| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
* Department of Animal Husbandry and Animal Health and
Department of Clinical Studies and
Department of Veterinary Microbiology, The Royal Veterinary and Agricultural University, 1870 Frederiksberg C, Denmark
Danish Institute for Agricultural Science, P.O. Box 50, 8830 Tjele, Denmark
Corresponding author: T. W. Bennedsgaard; e-mail: TorbenW.Bennedsgaard{at}agrci.dk.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Key Words: organic farming • herd health • milk production • effect of disease
Abbreviation key: ECM = energy-corrected milk
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Østergaard and Gröhn (1999) estimated the loss of production in energy-corrected milk (ECM) using a disease index for the time before or after treatment or the occurrence of other diseases on weekly production data from three research farms. Several diseases were separately evaluated using models of the lactation curve with linear, quadratic, and cubic continuous covariates for lactation stage in a repeated measures analysis with first-order autoregressive covariance structure. Breed, parity, and calving season were included in the models. The model contained random effects for each treatment group within trials. For mastitis they found a higher milk yield for multiparous cows 2 to 4 wk before treatment and for all parities a negative effect on milk yield from 1 wk before disease and the rest of the lactation. No correction was made for SCC or production in previous lactation. Rajala-Schultz et al. (1999) estimated the loss of milk yield after treated cases of mastitis using a mixed model with monthly test milk yield as the outcome variable and a first-order autoregressive correlation structure. Their analysis was based on data from 24,276 cows that were followed for a single lactation. Milk yield was evaluated in relation to two different baselines: production of healthy cows and production before clinical diagnosis and treatment. When compared with the cows own production, the analysis was related to three stages of lactation: before peak yield of the individual cow, from peak to 120 d postpartum and from 120 d postpartum to calving. In these models, a mastitis index was used to categorize test days in relation to the time period before or after mastitis treatment. Season, stage of lactation, and disease variables were included as fixed effects. Random effects at herd level were only included in the model for a subset of data because of the need for computer power. Cows treated for mastitis were found to have higher yields both before and after treatment compared with cows not treated for mastitis. When compared to the cows’ own production, a negative effect on production could be seen up to 4 wk before treatment.
The herds in the present study were all certified organic herds for the whole study period. In some studies (Vaarst et al., 1993; Hardeng and Edge, 2001), organic herds have been shown to have fewer veterinary treatments for mastitis. This might be a result of better udder health, but it could also be a result of the organic farmers being more reluctant to use antibiotics, which might eventually lead to chronic infections or treatment at a later time in relation to the initial infection. The regulations for Danish organic production allow only veterinarians to treat cattle older than 6 mo. The absence of treatments by the farmers is expected to result in more reliable registration of mastitis treatments because the treatment will be registered both in the herd and by automatic electronic reporting to the Danish Cattle database.
The aim of this study was to estimate the combined effect of the SCC and treatment for mastitis on milk production in a model allowing for correction for other factors affecting production, including information on the previous lactation, and to explore possible interactions between the parameters. New software and the general availability of more powerful computers allow for more complex models that account for herd- and cow-specific factors and analysis of the variance components by using multilevel modeling with random effects on herd and lactation level. To take into account the eventual change in the shape of the lactation curve due to health status and herd-specific factors, Singer (1998) chose a random coefficients linear model. This approach gives the possibility of making inferences on the effect on herd and cow level.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Records from cows calving between January 1, 1997, and March 1, 2001, were included in the analysis. Test days before 9 d postpartum were excluded because of the natural high SCC at the start of lactation. Records after d 305 were excluded because of few observations and possible effects of preferential treatment due to drying off or culling (Enevoldsen et al., 2000).
Test-day records (n = 128,153) were available. Records with missing information (n = 1858) on age at first calving, gestation length, days open antepartum, or milk production in previous lactation were excluded.
The last test day was excluded for cows culled between d 210 and 320 postpartum because these observations tended to show a sudden drop in milk production. The resulting dataset consisted of 118,427 records from 17,488 lactations of 9317 cows. Seventy-four percent of the cows were Danish Holstein-Fresian, 16% Jersey, and the last 10% primarily cross-breed cows.
Statistical Analysis
Because of the differences in the shape of the lactation curve and the inclusion of information on the previous lactation, the data were analyzed separately for parities 1, 2, and 3 or higher. Continuous variables were centered about the grand mean for each lactation to make interpretation of the results easier and decrease numerical problems associated with inclusion of polynomials (Kreft et al., 1995).
Only a limited number of observations are available from each lactation with monthly registrations. To use this information, including observations from lactations with few or missing test days, a very simple multilevel three-parameter piecewise linear random coefficients linear model was used to analyze the data. The MLwiN 1.1 (Institute of Education, London) was used for the analyses (Rasbash et al., 2001). The three levels in the model were herd, lactation, and test day. The monthly milk yields as ECM were used as the outcome variable (Sjaunja et al., 1990).
The shape of the lactation curve was modeled as a two-piece linear function of DIM. For DIM 
60 two variables, DIMun60 and DIM60 were calculated as DIMun60 = (DIM-60)/60 and DIM60 = 0; for 60 DIM 305, the two variables were calculated as DIM60 = (DIM-60)/245 and DIMun60 = 0. The model gives a lactation curve that "peaks" at d 60 postpartum. The baseline model is
 |
where
ECMijk represents ECM in kg/d on test day i of the jth lactation in the kth herd;
ß0 represents the expected milk yield at d 60 postpartum (DIM60 = 0 and DIMun60 = 0);
0k represents departure of the kth herd from the overall mean;
u0jk represents the departure of the jth lactation from the overall mean in the kth herd;
ß1DIMun60ijk and ß2DIM60ijk represent the overall mean slope before and after d 60 postpartum;
1kDIMun60ijk and 2kDIM60ijk represent the departure of the kth herd from the overall mean slope before and after d 60 postpartum;
µ1jkDIMun60ijk and µ2jkDIM60ijk represent the departure of the jth lactation in the kth herd from the overall mean slope before and after d 60 postpartum;
0ijk represents the residual variance; and the random variables 0k, µ0jk, 1k DIMun60ijk, µ1jkDIMun60ijk, 2k DIM60ijk, µ2jkDIM60ijk, and 0ijk are assumed normally distributed with zero mean. The variances for the random parameters were estimated. Random parameters within the same level were analyzed as correlated and covariances were estimated.
This model provides estimates of completely individual lactation curves for each lactation. Dependency between lactations from the same cow is accounted for by inclusion of variables on production in the previous lactation. The reason for rescaling the stages of lactation (DIMun60 and DIM60) was to allow for a more straightforward interpretation of the random effects as suggested by Singer (1998).
Effects of fixed and random variables on lactation and test-day level were tested in the model. Variables included information on days open, length of gestation and dry period, milk production, SCC and veterinary treatments in the previous lactation and in the dry period, as well as information on calving complications, SCC, and veterinary treatments in actual lactation. Polynomials and interactions between all variables were tested. Development over time and seasonal changes were modeled by a centered continuous variable for the time and linear, quadratic, and cubic terms for the calving month. The log-likelihood given by the iterative generalized least squares estimation of MLwiN 1.1 was used to include or exclude the variables by performing 
2-tests with a 5% alpha significance level. Excluded variables were reintroduced and tested in the final model (Goldstein, 1995).
A basic unstructured covariance structure was used.
The SCC were divided by 1000 and converted to the natural logarithm (LNSCC) to account for the skewness to the right of SCC distribution (Ali and Shook, 1980). SCC below 100,000 cells/ml was used as reference and set to zero because only a weak nonsignificant relation between SCC and production as kg ECM was seen throughout the lactation for SCC between SCC 0 and 100,000. The effect of the SCC was modeled by two variables. The first variable had the value ln(SCC) - ln(100) for SCC from 100,000 to 1,500,000 cells/ml and the value ln(1500) - ln(100) for SCC above 1,500,000 cells/ml. The other variable had the value ln(SCC) - ln(1500) for SCC above 1,500,000 cells/ml and was zero for SCC below 1,500,000 cells/ml. Interactions with DIM60 and DIMUN60 were tested, but only the interactions with DIM60 were retained in the models. These variables all showed a linear relationship to ECM.
Hortet et al. (1999) showed that an interaction between LNSCC and DIM and DIM*DIM gave a better fit when modeling SCC and milk yield. In their analysis they excluded test days <15 because they showed a stronger effect of SCC on milk yield only for second parity. In our dataset, the same effect was seen for DIM <16 in second and later lactations. Because an exploration of the data did not show any signs of this effect being a result of outliers or problems with the basic model the relationship was modeled by adding an extra variable for SCC for DIM<16 together with a dummy variable for DIM<16. These variables gave a good fit to the data. The deviation from the general trend might be a result of cows with a prolonged production of colostrum having lower production.
Mastitis treatment was modeled by a series of variables for treatment 0 to 14 d before calving, from calving to d 60 and from d 60 to 305. The variables take the value of 1 from the first test day after treatment. Separate variables for the second treatment in the lactation were also introduced. For the few cases of a third or fourth mastitis treatment two variables taking the value of 1 on the first test day after treatment only were introduced.
Interaction terms with the number of days after treatment and DIM were tested, and one of them was included for most variables related to treatment. Random effects were included in the models for SCC and mastitis treatment. Interaction terms between the variables for SCC and the variables for mastitis treatment were tested and found nonsignificant. To test this result, the model was run after exclusion of all cows treated for mastitis. The estimates for the effect of SCC on milk production were similar to the estimates with the full dataset.
The distribution of the residuals was tested for normality by graphical exploration in MLwiN.
The predictions from the model were analyzed for preferential culling of cows in relation to SCC and mastitis treatment by linear regression with predicted loss in kg/ECM per day as outcome and DIM at culling as explanatory variable. Cows not culled at d 305 after treatment were excluded from the analysis.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
. The herd size varied from 20 to 198 cows, and the mean herd size was 85 cows in 1997 and 98 cows in 2001. The median DIM for first mastitis treatment was 32, 38, and 34 in first, second, and later lactations, respectively. The variables of the final models are given in Table 2.
|
|
. In all parities, the negative effect of SCC below 1,500,000 cells/ml was lowest around d 60 and highest at the end of lactation, resulting in a steeper decline of the lactation curve (poor persistency). For higher cell counts, the effect varied less with stage of lactation. Figure 1 shows the relation between SCC, DIM, and loss of milk production in parity 1. Cows treated once for mastitis had lower production in all herd groups except for mastitis treatments of heifers in the first 14 d of lactation. For the heifers a small nonsignificant positive effect with a very large standard error was found for mastitis treatment in the first 14 d of lactation, and the variable was excluded from the model. In parities 2 and 
3, the negative effect of mastitis treatment was significantly reduced with time after treatment for treatments before d 15 for parity 2 and for treatments before d 60 for parities 3. Predictions for loss of milk production for first mastitis treatment at DIM -7, 15, 50, and 150 are shown in Table 3. Estimates of the effect of a second mastitis treatment during the lactation was only found significant in second and higher lactation and was smaller than the effect of the first treatment (-0.9 kg of ECM/d for treatments at d 16 to 60; -0.2 kg of ECM/d for treatments after d 60 in parity 2 and -0.8 kg of ECM/d for treatments after d 60 in parity 3). The inclusion of a variable for third and fourth treatments showed a very large standard error and resulted, in general, in nonsignificant positive estimates. Udder treatments with antibiotics at drying off were given to 15 and 20% of the cows that were not culled in first and later lactations. Udder treatment at drying off did not result in a significant change in milk production in the following lactation.
|
|
Combined Effect of SCC and Mastitis Treatment
Treatment for mastitis was, in general, followed by a change in SCC. The analysis showed that cows with low SCC at the last test day before mastitis treatment in general had higher SCC after treatment, whereas cows with high cell counts before treatment (>700,000 cells/ml) on average showed a decline in SCC after treatment. To obtain the final estimate for changes in milk production the treatment effect should be combined with the effect of the change in SCC. The resulting lactation curves for cows in parities 1 and 3 with three different levels of SCC before treatment at DIM = 50 and 150 are shown in Figures 2 and 3. The mean of the two previous test days was used to calculate the SCC before treatment.
|
|
Effect of Herd
No significant differences between groups of herds with low or high calculated bulk tank SCC, low or high prevalence of mastitis treatment, low or high milk production per cow, or combinations of these groupings could be found. The variance of the effect of mastitis treatment and SCC between animals was much larger than between herds, and no random effects related to SCC or mastitis treatment were significant on herd level when random effects on lactation level were included in the model. The inclusion of variables for SCC and mastitis treatment without random effects explained 4, 5, and 2% of the variation at DIM = 60 at test day, lactation, and herd level, respectively.
Culling and Production Loss
Cows treated for mastitis before DIM = 60 were culled earlier than other cows. A total of 17% of the cows in parity 3 or higher were culled before DIM = 200. Twenty-six and forty percent of the cows were culled before d 200 postpartum for the cows treated for mastitis before DIM = 15 with a SCC between 750,000 and 2,000,000 cells/ml or above 2,000,000 cells/ml before treatment (significant difference, P < 0.05). Analysis of the estimated loss of production for these cows including both fixed and random effects on lactation level for SCC and mastitis treatment showed that the cows culled early in the lactation have larger estimated loss of production (kg of ECM/d) due to mastitis and SCC (Figure 4).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Evaluation of Model
When estimating the effect of mastitis on milk production only few, if any, test days from the recent lactation before treatment will be available as baseline for the estimation. In this study, this problem was handled by including information on production from the previous lactation as an estimate of cow production capacity. However, in first lactation, no information about previous production was available. The positive nonsignificant estimate obtained for mastitis treatment before DIM = 15 might be an estimation problem caused by the lack of information about the production capacity of the cow. Inclusion of information on the genetic level of the heifers might change the estimate, but this was not available. The autoregressive models used in several studies (Østergaard and Gröhn, 1999; Rajala-Schultz et al., 1999) allows the use of information from the previous test days for the same cow but might not be able to take account for an eventual prolonged decrease in milk production caused by a chronic infection before treatment. The multilevel model combined with the estimation of both the effect of SCC and mastitis treatment and other diseases allows a separation of the effect of udder infections into the subclinical effect of increased SCC and the effect of udder infections, which the farmers decides to treat. The multilevel models take into account the general characteristics of the lactation curves on the herd level and fits individual lactation curves for each cow. The inclusion of random effects for all variables for SCC and most variables for the first mastitis treatment gives the possibility of studying the variation in the responses to treatment. The effect of early culling of cows treated for mastitis is also controlled by using a multilevel model because the estimates are calculated based on the development in production on herd and cow level. However, no prediction for the SCC exists for the culled cows. Therefore, the development in the combined effect of mastitis treatment and SCC will be biased towards a lower estimate of the losses in late lactation compared with a situation in which no cows were culled. The very small change in production following mastitis treatment for cows with high SCC might be a result of these mastitis cases being of chronic character with little systemic effect on the cow, or it might be cows with no clinical signs treated with antibiotics to control the SCC. The relation between early culling and production loss of these cows further reduces the difference in the long-term losses in the predicted lactation curves.
The grafting point for the diphasic model for the relation between SCC and production was set at 1,500,000. This is high in comparison with other studies. Dentine and Daniel (1993) found a greater slope above 837,000 cells/ml. Hortet et al (1999) used 600,000 cells/ml as exclusion criteria because of few observations above this level, and a higher variability in the response. The grafting point in this study was chosen after exploration of the residuals. The relationship between SCC and production was not fully linear, but the character of the relationship changed with stage of lactation with a lower grafting point in early lactation compared with the late lactation. This is in accordance with Jones et al. (1990). The effect of the SCC late in lactation decreased for SCC above 1,500,000 cells/ml. This is probably an estimation problem due to very few observations with high cell counts late in the lactation, and the estimate should be interpreted carefully.
Effect of SCC
Comparison of the effect of SCC on production between different studies can be difficult because of different outcomes and reference values. Hortet et al. (1999) used kilograms of milk as outcome, SCC below 50,000 cells/ml as reference group, and excluded records with cell counts above 600,000 cells/ml or following mastitis treatment. When running the same analysis on the dataset of this study for cows in parity 1, the estimates obtained are identical to the results of the French study within 0.01 kg of milk at a given SCC. For higher parities, the average estimate over the lactation was also nearly identical though the change in the effect over the lactation was smaller in this dataset leading to lower predicted losses at the end of the lactation. Changing the outcome from ECM to noncorrected milk changes the change in the effect of SCC with lactation stage from being substantial to being nonsignificant (0 kg of ECM/d at DIM = 60 to -0.8 kg of ECM/d at DIM = 305 for one unit change in ln(SCC) vs. 0.43 kg of milk/d at DIM = 60 to 0.40 kg of milk/d at DIM = 305 for one unit change in ln(SCC)). This difference is probably an effect of the SCC affecting the composition of the milk somewhat depending on the stage of lactation. The exclusion of records with SCC above 600,000 cells/ml or following mastitis treatment did not change the estimates of the model in this study significantly indicating a robust estimation of the different parameters in the model. Koldeweij et al. (1999) estimated the effect of mastitis by using the SCC as the explanatory variable. The production loss was estimated to 0.5 and 1.0 kg of ECM/d per unit ln(SCC) for primiparous and multiparous cows compared with 0.3 and 0.6 kg of ECM/d for SCC below 1,500,000 in the present study. The estimates also included the effect of mastitis, and no grafting point was used to allow for a steeper slope at higher SCC. The estimates in the present study are in accordance with the conclusions of the most recent review, which summarized the results of previous papers. The conclusion of this review was a loss of 0.4 kg of milk/d in primiparous cows and 0.6 kg of milk/d in multiparous cows per twofold increase in SCC above 50,000 cells/ml (Hortet and Seegers, 1998a). The fact that the estimates for the SCC variables did not change when excluding all records following mastitis treatment indicates that the effect of SCC and mastitis treatment should be modeled separately, though the combined effect of mastitis treatment and change in SCC should be calculated to get a reliable estimate.
Effect of Mastitis Treatment
In the most recent review of the effect of mastitis on milk yield, the total losses were summarized as 450 to 550 kg for treatments before peak, 200 to 300 kg on average for all the cases in first lactation, and 300 to 400 kg for later lactations with high variability (Hortet and Seegers, 1998b). A later study found rather similar estimates for first lactation, but higher estimates for parity 3+ with losses of 500 to 600 kg for treatment before peak and 300 to 400 kg after peak (Rajala-Schultz et al., 1999). These estimates are comparable to the estimates in this study for older cows, whereas the predicted loss for first-lactation cows were substantially lower in our study. For the farmer, the milk loss will be substantially larger because in Danish organic herds, the withdrawal time after antibiotic treatment is 9 to 12 d after last treatment for a typical mastitis case. This will result in the withdrawal of about 350 kg of milk for a third-lactation cow at peak production.
It is important to beware of the broad definition of mastitis used in this and other studies. Mastitis treatment depends primarily on the farmer’s selection of cows for treatment. This is also the case in this study, in which a veterinarian treated all cows. Cows treated might have had an IMI for a long period before veterinary treatment. The decision to treat the cows might be a result of clinical symptoms, an increase in SCC, a drop in milk production, or general problems in the herd with high bulk tank SCC leading to treatment of cows with high SCC. In contrast, the SCC of all milk-producing cows are measured at every test day. This might explain the lack of a significant interaction term between the effect of SCC and mastitis treatment in the model.
Effect of SCC and Mastitis Treatment in Previous Lactation
Higher SCC in the previous lactation resulted in higher predicted production in the following lactation, all other things being equal. This can be seen as an estimate of the amount of milk that is necessary to make the farmer keep the cow despite an elevated SCC. When the variables for SCC and mastitis treatment in the lactation of interest was removed from the model, the estimate for the SCC of the previous lactation was negative, which has been shown in other studies (Fetrow et al., 1991; Houben et al., 1993). The positive relation between mastitis treatment in the previous lactation and production is larger than the findings in other studies, where no significant carryover effect was seen, except for cows with more than two treatments that had lower production in the following lactation (Lucey and Rowlands, 1984; Houben et al., 1993). The result in our study might be a result of the culling strategy of the farmers, where cows with chronic or severe inflammatory manifestations in the udder after mastitis treatment are culled.
Effect on Herd Level
The variance of the effect of mastitis treatment and SCC between animals was much larger than between herds, and no random effects related to SCC or mastitis treatment were significant on herd level when random effects on lactation level were included in the model. In relation to the SCC, this probably reflects a general effect of the SCC independent of herd effects though different udder pathogens affect the SCC to different extent (Haas et al., 2002). In relation to the effect of mastitis treatment, the absence of a herd effect is more surprising. The farmer’s treatment decisions are complex and include many decisions of both tactical and emotional character leading to preferential treatment, resulting in some cows being treated for mild infections and other cows not being treated despite severe changes in the udder leading to drying of single quarters (Vaarst et al., 2002). Graphical exploration of the predicted losses at cow level showed that the losses varied considerably from year to year in the individual herds, and thus explains lack of consistent differences between farms. This might reflect changes in the character of mastitis in the herds or changing treatment policy caused by, e.g., quota restriction, and emphasizes the need for a continuous evaluation of the treatment policy in the herds.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Received for publication October 25, 2002. Accepted for publication April 7, 2003.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  N. Capion, S. M. Thamsborg, and C. Enevoldsen Conformation of Hind Legs and Lameness in Danish Holstein Heifers J Dairy Sci, May 1, 2008; 91(5): 2089 - 2097. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. De Vliegher, H. W. Barkema, H. Stryhn, G. Opsomer, and A. de Kruif Impact of Early Lactation Somatic Cell Count in Heifers on Milk Yield Over the First Lactation J Dairy Sci, March 1, 2005; 88(3): 938 - 947. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
) and SCC above 750,000 cells/ml before mastitis treatment at DIM 50 (
).
