Associations of cow-and herd-level factors during the dry period with indicators of udder health in early-lactation cows milked

This observational study aimed to determine the association of cow-level factors and herd-level housing and management practices during the dry period with indicators of udder health in early-lactation cows in automated milking system (AMS) herds. Data were collected from 166 commercial AMS dairy farms (mean ± standard deviation = 116 ± 111 milking cows; range = 39 to 1,200) across Canada between October 2018 and September 2020. Information on herd demographics, housing, and management practices was obtained on each farm using 2 surveys. On each farm, we selected all cows that had available Dairy Herd Improvement (DHI) somatic cell count (SCC) data for their last milk test before dry-off (>250 d in milk) and their first milk test after calving (5–45 d in milk). Data from 14,007 cows were included after excluding cows with a dry period of <30 d and >120 d. Using the SCC data, we calculated for each cow the somatic cell score (SCS) for the last milk test before dry-off (PreSCS) and the first milk test after calving (PostSCS), which we then averaged per herd at a test-day level. Intramammary infection (IMI) was estimated using cow SCC data. Each cow was classified as not infected (SCC <200,000 cells/mL) or infected (SCC ≥200,000 cells/mL) at her last milk test before dry-off and her first milk test after calving. Based on this classification, cows were further categorized as never infected, always infected, new IMI, or cured IMI. At the cow level, a higher PostSCS was associated with longer dry periods. The odds of having a new IMI were higher for cows of higher parity and that had lower 305-d milk yield before dry-off. Cows with lower parity were more likely to cure an IMI. At the herd level, a higher 305-d milk yield before dry-off was associated with a lower incidence of new IMI and a higher incidence of cured IMI. Separating cows into a different pen as preparation for dry-off tended to be associated with a lower PostSCS and incidence of new IMI. At dry-off, herds that used teat sealants and blanket antibiotic dry cow therapy also had lower PostSCS. During the dry period, housing cows in different groups was associated with a higher PostSCS and a lower incidence of cured IMI, while housing cows in both pack pens and stalls compared with only pack pens was associated with a lower incidence of new IMI. Finally, placing cows onto the AMS to be milked one or more days after calving tended to be associated with a lower PostSCS compared with placing them in the AMS within the first day postpartum. In summary, indicators of udder health in early-lactation cows in AMS herds were associated with several cow-level factors and herd-level housing and management practices before dry-off, at dry-off, during the dry period, and at the beginning of lactation. Thus, if some of the associations identified are causal, AMS producers may be able to improve udder health through modifications of housing and management practices


INTRODUCTION
During the dry period, dairy cows are at high risk of developing IMI (Halasa et al., 2009) caused mainly by environmental pathogens (Klaas and Zadoks, 2018), which increases their risk of clinical mastitis in the following lactation (Bradley and Green, 2004).Mastitis, both in its subclinical and clinical forms, causes substantial economic losses for dairy producers due to decreased milk yield, increased culling, discarded milk, and the costs associated with implementing preventive strategies (Aghamohammadi et al., 2018).Furthermore, it reduces cow welfare (Klaas and Zadoks, 2018), which is a topic of public concern (Wolf et al., 2016).Thus, a strategy frequently used for preventing IMI during the dry period is the use of dry cow antibiotic therapy at dry-off (Winder et al., 2019a).However, in recent years and due to concerns related to antimicrobial resistance, it has been recommended to reduce the use of antimicrobials in the dairy industry and prevent infections using other approaches, such as improving hygiene and implementing other management practices at the farms (WHO, 2016).
Several practices around the dry period have been previously associated with early-lactation udder health in dairy cows.The use of dry cow antibiotic therapy (Winder et al., 2019a;Kabera et al., 2021) and teat sealants at dry-off (Dufour et al., 2011;Rabiee and Lean, 2013;Pearce et al., 2023) are the most investigated strategies (McMullen et al., 2021).However, management practices before dry-off, such as the method employed for milk cessation (Vilar and Rajala-Schultz, 2020), and during the dry period, including daily checkups of the udders for mastitis detection (Dufour et al., 2011), type of bedding and its disinfection (Green et al., 2007), type of housing (Astiz et al., 2014), and the frequency with which pens are cleaned after cows calve (Dufour et al., 2011) have also been associated with indicators of udder health in early-lactation cows.At the same time, udder health is also influenced by cow-level factors such as parity (Dingwell et al., 2004), milk yield at (Rajala-Schultz et al., 2005) or near (Newman et al., 2010) dry-off, SCC in the previous lactation (Niemi et al., 2021) and before dry-off (Green et al., 2007(Green et al., , 2008)), number of infected quarters before dry-off (Sol et al., 1994), and dry period length (Pinedo et al., 2011;van Hoeij et al., 2016).
It is clear that early-lactation udder health is influenced by both cow-level factors and herd-level housing and management practices, not only during the dry period but also before dry-off, at dry-off, and at calving (Green et al., 2007(Green et al., , 2008)).More recently, however, it has been suggested that the risk of developing IMI during the dry period is influenced to a greater extent by herd-level factors and management practices than by cow-level factors (Klaas and Zadoks, 2018).In addition, most studies that have investigated this topic have been conducted in herds with conventional milking systems, and to our knowledge, only a few studies have examined the associations between cow-and herd-level factors during the dry period and early-lactation udder health in herds using automated milking systems (AMS; e.g., Niemi et al., 2020Niemi et al., , 2021)).Given that the adoption of technology and the use of AMS is increasing worldwide (Barkema et al., 2015), with 17% of all Canadian dairy herds milking cows with AMS in 2021 (Canadian Dairy Information Centre, 2021), there is a need to identify modifiable management practices that may influence the risk of IMI during the dry period in AMS herds.
Thus, the overall objective of this study was to determine the association of cow-and herd-level factors during the dry period with indicators of udder health in early-lactation dairy cows milked by AMS.Our first specific objective was to assess whether cow-level factors at dry-off (e.g., parity, breed, milk yield) were associated with indicators of udder health in early lactation.Our second specific objective was to associate herd-level housing factors and management practices in preparation for dry-off, at dry-off, during the dry period, around calving, and at the beginning of lactation with indicators of udder health in early-lactation cows in AMS herds.

MATERIALS AND METHODS
A retrospective cohort study was conducted using data from commercial AMS dairy farms collected between October 2018 and September 2020.These data were collected as part of a larger study undertaken to benchmark the herd-level housing and management strategies of AMS farms across Canada and examine their associations with milk production and quality (Matson et al., 2021).The study design was approved by the University of Guelph Animal Care Committee (AUP#3963), University of Guelph Research Ethics Board (REB 19-01-012), University of Calgary Research Ethics Board (REB19-0414), and University of Saskatchewan Research Ethics Board (Beh ID 1305).

Farm Recruitment and Cow Selection
A detailed description of farm recruitment is reported by Matson et al. (2021).Briefly, early in 2019, AMS producers across Canada were contacted through emails sent by the researchers in partnership with the national DHI organization (Lactanet Canada; Sainte-Anne-de-Bellevue, Quebec), to inquire whether they were interested in enrolling in the study.Farms were selected if they met the following criteria: (1) they were milking with an AMS, (2) they were enrolled in DHI milk recording, and (3) the farm owner or manager was willing to participate in the study and granted written consent to use their DHI records.In total, 200 AMS farms were originally recruited; however, for the present study we only used 166 farms, as some producers did not provide information about routine management practices of dry cows, withdrew from the DHI milk recording program during the study period, or lacked DHI SCC information for more than 50% of their cows.
For each farm during the study period, we selected all cows that had available milk production information for 2 milk tests: a milk test before dry-off and a milk test after calving.To do this, we chose (1) the last milk sample before dry-off that had available SCC data and was obtained when the cows were above 250

DIM, and
(2) the first milk sample after parturition that had available SCC data and was obtained when the cows were between 5 and 45 DIM.Because the study was conducted over a period of time longer than 1 yr, some cows had SCC information for more than one dry period.In these cases, we only used data for those cows from their most recent lactation, as this fit better with the time in which the second survey was conducted, which had most of the questions related to dry cow management.Initially, we had data available from 14,848 cows.However, cows with a dry period of less than 30 d (n = 291) and more than 120 d (n = 548) were excluded from the study, as those dry periods were considered too short and too long, respectively, for a normal dry period (Niemi et al., 2021).Data from 14,007 cows were included in our study after applying our exclusion criteria.The number of farms and cows available for this study was based on sample size calculations by Matson et al. (2021); therefore, a priori sample size and power analyses were not conducted for our specific objectives.

Herd-Level Management Practices Survey
Each farm was visited once between April 2019 and September 2019 by research staff from the University of Calgary (Calgary, AB, Canada), the University of Guelph (Guelph, ON, Canada), or the University of Saskatchewan (Saskatchewan, SK, Canada).During this visit, a structured survey was conducted with the farm owner or herd manager to collect information on herd demographics, housing, and routine management practices (as per Matson et al., 2021).A second followup survey was conducted online between May 2020 and August 2020 by researchers of the University of Guelph to determine changes in management practices during the first year of the study and collect additional information about dry cow management and housing.
Producers were asked in the surveys about routine management practices before dry-off, at dry-off, during the dry period, around calving, and at the beginning of lactation.Surveyed management practices before dry-off included whether cows were separated into a separate pen or another area as preparation for dry-off, which was categorized as yes or no.Dry-off management practices surveyed included whether producers used any type of treatment when they detected cases of IMI during dry-off, which was categorized as yes or no.Producers also reported the production level (kg/d) at which cows were typically dried off.Regarding the dryoff protocol, the use of dry cow antibiotic therapy was recorded and classified as blanket or nonblanket, which included farms that used selective dry cow antibiotic therapy and farms that did not use dry cow antibiotic therapy at dry-off.In addition, the use of teat sealants (i.e., including either internal or external teat sealants) at dry-off was recorded and classified as yes or no.Finally, producers were asked about the timing of dryoff as a function of the days of pregnancy; that value was subtracted from 280 d to calculate the target dry period length (number of days) for each farm.
The surveyed routine management practices during the dry period included the strategy used to group the dry cows, which was classified as either all together or by stage of dry period (i.e., far-off and close-up groups housed separately).Producers also indicated whether the cows stayed in the same group for the entire dry period (i.e., yes or no).Housing type was recorded and categorized as pack pens, stalls, and both.Bedding type was classified based on its composition as organic and inorganic (i.e., sand).Bedding management was also recorded and included the frequency in which new bedding was added (categorized into at least once a day, once a week, and once a month).The method used to ventilate the barn was categorized into ceiling fans, panel or basket fans, tunnel ventilation, or natural ventilation.Finally, the number of hours in which dry cows were exposed to supplementary light every day was recorded.
The surveyed management practices around calving included whether cows gave birth on the farms in a group pen, individual pen, or both (i.e., farms had both a group pen and individual pens).Information provided for early-lactation management practices included whether cows were placed on the AMS to be milked immediately (i.e., within the first day postpartum) or after the first day postpartum.

DHI Data, SCS, ΔSCS, and IMI Classification
The DHI data were collected for every enrolled cow on each farm between 6 mo before the farm visit until the end of the study.On average, we collected data for 642 ± 54.7 d (mean ± SD) for each farm, ranging from 385 to 701 d.These data provided information on SCC, milk yield (kg/d), and DIM on the day of the test, projected 305-d milk yield (kg), parity, breed, and dry period length for each cow.The data for each of these variables were averaged by herd at a test-day level (i.e., for the last test before dry-off and for the first test after calving).
For each cow, we calculated the SCS for the last milk test before dry-off (PreSCS) and the first milk test after calving (PostSCS), as log 2 (SCC/100,000) + 3 (Wiggans and Shook, 1987), where SCC was cells/mL of milk.Based on these SCS, we calculated for each cow the change in SCS during the dry period (ΔSCS) as follows: PostSCS − PreSCS (where a positive value indicates an increase in SCS across the dry period).We then calculated the average PreSCS, PostSCS, and ΔSCS for each farm.
Subclinical IMI was estimated using individual cow SCC measurements from the DHI data.Each cow was retrospectively classified according to her SCC as not infected (SCC <200,000 cells/mL) or infected (SCC ≥200,000 cells/mL; Dohoo and Leslie, 1991;Schukken et al., 2003).This was performed both at her last test before dry-off and her first test after calving.Based on this classification, we further categorized cows as (1) never infected, when a cow was not infected in any of those tests, (2) always infected, when a cow was infected in both tests, (3) new IMI, when a cow was not infected in her last test before dry-off, but was infected on her first test after calving, and (4) cured IMI, when a cow was infected in her last test before dry-off, but was not infected on her first test after calving.We then calculated the incidence risk of new IMI and cured IMI, separately, for all cows and each herd during the dry period.Incidence risk of new IMI was calculated as the number of cows with new IMI/number of cows at risk, where cows at risk was the sum of cows never infected and cows that developed a new infection.Similarly, the incidence risk of cured IMI was calculated as the number of cows cured/number of cows at risk, where cows at risk was the sum of cows that cured and cows that were always infected (Dohoo et al., 2009).

Statistical Analysis
Statistical analyses were performed with SAS (version 9.4; SAS Institute Inc.).Outcome variables were assessed at the cow and herd level, depending on our objectives with cow and farm, respectively, as the experimental units.Before analysis, all data were screened for normality and identification of outliers with the UNIVARIATE procedure.At the cow level, data for parity at dry-off and DIM at the last milk test before dry-off were right-skewed, so we categorized cows according to parity 1, 2, and ≥3 and DIM at the last milk test before dry-off as <305 or ≥305.
Separate multivariable models were built to address each of our objectives.All models were built using a similar methodology.Degrees of freedom for the fixed effects were estimated using the Kenward-Roger option in the MODEL statement.As a first step, univariable models were used to screen all explanatory variables as fixed effects.Only variables with P < 0.25 were retained for potential inclusion in the multivariable models (Dohoo et al., 2009).Correlations between kept continuous explanatory variables were assessed with the CORR procedure of SAS.Pearson correlation coefficients were calculated to detect collinearity.If 2 variables were highly correlated (r > 0.80), the one with the lower P-value in the univariable analysis or the most biologically plausible relationship was kept in the multivariable model.Associations between kept categorical variables were examined with the FREQ procedure of SAS.Multivariable models were then built including all selected explanatory variables.Manual backward elimination was used to remove variables with P > 0.1.Only variables with P ≤ 0.1 remained in the model and were considered significant at P ≤ 0.05 and tendencies at 0.05 < P ≤ 0.1.When performing backward elimination, confounding variables (i.e., a nonintervening variable that when removed from the model resulted in a ≥25% change in the estimate coefficient of an individual retained variable) were to be kept in the multivariable model.However, no confounding variables were detected during that step.Finally, for all remaining variables in the model, we examined 2-way interactions if biologically plausible.For all final models, we visually examined the normality of the residuals of each outcome and explanatory variable assessed to evaluate model fit.
To identify factors associated with the risk of IMI at a cow level, we performed a multivariable logistic regression analysis using the GLIMMIX procedure of SAS (distribution = binary, link = logit).Two separate models were built to assess our outcome variables of interest (i.e., new IMI and cured IMI).Categorical explanatory variables assessed included parity at dryoff, breed (Holstein or non-Holstein), and DIM at the time in which the last milk test was obtained before dry-off.Continuous explanatory variables examined included dry period length, number of days between the last milk test before dry-off and the date of dry-off, milk yield and projected 305-d milk yield on the day in which the last milk test was obtained before dry-off, and DIM at the time of the first milk test after calving.Farm was included in both models as a random effect.Odds ratios with 95% CI are presented for explanatory variables associated with the risk of new IMI and cured IMI.
To examine the association of cow-level variables with ΔSCS and PostSCS, we performed a multivariable regression analysis using the MIXED procedure of SAS.Separate models were built for each of these outcome variables.Both models included farm as a random effect.SCC at the last milk test before dry-off was categorized as low (<200,000 cells/mL) or high (≥200,000 cells/mL) and was included as a covariate in the model for ΔSCS.The covariate PreSCS was included in the model for PostSCS.Explanatory variables assessed were the same as in the cow-level multivariable logistic regression models.Given that the explanatory variables retained in the final multivariable models for PostSCS Multivariable regression analyses, using the MIXED procedure of SAS, were also used to examine the association of the various herd-level explanatory variables assessed in the surveys with herd-average incidence of new IMI, herd-average incidence of cured IMI, herdaverage PostSCS, and herd-average ΔSCS during the dry period.Explanatory variables included in the models are presented in Tables 1 and 2. Models were built separately for each of these outcome variables.In all models, region was treated as a random effect.The proportion of cows with high SCC (≥200,000 cells/mL) at the last milk test before dry-off was calculated for each herd and included as a covariate in the model for herd-average ΔSCS.In all the other models, herd-average PreSCS was included as a covariate.As with the analyses at the cow level, similar herd-level explanatory variables were retained in the final multivariable models for herd-average PostSCS and herd-average ΔSCS.Thus, we decided to report only the results for herdaverage PostSCS.

Descriptive Herd and Cow Statistics
A descriptive summary for the 166 dairy herds using AMS and 14,007 cows selected for this study are presented in Table 2 and 3  (≥90% of all milking cows in the herd were Holstein), whereas in 10.2% (n = 17) of farms, the milking herd was classified as non-Holstein (<90% of all milking cows were Holstein).Non-Holstein herds included Ayrshire (n = 1), mixed Ayrshire (n = 3), Brown Swiss (n = 1), Jersey (n = 7), mixed Jersey (n = 4), and other mixed-breed (n = 1) herds.Of all 14,007 cows included in this study, 13,022 (93.0%) were Holstein and 985 (7.0%) were cows from other breeds (i.e., Ayrshire, Brown Swiss, Canadienne, Fleckvieh, Jersey, and Milking Shorthorn).Based on their SCC at the last milk test before dry-off and first milk test after calving, 11.9% (n = 1,665) of all cows developed a new IMI during the dry period, 15.9% (n = 2,237) of all cows cured an IMI during the dry period, 7.5% (n = 1,045) of all cows were always infected, and 64.7% (n = 9,060) of all cows were never infected.Thus, 10,725 cows were at risk of developing a new IMI, and 3,282 cows had potential incidence of cured IMI.Based on these values, we determined that the incidence of new IMI from before dry-off to after calving in the cows recruited in this study was 15.5%, whereas the incidence of cured IMI was 68.2%.Herd-average SCS data and herd-average incidence (%) of new IMI and cured IMI are presented in Table 2.

Cow-Level Factors Associated with IMI
Cow-level factors associated with the risk of new IMI that were retained in our final multivariable model were DIM at both milk tests, parity, and projected 305d milk yield (Table 4).Cows for which the last milk test before dry-off was performed ≥305 DIM (359 ± 52.5 DIM, mean ± SD; n = 6,826) were 1.13 times more likely to develop a new IMI during the dry period compared with cows that were tested <305 DIM (279 ± 14.7 DIM, mean ± SD, n = 7,181).We also detected that cows in parity 1 and 2 were 2.17 and 1.54 times less likely, respectively, than cows in parity ≥3 to develop a new IMI during the dry period.Furthermore, the odds of a cow developing a new IMI during the dry period were 1.12 times lower for every one-standard deviation (SD) increase (2,422.0kg) in projected 305d milk yield at the last milk test before dry-off.This model also controlled for the DIM at the first milk test after calving; cows were 1.27 times less likely to have a  new IMI for every 1-SD increase (11.0 d) in DIM at the first milk test after calving.Parity at dry-off and DIM at the first milk test after calving were also associated with the incidence of cured IMI (Table 4).Cows in parity 1 and 2 were 1.68 and 1.46 times more likely, respectively, than cows in parity ≥3 to cure an IMI during the dry period.In addition, for every 1-SD increase (11.0 d) in DIM at the first milk test after calving, cows were 1.09 times more likely to cure an IMI.

Cow-Level Factors Associated with PostSCS
Cow-level factors associated with PostSCS that were retained in our final multivariable model were dry period length and DIM at the first milk test after calving (Table 5).When controlling for PreSCS, a longer dry period was positively associated with PostSCS; each 10-d increase in the duration of the dry period was associated with 0.027 higher PostSCS.In contrast, higher DIM at the first milk test after calving was negatively associated with PostSCS, because each 10-d increase in DIM was associated with a −0.25 lower PostSCS.

Herd-Level Factors Associated with IMI
Explanatory variables that were retained in our final multivariable models for herd-average incidence (%) of new IMI and cured IMI are presented in Table 6.When controlling for herd-average PreSCS, housing dry cows in both pack pens and stalls was associated with a −3.89-percentage point (p.p.) lower herd-average in-cidence of new IMI as compared with housing dry cows only in pack pens.In addition, every 1,000-kg increase in herd-average projected 305-d milk yield on the last milk test before dry-off was associated with a −1.2 p.p. lower herd-average incidence of new IMI.In contrast, not separating cows into a different pen or area as preparation for dry-off tended to be associated with a 2.65 p.p. higher herd-average incidence of new IMI.
In the case of herd-average incidence of cured IMI when controlling for herd-average PreSCS, not housing cows in the same group for the entire dry period was associated with a −7.67 p.p. lower herd-average incidence of cured IMI as compared with herds in which cows remained in the same group during the entire dry period.Furthermore, every 1,000-kg increase in herd-average projected 305-d milk yield on the last test before dryoff was associated with a 2.4 p.p higher herd-average incidence of cured IMI.

Herd-Level Factors Associated with PostSCS
Explanatory variables retained in the final multivariable model for herd-average PostSCS are presented in Table 7.When controlling for herd-average PreSCS, not using a teat sealant at dry-off was associated with a 0.19 higher herd-average PostSCS as compared with using teat sealants, whereas using blanket dry cow antibiotic therapy at dry-off was associated with a −0.31 lower herd-average PostSCS compared with not using blanket dry cow antibiotic therapy.Furthermore, not housing cows in the same group for the entire dry period was also associated with a 0.19 higher herd-   New IMI = a cow that was not infected (SCC <200,000 cells/mL) in her last test before dry-off, but was infected (SCC ≥200,000 cells/mL) on her first test after calving. 2 Cured IMI = a cow that was infected (SCC ≥200,000 cells/mL) in her last test before dry-off, but was not infected (SCC <200,000 cells/mL) on her first test after calving.average PostSCS compared with herds in which dry cows remained in the same group.Finally, not separating cows into a different pen as preparation for dry-off tended to be associated with a 0.15 higher herd-average PostSCS compared with moving cows into a separate pen.Similarly, placing cows onto the AMS to be milked within the first day after calving also tended to be associated with a 0.14 higher herd-average PostSCS as compared with placing cows onto the AMS after the first day after calving.

DISCUSSION
Dairy cows are susceptible to developing IMI during the dry period (Halasa et al., 2009).Early-lactation SCC, which is an indicator of these IMI (Schukken et al., 2003), has been with various cow-level factors and housing and management practices during the dry period (Dufour et al., 2011).However, most of this research has been conducted in herds with conventional milking systems.To our knowledge, this is the first study to examine the associations of both cow-level factors and herd-level housing and management practices around the dry period with indicators of early-lactation udder health in commercial AMS herds.

Descriptive Herd and Cow Statistics
As expected, most of the AMS herds enrolled in this study were located in Ontario (41.6%) and Quebec (25.3%), as these are the 2 provinces that hold the largest number of AMS farms in Canada (Canadian Dairy Information Centre, 2021).The average herd size of the enrolled farms was 111 lactating cows, which is similar to (105, King et al., 2016) and larger than (77, Deming et al., 2013;88, Matson et al., 2022) that reported in other studies conducted in Canadian AMS herds.In addition, 93% of the selected cows were Holstein,  Incidence of cured IMI during the dry period (%), calculated as the number of cows cured/number of cows at risk, where cows at risk was the sum of cows that cured and cows that were always infected.
which coincides with the percentage reported by the Canadian Dairy Information Centre (2021) for dairy farms in this country.The incidence risk of new IMI during the dry period at the cow (15.5%) and herd level (16.5%) were similar to the 13% to 16% risk of new IMI reported by Cameron et al. (2014) and Kabera et al. (2020), whereas Rowe et al. (2020) reported a risk of 20% to 21%.In the case of cured IMI, the incidence risk at the cow (68.2%) and herd level (69.7%) were lower than the 82% to 89% risk of cured IMI noted in previous studies (Cameron et al., 2014;Rowe et al., 2020).Differences between studies may be associated with the definitions of IMI used, the moment in which IMI was determined, and the microorganisms causing the IMI.In addition, in the cited studies, the IMI were detected at the quarter level, using bacteriological cultures of milk samples, and in cows from conventional milking systems.Thus, more research is needed to confirm whether the risk of new and cured IMI during the dry period is different in herds with AMS and conventional milking systems.

Cow-Level Factors Associated with Udder Health
Several cow-level factors were associated with indicators of early-lactation udder health in cows milked by AMS.As expected, cows in parity ≥3 were more likely to develop a new IMI during the dry period than cows in parity 1 and 2. This is consistent with research conducted in conventional milking systems, where cows of higher parity were more likely to develop a new IMI during the dry period (Dingwell et al., 2004), potentially due to anatomical changes in the udder, which would allow for easier penetration of pathogens through the teat canal (Dingwell et al., 2004;Green et al., 2007), or lesser immune function due to increasing age (i.e., immunosenescence; Green et al., 2007).Cows in parity 1 and 2 were also more likely to cure an IMI during the dry period than cows in parity ≥3, which is also consistent with previous research (Sol et al., 1994).Older cows may be less likely to cure an IMI due to the size of their mammary gland; as its size increases with age, the dose of the antimicrobial used may be insufficient to clear an infection (Barkema et al., 2006;Samson et al., 2016).Antimicrobial resistance may also play a role (Barkema et al., 2006) because its prevalence is higher in older cows (Sol et al., 2000), but it could also be simply caused by an immune system dysfunction (Samson et al., 2016).
Cows with higher 305-d milk yield before dry-off also had a lower risk of developing new IMI.This is consistent with the results of Pinedo et al. (2012), who suggested that a lower milk yield before dry-off in cows with a higher risk of developing new IMI after calving may indicate poor udder health already in that lactation.It could also be hypothesized that AMS herds with higher milk production at dry-off are better managed to achieve higher production levels.Consequently, im- proved udder health in early-lactation cows with higher milk yield at dry-off could reflect better management.
In contrast, studies conducted in herds with conventional milking systems have reported that IMI after calving are more frequent in cows with higher milk production at (Rajala-Schultz et al., 2005) or near dry-off (Newman et al., 2010), possibly as a result of delayed closure of the teat canal at dry-off due to increased internal intramammary pressure, which may lead to milk leakage and increase the exposure to pathogens (Vilar and Rajala-Schultz, 2020).At the same time, a higher concentration of lactoferrin, which helps prevent IMI due to its antibacterial properties, has been detected at dry-off in cows producing lower levels of milk (Vilar and Rajala-Schultz, 2020).However, other researchers reported no association between postpartum udder health and milk yield on the last test before dry-off (Green et al., 2007) and at dry-off (Natzke et al., 1975) or 305-d milk yield from the previous lactation (Green et al., 2007).Different results may be associated with the definition of IMI used, the moment in which milk yield was measured (i.e., at dry-off vs. at the last milk test before dry-off), the production level of the cows at dry-off, as well as the influence of environmental factors and management practices implemented on the farms (Green et al., 2007;McMullen et al., 2021).In addition, differences between studies conducted in farms with AMS and conventional milking systems may be associated with housing facilities.Farms that adopt AMS build new barns, retrofit their facilities to install the robots, or do a combination of both.Thus, dry cow facilities may also have been improved when farms transition to an AMS, leading to better overall facilities for transition cows, which may be associated with improved udder health and production.However, to our knowledge, this has not yet been investigated and, therefore, requires further examination.We detected that a longer dry period was associated with higher PostSCS.Pinedo et al. (2011) similarly reported higher odds of subclinical mastitis after calving, determined based on logSCC values, in cows with longer dry periods.It is possible that with longer dry periods, cows that are already susceptible to developing a new IMI during this time are at increased risk of infections because exposure to environmental pathogens is longer (Natzke et al., 1975).This may be more problematic in herds in which cows tend to have longer dry periods, as it can lead to higher stocking density in the dry cow pen, which has been associated with worse hygiene (Creutzinger et al., 2021) and higher SCC after calving, possibly due to increased contamination of the environment and exposure to pathogens (Green et al., 2008).Furthermore, the concentration of the dry cow antibiotic may decrease by the time of calving in cows that have longer dry periods, and therefore, its efficacy to prevent IMI around parturition may also decline (Weber et al., 2021).However, more studies are needed to confirm these hypotheses, especially because researchers have reported opposite results; early-lactation cows that had a short or no dry period had increased (van Hoeij et al., 2016) or similar (Church et al., 2008) SCC.Different results may be due to sample size and other cow-level factors that may have a greater effect on early-lactation udder health than the dry period length itself, such as parity and milk production at dry-off (Green et al., 2008).Moreover, the use of dry cow antibiotic therapy may have influenced some previous results (McMullen et al., 2021), where the use of dry cow antibiotics was confounded with having a dry period (van Hoeij et al., 2016).
Cows for which the last milk test before dry-off was obtained ≥305 DIM were also more likely to develop a new IMI during the dry period compared with cows tested <305 DIM, which could have been associated with lactation length.Although we did not use lactation length as an explanatory variable in our analyses due to the lack of information on the date of dry-off for several cows, the DIM at the last milk test before dry-off could be indicative of this; cows tested ≥305 DIM were dried off at 379 ± 54.2 DIM (n = 5,909), whereas cows tested <305 DIM were dried off at 305 ± 18.7 DIM (n = 6,186).Niemi et al. (2021) similarly reported that cows with a longer lactation were more likely to have SCC ≥200,000 cells/mL at their first milk test after calving.In contrast, Ma et al. (2022) extended the cows' voluntary waiting period to prolong their lactation and noted no effect of lactation length on the proportion of cows with cured and new IMI after calving and on the occurrence of high SCC in both the last milk test before dry-off and the first milk test after calving.Alternatively, it has been hypothesized that a longer lactation may help reduce the risk of new IMI in the following lactation as a result of lower milk yield at dry-off (Burgers et al., 2021).Therefore, more research is needed to examine the effect of lactation length on udder health during the dry period and subsequent lactation in AMS herds.
Finally, higher DIM at the first milk test after calving was associated with lower PostSCS, which was expected because the immediate postpartum period is known to be characterized by high levels of SCC in noninfected quarters that decline within 2 wk after calving and may also be indicative of the high rate of IMI observed in postpartum cows (Barkema et al., 1999).In the present study, cows were less likely to have a new IMI and were more likely to cure an IMI as the DIM at the first milk test after calving increased.One possible explanation for these findings is that cows with IMI at calving may have been treated by the AMS producers at that time and would have, therefore, had more days to clear the infection when their first milk test was obtained further away from calving.Consequently, cows with new IMI would have been less likely to be diagnosed as such and would have been more likely to be categorized as never infected when they were tested later.At the same time, infected cows at dry-off that were still infected at calving would have had more days to clear the infection before we estimated subclinical mastitis and, thus, would have been more likely to be categorized as cured IMI.

Herd-Level Factors Associated with Udder Health.
In line with the results of our cow-level analyses, a higher herd-average 305-d milk yield before dry-off was associated with a lower herd-average incidence of new IMI.In addition, a higher 305-d milk yield before dryoff was associated with a higher herd-average incidence of cured IMI.It could be hypothesized that herds with lower 305-d milk yield before dry-off and lower incidence of cured IMI may have been herds where late-lactation cows had multiple quarters infected (Barkema et al., 2006) or had several cows with IMI caused by major pathogens (Gonçalves et al., 2018) or with chronic IMI, which may lead to fibrosis in the udder affecting its capacity to synthesize milk (Gonçalves et al., 2020) and reducing its response to treatment (Barkema et al., 2006).
Before dry-off, separating cows into a different pen tended to be associated with a lower herd-average postSCS, and a lower herd-average incidence of new IMI.One possible reason for this finding is that AMS producers who separate cows into different pens in preparation for dry-off may implement other management practices that are associated with improved udder health during the dry period.For example, they may move cows into a different pen to change their lactation diet to an energy-restricted diet or the dry cow diet, which is an effective method of gradually reducing milk yield before dry-off (Vilar and Rajala-Schultz, 2020), and has been associated with decreased milk leakage during the first days after dry-off and a tendency for reduced risk of IMI during the dry period (Tucker et al., 2009).
At dry-off, using blanket dry cow antibiotic therapy was associated with lower herd-average PostSCS compared with using nonblanket antibiotic therapy.Similarly, Niemi et al. (2021) reported lower SCC in postpartum cows that received blanket dry cow therapy than selective therapy.Sigmund et al. (2023) also noted a higher reduction of SCC after calving in cows that received dry cow antibiotic therapy compared with cows not treated at dry-off.In contrast, Vanhoudt et al. (2018) detected no changes in herd-level SCC dynamics during the dry period after changing blanket for selective dry cow therapy in the Netherlands.A recent review and meta-analysis also noted no differences in SCC after calving when blanket or selective dry cow therapy in combination with an internal teat sealant in noninfected quarters were used at dry-off (Kabera et al., 2021).Different results among studies may be associated with the study population, the criteria used to select which quarters and cows need to be treated, how the selective treatment is applied (i.e., at the quarter or cow level), the use of teat sealants at dry-off, and the definition of IMI used and its prevalence before dry-off (Dufour et al., 2011;Winder et al., 2019a;Kabera et al., 2021).Furthermore, in the present study, only a few herds did not use dry cow antibiotic therapy.For this reason, we combined them in our analyses with the herds that used selective dry cow therapy.Consequently, we compared herds using blanket dry cow therapy against herds that used selective or no antibiotic therapy at dry-off, which could have led to the higher PostSCS we detected in the herds that did not treat all quarters and all cows.In the case of IMI, we did not observe an association between using blanket dry cow therapy and the incidences of new and cured IMI.We had expected to detect a lower incidence of new IMI in herds using blanket dry cow therapy, as this strategy has been associated with a lower IMI incidence (Dufour et al., 2011;Winder et al., 2019a).It is possible that those herds that did not use blanket dry cow therapy were herds where cows had good udder health at dry-off (Vilar et al., 2018) and that used adequate criteria for selecting which cows did not need to receive dry cow antibiotic therapy, implemented good management practices at dry-off, and had a high level of hygiene (Vanhoudt et al., 2018;Weber et al., 2021).
Not using teat sealants at dry-off was also associated with higher herd-average PostSCS.Similarly, lower SCC (Godden et al., 2003) and linear scores (LS; Runciman et al., 2010) have been reported in cows treated with an internal teat sealant plus dry cow antibiotic therapy than in cows treated with only antibiotics.However, other researchers have noted no associations between SCC and the use of internal teat sealants in herds that used blanket, selective, or no antibiotic therapy at dryoff (Niemi et al., 2020) or between LS and the use of internal teat sealants in combination with dry cow antibiotic therapy compared with the use of antibiotics on its own (Rabiee and Lean, 2013).Thus, further studies should be designed to examine the effect of using teat sealants at dry-off-when administered alone and in combination with dry cow antibiotic therapy-on SCC after calving.Contrary to what we expected, we did not detect an association between using teat sealants and the incidence of new IMI.According to 3 meta-analyses, using internal teat sealants alone (i.e., without using antibiotic therapy) helps prevent new IMI during the dry period (Rabiee and Lean, 2013;Dufour et al., 2019;Pearce et al., 2023).Differences between our results and the ones reported in these studies may have been caused by the type of teat sealant used, because the AMS producers in the present study used either internal or external teat sealants.Although both types protect the teat end and prevent bacterial access to the udder during the dry period (Winder et al., 2019b), their effectiveness in preventing IMI differs (Lim et al., 2007), and we were unable to differentiate between them in our analyses.Furthermore, our definition of new IMI may have also influenced our results, as it was based on changes in SCC between milk tests and not on actual bacterial presence.
During the dry period, herds in which cows did not stay in the same group had a higher herd-average postSCS and a lower herd-average incidence of cured IMI than herds in which cows remained in the same group.To our knowledge, the effect of housing cows in the same or different groups during the dry period on early-lactation udder health has not been previously investigated.However, because cases of environmental mastitis are higher during the dry period and around calving (Klaas and Zadoks, 2018), our results may be associated with the hygiene of the environment (Green et al., 2007;McMullen et al., 2021).It could be hypothesized that AMS producers that have different groups of dry cows may have more difficulties maintaining good environmental hygiene in all pens, as they would need to clean and disinfect more areas.In addition, differences in space allowance could have influenced our results, as providing less space in prepartum pens has been associated with worst cow hygiene (Creutzinger et al., 2021) and higher SCC after calving (Green et al., 2008).However, Sol et al. (1994) reported that farm hygiene did not influence the bacteriological cure rate of IMI during the dry period in cows treated with dry cow antibiotic therapy.It is also possible that stress caused by regrouping could affect udder health because it can increase cortisol concentrations (Silva et al., 2013), which leads to immunosuppression (Aleri et al., 2016) and has been associated with increased SCS in lactating cows (Ebinghaus et al., 2020).However, we did not examine these factors in our study; thus, the impact that regrouping, space allowance, and cow and environmental hygiene during the dry period may have on udder health in AMS herds requires further examination.
Interestingly, housing dry cows in both pack pens and stalls was associated with a lower herd-average incidence of new IMI compared with housing dry cows only in pack pens, which may have been caused by worst environmental hygiene and higher exposure to pathogens in the herds where cows were housed only in packs.The use of both housing systems was more likely to occur in herds that manage dry cows in far-off and close-up groups, as we detected an association (P = 0.015) between these 2 explanatory variables.This coincides with a recent study conducted in Canadian dairy farms that determined that far-off cows are housed more frequently in freestalls, whereas close-up cows are more frequently housed in straw-bedded loose packs (Couto Serrenho et al., 2022).Although straw yards provide many benefits in terms of better cow comfort, decreased lameness, and increased time spent performing natural behaviors compared with stalls, they may also constitute a risk factor for poor udder health due to difficulties in maintaining them clean (Leso et al., 2020), as disinfection is not feasible (Klaas and Zadoks, 2018).Researchers have shown that lactating cows housed in straw yards had worst hygiene, higher SCC, and higher incidence of clinical mastitis compared with cows housed in stalls (Fregonesi and Leaver, 2001).However, to our knowledge, studies comparing the use of stalls and packs on udder health have only been conducted on lactating cows.Thus, more research is needed to understand how the housing system during the dry period may affect early-lactation udder health in AMS herds, with a particular focus on comparing stalls and different types of bedded packs.
Finally, placing cows onto the AMS within the first day after calving tended to be associated with a higher herd-average PostSCS than placing them onto the AMS after the first day postpartum.A plausible explanation for this finding is the attitude of producers toward preventing and controlling mastitis in their herds (Dufour et al., 2011).It could be hypothesized that in those herds that take a longer time in placing cows on the AMS, the producers may pay more attention to the health of the cows (e.g., conduct daily inspections of the udders; Dufour et al., 2011), helping them detect and treat mastitis before herds where cows are taken immediately to the AMS.However, to our knowledge, no previous study has examined this and, thus, requires further investigation.A second hypothesis could be that herds in which cows are placed in the AMS later after calving may leave the cow-calf pair together for a longer time, allowing calves to suckle from their dams during this time.It has been reported that this practice is associated with a reduced risk of mastitis in postpartum cows, possibly due to the removal of residual milk from the mammary gland by the suckling calves or the inhibition of bacterial growth due to the presence of lysozymes in their saliva (Beaver et al., 2019).Although we asked in the surveys how long after birth calves were removed from their dam, we did not include this variable in our analysis because less than 10% of the producers removed them after the first day of birth.Therefore, more research is needed to identify the reasons for the improved udder health observed in cows that are not moved immediately to the AMS after calving.

Strengths and Limitations
A strength of the present study is that we assessed housing and management practices during the dry period on a large sample of commercial AMS herds (n = 166) across Canada, including a large sample of dairy cows (n = 14,007) in our analyses.In doing so, we could assess several variables that may have been associated with early-lactation udder health at the herd and cow level, in contrast with what happens in controlled studies, where one or a few management practices are typically examined while other factors are kept constant and, therefore, may not be entirely representative of what occurs in practice in dairy herds.By using data from a large number of commercial farms across Canada, we also included herds with different breeds, climates, feeding and management practices, and housing environments.Therefore, our results could apply to any Canadian AMS herd that implements housing and management practices similar to those of the farms enrolled in this study.
Despite many detected associations of indicators of early-lactation udder health with various cow-level factors and herd-level housing and management practices, some limitations in this study should be considered when interpreting our results.First, only farms enrolled in DHI milk recording were included in this study.Thus, our results do not represent AMS farms that have chosen not to participate in a milk recording program.Second, to assess the udder health of each cow, we used the last milk test before dry-off that had available SCC data and was obtained when the cow was above 250 DIM.However, this milk test was not necessarily the cow's last test before dry-off; it was the last test with SCC data available.Therefore, it may not have been that representative of what actually occurred late in lactation and during the dry period.For example, some of the cows categorized as cured IMI may have been cows that received treatment after the test we selected and cleared the IMI before their last test before dry-off.Thus, they may have cured the IMI late in lactation and not during the dry period.Third, the occurrence of IMI was determined based on SCC, which is an indicator of IMI (Dohoo and Leslie, 1991;Schuk-ken et al., 2003), but not by bacteriological cultures of quarter milk samples as recommended by the National Mastitis Council (NMC, 2012), which may have led to misclassification bias.For instance, some of the cows categorized as always infected may have been cows that cured an IMI during the dry period and then became reinfected in the same quarter or developed a new IMI in another quarter.However, because we did not know which quarter was infected and which microorganism was causing the infection, we were unable to detect this.In addition, as with any cut point, some degree of misclassification may have also occurred when we estimated the presence of subclinical IMI based on SCC ≥200,000 cells/mL.For example, a cow with SCC of 199,000 cells/mL on their last milk test before dry-off and SCC of 201,000 cells/mL on their first milk test after calving would have been classified as a cow with a new IMI, although it may have very likely been a normal biological variation.Despite this, using such a cut point is well accepted in the literature (Dohoo and Leslie, 1991;Schukken et al., 2003).Fourth, although we excluded from our study cows that had dry periods considered too short (i.e., <30 d) and too long (i.e., >120 d) for a normal dry period, we did not have information about the dry-off date for some cows.Consequently, we may have unknowingly included in our analyses cows with unusually short or long dry periods.Finally, some relevant aspects of udder health management in the participant herds were not able to be collected as part of this project; for example, which of the cows selected in our study received antimicrobial treatment for IMI around the dry period.

CONCLUSIONS
Improved indicators of udder health in early-lactation cows in AMS herds were associated with lower parity, shorter dry periods, and greater 305-d milk yield before dry-off at the cow level.At the herd level, improved udder health was associated with greater 305-d milk yield before dry-off, as well as with housing and management practices around the dry period, including separating cows into a different pen as preparation for dry-off, using teat sealants and blanket antibiotic dry cow therapy, housing cows in the same group throughout the dry period, housing dry cows in both pack pens and stalls, and placing cows on the AMS to be milked after the first day postpartum.Thus, if these associations are causal, AMS producers may be able to improve the health and welfare of early-lactation cows by modifying housing and management practices in preparation for dry-off, at dry-off, during the dry period, and at the beginning of lactation.
Wagemann-Fluxá et al.: EARLY-LACTATION UDDER HEALTH IN ROBOT HERDS and ΔSCS were similar, we report in the results only those factors associated with PostSCS.
, respectively.Enrolled farms were located in Ontario (n = 69), Quebec (n = 42), Western Canada (n = 43), and Atlantic Canada (n = 12).In 89.8% (n = 149) of enrolled farms, the milking herd was composed predominantly of Holstein cows Wagemann-Fluxá et al.: EARLY-LACTATION UDDER HEALTH IN ROBOT HERDS Wagemann-Fluxá et al.: EARLY-LACTATION UDDER HEALTH IN ROBOT HERDS Wagemann-Fluxá et al.: EARLY-LACTATION UDDER HEALTH IN ROBOT HERDS and 95% CI for categorical variables from the referent state or for a 1-SD increase in the continuous variables (projected 305-d milk yield [kg] [1 SD = 2,422.0kg] and post-DIM [1 SD = 11.0 d]). 5 Pre-DIM = days in milk at the last milk test before dry-off.d milk yield (kg) at the last milk test before dry-off.9 Post-DIM = days in milk at the first milk test after calving.

Table 1 .
Categorical housing and management variables at dry-off, during the dry period, at calving, and the beginning of lactation in 166 dairy herds using automated milking systems across Canada

Table 2 .
Descriptive statistics and continuous housing and management variables at dry-off and during the dry period in 166 dairy herds using automated milking systems across Canada 2Reported by the producers.3Dry period length = calculated based on DHI data from the cows selected per herd.4 PreSCS = SCS on the last milk test before dry-off, calculated as log 2 (SCC/100,000) + 3. 5 PostSCS = SCS on the first milk test after calving, calculated as log 2 (SCC/100,000) + 3. 6 ΔSCS = change in SCS during the dry period calculated as the difference between SCS at the first milk test after calving (PostSCS) and last milk test before dry-off (PreSCS).

Table 3 .
Descriptive statistics of 14,007 dairy cows from 166 herds using automated milking systems across Canada

Table 4 .
Final multivariable logistic regression model for cow-level factors associated with the occurrence of new and cured IMI during the dry period for 14,007 cows from 166 dairy herds using automated milking systems across Canada Variable

Table 5 .
Final multivariable linear model for the factors associated with cow-level SCS in the first milk test after calving (PostSCS 1 ) during the dry period for 14,007 cows from 166 dairy herds using automated milking systems across Canada

Table 6 .
Final multivariable linear model of the herd-level factors associated with herd-average incidence of new and cured IMI during the dry period in 166 dairy herds using automated milking systems across Canada 2

Table 7 .
Wagemann-Fluxá et al.: EARLY-LACTATION UDDER HEALTH IN ROBOT HERDS Final multivariable linear model of the herd-level factors associated with herd-average SCS in the first milk test after calving (PostSCS 1 ) during the dry period in 166 dairy herds using automated milking systems across Canada