The use of early lactation milk protein content to predict subsequent fertility performance and likelihood of culling, in commercial dairy cows

A dataset of 1,846,990 completed lactation records was created using milk recording data from 8,967 commercial dairy farms in the United Kingdom over a fi ve year period. Herd-specifi c lactation curves describing levels of milk, fat and protein by lactation number and month of calving were generated for each farm. The actual yield of milk and protein proportion at the fi rst milk recording of individual cow lactations were compared with the levels taken from the lactation curves. Logistic regression analysis showed that cows producing milk with a lower percentage of protein than average had a signifi cantly lower probability of being in-calf at 100 days post calving and a signifi cantly higher probability of being culled at the end of lactation. The culling rates derived from the studied database demonstrate the current high wastage rate of commercial dairy cows. Much of this wastage is due to involuntary culling as a result of reproductive failure.


INTRODUCTION
As farmers strive to reduce costs and maximize their income, average herd size and lactation yield in United Kingdom herds have increased steadily (Dairy Council, 2003).Milk yields are frequently the overriding focus of management and genetic selection (Lucy, 2000).In common with intensive systems elsewhere, the general rise in milk production in the United Kingdom is associated with a decline in the levels of fertility (Royal et al., 2000;Westwood et al., 2002), therefore annual culling rates now exceed 30% of cows in many dairy herds, with failure to conceive being a major reason (Dairy Council, 2003;Dairy Research International, 2003;Gröhn et al., 2003;Rajala-Shultz and Fraser, 2003).Higher milk yields increase the likelihood of extended periods of negative energy balance in the early lactation (Beam and Butler, 1999;Royal et al., 2000); the consequences of extended negative energy balance are longer post-partum anoestrus intervals (Beam and Butler, 1997), reduced embryo survival (Dunne et al., 1999) and reduced display of oestrus (Washburn et al., 2002); a common economic consequence of excessive delay in conception often result in a decision to cull the cow (Arbel et al., 2001;Esslemont and Kossaibati, 2002).Protein content of milk is linked to the energy status of the dairy cow, thus, during times of energy defi cit milk protein proportion has been shown to decrease (Grieve et al., 1986;Heuer et al., 1999;Reksen et al., 2002).The aim of this study was to test the association between the deviation statistics of the cows' milk components and the subsequent fate of cows, in order to build a potential indicator for cows performing outside adjusted herd average norms.The value of this statistic as a management indicator was also assessed.

Herd-specifi c lactation curves
The concept of the lactation curve developed by Wood (1969) is applied in the InterHerd TM software to generate herd-specifi c lactation curves through the analysis of historical milk recording data for each herd (PAN Livestock Services, 2003).The central equation used by Woods to describe the lactation curve is: (-cn)  where: Y n -the average daily milk in nth week, an is the initial milk yield just after calving; b -the inclining slope parameter up to peak yield; cn -the declining slope parameter; m i -month of production adjustments.
Similar formulae were used to describe the production of milk protein and fat.Herd-specifi c regression using historical milk recording data for cows in three lactation groups (Parities 1, 2 and 3+) were performed to fi t the constants in Woods' equation.The resulting formulae were then used to generate herd-specifi c curves that show production of milk, protein and fat for the average cow in that herd, adjusted for lactation number, days since calving and month of production.TENA-MARTINEZ M.J. ET AL.

The dataset
Complete lactation records were obtained from National Milk Records (NMR) from commercial farms between September 1997 and October 2003.The dataset contained information for individual lactation including the fate, herd, actual and herd-average lactation information.If a cow re-calved at the end of the parity, its pregnancy status at 100 and 200 days post-calving was derived from the re-calving date.The actual somatic cell count (SCC) was also included, being a known factor infl uencing culling decisions.The dataset was created according to the following criteria for inclusion in the analysis: A cow's parity record required a milk recording between day 15 and day 80 of the parity with full milk composition results.The cow must also have completed the parity with one of two possible fates (outcomes): i. re-calved (the cow calved again to start a new parity) and ii.culled (the cow left the herd without calving again).Parities that ended in the death of the cow (2% of all parities) were excluded from the analysis.
For the fi rst milk recording in each parity between day 15 and day 80 (1stMR), deviation statistics were calculated for milk yield and proportions of milk fat and protein.These were derived by subtracting the herd average value, with adjustment for the animal's parity, month of calving and days since calving (the 'adjusted herd average'), from the actual recorded value.

Descriptive statistics
The initial dataset comprised of 1,846,990 parity records from 921,242 different cows on 8,967 farms.Table 1 details the distribution of records in the dataset by parity, fate and pregnancy status at 100 and 200 days post-calving.The culling percentage ranges from 16% of cows in parity 1 to 38% of cows in parity 4 or higher (4+).The overall cull rate was 26%.
Up to and including parity 3, around 40% of cows had conceived within 100 days of calving, rising to around 70% within 200 days.For cows in parity 4+ these fi gures were much lower, 28 and 50%, respectively.Based on the re-calving percentages showed in Table 2, an estimated cohort of 100 cows in parity 1 that would survive to subsequent parities, it can be assumed that more than 50% of cows would fail to survive beyond parity 3.

Logistic regression analysis
Binary logistic regression models were developed using SAS (1999) with fate, either culled or re-calved, as the dependent variable.A range of potential explanatory variables, as shown in Table 2, were tested.Different combinations of the variables were tested for their ability to predict the likelihood that a cow is subsequently culled.Models were built manually in a stepwise manner.Experimentation with different order of variable addition and subtraction was carried out.The signifi cance of different factors in the model was assessed using the change in deviance due to the addition factor to the model.This is based on an approximated χ 2 test (Collett, 1999).Having found the combination of the variables that best predicted culling outcome the association of these variables with fertility (the probability of being in-calf at 100 days post-calving) was then assessed.All the analyses were conducted separately for four parity sub-groups: parity 1, parity 2, parity 3 and parity 4 or higher.LogActual SCC logarithmic transformation of the somatic cell count recorded at the 1stMR Herd identifi er a variable identifying the specifi c herd in which the cow was located

RESULTS
A number of variables were excluded from the model due to the limited impact that their inclusion had on the dependent variable.These excluded variables were herd identifi er, fat/protein ratio and fat deviation.
An initial analysis of the relationship between protein proportion and culling probability, without consideration of a cow's yield, suggested that cows with higher protein proportions were more likely to be culled.This contradicted the expectation that cows producing milk with lower protein proportions would have TENA-MARTINEZ M.J. ET AL.
higher culling probability because of the link with negative energy balance and poor fertility.However, the data showed a negative correlation between the protein proportion and yield of milk at 1stMR.This was as expected, because cows with high milk yields generally produce milk with lower proportions of protein.Since farmers resist culling cows with higher yields there is a potential for confounding of the effects of milk yield and protein proportion on culling probability.Therefore it was necessary to include both milk yield and protein proportion in all models.The resulting models predicted increasing culling probability with both decreasing yield and decreasing protein proportion.
The use of deviation statistics related to milk yield and protein proportion, rather than the measures per se, resulted in a better fi tting model.Adding either of the SCC variables signifi cantly improved the model showing these are signifi cant factors affecting the probability of a cow being culled.Inclusion of the logarithmic transformation of the SCC at the 1stMR resulted in a better fi tting model than using the logarithmic transformation of the average SCC from milk recordings to day 305 of the lactation.
All the variables included in the fi nal models were highly signifi cant, showing that a combination of the deviation statistic for milk protein proportion and milk yield combined with the log of the actual SCC provided the best prediction of fate of the cow.The parameters of the models for likelihood of culling for parity groups 1, 2, 3 and 4+ are displayed in Table 3.The same three explanatory variables were also used in a logistic regression model describing the likelihood of a cow being in-calf at 100 days post calving.The parameters of these models are displayed in Table 4. Figure 1 shows graphically the relationship between the protein deviation statistic and culling probability, based on the regression model equation, for fi ve scenarios illustrating different combinations of yield deviation and SCC level (Table 5).
Figure 2 shows graphically the relationship between the protein deviation statistic and probability of being in-calf at 100 days post calving, based on the Cows with protein proportions below the adjusted herd average (a negative deviation) have an increased probability of culling.The culling probability in all parity groups was also higher in the cows with higher SCC and below adjusted herd average yields recorded at the 1stMR.Figure 1 shows that the effect of protein deviation differs with parity number.Figure 3 compares the parity groups for a single scenario (average yield, medium SCC).The steeping curves in Figure 3 show that the effect of the protein deviation on the probability of culling strengthens as parity number increases.The effect of protein deviation on culling probability in parity 1 is slight, though still statistically signifi cant, whereas the effect is markedly stronger in parity 4+ cows than in other parity groups.
Cows with protein proportions below the adjusted herd average (a negative deviation) have a decreased probability of being in-calf at 100 days.For cows in parity 4+, the in-calf at 100 days post-calving probability was much lower for cows with below adjusted herd average yields recorded at the 1stMR.In other parity groups (1-3) the in-calf at 100 days post-calving probability was slightly lower than average for cows with higher SCC.
Figure 4 compares the in-calf at 100 days post-calving probability between parity groups for a single scenario (average yield, medium SCC).The curves for parity 1 to 3 cows in Figure 4 have similar gradients, showing that the effect of the protein deviation on in-calf probability is similar in these parities.The curve for parity 4+ cows is slightly steeper, suggesting a stronger effect.

DISCUSSION
This analysis, based on a very large dataset of recent lactation records derived from over 40% of commercial herds in Great Britain, highlights the very high culling rates that persist in modern dairy systems.That the majority of cows fail to reach their fourth lactation represents a massive waste of resources.A high culling rate entails the need to rear more replacements and many of them will have a reduced life as productive milking cows.
The analyses of herd-specifi c lactation curves, based on historical milk records of all cows on a farm, provided a basis for within-herd comparison of yields of milk and its constituents.This enables comparison of each cow's actual production on any milk recording date against a herd standard that is adjusted for the parity, month of calving and days into lactation.
Use of deviation statistics accounted for variation between herds, parity and month of calving.It is of note that a herd identifi er did not add a signifi cant level of explanation to any model containing deviation statistics.The use of such deviation statistics in studies that compare performance of cows in a number of different herds, such as that employed for sire proofi ng (Wall et al., 2003), may provide a clearer indication of the true differences between progeny.
The fi ndings of this study support the link between milk protein proportion and energy metabolism in the dairy cow (Grieve et al., 1986;Heuer et al., 1999;Reksen et al., 2002).A negative deviation in protein proportion in early lactation is believed to refl ect a shortage of energy that will have a negative impact on fertility (Beam and Butler, 1997;Dunne et al., 1999;Washburn et al., 2002).This analysis demonstrates a link between the protein deviation statistic and the probability of a cow being in-calf at 100 days after calving.Poor fertility performance is known to be a major cause of culling, hence the association between the protein deviation statistic and the culling probability shown in this research.
Actual milk fat proportion, fat deviation and fat:protein ratio variables were not signifi cantly associated with culling probability suggesting that the milk protein proportion is a superior indicator of negative energy balance than milk fat parameters.
It was observed that the consequences of protein deviation, in terms of culling probability, differ markedly between parities.In parity 1, the effect of protein deviation on culling probability, though statistically signifi cant, is practically negligible (Figure 2a).For older cows, particularly those in parities 4 and above, a negative protein deviation materially increase the probability of being culled (Figure 2d).The protein deviation has a similar effect on the probability of being in-calf at 100 days after calving in all parities.It therefore follows that failure to conceive early in lactation results in a culling decision more readily for older cows than younger ones, especially fi rst parity animals.It is possible that farmers are more willing to re-serve younger cows more often than older ones.In addition, farmers cull fi rst parity cows for a wide range of reasons, for example behavioural or conformation problems, which may dilute any effect of fertility problems on culling probability.The markedly higher impact of protein deviation on culling probability in the parity 4+ group suggests that fertility plays a more prominent role in culling decisions in older cows.It is possible that the protein deviation is also linked to other factors leading to culling decisions, such as chronic mastitis and lameness.Within the current database there is potential to explore further the relationship between protein deviation and SCC profi les.

CONCLUSIONS
The results of these analyses show that milk protein, milk yield and somatic cell counts obtained from routine milk recording in early lactation (from day 15 to day 80 after calving) are strongly associated with the fertility performance and subsequent fate of the dairy cow.
If the very high culling rates currently observed are to be reduced, more attention should be focused on managing the early lactation cow.In herds with high culling rates, farmers should target minimizing negative energy balance in early lactation, so that fertility is optimized, rather than maximizing milk output.On most farms newly calved cows are all managed together and fed for maximum yield irrespective of the age of the animals concerned.This research suggests that culling rates, particularly in older cows, would be signifi cantly reduced by managing these animals separately in a less demanding system.
The protein deviation statistic provides a practical tool for the monitoring of energy balance in the commercial dairy herd.Milk recording organizations and farm advisers can readily identify animals or groups of animals that are producing milk with lower protein proportions.Further investigation is needed to identify critical levels of protein deviation to trigger management interventions.The fact that the milk recording data are routinely, and rapidly available should make this a more effective early warning of energy defi cit than alternatives such as body condition scoring and metabolic profi les which require additional time and effort from the farmer.

Table 1 .
Distribution of cow records by parity in the dataset

Table 2 .
Explanatory variable tested in the analysis Actual yield the recorded yield of milk (kg) at the 1stMR Actual protein proportion the recorded proportion of milk protein at the 1stMR

Table 3 .
Binary Logistic Regression Models describing the probability of culling