Genetic evaluation for stillbirth and preweaning mortality in Australian dairy cattle

The welfare of calves is important to both farmers and consumers. Practices that increase the proportion of calves born alive and enable them to thrive through to weaning contribute to improved sustainability. Stillbirths (SB) are calvings where the calf dies at birth or within 24 h after birth. Preweaning mortality (PWM) refers to calves that die after the first day of life but before weaning based on termination data. Both SB and PWM are binary traits characterized by low heritability. Data collection for these traits is incomplete, compared with traits such as milk yield in cows. Despite these challenges, genetic variation can be measured and used to produce breeding tools, such as EBVs, to reduce calf mortality over time. The aim of this study was to compare the performance of various linear models to predict SB and PWM traits in Holstein and Jersey cattle and evaluate their applicability for industry-wide use in the Australian dairy industry. Calving records from


INTRODUCTION
Healthy calvings producing healthy calves initiate new lactations and produce the next generation of replacement heifers, which are both vital to sustainable and enduring dairy farm businesses.Farmers and consumers share an interest in healthy calves, as their welfare is important.Ideally, calves are born alive and without much human interference, but this is not always possible.Calvings that produce calves that die at birth or shortly thereafter are described as stillbirths (SB).The SB rate is estimated to be 3% to 10% in dairy cattle depending on parity and breed of the dam and sex of the calf (Cuttance et al., 2017;Cuttance and Laven, 2019;Sigdel et al., 2022).In a detailed study of Iranian dairy cattle, Bahrami-Yekdangi et al. (2022) reported that the main effects associated with SB were calf sex, gestation length, hypocalcemia at calving, calf birth weight, calving season, twins, dry period length, and parity.This list highlights that a successful calving actively involves both calf and dam.Consequently, we define the traits separately, as SB direct and SB maternal to potentially select for calves that are born alive and dams that deliver live calves.
Studies in Denmark, Israel, and the United States have shown that calf mortality is higher in the first 2 mo of life (Norberg et al., 2013;Neupane et al., 2021;Weller et al., 2021).Furthermore, mortality is lower during the calf and heifer rearing period compared with SB and is estimated to be 3% to 6% losses in dairy cattle depending on the breed and period of measure (Fuerst-Waltl and Sørensen, 2010;Cuttance et al., 2017;Abuelo et al., 2019;Neupane et al., 2021).Disease is considered the most significant cause of mortality in dairy calves and is influenced by colostrum management, feeding protocols, and housing, among other factors (Verdon, 2021;Dairy Australia, 2022).Heifer mortality is reported to vary between rearing periods with the probability of death being higher early in life (Hansen et al., 2003;Neupane et al., 2021;Weller et al., 2021).For this reason, calf loss during the period before weaning is the focus of this study and is described as preweaning mortality (PWM).Preweaning mortality is a novel trait, as it is routinely evaluated in fewer countries compared with stillbirth.However, there are examples of calf survival EBVs in Nordic and US dairy genetic evaluations (Norman et al., 2020;Council on Dairy Cattle Breeding, 2024;Nordic Cattle Genetic Evaluation, 2024).
Data availability challenges the feasibility of genetic evaluation for calf mortality, especially PWM.In many countries, including Australia, farmers are not required to record calvings, calf identity, or terminations, and this can lead to incomplete recording or poor-quality records.Despite data complexity, sire selection that includes genetic evaluations for calving ease (CE) has reduced the proportion of cows and calves experiencing calving difficulties (Cole et al., 2020).Some genetic evaluation units have extended their services to successfully include SB and measures of calf livability with heritability estimates between 1% and 7% depending on trait, breed, parity, and model.For example, Interbull genetic evaluation forms report the heritability of SB direct in Holsteins to be 0.05 in USA and 0.01 in Germany (Interbull, 2022).From these same forms, the heritability of SB maternal is generally higher (0.07 in the United States and 0.03 in Germany).These estimates are similar to the summary compiled by Cole et al. (2007) that suggested a typical range of 0.01 to 0.05 for SB direct and 0.01 to 0.06 for SB maternal.For calf survival, recent research reports heritability estimates of between 0.01 and 0.06 in the United States and Israel, with the higher estimates coming from models that include a genomic relationship matrix instead of a numerator relationship matrix (Gonzalez-Peña et al., 2019;Neupane et al., 2021).
In Australia, heifer calves are typically weaned at 8 to 12 wk of age, and the period between birth and weaning is considered to be a period of high disease risk (Dairy Australia, 2022).Furthermore, a substantial group of surplus Australian heifers are sold from the herd at around 6 mo of age to meet a specific market demand.There is a lack of precision in recording termination reason, so, to avoid inadvertently including sold heifers with dead heifers, this study focused on the preweaning period.
The aim of this work is to estimate genetic parameters and EBV for SB direct, SB maternal, and PWM for Australian Holstein and Jersey cattle using is available in DataGene's Industry for Good Centralized Data Repository (https: / / www .datavat.com.au/).

Data
Farmer-recorded calving data, calf identity, pedigree, culling records, and EBV used in this study were obtained from DataGene Pty Ltd., Melbourne, Australia.Calving events and calves born between January 1, 2000, and April 14, 2022, were collated.Both SB and PWM are distinct binary traits and were coded in this dataset as 0 for dead and 100 for alive so that higher EBV of the trait are desired and estimates are produced as a percent that is easy to understand.Farmers record SB observations along with other traits such as CE and calf size (CS) when the calving event is entered into the herd management system, and this is routinely exchanged with the Centralized Data Repository.Preweaning mortality was estimated using recorded termination dates because weaning dates and other life events were poorly recorded in precalving females.In a pragmatic approach, PWM was defined as calves born alive with a termination date before d 84 (12 wk) of life.
Calving ease scores have evolved over time and vary between herd management programs (McClintock, 2004;Eaglen et al., 2012), so recoding was required.Data for CE were unified to a single system as illustrated in Table 1.Due to the low frequencies of moderate difficulty, high difficulty, surgery, and malpresentation, these scores were grouped, leaving 3 levels.Dystocia was defined as levels 2 and 3.
Calf size was scored as tiny, small, average, big, or huge.Calf weights were not available, as this information is not routinely recorded for Australian dairy herds.Due to the low frequency of tiny and huge calf sizes, CS scores were grouped to 3 levels so that 1 is tiny or small, 2 is average, and 3 is big or huge, as described by McClintock et al. (2003).To enable some testing of alternative CS phenotypes to explore the nonlinearity of this trait's optimum, CS was recoded using 5 strategies: (A) 1 = tiny/small, 2 = average, 3 = big/huge (B) 0 = non-average, 1 = average (C) 1 = tiny/small, 2 = average, big/huge removed (D) Tiny/small removed, 2 = average, 3 = big/huge (E) Applying estimated calf weights as 35 kg = tiny/ small, 40 kg = average, 45 kg = big/huge We also obtained the main pedigree file from DataGene, and it was pruned to include animals related to those in the dataset.The number of animals in the trimmed pedigree was 1,080,458 with records up to 19 generations.

Data Editing
The availability of well-recorded calving observations is a key challenge in the genetic evaluation of calf-related traits, and this is highlighted by the proportion of records retained for this analysis.A series of edits was applied to assign variables and remove unexpected records such as calvings from cows more than 12 yr of age and herdyear-season (HYS) where all calvings were recorded on the same date, suggesting lower-quality data recording.A total of 54% of the calvings produced female calves.Only data for female calves were retained, mainly because the quality of female recording is higher and termination dates were available that were later required for PWM analysis.To avoid using biased data from HYS where SB and calf losses were not well recorded, only HYS were included with calvings that produced at least 1 alive and 1 dead calf for SB.Similarly, for PWM, HYS with at least 1 alive and 1 dead calf were used in each analysis.We acknowledge that this is a harsh edit, considering the mean PWM is only 2%, but it is vital to ensure that data used for this study are from herds that provided goodquality data and we could be confident that calf losses were being recorded.For example, if females were first recorded at the time of first calving, which is common in Australia, it is impossible to know whether some females from that HYS died as calves.Calvings producing multiple calves were excluded, to be able to compare results with similar studies (Hansen et al., 2004;Pryce et al., 2006;Fuerst-Waltl and Sørensen, 2010).A minimum of 10 observations per sire and 5 per HYS was applied to reduce cases of singularities.The proportions of retained records after each data cleaning step are described in Table 2.
In total, 354,689 Holstein calving records in 9,634 HYS groups with 4,648 sires with both CE and fate records were used for the SB analysis.There were an additional 73,525 Jersey calving records in 3,491 HYS groups with 777 sires.A subset of these data contain-  (177,864) were Holstein and the remaining 14% (29,612) were Jersey.

Genetic Analyses
Both the calf and the dam contribute to a successful calving, and farmers would likely want to breed calves that can survive and in future become cows that deliver live calves.Animal models were used to estimate the genetic effects for SB direct (trait of the calf) and SB maternal (trait of the cow).
Models.Stillbirth, recorded as a binary trait (dead or alive) and CE, grouped into 3 levels (no assistance, slight assistance, and moderate or high assistance) were first analyzed separately using animal models.Both direct and maternal models were applied.
The estimated variance components in univariate linear models were used as starting values for bivariate linear models.The general bivariate direct and maternal models used were, respectively, and where y was the vector of phenotypic records for each trait (y 1 was SB: 0 for dead and 100 for alive calves; and y 2 was CE); b was the vector of fixed effects of HYS and parity group; u (or m) was the vector of animal, sire (or dam) random effects related to the pedigree structure depending on the model; c was the vector of random permanent environment effects of dam, but the variance explained by this effect was zero and therefore dropped from the model; e was the vector of random residual effects; and X, Z, Q, and W were design matrices relating calving observations to the corresponding effects.The subscripts associate the vectors or matrices to trait 1 (SB) or 2 (CE).We assumed the distributions for the random effects in the direct and maternal models and the residual terms in these 2 models to be, respectively, and , where σ u 2 in the animal (and sire) model corresponds to the animal (and sire) genetic variance; σ m 2 corresponds to maternal genetic variance; σ describes the covariance between different effects; A was the additive relationship matrix calculated from the pedigree information; and I was the identity matrix.The subscripts associate the variances and covariances to the effects as well as to traits 1 and 2.
Preweaning losses were lower than the stillbirth rate, the dataset was smaller, and the maternal effect was less prominent compared with SB, so a sire model was used, ignoring the maternal effect.The same general model and fixed effects as SB were used for PWM, except trait 1 was PWM and trait 2 was CE.
Calf size was added as a third trait in a multivariate analysis, with SB and CE, using an expanded version of the general model above.The bivariate models were used to estimate starting values for the multivariate models.Calf size is an important contributor to SB and was added so that breeding for fewer stillbirths while we include CE as trait 2 will not result in unintended consequences, such as very small calves.Additionally, where SB are not recorded well, including CS and CE data could be useful in the genetic evaluation of SB.
Some further models were tested to better understand the genetic relationships between the calving traits in our study.For example, a univariate animal model incorporating maternal genetic effects was used to estimate the significance of the covariance between the direct additive and maternal effects for SB.Second, a bivariate sire model incorporating SB and PWM was used to estimate the correlation between these 2 traits.
The data for Holstein and Jersey breeds were analyzed separately, as there was limited across-breed genetic con-nectedness in the pedigree file and we suspected that susceptibility to disease and the effects of CE and CS may be underlying contributors and could be different across breeds.For all models, version 4.2.1.188 of ASReml was used to calculate variance components, genetic and phenotypic correlations, heritability, and EBV (Gilmour et al., 2015).
Reliability of Predictions.The reliability of genetic prediction for SB and PWM was calculated using the standard errors of EBV, as where PEV i is the prediction error variance (squared error of the EBV i for animal i in the pedigree), and σ u 2 is the estimated genetic variance in the prediction model.The EBV and reliabilities for SB and PWM in Holstein and Jersey breeds were reported for sires with an EBV reliability of at least 0.40, so that low reliability EBV for sires with little or no progeny performance information or poor pedigree relatedness were removed from further analysis, noting that sires with EBV reliability of 0.4 had approximately 40 calving observations in the dataset.
Coefficient of Genetic Variation.To compare the potential for the genetic improvement of calf traits against other traits and other populations, the coefficient of genetic variation (CV%) was calculated as where σ a 2 was the additive genetic variance and X̅ was trait average, which was calculated for a binary trait as with the upper and lower boundaries, minimum (min) = 0 and maximum (max) = 1, and X̅ was the trait mean (Burdon, 2008).

Approximate Genetic Correlation Between Calf Survival EBV and a Selection Index.
Records for dead calves are not available for traits in a selection index of total economic merit, so it is difficult to directly calculate the genetic correlation between the early-life traits we present here and traits later in life (Weller et a., 2021).As we wanted to understand the consequence of selecting on Australia's national selection index, the Balanced Performance Index (BPI), this relationship was approximated using a Pearson correlation adjusted for the reliability of both EBV and selection index.The EBV for SB direct, SB maternal, and PWM for sires with reliability ≥0.40 were compared with April 2022 published EBV obtained from DataGene Pty Ltd.This date was selected as the closest official genetic evaluation to the date of data extract, to preserve a close alignment of data used in this analysis and the official evaluation.Sire EBV for SB and PWM were correlated with sire BPI using the tidyverse package v2.0.0 (Wickham et al., 2019) in RStudio 2022.02.4 (RStudio Team, 2022) with R v4.0.3 (R Core Team, 2020).To correct the correlations for the reliabilities, we applied the method of Calo et al. (1973): where rb,t is the approximate genetic correlation between BPI and each calf trait; REL bi and REL ti are the reliabilities for BPI and calf trait for sire i, respectively; and r b,t is the Pearson correlation between BPI and each calf trait EBV.Genetic Trends.The mean EBV, range, and SD of a subset of sires with reliability ≥0.40 in the dataset obtained by grouping by year of birth were used to investigate the genetic trend for Holstein and Jersey breeds.

Calving and Stillbirth Records
Farmers are contributing high-quality calving records that include calf fate for more than 700 herds each year.This is half of the number of herds that were recording 20 years ago; however, the total numbers of herds and cows in Australia have fallen by 64% and 36% respectively in the same period (Dairy Australia, 2021).In recent times, one-quarter of the cows recorded in herd recording schemes have calving records that included CE, CS, and SB.

Incidence of Stillbirth
Almost 7% (6.8%) of calvings were reported to produce a dead calf in Australia, higher for male calves (8.2%) compared with females (5.3%) and higher in calves born from Jersey dams (8.2%) compared with Holstein dams (6.4%).This study focused on the fate of female calves from Holstein and Jersey dams born in the past 20 yr, as described in more detail in Table 3.In this edited dataset, 3.9% of females were stillborn.
Calves experiencing dystocia during birth were at greater risk of being stillborn, with almost one-third of Holstein female calves with dystocia being born dead.Primiparous Holstein cows were twice as likely to pro-duce a stillborn female calf, compared with multiparous Holsteins (7.1% compared with 2.9%).However, SB were more evenly distributed across primiparous and multiparous Jersey cows.Holstein and Jersey female calves that were born either big/huge or tiny/small had higher SB rates compared with average-sized calves, at 7.5%, 7.8%, and 3.4% respectively.

Incidence of Preweaning Mortality
Australia's pasture-based dairying environment leads to compact calving periods.For example, almost onequarter of the calves in this analysis were born in the month of August.Despite the hectic nature of intensive calving seasons and the pressure this can place on calf rearing staff and facilities, 98% of recorded calves lived at least to weaning, as shown in Table 4. Losses were greater in Jerseys (4.1%) compared with Holsteins (1.9%).Compared with SB, we observed smaller differences in PWM between calves born to primiparous and multiparous dams.
Although these broad descriptions of the data are interesting, especially for farmers, the relatively low incidence of SB and PWM combined with incomplete recording of calf traits means that further data editing and modeling are required to detect differences that may be used for genetic selection to reduce SB and PWM.In this next section, results from genetic models are presented to compare predictions for individuals after adjustments are made for important fixed effects in each trait.

Heritability of Stillbirth and Preweaning Mortality
The heritability of SB direct, SB maternal, and PWM varied by model and between breeds but are broadly in line with results from similar studies (Neupane et al., 2021;Interbull, 2022).The differences are most easily described on a trait-by-trait basis, as shown in Table 5.
Stillbirth Direct.The heritability for SB direct using a bivariate linear animal model that included SB and CE was 0.015 ± 0.002 for Holstein and 0.024 ± 0.005 for Jersey.Bivariate SB direct animal models perform similarly to univariate animal models but have the advantage of increasing the amount of information available to the model and are expected to have more accurately estimated parameters in less heavily edited datasets that are commonly used by genetic evaluation centers.The heritabilities of Holstein SB direct bivariate animal model and univariate animal models were 0.015 ± 0.002 and 0.014 ± 0.002, respectively.Some bivariate and multivariate models that added CS as trait 2 or 3 resulted in unexpected correlations, suggesting that the association between CS and SB or CE was complex.In Holsteins, regrouping CS using 2 levels (non-average = 0 and average = 1) produced positive phenotypic (0.085 ± 0.001) and genetic (0.284 ± 0.065) correlations with SB in a bivariate model.For Jerseys, this method of regrouping produced mildly negative correlations, whereas using 3 levels of CS (tiny/small, average, and big/huge) produced neutral phenotypic (0.007 ± 0.001) and strongly positive genetic (0.463 ± 0.063) correlations.Both genetic and phenotypic variances of SB were higher when CS was included.The associations between these traits are complex mainly because intermediate size is favorable, which may explain why the multivariate animal model, including CS, had a lower SB heritability (0.009 ± 0.001) compared with the bivariate model (0.015 ± 0.002) in Holsteins as well as in Jerseys (0.004 ± 0.001 compared with 0.024 ± 0.005).
Stillbirth Maternal.In the population that included primiparous and multiparous dams, the heritability estimates for SB maternal using a linear bivariate maternal model that included SB and CE, were 0.012 ± 0.001 and 0.005 ± 0.002 in Holstein and Jersey respectively.Inter- estingly, heritability estimates were similar when SB is considered as a maternal trait (0.012 ± 0.001) rather than a direct trait of the calf (0.015 ± 0.002) for Holsteins, but they are lower in Jerseys (0.005 ± 0.001 compared with 0.024 ± 0.005).For Holsteins, the heritability estimates were higher for models that include only primiparous dams (0.068 ± 0.008) compared with those that include all dams (0.012 ± 0.001), but these same differences were not observed in Jersey dams, suggesting that CE in Holstein dams is a major contributing factor to SB, especially when the dam is a heifer.Preweaning Mortality.Despite the smaller dataset and lower incidence of PWM, the bivariate linear sire model that included PWM and CE returned heritability estimates of 0.021 ± 0.003 for Holstein and 0.050 ± 0.013 for Jersey, and these results were similar to the univariate sire model results.
Heritability estimates and genetic and phenotypic correlations between CE and each trait are presented in Table 5.For both SB direct and SB maternal, the relationships with CE traits were moderately strong and in the expected direction.For example, as calving becomes more difficult risk of SB increased.This genetic relationship is most convincing in Holstein animals (−0.700 ± 0.039) compared with Jerseys (−0.398 ± 0.140).When the maternal genetic effects were added to the model, the indicative genetic correlation between SB direct and SB maternal varies by breed, where the relationship is slightly positive in Holsteins (0.225 ± 0.137) but negative in Jersey cattle (−0.453 ± 0.212), although these results should be treated with caution due to high standard errors associated with the estimated correlation coefficients.Further, the goodness of fit of models with and without considering genetic covariance between maternal and direct SB was compared using a likelihood ratio test, suggesting that there is no difference and the covariance can be ignored (i.e., it is not different from zero; Self and Liang, 1987).
In the case of PWM, the correlations with CE were also difficult to interpret with confidence due to the large standard errors.Our confidence in the genetic correlations is highly affected by the heritability of calf traits, which is low.Although it may seem obvious, it is worthwhile to note that calves that died during calving do not have a record in the PWM dataset.Many of the dead calves would have had difficult calvings, which may also influence the correlation estimate between CE and PWM.

Stillbirth and Preweaning Mortality Predictions
Sire EBV were estimated using bivariate animal models for SB direct and maternal and bivariate sire models for PWM.The SD of sire EBV vary between 0.67 and 3.28 for 3 traits in 2 breeds, as shown in Table 6.As each 1% change in EBV represents a 1-percentage-point difference in deaths, genetic variation is sufficient to expect dairy farmers and breeding program managers to be interested in these traits.The mean reliabilities for all Holstein sires are 0.37, 0.37, and 0.25 for SB direct, SB maternal, and PWM, respectively.In Jersey sires, the mean reliabilities are 0.34, 0.20, and 0.30 for SB direct, SB maternal, and PWM, respectively.

Coefficient of Genetic Variation
The CV% is a function of the additive genetic variance and frequency of calf deaths and was calculated for each trait in this study on a scale of 0 to 1, using an equation developed specifically for binary traits (Burdon, 2008).In Holsteins, the CV% was 11.7 and 14.5 for SB and PWM compared with 15.4 and 23 for Jerseys as shown in Table 7.As this coefficient is most useful when compared with other traits and populations, Table 7 also reports similar coefficients for health traits from another study for comparison purposes.

Correlation Between Stillbirth and Indexes
It is helpful to understand the relationship between the new calf traits in this analysis and indexes commonly used in breeding programs so that antagonistic relationships are identified and to assist in explaining genetic trends for calf traits.Multitrait selection indexes were not directly included in the models, and dead calves do not have records used to evaluate traits in selection indexes, so an approximation was necessary.Table 8 shows the estimated correlations between calf trait EBV and Australia's national index, the BPI, for sires with EBV reliability of at least 0.40.In the case of SB direct, this is equivalent to about 40 calving observations (n = 1,149 Holstein and 181 Jersey sires).The most notable observation from this analysis is the mainly positive relationship with BPI that suggests current selection practices are favoring improved calf traits.It is likely that this simple correlation is overestimating the genetic correlation and that potential bias may exist due to direct or indirect selection (genetic trend in both traits), so caution is required when interpreting these results.Additionally, it is possible that preferential treatment of genetically superior animals could bias the EBV estimates of calf traits.Further research is required to understand the strength of these relationships.

Genetic Trends for Calf Traits
Consistent with the estimated genetic correlations with BPI, the genetic trends suggest that sires were improving for both SB traits, despite the lack of specific EBV to enable direct genetic selection.Compared with 20 years ago, Figures 1 and 2 show that Holstein sires born in 2018 produced on average 2 fewer stillborn calves per 100 and daughters that delivered 2 fewer stillborn calves per 100 calvings.Based on average sire EBV for each year of birth, virtually no genetic improvement in PWM was detectable during the same period in Holsteins, as illustrated in Figure 3.No obvious trends occurred for calf traits in the Jersey breed, so similar graphs are not reported here.

DISCUSSION
This study provides new insights into the incidence of dairy calf mortality in a pasture-based dairying environment and adds to the global knowledge base of genetic variance in dairy calf health traits and relationships between calf traits.In the past 2 decades, the incidences of SB and PWM have been 3.9% and 2.2% in female dairy calves with Holstein or Jersey dams in Australia.The genetic CV% for SB and PWM in Table 7  Holstein (11.7 and 14.5) and Jersey (15.4 and 23.0) suggests that considerable genetic variation in calf health traits exists, and it is likely that the proportionally high residual variability contributed to the lower heritability result (Berry, 2018).The CV% was similar for calf traits and previously reported health traits.As health traits are commonly considered in breeding programs, the low heritability should not deter industry efforts from also breeding for improved calf health traits.Consistent with Yao et al. (2014), the Holstein genetic trends for SB direct and SB maternal have improved since the year 2000.The genetic trend for PWM was stable, suggesting that current selection practices have no obvious effect on calf survival through the rearing period to weaning and that any changes to phenotypic calf survival are a function of preweaning management.Although we set the reliability threshold to be higher than 0.4, we should note, the lower average reliability of EBV for PWM could cause the genetic trend to become less obvious in PWM compared with SB.

Recording
Close to one-fifth of Australian dairy farms contributed usable SB records in 2022 for at least some cows, but only 6% of Australian cows have SB records of dead or alive calves.This under-recording is likely due to competition for labor during the intensive seasonal calving seasons common to Australia's pasture-based dairy system, the low value placed on this information by farmers, as no management reports are available to monitor and improve performance, and a sense of helplessness when it comes to preventing stillborn calves.The proportion of cows with SB records varies by herd size, with more frequent records coming from smaller herds.This underrecording of SB offers opportunities to make consider-  8. Pearson correlation between calf trait EBVs and Australia's national selection index, known as the Balanced Performance Index (BPI) for sires with EBV reliability ≥0.40 and estimated genetic correlations using Calo's formula (Calo et al., 1973) Breed EBV  able increases to the number of observations available for analysis.One strategy could be an investigation of the data collection hurdles, specifically on large farms that already contribute some data.Targeting a defined group of farms that are already connected to industry databases could be a practical approach to identifying and prioritizing activities that target data recording practice change.
In a review of perinatal mortality in dairy cattle, Cuttance and Laven (2019) provide strong evidence of lack of standardized recording practices of calf traits.Data from current systems is not of sufficient quality to be sure that groups of animals are completely recorded; for example, it can be unclear whether only live calves in a particular HYS were reported in the data.This can lead to severe data editing to ensure evidence of both live and  dead calves in each group, as was the case in this study, and could mean the dataset is not as representative as it otherwise could have been.
Collaborative efforts between farmers and industry that improve the recording of SB are consistent with global dairy industry efforts to promote sustainability and improve standards of animal welfare.Furthermore, Verdon (2021) recently reported significant consumer interest in calf welfare and highlighted the need for improved measurement and monitoring of calf health and welfare metrics.Recording and monitoring SB and other calf health records will provide valuable industry insights to help progress these priorities as well as enable more reliable calf trait genetic evaluations.

Stillbirth
Stillbirth rates in this dataset were within the ranges of previously published studies mentioned in the Introduction.The benefit of reduced SB to calf welfare is quite clear; however, it is also useful to consider the benefit to the dam.Bicalho et al. (2007) found decreased pregnancy and increased median days open in dams that delivered a stillborn calf.There was evidence that milk yield in early lactation was lower following a SB, at least in younger cows (Berry et al., 2007;Eaglen et al., 2011).Although the underlying mechanism is not well understood, this could be associated with higher rates of SB because of dystocia, which has been reported to negatively affect milk production and fertility of the dam (Barrier and Haskell, 2011).Multiple approaches to modeling the direct and maternal effects of SB are possible, as discussed by Hansen et al. (2004), but we chose to treat the effects as 2 separate traits that could later be combined into an index with separate economic values.Using the selected bivariate animal model, the heritabilities of the 2 SB traits are similarly low in Holsteins.The heritabilities of SB direct (0.015 ± 0.002) and SB maternal (0.012 ± 0.002) for Australian Holstein cattle were within the ranges reported in other countries once the parity of the dam was included as a fixed effect in the model (Jamrozik et al., 2005;Cole et al., 2007;Neupane et al., 2021).Fewer studies of SB in Jerseys have been published, but the heritability estimates (0.024 ± 0.005 direct, 0.005 ± 0.002 maternal) in this analysis were similar or slightly higher than an earlier Australian analysis by Scott (2017) where a sire-maternal grandsire model was used.Consistent with Scott (2017), we found that the heritability of the maternal SB trait in Jerseys is markedly lower than in Holsteins.One reason for this could be the lower incidence of dystocia in Jerseys, suggesting that SB in Jerseys is primarily a trait of the calf rather than the cow.It has been suggested that SB in Jersey cattle is influenced by inbreeding, and this is supported by Scott (2017), who found a significant positive relationship between genomic inbreeding and SB that was similar in scale to the effect of a moderate level of calving difficulty.It will be important for farmers with Jersey cattle to increase the recording of calving data so that calving trends and inbreeding can be carefully monitored over time.
In this analysis, we focused on stillborn dairy calves born to Holstein and Jersey dams because this provided the most data.Many countries have seen an accelerated growth in the use of sexed semen to produce herd replacements and beef semen to produce a terminal beef-cross calf that is desirable to the market and reduces the number of low-value dairy bull calves (Pereira et al., 2022).Both strategies are likely to change the rate of SB.There is a paucity of available data related to CE and SB in calves sired by beef-breed bulls from dairy-breed cows.The lack of available data will make trend monitoring difficult, which is why the progeny of beef-sired calves born from dairy dams should also be recorded.

Preweaning Mortality
The PWM heritability estimates in this study (0.021 ± 0.003 Holstein, 0.050 ± 0.013 Jersey) were higher than estimates for the period of d 5 to weaning reported in Holsteins by Henderson et al. (2011) but within the ranges for survival to older ages more commonly found in the literature.Researchers reported heritability in Holsteins of 0.007 ± 0.001 for heifer livability from d 2 to first lactation in the United States (Neupane et al., 2021) and 0.011 for survival from d 3 to d 365 in the Netherlands (van Pelt et al., 2012).Lynch et al. (2022) reported the heritability of calf disease (not death) within the first 6 mo of life to be 0.01 to 0.04 for diarrhea and respiratory problems in Canadian dairy calves.
Farmers may argue that the large environmental variances observed in SB and PWM mean that better calf care is an easier way to improve both traits on a dairy farm.Historically, this argument could have also been applied to other traits such as daughter fertility, where substantial genetic gains have been made (Berry et al., 2014).However, we have shown that the genetic CV% is similar to health traits, such as mastitis resistance (Abdelsayed et al., 2017;Table 7) that are already part of breeding programs.Although feedback from farmers implies this is a trait they would like to improve (by any means possible), encouraging farmers to collect data may improve calf care, with the added benefit of contributing to better genetic predictions.
Should additional data become available, 4 strategies could improve the predictions of PWM and potentially preweaning morbidity: (a) prepare predictions based on the disease agent, such as calf scours or pneumonia, as proposed by Lynch et al. (2022); (b) consider calf sur-vival over a longer period of time, such as from d 2 to first lactation, as proposed by Neupane et al. (2021); (c) apply a user-friendly scoring system for recording calf vitality, as previously proposed by the authors (Axford et al., 2022b); and (d) standardize calf recording, as advocated by Cuttance and Laven (2019).

Trait Relationships
Stillbirth is a complicated trait, as both calf and dam factors contribute to the arrival of a live calf.Although the fetus may initiate parturition, as discussed in more detail by Eaglen et al. (2012), there is a complex web of physiological and physical processes in action that involve both fetus and dam.However, attempting to understand the relationships that exist between traits is useful to predict the potential trade-offs between traits and inform strategies for inclusion in breeding programs.
Calf Traits.A difficult calving is considered a painful experience for both calf and cow and, if calf death occurs, it is usually shortly before, during, or a short time after parturition.Meijering (1984) explains that hypoxia causing severe acidosis will reduce calf vitality and ultimately cause calf death during some difficult calvings.Especially in Holsteins, SB direct and SB maternal were strongly and negatively correlated with CE (approximately −0.7), indicating a strong link between easier calvings and more live calves.These estimates were similar in direction but higher than those reported by Eaglen et al. (2013) in first-parity Holstein calvings (−0.38).
Dystocia is more common in first parity compared with later parities; however, the estimated genetic correlation between CE in parity 1 and parity 2 is high in Australian Holsteins (0.85 for direct and 0.93 for maternal CE; M. Haile-Mariam, Agriculture Victoria, Australia, unpublished data), implying that CE can be considered as the same trait across parities.These are higher than correlations found in other countries; for example, 0.75 (direct) in Israel (Weller and Ezra, 2016) and 0.80 (direct) in the United Kingdom (Eaglen et al., 2013).Norman et al. (2009) reported very high correlation (≥0.96) between predicted transmitting abilities for gestation length based on heifer and cow breedings, despite differences between the heritability of each trait.
Calving ease EBV is commonly included as an independent trait in on-farm breeding programs, especially in heifers, to reduce dystocia.Fewer SB occur with easier calvings, so we would expect a trend toward lower incidence of SB as the genetic trend for CE has increased, favoring easier calvings (DataGene, 2023).However, the incidence of SB has remained stable in female calves and increased in male calves in recent years (Axford et al., 2022a), suggesting that more factors are involved.In an earlier Australian study, McClintock et al. (2003) previously reported a very high correlation between CS and dystocia in mature cows (0.81), and the incidence of SB is similar between tiny/small calves and big/huge calves, so we expected CS to be a useful trait to include in the multitrait model, but this was not the case in the current analyses.The models did not perform better by adding CS, even after experimenting with different methods of categorizing size, suggesting that adding CS into the model is not straightforward.It is generally not desirable to have calves that are too big or too small.A large calf may be difficult for the dam to deliver, but a small calf may be more challenging to successfully rear, and both large and small calves have a higher incidence of SB compared with normal-sized calves.Breed differences in the genetic relationship between SB and CS were observed after applying multiple grouping strategies to this dataset.Based on the genetic correlations between these 2 traits, CS correlation with SB is complex.In Holsteins, average is more favorable than tiny/small or big/huge size.However, in Jersey, a higher risk of SB occurs in tiny/small calves compared with big/huge calves.Calf size is a subjective score in some countries, including Australia, with little guidance offered to farmers on how to classify CS.It is unclear from these results whether the genetic variation in size is already accounted for by CE.By not accounting for CS in the model, it is possible that selection with SB EBV could lead to an increase in the proportion of calves with desirable average size because of correlated responses.New on-farm initiatives to record calf birth weights would enable further analysis that could resolve this question.
The relationship between SB direct and CE in Jersey cattle is not consistent with Holsteins, suggesting possible breed differences.The correlation is moderately negative (−0.398 ± 0.140), as expected.However, the direction is different for SB maternal, where a cow experiencing dystocia is more likely to deliver a live calf (0.558 ± 0.219).A high standard error is associated with these correlations, likely the result of lower incidence of dystocia combined with a smaller dataset, but this is an indication that factors beyond CE influence SB.Large Jersey calves are more likely to be born alive but may increase dystocia.In contrast to Holstein and Brown Swiss, where SB increased with CE score, Yao et al. (2014) found that SB was similar for Jersey calvings described as needing assistance, requiring considerable force, or characterized by extreme difficulty.A further possibility could be differences in management.For example, Jersey cattle are assumed to be easy calvers, and calvings may not be monitored in the same way that Holstein calvings are monitored, so the consequences of dystocia could be more severe due to delay in assisting the calving.These subtle management differences are difficult to isolate from "true" genetic effects.The results in this study were consistent with other studies of Jersey animals, such as Olson et al. (2009) who advocate for more complete recording of SB in Jerseys to better understand breed differences.Scott (2017) found that genomic inbreeding was a strong predictor for SB in Jersey cattle and warrants further study.Another Australian study reported genetic variance in the gestation length of Holsteins, with direct heritability of 0.28 in heifers and 0.36 in cows, and this trait could also influence SB (Haile-Mariam and Pryce, 2019).In all cases, increasing the quantity and quality of calving and fate records is important, especially for Jerseys.
The genetic relationships between PWM and CE and PWM and SB direct were unclear due to the high standard errors for both Holsteins and Jerseys.
In this study the genetic correlation between SB direct and SB maternal suggests a positive relationship in Holsteins and negative relationship in Jerseys.However, these results require caution when interpreting, as there was no significant difference between models that fit or ignore the covariance between the maternal and direct effect.
Calf Traits and Balanced Performance Index.Based on estimated genetic correlations, selecting animals with higher BPI favors improved direct and maternal SB in Holsteins, presumably through correlated responses with economically important traits such as cow survival that have a heavy weight in the index (Byrne et al., 2016).However, the relationships between BPI and SB traits in Jersey cattle were not as convincing, despite cow survival having the same economic weight in BPI for both breeds.At least in Holsteins, the sire EBV correlations were much stronger than expected, which suggests possible preferential treatment of calvings involving dams and calves of high genetic merit.In a tight calving pattern, where labor resources are limited, animals are often treated homogeneously.However, it is possible for groups within a season to be treated differently.For example, early-born calves may be housed in a cleaner environment and receive milk for a longer period than late-born calves within the same HYS.This would have an effect if early-born calves were the progeny of high-BPI parents.In recent times, the increased use of fixedtime AI programs with sexed semen in heifers may mean that a greater proportion of high-BPI calves are born earlier in the calving period, and it is possible that this may bias the results.It is illogical to suggest that preferential treatment would vary by breed, so these factors are unlikely to explain breed differences.Further studies of multivariate models including calf survival and cow survival, multi-breed models that include crossbred animals, and a larger dataset of Jersey animals would all be useful to explore this in more detail.

Application of Genetic Predictions for Calf Traits
The EBV for SB and PWM from this study are useful for monitoring genetic trends and ranking sires; however, the reliability (~0.2-0.4) is far lower than other traits that are commonly used in breeding programs.As breeding decisions have long-lasting consequences in a herd, it is important that farmers and breeding program managers can be confident to use EBV in sire selection, and, at this level of reliability, it is unlikely that these 3 calf traits will be well accepted for individual sire selection.Additional research is required before release in an official evaluation, and we recommend that (a) genotypes of live and dead calves be added to evaluation models, to be able to produce genomic EBV; (b) genomic EBV be validated on an independent dataset; (c) a multitrait model that includes correlated traits be tested to improve the accuracy of prediction; and (d) the genotypes of animals be explored to find specific haplotypes that may cause SB, especially in Jerseys.
Despite the lower heritability, the distribution of SB and PWM EBV suggests potential to improve calf welfare by lowering the incidences of SB and PWM through genetic selection.These data suggest that a modest strategy of selecting bulls that are 1 SD above the mean for SB direct would produce 1% or almost 1,000 fewer stillborn calves in Australia each year.The impact of SB maternal is less immediate because the benefits appear once the sire's daughters calve for the first time, but the trait has the advantage of being expressed repeatedly with each consecutive calving event.Alternatively, avoiding bulls with extremely poor SB EBV is also expected to have a positive effect on calf health.For example, there were 12 Jersey bulls with EBV 3 SD below the mean that were predicted to produce a mean of 7.6% more SB.If the overall breeding objective is to improve profit per cow per year, as is the case with the BPI, the benefit of breeding for improved calf health needs to be offset by any reduction in genetic gain for profit, which can be achieved by correctly accounting for these traits in an economic index (Byrne et al., 2016).Despite the incomplete recording of calvings and calf health events, these results suggest the existence of sufficient measurable genetic variance to warrant continued industry efforts toward increasing the quantity of calving and calf health data for genotyped animals and exploring options to include calf traits in dairy breeding programs.

CONCLUSIONS
The incidences of SB and PWM in female dairy calves with Holstein or Jersey dams in Australia over the past 2 decades are reported at 3.9% and 2.2%, respectively.The genetic variation, as indicated by the CV% in Holsteins , suggests potential for improvement in calf health traits.Calving ease is strongly correlated with SB in Holsteins, where SB direct and maternal exhibit a strong and negative correlation with CE (approximately −0.7).The relationship with CS appears to be breed dependent, with average-sized calves being more favorable in Holsteins, whereas Jersey SB was higher in smaller calves compared with larger calves.However, in Australia improvement in data recording practices is needed, to facilitate routine genetic evaluations.

NOTES
This research was funded by DairyBio (Melbourne, Australia), an initiative of Dairy Australia (Melbourne, Australia), Agriculture Victoria (Victoria, Australia), and the Gardiner Foundation (Melbourne, Australia), and contributed to a PhD project at La Trobe University (Melbourne, Australia).The authors thank DataGene (Melbourne, Australia), herd recording centers, and dairy farmers for recording calving data used in this research.The authors appreciate the input of anonymous peer reviewers, whose advice improved the clarity of the manuscript.Because no human or animal subjects were used, this analysis did not require approval by an Institutional Animal Care and Use Committee or Institutional Review Board.The authors have not stated any conflicts of interest.
Axford et al.: GENETIC EVALUATION FOR CALF MORTALITYTable

Figure 1 .
Figure 1.Genetic trend for stillbirth direct EBVs for Holstein sires with reliability ≥0.40.Median (midline) and upper and lower quartiles marked by the box, whiskers denote 1.5 × interquartile distance, and dots are outliers.
Figure 2. Genetic trend for stillbirth maternal EBVs for Holstein sires with reliability ≥0.40.Median (midline) and upper and lower quartiles marked by the box, whiskers denote 1.5 × interquartile distance, and dots are outliers.

Figure 3 .
Figure 3. Genetic trend for preweaning mortality EBVs for Holstein sires with reliability ≥0.40.Median (midline) and upper and lower quartiles marked by the box, whiskers denote 1.5 × interquartile distance, and dots are outliers.

Table 1 .
Axford et al.:GENETIC EVALUATION FOR CALF MORTALITY Unifying 2 calving ease scoring systems

Table 2 .
Effects of data cleaning and editing tasks on the number of records in the dataset

Table 3 .
Axford et al.: GENETIC EVALUATION FOR CALF MORTALITY Reported rates of stillbirth, dystocia, and calf size in Australian female dairy calves with Holstein and Jersey dams, 2000-2021 (n = 903,900 calves)

Table 5 .
Axford et al.:GENETIC EVALUATION FOR CALF MORTALITY Estimated heritability ± SE (diagonal; SE in parentheses), genetic correlation ± SE (upper triangular), and phenotypic correlation ± SE (lower triangular) in Holstein and Jersey breeds using a linear bivariate animal model for stillbirth (SB) and calving ease (CE), and a linear bivariate sire model for preweaning mortality (PWM)

Table 6 .
for bothAxford et al.:GENETIC EVALUATION FOR CALF MORTALITY Mean EBV and reliability for 1,192 Holstein and 181 Jersey sires where trait 1 is stillbirth (SB) direct, stillbirth maternal or preweaning mortality and trait 2 is calving ease (CE) and sire stillbirth direct reliability ≥0.40

Table 7 .
Examples of coefficient of genetic variation percentage (CV%), genetic standard deviation (SD), and heritability (h 2 ) for various binary traits (recorded as 0 and 1 scores) in dairy cattle Axford et al.: GENETIC EVALUATION FOR CALF MORTALITY