Discussion Directory UMM :Data Elmu:jurnal:L:Livestock Production Science:Vol65.Issue1-2.Jul2000:

C . Heuer et al. Livestock Production Science 65 2000 91 –105 99 Table 6

4. Discussion

Coefficients of the final model selected for prediction of herd a mean energy balance This study primarily investigates the hypothesis Parameter Coefficient 95 Conf. interval P that EB in early lactation calculated from cow side Constant 217.77 195.20, 240.34 0.0001 data and averaged at the level of the herd can be Week 2 2 31.92 2 35.68, 228.16 0.0001 used to evaluate the performance of a common dairy Week 3 2 20.64 2 23.07, 218.21 0.0001 ration to provide energy for lactation. For the Week 4 2 15.56 2 17.89, 213.23 0.0001 purpose of this study we assumed that the energy Week 5 2 11.53 2 13.67, 29.39 0.0001 content of the common ration of this experiment Week 6 2 7.98 2 9.85, 26.11 0.0001 Week 7 2 10.62 2 12.73, 28.51 0.0001 grass–maize silage plus concentrate calculated Week 8 2 7.21 2 8.69, 25.73 0.0001 according to the current feed evaluation system Van Week 9 2 5.29 2 6.86, 23.72 0.0001 Es, 1978 was an appropriate ‘gold-standard’ for true Week 10 2 3.95 2 5.19, 22.70 0.0001 EB. Even though ration components in 1996 may Week 11 2 2.66 2 3.94, 21.38 0.0001 differ from those used before 1978, energy intake Week 12 0.00 Parity 1 2 34.91 2 40.39, 229.43 0.0001 and energy required for lactation appeared to corre- Parity 2 2 7.20 2 11.65, 22.75 0.0023 late well r .0.66 when the evaluation system was Parity 3 2 6.67 2 11.74, 21.60 0.0120 applied to today’s rations Saama and Mao, 1995. Parity 41 0.00 Therefore, the chosen energy evaluation system was Milk 2 2.11 kg 2 2.34, 21.87 0.0001 likely to be a valid standard to estimate prediction Protein 2 15.36 2 19.76, 210.96 0.0001 FP 2 49.24 2 53.01, 245.47 0.0001 coefficients for the model parameters. The work of a 2 Saama and Mao 1995 also indicated that there may n 5771, R 50.25; milk, milk yield; protein, milk protein; be a constant difference between the calculated fat, milk fat; FP, fat–protein-ratio. energy intake based on the system and calorimetric measurements. This might have affected the intercept of our model, but the parameter coefficients would most likely be unaffected. 4.1. Model development Calculated energy balance is the difference be- tween energy intake and energy requirement. The calculation of energy requirement was based on body weight, parity, milk yield, and concentrations of milk fat and milk protein. Hence, the total energy require- 2 ment correlated well with milk yield r 5 0.90 and 2 fairly with body weight r 5 0.53. Since milk yield and milk solids represented the requirement part in both calculated and predicted EB, a close correlation was expected between calculated and predicted EB, Fig. 1. Predicted energy balance by lactation number parities 1, merely by definition. However, the correlation co- 2, 3 or 4 and higher. The horizontal line indicates the median, the 2 box 25 to 75 of the data, the vertical lines typical values, and efficients between EB and energy requirement r 5 2 stars depict possible outliers. 0.06, EB and milk yield r 5 0.04 or EB and milk 2 solids r 5 0.001 to 0.06 were small. Therefore, the quantity of milk and milk solids could only explain a the herd size was 20 cows the standard deviations small part of the association between milk test day were 6.7–8.2 MJ. Fig. 3 shows the loss of precision information and EB. A larger part must have been in relation to mean predicted EB by week of due to energy intake, the generally unknown parame- lactation. ter in the equation of EB. 100 C . Heuer et al. Livestock Production Science 65 2000 91 –105 Fig. 2. Plot of residuals observed minus predicted energy balance versus predicted values of model coefficients from Table 5; standard deviation indicated as horizontal band above the zero line 771 weekly measurements, 72 cows. Fig. 3. Herd means of post partum energy balance: observed and predicted means of the total of 771 observations 72 cows, and predicted means minus standard deviations of reduced sample or herd sizes. Standard deviations calculated from the means of 200 repeated random selections. C . Heuer et al. Livestock Production Science 65 2000 91 –105 101 Table 7 Precision of predicted herd mean energy balance EB as MJ NEL at decreasing sample n and herd size in comparison to observed values; population standard deviation of 200 predicted herd means obtained by repeated random sampling from the pool of 771 milk tests 72 cows, 11 weekly tests, 2–12 weeks p.p. 2 3 4 5 6 7 8 9 10 11 12 Mean observed EB n 5771 2 20.0 2 13.7 2 10.7 2 6.5 2 3.7 2 4.1 0.3 2.7 4.9 7.5 11.4 Mean predicted EB n 5771 2 22.6 2 14.0 2 10.9 2 6.7 2 3.9 2 4.3 0.1 2.5 4.7 7.3 11.2 Standard deviation of predicted herd level EB: 3 months, 72 cows n 5209 2.0 1.9 1.9 2.2 2.2 2.4 2.2 2.3 2.2 1.9 2.1 1 month, 72 cows n 570 4.3 4.1 4.5 4.8 4.7 4.5 4.0 3.9 4.0 4.0 4.2 1 month, 40 cows n 540 6.2 6.1 4.4 5.9 5.8 6.5 5.3 5.6 5.9 5.1 5.4 1 month, 20 cows n 520 6.9 8.2 6.7 9.3 8.1 7.6 8.1 7.3 8.3 7.7 7.8 Average number of cows in the sample: 3 months, 72 cows n 5209 17.3 23.7 24.0 17.7 18.3 18.0 17.9 18.1 18.1 17.8 18.1 1 month, 72 cows n 570 4.7 6.5 6.5 6.6 6.6 6.5 6.5 6.5 6.6 6.5 6.6 1 month, 40 cows n 540 2.8 3.6 3.8 3.9 3.5 3.6 3.8 3.7 3.7 3.9 3.9 1 month, 20 cows n 520 1.5 1.9 1.9 1.8 1.8 2.0 1.7 1.9 1.8 1.9 1.9 The level of EB, time of return to positive EB, and ketonemia with lower reliability than KET or NP. time to maximal dry matter intake were similar as in But it detected calculated low EB more reliably than the study of Gravert et al. 1986 who evaluated KET or NP. In the classification of ‘normal’ EB, ketone bodies as indicators of EB. Senatore et al. however, BHB, KET and NP performed slightly 1996 used a total mixed ration resulting in a much better higher specificity than the fat–protein-ratio. higher and longer energy deficit in a study of the Poor relationships between BHB and EB were also effect of energy deficit on fertility. Both studies reported by Erfele et al. 1974. BCS was a poor calculated the diets to meet requirements. The energy diagnostic aid for calculated low or normal EB in deficit of the cows in our study may therefore be this study low sensitivity, low specificity. categorised as moderate. We deem the data base In the multivariate prediction model, cow was not suitable for model development because a moderate included as fixed or random effect, but was corrected post partum energy deficit may be expected in most for in the error structure. We intended to develop a high yielding dairy herds. The model is supposed to population not cow specific prediction equation for be applied for such herds. EB Diggle et al., 1994. Modelling criteria were a 2 As in similar studies Gravert, 1991; Dirksen et good fit high R , low residual variation, a low al., 1997; Geishauser et al., 1998, the variance of degree of collinearity among covariates, and a low sensitivity and specificity of ketonemia might have number of parameters. Increasing the number of 2 been biased because test results of subsequent sam- parameters in the model would increase the R but ples from the same cow were not independent. make the model too similar to the data and supposed- Correlated samples are likely to produce smaller ly less applicable to other herds and rations. In fact, variances McDermott et al., 1994. Nevertheless, additional interaction terms yielded only marginal hyperketonemia, indicated by blood BHB .1.2 gains for the precision of prediction. Therefore, mmol l, was detected by field tests KET, NP interaction terms were excluded from the final slightly more reliably higher sensitivity than and prediction model. A high correlation between milk normal ketonaemia as reliable similar specificity as fat and the fat–protein-ratio was the reason for reported earlier Dirksen and Breitner, 1993; Nielen substantial collinearity. Because the fat–protein-ratio et al., 1994; Geishauser et al., 1998. Sensitivity and predicted EB better than milk fat, the latter was specificity of the fat–protein-ratio in milk were omitted. slightly higher than reported Duffield et al., 1997. The final model showed that lactation stage, In our study, the fat–protein-ratio detected hyper- parity, milk yield, and especially the fat–protein- 102 C . Heuer et al. Livestock Production Science 65 2000 91 –105 ratio were strong predictors of EB. The consideration parity and week of lactation. However, the coeffi- of the fat–protein-ratio almost fully replaced further cient indicated a drop in EB of 2.11 MJ NEL with tests of ketone bodies in blood or milk, despite every additional kg standard milk provided FP 6.9–10.0 sub-clinical ketosis in the first 5 weeks of remains constant. Applying the prediction model to lactation. The level of ketonemia in this study was, herds with higher milk production would therefore however, somewhat lower than in other studies of result in more negative EB compared to herds with high producing dairy herds Dohoo and Martin, lower production. And vice versa, a herd that 1984; Nielen et al., 1994; Duffield et al., 1997. A decreased milk production, for example as a conse- mild state of negative EB may more often or more quence of a severe energy deficit, would exhibit high readily result in an elevated fat–protein-ratio than in EB. Consequently, the model predictions cannot be solely increased ketone body levels in blood or milk. reasonably interpreted without simultaneous consid- Ketone bodies may rise when negative EB is increas- eration of the average level of production. Only if a ingly de-compensated. All cows in this study were herd maintains its average production level could a able to compensate negative EB, because ketone lower model prediction be regarded as a relative bodies rarely accumulated to high concentrations. An decrease of EB. instant increase of the fat–protein-ratio in response As the diagnosis of EB by single test parameters of low EB is in agreement with earlier observations was affected by lactation stage, parity, and milk Grieve et al., 1986; Gravert, 1991. Hence, adipose yield, the interpretation of individual results of milk cells mobilise while mammary cells synthesise tri- or blood tests with respect to EB should take these glycerides in response to negative EB. The differ- effects into account. Nevertheless, even after consid- ence in response was called ‘acetate partitioning’ by eration of these effects, some 75 of the variability 2 Kronfeld 1982 who suggested that, in early lacta- in EB remained unexplained R 5 25. However, tion, acetate is converted to milk fat to a greater given that microbial fermentation, rumen absorption extent by the mammary gland than by adipose of volatile fatty acids, digestive interactions between tissues, and that it is reversed in later lactation. The rumen and abomasum, intestinal absorbtion, regula- suggestion was confirmed by MacNamara et al. tory mechanisms of the intermediate metabolism, 1995 who estimated that, in early lactation, the passage from blood to mammary parenchyma, and mammary gland synthesised milk fat while adipose synthesis of lactose, fat and protein by the mammary tissue mobilised fat in response to added dietary fat; gland may all affect the association between feed- the opposite was observed in later lactation. NEL and milk constituents, we regard the strength of Milk protein was also a highly significant predic- the observed association as relatively high. More- tor but had relatively little bearing on EB. Milk over, the cow level model was only used as a tool to protein percent correlates positively with energy identify important prediction parameters. At this intake Magdus et al., 1988. Therefore, low energy step, a moderate model fit may be sufficient. There intake results in a higher fat–protein-ratio. Conse- was no specific interest in cow-level prediction of quently, part of the variation due to milk protein was negative EB in this study. More crucial was the probably represented by the fat–protein-ratio in our certainty or uncertainty of herd level prediction data. Because of its known relationship with energy because the model was developed to estimate aver- intake, we included milk protein in the final model age herd EB in order to provide a monitoring tool for even though it only contributed 0.1 to the model the provision of net energy through the ration. 2 R of this study. The model coefficient for milk yield Because Fig. 2 suggests that the 75 of the remain- was negative indicating lower EB with increasing ing variation in EB not explained by the predictors yield. This negative effect was partly compensated was random variability, the precision of the predicted for by an inverse relationship between milk yield and herd means was expected to be considerably more milk protein r 5 20.37. The impact of milk yield precise than the predictions for individual cows. 2 on the variation of EB was relatively small R 5 With the exception of week 7, observed and 0.9 when compared to a model only including predicted herd means increased steadily Table 4. C . Heuer et al. Livestock Production Science 65 2000 91 –105 103 The dip in week 7 was presumably caused by feed November 26. Therefore, the difference in week 2 changes due to the experimental diets starting in was the sum of two possible effects: a smaller week 6. We did not correct the week 7 coefficient for sample and calving seasonality. Seasonal effects this effect because the herd simulation intended to were probably small because the cows were estimate the precision of prediction, and this was synchronised, kept indoors and received the same unlikely to change after correction. However, before ration throughout the study period. We therefore applying the prediction equation of Table 4 to other believe that the difference was most likely caused by herds and rations, it may be reasonable to adjust the sampling variation. coefficient for week 7, e.g. by replacing it with the The standard deviations S.D. of herd means average of the adjacent weeks 6 and 8. increased with decreasing number of cows in the Cows in first lactation were more severely and sample. When EB prediction was based on only one longer in negative EB than cows in lactation 2 or 3 of 11 available weekly milk tests per cow n 570, Fig. 1. It was previously reported that EB de- the weekly means were based on 4–7 cows and the creased with increasing parity Lean et al., 1991; herd means varied by an equivalent of 61.5 kg Duffield et al., 1997, but in agreement with our standard milk Table 5. In order to have 70 cows findings, Dirksen and Breitner 1993 observed 2–12 weeks in milk at one point in time, the total higher BHB levels postpartum in primiparous than in herd size would have to be |280 cows 4370 cows, multiparous cows. A potentially greater risk of first non-seasonal calving. Compared to the predictions calving heifers than older cows for severe negative for individual cows S.D.59.3 MJ day, the vari- EB renders more investigative work. ability of the herd means of these 72 cows was less It was not surprising that BCS was a poor than half as large S.D. |4.3 MJ day. predictor of EB in our study. Although the propor- A test day was defined as one weekly average tion of cows in relatively obese condition BCS milk test result randomly selected from four avail- 3.5 was 41.5 in lactation week 2 and decreased to able results within a month. Pooling 3 subsequent 12.5 in week 11, the variability in BCS throughout test days improved the precision further by over the first 12 weeks of lactation was probably too small 100 S.D. |62 MJ, 0.7 kg standard milk. On the to reveal significant associations with EB. Using other hand, EB prediction based on 1 test day of 40 body condition loss instead of BCS did not improve cows projected total herd size of 160 cows reduced the model either data not shown. Moreover, the the precision by 33 compared to 70 cows with 1 relationship between BCS at calving and milk pro- test day. A herd with 20 fresh cows, equivalent to a duction in the first 12 weeks postpartum is not clear total herd size of 80, did often not have enough cows Garnsworthy, 1988, the risk of ketonemia and to compute weekly EB means for each of the 11 ketosis was the same for cows in obese or in normal weeks S.D.567.5 MJ, 2.5 kg standard milk. If condition at calving Smith et al., 1997, and BCS mean EB was based on one or two cows, individual was regarded as not sufficiently sensitive to reveal cow effects made the mean highly variable. Hence, substantial changes in EB Senatore et al., 1996. herd size was a limitation to herd level prediction of EB based on test day information. Although pooling 4.2. Prediction of herd level EB subsequent test days efficiently increased precision, it may not be desirable to evaluate new rations as Overall means of observed and predicted EB were late as 2–3 months after the feed change. Collapsing nearly identical, indicating that the prediction was the initial 12 weekly into 3 monthly means may not unbiased. However, predicted EB was slightly lower be advisable either, because EB was particularly than observed EB in lactation week 2 when only 52 critical from week 2 to 7, and our analysis suggested cows were in the study. The 20 cows with missing that stratification by week postpartum had profound observations in week 2 had calved between Sep- effects on the prediction of EB. The solution will tember 10 and October 3, and the other 52 cows with therefore have to be a compromise between pooling complete data had calved between October 4 and and wider precision limits. 104 C . Heuer et al. Livestock Production Science 65 2000 91 –105 Observed and predicted EB were in this study References from one herd. Whether prediction may be extrapo- lated to other herds under different conditions ra- Diggle, P.J., Liang, K.-Y., Zeger, S.L., 1994. Analysis of Longi- tudinal Data, Clarendon Press, Oxford. tions, seasons cannot be concluded from this study. Dirksen, G., Breitner, W., 1993. A new quick test for semiquantita- Therefore, the validation of EB prediction using data tive determination of beta-hydroxybutyric acid in bovine milk. of other feeding trials is an important next step. J. Vet. Med. A 40, 779–784. Dirksen, G.U., Hagert-Theen, C., Alexander-Katz, M., Berger, A., ¨ ¨ 1997. Stoffwechseluberwachung bei Kuhen in der Hochlakta-

5. Conclusions