = effect of lactation m (2 classes), [natural logarithm of (See + 1)] - 1.5, regression coefficient for milk yield on LNSee, regression coefficient for milk yield on LNSee squared, = regression coefficient for LNSee cubed, and residual.

=

= = =

=

Various modifications to this basic model were tested, including use of age as a covariate, replacing the categorical factor for stage in lactation with a continuous variable for 1
No. la, 1990

2796

BARTI..ETT ET AL.

TABLE 1. Summary of Michigan DHIA data used for the analysis. All ~ Variable l

Mean

SCC

4.27 -.31

LNSCC Milk

24.34

Test month Age Calving month DIM Stage

6.55 47.58 6.58 171.32 5.89

Heif~

Older cows4

SD

Mean

SD

Mean

SD

8.33 .85 8.63 3.43 22.53 3.40 108.63 3.01

3.01 -.50 21.91 6.58 27.96 6.51 174.61 5.95

6.24 .75 6.13 3.44 5.59 3.31 111.65 3.03

4.97 -.21 25.69 6.54 58.45 6.62 169.50 5.85

9.2 .9 9.5 3.4 21.0 3.4 106.9 3.0

lWhere: SCC = Somatic cell count/loo,OOO, Milk = aetuaI daily milk per cow (in kg), Test month = 1 to 12 (I = January), Age =age in months at lactation start, Calving month =month of calving, 1 to 12 (1 =January), DIM =days in milk at test, and Stage = DIM 3O-d intervals (stage 1 = 1 to 29 DIM). 2397,172 records. 3141,481 records. 4zS5,691 records.

identification and animal number were absorbed during the analysis, so that analysis of residuals was not practical. For each of the 397,172 milk test records, the predicted (Model A) daily milk production with a see of 0 was subtracted from the value of daily milk predicted with the use of the actual see score. In this way, a predicted difference (DIFF) was obtained that reflected kilograms of daily milk lost associated with Sec. Predicted loss in daily milk was also calculated using Michigan's DHIA standard formula based on the work of Raubertas and Shook (8). A herd average value for each of these two estimates of milk loss associated with see was compared using correlation analysis. RESULTS

A summary of the dependent and independent variables is shown in Table 1. Heifers and cows differed considerably in their mean sec and milk production. The analysis of covariance models are summarized in Table 2. The estimated regression coefficients are given in Table 3. All explanatory variables contributed to the model. The coefficient of determination (R2) was .78 for both models, indicating that 78% of the variation in milk production was explained by the model. The coefficients associated with stage in lactation (fable 3) follow a relatively typical lactation curve, in that they start relatively high, rise to a peak in the 2nd mo, and then gradually decline. Although the summary of our data Joumal of Dairy Science Vol. 73, No. 10, 1990

shown in Table 1 clearly indicates that cows produced more milk than heifers, the regression coefficients from our statistical model would appear to indicate that first lactation heifers have almost identical milk production compared with older cows (.1726 VS. 0 in Model A), as seen in Table 3. When months of age (linear and quadratic) were added to Model A shown in Tables 2 and 3, both contributed to the model (P = .0001); however, the overall increase in R2 from adding these two variables was very small (from .783 to .785). It was therefore concluded that age did not contribute to the usefulness of the model; therefore, it was eliminated as an explanatory variable. Another modification to Model A was attempted whereby the categorical variable for stage in lactations (10 classes) was replaced by the continuous variable (linear and quadratic terms) for number of days postcalving. This modification caused the R2 to drop slightly from .783 in the original model to .778. The last modification (Model B) was inclusion of the interaction between LNSee and the lactation factor. This resulted in an increase in R2 of only .0003. Although it is not the primary purpose of this study, the effect of lactation on milk loss associated with see can best be evaluated by Model B. Given our mean LNSee of -.31, Model B shows that a heifer would lose .92 kg DIFF associated with sec, compared with a cow that would lose 1.52 kg. Thus, a cow with our project average LNSee of -.31 lost 1.65

MILK. PRODUCTION AND SOMATIC CELL COUNT

(1.52/.92) times as much milk. associated with see as did a heifer with this same LNSee. The herd average values of DIFF were compared with the herd average values predicted by Raubertas and Shook (8) and were found to be highly correlated (r = .98, P = .(01). Our study showed that the mean herd lost an estimated average of 1.17 kg of milk/cow per d associated with see scores greater than O. An almost identical 1.20-kg milk. loss would have been predicted by Raubertas and Shook (8). The average animal tested produced 24.34 kg of milk/d. Based upon our predictive model A, the average cow would have produced 1.17 more kg of milk if the see for all tests were O. We can therefore estimate that elevated see was associated with approximately 5% (l.17/ (1.17 + 24.34» milk. loss. DISCUSSION

Because the mean difference between our estimate of herd production loss and that of Raubertas and Shook (8) was only .03 (1.20 1.17) kg of milk/cow per d, our study confinns the current use of DmA estimates of lost milk. production associated with see. Raubertas and Shook (8) used a model with total lactation yield and lactation average log see, whereas our model had one record per milk test. Therefore, in Raubertas and Shook's model, equal weight was given to short lactations as was given to long lactations. Raubertas and Shook also excluded lactations in which the cow was culled before being dried off. Michigan DmA records indicate that approximately 34% (59,411/172,086) of Michigan cows on DmA testing were culled sometime during 1986. Of all heifer lactations culled, at least 5.8% (987/17,081) were culled explicitly for mastitis. Of all cow lactations culled, 9.8% (4160/42,360) were culled explicitly for mastitis. Because of the difference in mastitis culling policy between heifers and cows, exclusion of culled heifers and cows by Raubertas and Shook could represent a bias. Our study design was similar to that employed by Jones et al. (5) who studied 34 dairy herds in Virginia. It is difficult to compare our results to those of Jones et al. (5) because the coefficients for stage in lactation, month of calving, and lactation were not shown. However, some comparison can be made if their

2797

model is applied to our data and mean values are assumed for stage in lactation, see, and month of calving. Under such circumstances, their model would predict production of 26.2 kg and ours would predict 33.3 kg. This difference could relate to their sampling procedure within three comparatively low production groups, and to other differences been the dairy herds and environments in Virginia and Michigan. Our methodology was designed to estimate herd milk loss associated with see and not to evaluate the independent effect of parity (lactation). Our model mandates that the parity effect be based upon "crossover" cattle that were represented in the data as a heifer and then as a cow during the 12-mo period of study. The evaluation of parity is thus a "within-cow" effect in that it compares milk production of these animals when they were first-lactation heifers to what it was when they became cows. In such cases, one would expect the effect of lactation to be slight, as this variable would only represent a difference of several months of age. Also, this subset of "crossover animals" is somewhat unique in that it contains a preponderance of high producing animals (compared to herdmates) or else they would not have been allowed to remain in the herd and begin their second lactation. Additionally, for the "crossover animals", we would expect some confounding between stage in lactation and parity, because "crossover animals" would contribute data only for the last part of their heifer lactation and only for the first part of their second lactation. There might also be some confounding between parity and month of calving, because the crossover lactations would necessarily tend to end their heifer lactations soon after March 1985 and start their second lactation sometime shortly before March 1986. Our data showed that cows had a higher mean value of see than heifers. There is some evidence that age per se is not associated with an increased see (3, 6, 9). Rather, the increased see in older cattle are primarily due to increased prevalence of infection and permanent glandular damage from previous infections. The Raubertas and Shook model (8) showed that, given the same see, cows lost twice as much milk associated with elevated see as did heifers. Our results (Model B) indicated a difJournal of Dairy Science Vol. 73,

No. 10, 1990

2798

BARn.ETI BT AL.

TABLE 2. Summary of analysis of covariance-determinants of individual daily milk weight. l Source of variation Herd ID Cow within herd Stage in lactation Month of calving Lactation LNSCC LNSCC2

LNSCC3 (LNSCC x Lactation) BI1'or

Type


503 53,033 9

8,951,632 21,784 327 39,778 129 2473 (8555) 6,416,596

II I 1

1 1 (I)

343,611 (343,610)

Type m3 probability

m SS2 (8,908,095) (21,914) (1845) (29,292) (685) (2712)

.0001 .0001 .0001 .0001 .0086 .0001 (.0001)

(6,408,040)

lDependent variable = individual cow daily milk weights (in kg). 2SS = Sums of squares. (Figures in parentheses refer to model B). 3The probability for all Type I F values was .0001. All variables in model B were significant at P = .0001.

ference of only 1.65 times as much milk lost by cows. The previous discussion regarding the effect of parity would also be relevant to understanding the effect of parity on DIFF. Jones et al. (5) agreed with Raubertas and Shook (8) in finding that, given equal sec scores, heifers lost approximately half as much milk as did older cows. One possible explanation for this difference is revealed in Figure 2 of the paper by Jones et al. (5); that figure shows that the difference in milk. loss between heifers and cows was greatest in low producing herds «6500 kg/yr) and only slightly different in medium and high producing herds. Eightyfour percent of the herds we studied were in the medium or high producing group as defined by Jones et al. (5), where the difference in milk loss between heifers and cows was smallest. It also appears that Jones et al. (5) constructed a separate statistical model for heifers and for cows. This might also account for some of the observed differences between our results. The inclusion or exclusion of culled cows is not explicitly stated by Jones et al. (5). Inclusion of the quadratic and cubic terms for LNSee did not contribute substantially to the predictive ability of our model (fables 2 and 3); and, therefore, might not be necessary as Raubertas and Shook (8) suggest. Although the quadratic and cubic forms of LNSee contributed significantly to the model (fable 2), the overall contribution of the quadratic and cubic terms to the sum of squares is vel)' small compared to the linear term for LNSee. The graphic relationship between LNSee and milk production is shown in Figure 1. Journal of Dairy Science Vol. 73,

No. 10, 1990

The test date was intentionally omitted from our model. In actuality. month of calving, stage in lactation, and month of test date are redundant, because given a relatively standard gestation length, any two of these variables determines the third. In formulating a statistical model, the month of calving and the stage in lactation were included and month of the milk test excluded so that our results could more easily be compared with previously published work. Because the available sec data were in units of 100,000 cells/mI, the LNSeC data were somewhat discrete in nature, especially when the value of sec was small. We believe that this limitation should not have represented a bias, but may have decreased the overall predictive ability of the model. Small differences of inconsequential biological significance frequently had statistical signif-

26.0

.~

25.~

t

no

'f

24.5

U

1!

'l;

.,.

24D

:0::

23.5

-\.5

"1.0

• .5

0

In (See t 1.0) -1.5

Figure I. Relationship between SCC and milk production, adjusted for a grand mean of 24.32 kg. Includes linear. quadratic, and cubic terms of LNSCC.

2799

MILK. PRODUCTION AND SOMATIC CEll COUNT

TABLE 3. Determinants of individual cow daily milk weight. Estimate

Parameter

Model A

Model B

15.549 17.673 16.228 14.178 12.192 10.284 8.370

15.530 17.649 16.210 14.164 12.180 10.275 8.365 6.345

Standard error of estimate

Model A probability) (Model B)

Model A

Model B

Stage )

2 3 4 5 6 7

8

6.348 4.006

9 10 Month of calving

o

Ian

.190 -.701 -2.754 -3.664 -1580 -.607 -.829 -1.417 -1.522 -1.129 -.412

Feb Mar

Apr May 100

lui Aug Sep Oct Nov

o

Dec

Lactation First Older LNSCC LNSCC LNSCC LNSCC x Lactation First Older

4.004

o

0.190 ~.688

-2.760 -3.682 -1.609 ~.620 ~.841

-1.425 -1.532 -1.133 -.418

.0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001

.035 .032 .031 .031 .031 .031 .031 .031 .031

.035 .032 .031 .031 .031 .031 .031 .031 .031

.0185 (.0181) .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001

.081 .109 .145 .162 .133

.081 .109 .145 .162 .133

o

o

0.427

.0001

o

-.827 .023 -.066

-.982 .055 -.690

.0001 .0086 (.0001) .0001

.173

.500

.106

.106

.093 .088 .083 .078 .071

.093 .088 .083 .078 .071

o

.041

o

.043

o

.018 .009

.019 .009 .006

.006

.0001

.023

0

0

IType m probability for T test for HO; parameter = O. Probabilities are shown in parentheses only if they differ from Model A probabilities.

icance since we had a very large sample size for this study. The t value for 120 df is 3.291 for P = .0005. Given an n of 300,000, any difference between two means in excess of .025 kg would be found to be significant at P<.OOO5. Clearly, .025 kg of milk is of little biological significance to most dairy farmers.

account for any possible bias due to culling or length of lactation. In conclusion, on a herd basis, our results regarding the association between milk production and sec agree very closely with those reported by Raubertas and Shook (8), which are accepted by many DInA processing centers.

CONCLUSIONS

ACKNOWLEDGMENTS

The widespread use of the hypothesized relationship between sce and lost milk production warrants analysis in numerous geographic areas by a variety of experimental designs. Analysis of our Michigan data by daily test results, rather than by lactation, enabled us to

The authors thank RoseAnn Miller and James Stoltz for their technical expertise as computer systems analysts and George Shook for his review and consultation. This project was funded in part by grant 83-CRSR-2-2203 from the USDA. loumal of Dairy Science Vol. 73.

No. 10, 1990

2800

BAR1LEIT ET AL.

REFERENCES 1 Bradley, R. A. and S. S. Srivastava. 1979. Correlation in polynomial regression. Am. Stat. 33: 1l. 2 Crist, W. L. 1986. The new DHIA somatic cell count score: What it is and bow 10 use it. Nat!. Mastitis Counc., Inc., Arlington, VA. 3 Dutsehaever, C. I., and G. C. Ashton. 1m. Variations of somatic cells and neutrophils in milk throughout lactation. J. Milk Food Techno!. 35:197. 4 Janzen, J. J. 1970. Economic losses resulting from mastitis. A review. J. Dairy Sci. 53:115l. 5 JODeS, G. M., R. E. Pearsoo, G. A. Clabaugh, and C. W. Heald. 1984. Relationships between somatic cell counts and milk production. J. Dairy Sci. 67:1823. 6 MarsbaIl, R. I., and J. E. Edmondson. 1962. Value of California mastitis test records 10 the practitioner. J. Am. Vet. Med. Assoc. 140:45. 7 :Meijering, A., H. J. Jaartsveld, M.W.A. Verstegen, and M.1.M Tielen. 1978. The cell count of milk in relation

Journal of Dairy Science Vol. 73,

No. 10, 1990

10 milk yield. J. Dairy Res. 45:5. 8 Raubertas, R. p~ and G. E. Shook. 1982. Relationship between lactation measures of somatic cell concentration and milk yield. J. Dairy Sci. 65:419. 9 Reichmuth, J. 1975. Somatic cell counting - Interpretation of results. Page 93 ill Proc. Seminar Mastitis Control 1975, Int. Dairy Fed. Doc. 1975. 10 SAS~ Usee's Guide: Statistics Version 5 Edition. 1985. SAS Inst., Inc., Cary, NC. 11 SchIl1tz, L. H. 1977. Somatic cells in milk, physiological aspects and relationship lD amount and composition of milk. J. Food Prot 40:125. 12 Schultz, L. 8., R. W. Brown, D. E. Jasper, J. W. Mellenberger, and P. D. Thompson. 1978. Current concepts of bovine mastitis. Nati. Mastitis Counc., Inc., ArlinglOn, VA. 13 Ward, G. E., and L. H. Schultz. 1972. Relationship of somatic cells in quarter milk to type of bacteria and production. J. Dairy Sci. 55:1428.