Genetic parameters and evaluation of mortality and slaughter rate in Holstein and Jersey cows

The longevity of dairy cattle has economic, animal welfare, and health implications and is influenced by the frequency of mortality on the farm and sale for slaughter. In this study cows removed from the herd due to death or slaughter during the lactation were coded 1 and cows that were not terminated were coded 0. Genetic parameters for mortality rates (MR) and slaughter rates (SR) were estimated for Holstein (H) and Jersey (J) breeds by applying both linear (LM) and threshold (TM) sire models using about 1.2 million H and 286,000 J cows. Estimated breeding values (EBV) for MR and SR were predicted using animal models to assess the opportunity for selection and genetic trends. Cow termination data, recorded between 1990 and 2020 on a voluntary basis by Australian dairy farmers, were analyzed. Cow MR has increased from below 1% in the 1990s to 4.1% and 3.6% in recent years in H and J cows, respectively. Most dead cows (~36%) left the herd before 120 d of lactation, while cows that were slaughtered left the herd toward the end of the lactation. Using the LM, heritability (h 2 ) estimates for MR were lower (1%) than those for SR (2%–3.5%). When h 2 were estimated using a TM, the estimates for both traits varied between 4% and 20%, suggesting that the difference in incidence level is one of the reasons for the difference in the h 2 values between MR and SR. Early test-day milk yield (MY) and 305-d MY (305-d MY) have unfavorable genetic correlations (0.32–0.41) with MR in both breeds. The genetic correlations of calving interval with MR were stronger (0.54–0.68) than with SR (0.28–0.45) suggesting that poor fertility can serve as an early indicator of poor cow health that may lead to increased risk of death. High early test-day somatic cell count is genetically associated with increased likelihood of slaughter (0.24–0.46), but not with increased


INTRODUCTION
Longer productive life increases profitability by reducing replacement cost and by increasing milk production of the herd as a result of having more older cows, and by increasing the opportunities for voluntary culling, as opposed to involuntary culling.Furthermore, longer productive life is associated with improved animal health and welfare (Barkema et al., 2015;Schuster et al., 2020) and reduced environmental footprints of dairy production (Pryce and Bell, 2017).The 2 main costly reasons that result in reduced productive life of cows in addition to voluntary culling are increased involuntary culling (i.e., sale for slaughter) and mortality (i.e., death on the farm).
The introduction of genetic selection for longevity has resulted in favorable genetic trends (e.g., Maia et al., 2014;VanRaden et al., 2016;van Pelt et al., 2016;Schuster et al., 2020).However, research in the United States (Miller et al., 2008;Tokuhisa et al., 2014;Van-Raden et al., 2016), in Denmark (Maia et al., 2014), in Israel (Weller et al., 2023), and elsewhere (Compton et al., 2017) shows that cow mortality has increased recently.The effect of increased adult cow mortality on farm profitability is greater than that of slaughter because farmers incur disposal costs instead of income from sale of culled cows.Information on the extent of death losses in Australian dairy herds is limited.Stevenson and Lean (1998) reported a mortality rate (MR) of 4.3% based on data collected by closely monitoring 8 dairy herds.More recently, Beggs et al. (2019) reported a MR of 1.9% after monitoring 50 Australian dairy herds for a year.Data voluntarily recorded by farmers and milk recording organization on cow termination dates and reasons could be useful to quantify the level of cow losses and to explore opportunities to reduce cow MR by improving herd management regimens and selecting sires that produce daughters that are less prone to MR (Miller et al., 2008).Longevity in some countries such as the United States is a combination of productive life and livability (VanRaden et al., 2016).The justification for including cow MR in the economic selection index (e.g., net merit index) in addition to productive life, is that it encompasses all termination reasons (including death) because the genetic correlation between cow livability and productive herd life is less than unity (genetic correlation of ~0.7) in different dairy breeds in the US herds (VanRaden et al., 2016).Maia et al. (2014) reported that the correlation between the sire components for MR and slaughter rate (SR) were small but negative in 3 Danish dairy cattle populations.
In Australia, data on termination reasons and dates have been recorded by farmers for the last 3 decades.The data that was recorded before 1990 was found to be less complete and unreliable for genetic evaluation of survival (Madgwick and Goddard, 1989).Since the 1990s termination data are available for about 60% of the herd-years.The termination data were recently analyzed to estimate age at culling and change in culling reasons over time (Wondatir Workie et al., 2021).However, the data has not been used to assess if trends in MR over time have changed compared with overall culling rates which is important given reports elsewhere (Miller et al., 2008;Maia et al., 2014;Tokuhisa et al., 2014;VanRaden et al., 2016).Furthermore currently, the termination data are not directly used for genetic evaluation of survival (Madgwick and Goddard, 1989) because the survival status of cows at the end of each lactation is determined by looking at the re-calving pattern of cows and their herd-mates (Madgwick and Goddard, 1989;Visscher and Goddard, 1995).The genetic evaluation for survival which is unadjusted for MY that is carried out by DataGene Ltd. (https: / / datagene .com.au,Melbourne, Australia) could potentially be improved if the termination data were combined with the survival data.Furthermore, knowledge of the genetic and phenotypic factors that affect MR and culling rate and differences between breed groups can be used to develop approaches to improve longevity, animal welfare, and health and reduce cow MR.To examine the importance of the MR and overall culling rate we divided cow termination reasons recorded by Australian farmers as MR and SR following the definition of Fetrow et al. (2006) and Maia et al. (2014).
The objectives of this study were: (1) to assess the quantity and quality of data on cow MR and SR by estimating the effect of fixed factors and comparing the estimates with results in the literature; (2) to quantify the extent of genetic variation in cow MR and SR for pure Holstein (H) and Jersey (J) breeds; (3) to estimate the correlation of cow MR and SR with production and other functional traits for H and J breeds; (4) to assess if a genetic evaluation for MR is feasible using the available data from Australian dairy herds.

Data Source
This study used already-collected data from a database used for genetic evaluation, and no handling of the already-recorded cows was required.Therefore, this analysis did not require approval by an Institutional Animal Care and Use Committee.Data for this study were obtained from Australian dairy herds that participated in milk recording and collecting other data including pedigree (animal, sire, dam, and maternal grandsire) from a database that is maintained by DataGene Pty Ltd., (Melbourne, Australia).Other than parents, the main items in cow pedigree data are breed, birth and termination dates and termination reasons.The 2 broad categories for cow termination are death (milk fever, accident, calving, paralysis, mastitis, Johne's diseases, and other unspecified causes) or slaughter (infertility, mastitis, low production, old age, type defects, poor temperament, difficulty of milking and other unspecified reasons).Cows sold for dairy purposes and those that belonged to herds that withdraw from data recording were assumed to have not been terminated.Only cows with at least one calving record were selected for this study.
The cow pedigree data were merged with lactation and calving data based on unique cow identification numbers.The merged data set included about 7.9 million purebred and crossbred cows from 22,374 herds.The data of cows that were born between 1988 and 2017 and that calved between 1990 and 2021 were used for this study.Data recorded before 1990 were less complete (Madgwick and Goddard, 1989) and were not used.All cows that calved for the first time between 17 and 39 mo of age and all their subsequent calving records were selected for further analyses.
As the recording of culling data are voluntary in Australia, we selected cows from herd-years where a minimum percentage of cases (death or slaughter) were recorded.Herds with less than 30 calvings, that recorded less than 0.5% cases (death or slaughter of the total) or more than 50% cases in a year were excluded.As a result of these selection criteria, 45% of herd-years and 17% of calvings were removed.After these edits, 5.77 million cows with 17.93 million calvings were selected for further analysis.In the selected data set about 15% of cows had some of their calvings in the herd-years that were nominated for exclusion.All the data of these cows were then removed from the data set.This was done to make the analyses based on cows for which we have complete data from first calving to the end of their herd life (death or slaughter or unknown fate).This is expected to minimize the effect of pre-selection as well as under-reporting of MR or SR on the results of the analyses.

Data Selection and Definition of Traits
The description of the data used for genetic parameters and evaluation for H and J breeds is shown in Table 1.For estimating genetic parameters, cows that have known sires and dams were selected.Furthermore, H and J sires with at least 20 and 3 daughters, respectively, were selected for estimating genetic parameters.To construct data set of manageable size for estimating genetic parameters for the H breed, only data from large herds with at least 50 calvings in a year were selected.However, when selecting data for genetic evaluation such restrictions were not applied because we wanted to get EBV for bulls with high reliability by using the mortality and slaughter data of all their daughters.Thus, in the data set used for genetic evaluation, cows with unknown dams and sires with few daughters were not excluded.
In this study MR or SR were coded 1 when a lactation ends in death or slaughter.Based on this, 3 traits associated with the fate of cows were defined.First, for all cows in the selected herd-years, the MR was defined by coding cows that died as 1 and coding cows that were slaughtered, still in the herd, and transferred to other herds as 0. Second, SR was defined by coding cows that were slaughtered as 1 and coding cows that died, still in the herd, and transferred to other herds for dairy purposes as 0. Third, culling rate was defined by coding cows that died and slaughtered as 1 and cows still in the herd and transferred to other herds for dairy purposes as 0. The third trait which includes both MR and SR, as defined above, is similar to the current definition of the trait used for genetic evaluation of survival in Australia (Madgwick and Goddard, 1989).However, this trait was only used for calculating EBV for culling rate of cows and then these EBV were compared with EBV for MR and SR.In all cases above if the fate of a cow could not be determined when the cow was in the herd-years selected for this study, the cow was included with its fate coded missing.
For cows selected for this study, their 305-d milk yield  yield, first-testday MY, first-test-day SCC, and calving interval (CIN) data of all lactations were extracted.This was done to assess the genetic relationship between the traits of primary interest (MR or SR) and the other economic traits.In the merged data set on average about 8% and 5% of the cows were included with missing fate data but with valid test-day MY and 305-d MY data, respectively.Similarly, approximately 15% and 34% of the cows with valid fate data were included without valid MY, and CIN data, respectively.

Data Analyses
First, the incidence of MR and SR were calculated for cows of different breeds in the data set to explore differences among breeds and identify the breeds for further genetic analyses.The frequency showed the mean MR rates were the highest in H and J breeds, so the data of the 2 breeds were selected for further analyses.
For genetic analyses in addition to the data on different traits, pedigree data that was used for genetic evaluations of Australian dairy cattle by DataGene was also accessed.The pedigree data were used when fitting genetic models and was also used to calculate inbreeding coefficient (F) of animals.The data on J and H cows were used for estimating genetic parameters and evaluation.All data analyses were carried out using ASReml (Gilmour et al., 2021).The data analyses were performed using sire models to estimate genetic parameters which included up to 3 traits.This was done to minimize the requirement for huge computational resources.EBV for MR and SR were estimated fitting animal models to obtain EBV for all animals using all the available data.
Heritability Using Linear and Threshold Models Based on All Parities, Parity 1, 2, and 3 Data.The H and J data were analyzed separately to assess if there are differences in MR and SR between the 2 main dairy breeds.When analyzing MR and SR the fixed effects fitted were dry-off of year of lactations or termination year of cows and dry-off of month of lactations or termination month of cows, age at calving within parity, F and herd-year-season of calving.The model included regression on age at calving (linear and quadratic) and F (linear).The random effect fitted was the sires of the cows with the numerator relationship between the sires and their ancestors assuming the use of a larger data will provide more accurate estimates of genetic parameters.For the threshold model (TM) analysis, a binomial distribution assuming a LOGIT link function which implies an underlying logistic distribution with an error variance of π 2 /3 was used in ASReml (Gilmour et al., 2021).
The model for estimating h 2 using the linear model (LM) and TM can be written in matrix notation as: where y is the vector of observations (death or slaughter) for the LM and a vector of unobserved liabilities for MR or SR for the TM; X is an incidence matrix relating observations to fixed effects described above, b is a vector of fixed effects; Z is an incidence matrix relating observations to sire effects; s is a vector of sire additive genetic effects; e is a vector of random residu-als.For the LM it was assumed that the random effects followed normal distributions of s~N 0, Aσ s 2 ( ) and e~N 0, Iσ e 2 ( ) , where σ s 2 , and σ e 2 represent additive genetic, and residual variances, respectively; A represents the numerator relationship matrix for sires; and I is the identity matrix.For the TM, the random effects were assumed to have followed normal distributions of s~N ) and e~N(0, Iπ 2 /3), where σ sT 2 , and π 2 /3 rep- resent additive genetic and residual variances, respectively.In the above model although repeated cow records were used, the permanent environmental effect of cows (PE) was not fitted in the final model because cows coded 1 for either MR or SR did not have repeated records (Ødegård et al., 2006) as a result the variance for PE converged to zero.

Correlation of MR and SR with Other Economic Traits.
Following the estimation of h 2 for MR and SR, genetic correlation of , with first test-day milk and Ln SCC (natural Log-transformed SCC), and CIN were estimated.The analyses of MR and SR using an LM was a series of 3 trait models by adding one of the continuous traits (e.g., MY, Ln SCC, CIN) at a time to a model that included MR and SR.
The 3-trait linear model used for estimating correlations can be written in matrix notation as: where y t is the observation on the 3 traits, X t is the incidence matrix linking the observations to the fixed effects (all fixed effects in Model 1 fitted for MR and SR and HYS, age at calving (linear and quadratic) within parity, F, and month of calving) for the other trait t (e.g., MY, CIN, Ln SCC), b t is the vector of all fixed effects for trait t, Z t is incidence matrix linking the observations to the random effects for trait t, s t is vector of random sire effects for trait t, and e t is the vector of random residual effects for trait t.For the economic traits such as MY, Ln SCC, and CIN.PE of cows was also fitted to account for repeated records and C is the incidence matrix for the random PE.
The expectations (E) are: E(y t ) = X t b t , E(s t ) = 0 and E(e t ) = 0, E(pe) = 0 with the following variancecovariance matrix: Var(s) = G t ⊗A s for sire model, Var(pe) = Ipe t and Var(e) = R t ⊗I e , where G t and R t , are genetic and residual variance-covariance matrices of size 3, respectively.Furthermore, A s and I are numerator relationships based on sires and identity matrices, respectively, and ⊗ is the Kronecker product function.
For these analyses the following genetic and residual variance-covariance matrices were assumed: where G t and R t are symmetric matrices of the genetic and residual variance-covariance, respectively, and T1, T2, and T3 refers to traits 1 to 3. It is worth noting that the way the 2 traits (MR and SR) are defined restricted the residual covariance between the 2 binary traits to be fixed at zero because records coded 1 for one trait were by default coded 0 for the other trait.In addition, residual covariance between the 2 binary traits and CIN were also fixed at zero because records coded 1 (dead or slaughtered) have no valid CIN records.These restrictions were necessary for the convergence of the models.
In addition to the above 3-trait LM analyses, bivariate analyses of MR or SR on the underlying logistic scale (threshold) and one of the continuous economic traits on the normal scale (TLM) were performed.The objective of performing these bi-variate TLM was to obtain estimates of genetic correlations of MR or SR with the other economic traits that are less affected by incidences of MR or SR because incidences levels could have large effect on the genetic variance (Dempster and Lerner, 1950) and therefore genetic correlation.Fitting the bi-variate TLM in ASReml (Gilmour et al., 2021) required that the residual variance for the binomial traits to be fixed at 1.0 as described above for the single trait TM analyses.Furthermore, because of this restriction, the residual correlations of MR and SR with the continuous economic traits were not included in the results table .Genetic Correlation Between MR or SR in the First 3 Parities.To explore if early MR or SR (e.g., during the first parity) is the same trait as that later in life (during the third parity) genetic correlation between the first 3 parities were calculated.For this analysis we used a 3-trait linear sire models where MR or SR in the first, second and third parity were considered as 3 different traits.We also estimated genetic correlation between MR and SR within parity using the data of the first 3 parities.
EBV for MR or SR and Their Correlations with Other Economic Traits.EBV for MR and SR were calculated using a linear animal model assuming the genetic parameters estimated above (3-trait model).The continuous trait selected for all analyses was 305-d MY as it was more heritable and had a slightly stronger genetic correlation with MR and SR than the other traits considered in this study.First, EBV were calculated for each breed separately assuming that the genetic parameters estimated using data of each breed are true values.Second, we also estimated EBV using the data of both breeds and including data from all cows to mimic the current genetic evaluation of dairy cattle in Australia which includes data of all breeds.These data included 20 different breed groups with at least 22,148 cows/records.For this analysis genetic parameters estimated using H data were used.
To assess the accuracy of EBV for MR and SR, EBV (i.e., parent average) for bulls that were born after 2013 were predicted and correlated with the corresponding bull solutions that were obtained based on their daughter's mortality and slaughter data.The bull solutions were obtained from a single trait model that fitted all fixed effects (i.e., Model 1) and the random effect of the bulls without including pedigree information.
The solutions for bulls with at least 25 progeny/ records obtained from the linear animal model analyses were used to calculate the mean and the standard deviation (SD) and are presented to show the extent of variability and opportunity for selection for J and H bulls.The presence of genetic trend was assessed for both breeds based on bulls with at least 25 or more progeny/records.The EBV for MR and SR from the current analyses for sires were also correlated with EBV for several traits and the 2 economic indexes (balanced performance index, BPI and health weight index, HWI) that were published in August 2022 by DataGene (https: / / datagene .com.au).The correlations of bull EBV for MR and SR with other economic traits routinely evaluated by DataGene were calculated for H and J bulls separately.The number of H bulls used for calculating correlations ranged from 9,892 (MY, fertility, and survival traits) to 9,869 (type traits).The number of J bulls varied from 1,977 (MY traits, fertility, and survival traits) to 1,958 (type traits).The bulls selected for calculating correlations were those with at least 40% reliability for all traits evaluated by Data-Gene.

Incidence of MR and SR and Fixed Effects
Table 1 shows the mean MR and SR in H and J cows which were used for genetic parameter estimation and evaluation.Both MR and SR were slightly higher in the data that were selected for estimating genetic parameters than the larger data used for evaluation.In H cows, the average age at death was longer at 79.6 mo than the average age at slaughter which was 73.7 mo.The difference between the average age at death and slaughter was smaller in J cows with 78.7 (death) and 74.9 (slaughter) months of age.On average the number of lactations completed was higher in J (5.7) than in H (5.3) because CINs of J cows were shorter.The lifetime MR in H cows (16.1%) was slightly higher than in J (14.7%) cows.
A clear distinguishing feature of cows that were removed from herds due to MR and SR is shown in Figure 1, where more deaths (36% in H or 37% in J) occurred before 120 d of lactation.However, most cows that were slaughtered were removed from the herd after mid-lactation (between 181 and 280 DIM) in both breeds.
Over the years, average MR increased from 0.78% for lactation that started in 1990 to 4.14% in 2016 in H cows.In J cows, MR increased from 0.75% in 1990 to 3.6% in 2016.After 2016 there was no significant increase in MR in both breeds.SR, recorded in the termination database, also increased from 4% in 1990 to 22% in H.In J, SR increased from 5% in 1990 to 20% in 2016 and remained the same or only increased slightly after 2016.Frequency of deaths increased in both breeds with parity from (2.0%-2.3%) in the first parity to 6% in the eighth and later parities.Frequency of death by termination month were high between August and November October (5.0%-6.9%)and low between May and July (1.0%-2.1%).The fixed effect of month and year of termination (cow or lactation), and age at calving within parity were significant in addition to herd-year-season of calving which had highly significant effects on both MR and SR.Inbreeding (F) of cows had significant effect on both MR and SR.A 1% increase in F increased MR by 0.07% and 0.05% in J and H, respectively, and the increase in SR was higher at 0.17% and 0.15%.

Heritability of MR and SR
Table 2 shows h 2 of MR and SR estimated using LM and TM.The h 2 of MR were lower than that of SR for both J and H breeds when estimated using both LM or TM with the exception of the unexpectedly higher h 2 for MR in parity 2 J cows.As expected, h 2 values for H breed were estimated with smaller standard errors than for J breed.Although incidence rates increased with parity, h 2 estimates did not show any clear trend.

Genetic and Environmental Correlations of MR and SR
The correlation of MR and SR with 305-d yield (milk, fat, and protein) and with first test-day MY and Ln SCC and CIN are shown in Table 3.The pattern of correlations of MR with MY traits is different to that of SR.The genetic correlation of MR with MY traits is positive (unfavorable) but that of SR is negative (favorable) in both breeds.Among MY traits, 305-d PY had slightly higher correlation with SR than the other yield traits in H but in J, both 305-d PY and FY have stronger correlations.High daily MY early in the lactation is associated with increased likelihood of death but not with increased or decreased chance of slaughter.The direction of the residual and genetic correlations between all MY traits and SR is the same but in the case of MR the direction of the residual correlation is negative and that of the genetic correlation is positive (Table 3).
High daily Ln SCC early in the lactation is not associated with likelihood of death but it does increase the chance of being slaughtered.In both breeds, cows with long CIN (poor fertility) have an increased chance of dying (Table 3) but the association of CIN with SR was weaker.The difference was obvious in J where the genetic correlation between SR and fertility was half of that between MR and fertility (Table 3).
Genetic correlation between MR and MY traits were slightly lower when estimated using the bivariate TLM than the 3-trait LM models (Table 3).However, the genetic correlation of 305-d yield traits with SR were stronger when estimated using the bivariate TLM than when estimated using LM (Table 3) in both breeds.The genetic correlation of early Ln SCC was stronger when estimated using the bi-variate TLM than when estimated using LM (Table 3) in both breeds for both MR and SR.Overall, the genetic correlation of MR and SR with the other economic traits (Table 3) from LM and TLM were similar.Although the TLM is less affected by incidence level, it may not necessarily provide more accurate estimates than LM because of the difficulty to clearly separate the residual correlation from the genetic correlations.The modeling approach of TLM implemented in ASReml (Gilmour et al., 2021) may have difficulty to properly separate the genetic correlation from the residual correlation in particular when the 2 correlations have the same direction.

Genetic Correlation Between MR or SR in the First 3 Parities
Table 4 shows the genetic correlation between MR or SR in the first 3 parities.In H as expected, the genetic correlation between first and third parity are the lowest for MR and SR.In J where the correlations were estimated with higher standard error, the genetic correlation between parity 1 and 2 was the lowest for MR.The genetic correlations between MR and SR in the first 3 parities for H were similar to the effectively zero correlation estimated based on the across parity data.In J, the genetic correlations between MR and SR in Table 4 were not consistent and were associated with large standard errors but were still close to the genetic correlation (approximately −0.31 ± 0.06) estimated based on across parity data.

EBV for MR or SR and Their Correlations with Other Economic Traits
Table 5 shows mean EBV for MR and SR for J and H bulls with their SD from multitrait and single trait models for bulls with at least 25 progeny/records.Correlations between EBV from multitrait and single trait models were higher for SR (0.98-0.99) than for MR (0.91-0.97).The SD of bull EBV were slightly higher when the estimates were from multitrait models than single trait models for both traits.Overall, the scope for selection for MR is rather limited because the SD of EBV for MR is only 13% to 14% of the EBV for SR. Figure 2 shows the genetic trend for MR for H and J breeds where there was a significant and continuous deterioration until 1998.After 1998 the genetic trend in H improved while the trend in J stopped deteriorating but remained at elevated levels.However, genetic trends for SR showed a declining trend over the years (results not plotted).
The correlations between bull EBV from the multibreed and single-breed analyses were high for both breeds and both traits (i.e., MR and SR).In H bulls with at least 25 progeny/records, the correlations between EBV from H only and all breed analyses were 0.95 and 0.96 for MR and SR, respectively.The corresponding values for J bulls were slightly lower at 0.88 for MR and 0.93 for SR.When calculated based on bulls with 25 or more progeny/records, EBV for overall culling rate were more favorably correlated with EBV for MR (0.44) and SR (0.92) in H than in J.The corresponding values for J bulls were 0.02 for MR and 0.92 for SR EBV.
Table 6 shows the correlation between solutions estimated based on their progeny data and their corresponding parent average (PA) for MR and SR for bulls born after 2013.The correlation between bull solutions and PA are lower for MR than for SR when the correlations were calculated for bulls with fewer number of progeny/records.
Correlations of EBV for MR and SR for H and J breeds with EBV for other economic traits including the 2 economic indexes (BPI, HWI) are shown in Table 7 where the correlation value is below −0.3 (favorable) or above 0.3 (unfavorable).In H, EBV for MR have positive (≥0.3) unfavorable correlation with EBV for MY, angularity, body depth and dairy strength and negative favorable correlation (≤−0.32) with calving ease, fertility (expressed as 6-wk pregnancy rate), and the HWI (Table 7).In J, EBV for MR have positive unfavorable correlation (0.43-0.52) with EBV for milk and PY and angularity.EBV for 305-d fat percent with EBV for MR (Table 7) were more negative in J (−0.36) than in H breed (−0.23).To verify these results, genetic correlations between 305-d fat percent and MR were estimated using covariance analyses.The genetic correlation estimates based on covariance analyses were less strong (−0.19 ± 0.05 in J) and (−0.13 ± 0.0.03 in H) compared with correlations between the EBV in Table 7.
The EBV correlation of SR with all traits in Table 7 and the 2 economic indexes (BPI, HWI) are negative (favorable) for both breeds except the positive correlation with body depth in H.As expected, the strongest negative correlations of EBV for SR in both breeds were with survival EBV and the 2 economic indexes (BPI, HWI).Table 7 shows poor temperament leads to increased chance of slaughter in J breed, but a poor score for likeability (a subjective score of cows by farmers) leads to an increased chance of slaughter in both breeds.

DISCUSSION
Reduced cow longevity, in particular due to increased cow MR has economic, animal welfare and health implications.This study was conducted to quantify death losses and genetic variation in cow MR and to assess if there are unfavorable genetic correlations that can lead to increased cow MR.The results highlighted that there are both phenotypic and genetic differences between cows that are removed from the herds due to death and slaughter.Most cows that died, leave the herd before reaching mid-lactation in contrast to most cows that are slaughtered, which leave the herd toward the end of the lactation.Genetically cow MR is more correlated with high early MY and 305-d MY whereas cows that are slaughtered are the ones that have low 305-d PY, MY and FY and have high early Ln SCC.The stronger genetic correlation between MR and CIN suggests that  cows with poor fertility are more likely to die and also genetically cows with poor fertility have an increased risk of being slaughtered.Although the h 2 of cow MR is low, the positive genetic correlation between MY and MR may have contributed to the increased trend in cow MR over the years in Australian dairy herds.However, recently both the genetic and phenotypic trend for MR and SR have declined in H and have become stable in J.This could be due to the strong selection for improved fertility and survival and less selection emphasis for MY (Pryce et al., 2009).
The incidence of cow MR in Australia dairy herds has increased over the last 20 years until 2016 which is similar to trends in several other counties (Thomsen and Houe, 2006;Miller et al., 2008;Maia et al., 2014;Compton et al., 2017).The present results show that MR in Australian herds are lower than in other comparable countries, though the data edit criteria that set 0.05% as a minimum for both MR and SR instead of only for MR could have contributed to low MR observed in current data.However, any downward bias due to this is small given that the estimates are comparable to the only MR estimates available based on Australian herds (Beggs et al., 2019).Beggs et al. (2019) based on data from 50 herds reported an annual MR of 1.9% which is equivalent to 2.1% on lactation basis assuming an average CIN of 403 d.
The current study suggests that the increased trend for MR in the past was possibly due to selection for increased MY without considering fertility and health traits (Rauw et al., 1998).In addition to intense selection for increased MY, the increased MR in Australian herds could be associated with increase in herd size as observed by Shahid et al. (2015) and Alvåsen et al. (2012).Although herd sizes have increased in Australia in recent years (Dairy Australia, 2022), the contribution of increased herd size to increased MR is less than 0.5% in current data.Further analyses of the current data showed that cow MR varied from 2.96% in herds with less than 100 calvings per year to 3.44% in herds with more than 500 calvings.This is smaller than the contribution of increased herd size to increased SR (14.5 in smallest herds to 18.0% in the largest herds).Although large herds are likely to give less attention to individual cows which may contribute to increased MR (Alvåsen et al., 2012) Economic Index that combines traits that contribute to the farm business (Richardson et al., 2022). 3 Economic Index with more emphasis on fertility, mastitis resistance and feed efficiency (Richardson et al., 2022).
that large herds have more capacity to provide better veterinarian services and management compared with small herds with limited resource (Beggs et al., 2015).The h 2 of MR in Australian dairy cattle is low (~1%) and is similar to those observed in other countries (Miller et al., 2008;Maia et al., 2014;VanRaden et al., 2016;Weller et al., 2023).The h 2 of cow MR from LM are similar to estimates reported by Miller et al. (2008) andVanRaden et al. (2016).The h 2 estimates from TM are also similar to those estimated by Tokuhisa et al. (2014) which ranged from 0.04 to 0.07 based on data from US dairy herds.The h 2 estimates for SR are similar to previous estimates for cow survival based on Australian dairy data (Madgwick and Goddard, 1989;Visscher and Goddard, 1995;Haile-Mariam et al., 2003).The similarity of the h 2 estimated using TM for MR and SR (varied between 4% and 20%) compared with when estimated from LM suggests that the differences in incidence level is one of the reasons for the difference in h 2 values between the 2 traits (Lett and Kirkpatrick, 2018).The h 2 of MR in J based on parity 2 data estimated using TM was unexpectedly high (Table 2).In addition, the lower-than-expected genetic correlation between parity 1 and 2 compared with that between parity 1 and 3, for example, is difficult to explain.
Genetically cow termination due to MR and SR which is negative in J is in agreement with Maia et al. (2014) and also the zero correlation in H is not significantly different from the estimates by Maia et al. (2014) given the standard errors associated with both estimates.When the genetic correlations were estimated assuming both traits to be binomial using a TM the genetic correlation was also close to zero in H but slightly more antagonistic (−0.40) in J.The negative genetic correlation between MR and SR in J is possibly the outcome of a strong positive correlation between MY and MR on the one hand and a strong negative correlation between MY and SR, on the other hand, (Table 3 and 7) compared with the moderate correlations in H.However, restricting the error correlation to 0 which was done to facilitate convergence may also have contribute to the negative or near zero genetic correlations between the traits.
Genetically high MY and poor fertility lead to increased risk of death (Table 3) and the medium to high correlation of EBV for MR with EBV for MY, angularity, body depth and fertility (Table 7) all suggests that death is the outcome of negative energy balance due to higher metabolic demand of high-yielding cows.These results also support that the negative effect of energy balance which may initially affect health and fertility could ultimately lead to increased likelihood of death (Rauw et al., 1998;Ingvartsen et al., 2003;Oltenacu and Broom, 2010).Results at the phenotypic level in several studies (e.g., McConnel et al., 2008;Alvåsen et al., 2012;Shahid et al., 2015) also show that the likelihood of death increases for cows with long previous CIN compared with cows with average or below average CIN.The genetic correlation between MR and CIN in the current study are slightly stronger than the genetic correlation between fertility and MR (−0.45) reported by VanRaden et al. (2016).This show that fertility in pasture-based dairy production system is a key trait with significant effect on culling decisions (Wondatir Workie et al., 2021).
While high genetic potential for 305-d MY leads to increased likelihood of death which agrees with literature (e.g., Dematawewa and Berger 1998;Rauw et al., 1998;Tokuhisa et al., 2014;Tsuruta et al., 2017) low genetic potential for PY in H and lower PY and FY in J increase the chance of being removed from the herd due to slaughter showing the value attached to PY in the Australian economic indexes (Pryce et al., 2009;Richardson et al., 2022).In both breeds, the genetic correlation estimates show that cows with high level of SCC early in the lactation and poor fertility have a higher likelihood of being slaughtered agrees with the literature (e.g., Haile-Mariam et al., 2003).Although a higher early SCC leads to an increased likelihood of SR, it does not appear to increase the likelihood of death (Table 3).This could be because we used only the first test-day SCC for the (co)variance analyses (Table 3).Correlations between EBV for MR and Ln SCC over the whole lactation in Table 7 at least for J breed agree with VanRaden et al. ( 2016) who reported a correlation of 0.25 between EBV for MR and SCS based on US data.
The genetic correlation between 305-d MY and MR estimated in the current study are more unfavorable than estimates based on US herds which varied from −0.01 to 0.33 in different US regions (Tokuhisa et al., 2014;VanRaden et al., 2016;Tsuruta et al., 2017).However, part of the reason for the difference in the estimates by Tokuhisa et al. (2014) and Tsuruta et al. (2017) and that of the current study may be related to the data edit procedures.In the current study cows with termination and calving data even without MY data were included.The finding that the direction of the genetic correlation between MY and MR is positive and the residual correlation is negative agrees with Dematawewa and Berger (1998) who concluded that management practices may keep MR of high producing cows lower even though those cows have lower genetic potential for survival than cow with low production.This also agrees with Alvåsen et al. (2012) who observed that MR in cows with higher MY are lower than cows with lower yield.This could also mean healthy cows produce more milk and avoid death, but unhealthy cows may stay in the herd producing less milk until they die due to poor health.
The data on cow MR and SR are not explicitly used for genetic evaluation of survival in Australia (Madgwick and Goddard. 1989;Visscher and Goddard 1995).The current study demonstrated that these data can be used for improving genetic evaluation of survival.The genetic correlation estimates among milk yield, MR, SR, and fertility traits (Table 3) and the correlation of EBV for MR with traits such as MY, type traits, fertility, and survival.Table 7 suggests that these data can also be used to identify early predictors of cow health, welfare and possibly resilience (Hine et al., 2019).The mortality data rather than the slaughter data could be more useful to develop EBV for resilience because it is closely related to functional traits such as health and fertility (Hine et al., 2019;Schuster et al., 2020;Schmidtmann et al., 2021;Bengtsson et al., 2022).The current mortality data can potentially be used for identifying new more heritable traits that can be measured earlier in life that may have stronger genetic correlation with animal health, welfare, and reproduction and overall resilience (Hine et al., 2019;Schuster et al., 2020;Schmidtmann et al., 2021;Bengtsson et al., 2022).Examining the genetic correlation of MR and SR with indicators of resilience (e.g., Poppe et al., 2020) could be an area of future research.
One of the objectives of the current study was to assess the feasibility of genetic evaluation for MR in addition to the current genetic evaluation for cow survival that is based on overall culling including cow mortality.Although the economic cost of MR is higher than SR the incidences in Australian dairy herds are still low.The accuracy of prediction based on PA is lower for MR compared with SR (Table 6).Although there is a potential to increase accuracy using genomic data, as demonstrated by VanRaden et al. (2016) the potential to increase reliability by using genomic data are likely to remain low given the size of the reference population in Australia.Furthermore, in H the strong selection pressure on fertility and survival and less pressure on MY means the genetic and phenotypic trend for MR is improving which agrees with the trend in the US (VanRaden et al., 2016).The genetic trend in J has also stabilized in recent years which means genetic evaluation for MR may not be a priority in Australia.Nevertheless, collection of a more comprehensive data on cow termination including cow mortality can be used for monitoring welfare, health and resilience of cows particularly if it can be used with some early relatively inexpensive predictors.

CONCLUSIONS
Death losses in Australia dairy herds have increased from below 1% in lactations that started in 1990 to up to 4% in recent years which are still low compared with similar dairy industries around the world.The h 2 of MR is about 1%.Milk yield has unfavorable genetic correlation (0.32-0.41) with MR in H and J.The genetic correlation of fertility with MR is stronger than with SR.High early test-day SCC is genetically associated with increased likelihood of slaughter.The genetic correlation of removal of cows from the herd due to MR and SR is negative (−0.3) in J but zero in H. Overall, strong selection for improved fertility and survival and less selection emphasis for MY, has led to improved genetic trend for MR in H and stabilized trend in J. Genetic evaluation for cow MR is feasible but the scope for selecting on MR EBV is lower than on SR EBV.The continued genetic evaluation for survival based on mortality and slaughter data could be sufficient in the current selection circumstances where breeding objectives are broadly defined.However, all Australian farmers should be encouraged to carry more complete recording of termination data for continuous monitoring of the trend in MR and SR.The data on mortality is a valuable resource for improving animal health and welfare and should analyzed frequently.

Figure 1 .
Figure 1.Mortality and slaughter rate by DIM classes for Holstein (H) and Jersey (J) breeds.

Figure 2 .
Figure 2. Genetic trend in cow mortality based on Jersey (J) and Holstein (H) bulls with 25 or more records per progeny.
Haile-Mariam et al.: GENETICS OF COW MORTALITY AND SLAUGHTER RATE

Table 1 .
Haile-Mariam et al.: GENETICS OF COW MORTALITY AND SLAUGHTER RATE Description of the data of mortality (MR) and slaughter (SR) rate used for estimating genetic parameters and genetic evaluation for Holstein and Jersey breeds

Table 2 .
Incidence levels (%) and heritability (h 2 ) of mortality and slaughter rate from linear (LM) and threshold (TM) models for all parities and first 3 parities in Holstein and Jersey from single trait analyses ± SE

Table 3 .
Genetic and residual correlations of mortality and slaughter rate with production and functional traits in Holstein and Jersey cattle estimated from linear and threshold models ± SE 1 305-d milk yield (MY), fat yield (FY), and protein yield (PY).2The earliest test daily milk yield.3Theearliest test daily SCC (natural log).

Table 4 .
Haile-Mariam et al.: GENETICS OF COW MORTALITY AND SLAUGHTER RATE Genetic correlations (± SE) between mortality (above diagonal) and slaughter rates (below diagonal) in the first 3 parities for Holstein and Jersey breeds

Table 5 .
Means (SD) of EBV for mortality and slaughter rate for Holstein and Jersey bulls calculated using linear animal model from multivariate analyses with 305-d milk yield (from all breed analyses) and univariate analyses of each trait 1Between EBV from multivariate and univariate models for mortality and slaughter.

Table 6 .
Correlations between parent average (PA) for mortality and slaughter rate (i.e., excluding their daughter's data) and solutions predicted for bulls (Sol) based on the mortality and slaughter data of their daughters by number of progeny/records

Table 7 .
, in Australia it was noted Haile-Mariam et al.: GENETICS OF COW MORTALITY AND SLAUGHTER RATE Correlations 1 between EBV for mortality and slaughter rate and EBV of some other economic traits for Holstein and Jersey bulls 2