A randomized controlled trial to study the effects of an automated premilking stimulation system on milking performance, teat tissue condition, and udder health in Holstein dairy cows

The objectives were to examine the effect of an automated premilking stimulation (APS) by means of a high pulsation frequency (300 cycles/min) without a reduction of the vacuum in the pulsation chamber or claw piece on (1) milking performance, (2) teat tissue condition, and (3) udder health in dairy cows. In a randomized controlled field study, Holstein cows (n = 427) from 1 commercial dairy farm with a milking schedule of 3 times per day were assigned to treatment and control groups over a 90-d period. Treatments consisted of a maximum of 80 s (APS80) or 99 s (APS99) of mechanical stimulation at a pulsation rate of 300 pulses per minute and a ratio of 25:75 (no reduction of the pulsation chamber or milking vacuum). Cows in the control group (CON) received traditional pre-milking stimulation by means of manual forestripping for 8 s. Milking characteristics were documented with on-farm milk meters. Short-and long-term changes in teat tissue condition induced by machine milking were assessed visually on a weekly basis. Composite milk samples were analyzed once per month to determine somatic cell count. Generalized linear mixed models were used to study the effect of the treatment on the outcome variables. We observed no meaningful differences in milk yield or milking unit-on time. Least squares means and their 95% confidence intervals (95% CI) for cows in the APS80, APS99, and CON groups were 13.5 (13.1–14.0), 13.2 (12.8–13.7), and 13.2 (12.8–13.7) kg for milk yield and 222 (213–231), 219 (210–228), and 223 (214–232) s for milking unit-on time, respectively. The effect of treatment on bimodality was modified by milk yield such that the odds of bimodality increased in the treatment groups with increasing milk yield. Compared with cows in the CON group, the odds ratios


INTRODUCTION
Optimal premilking udder preparation is a prerequisite for the harvest of high-quality milk from cows as completely and quickly as possible while minimizing the effects on the canal, surrounding tissue, and skin of the teat.This preparation comprises premilking teat sanitization, including disinfection, cleaning, and drying of teats, and premilking stimulation (NMC, 2013).Premilking stimulation consists of tactile stimulation and timing of the attachment of the milking unit.Through the activation of pressure-sensitive mechanoreceptors primarily located in the tip of the teat, tactile stimulation initiates the release of oxytocin from the pituitary gland into circulation (Bruckmaier and A randomized controlled trial to study the effects of an automated premilking stimulation system on milking performance, teat tissue condition, and udder health in Holstein dairy cows M. Wieland,1 * C. M. Geary, 1 D. V. Nydam, 1 P. D. Virkler, 1 M. Zurakowski, 1 R. D. Watters, 1 and R. Lynch2 Blum, 1996).The timing of milking unit attachment (i.e., preparation lag time) allows this hormone to reach the mammary gland via the blood and stimulate receptors of myoepithelial cells.Once oxytocin surmounts a threshold of about 3 times the baseline concentration (Bruckmaier et al., 1994), myoepithelial cell contraction leads to forceful mechanical ejection of the alveolar milk into the duct system, which makes it available for harvesting (Bruckmaier and Blum, 1998).
Historically, tactile stimulation has been achieved through manual forestripping by the human hand, and in most conventional dairy operations, this traditional practice is used to initiate the milk-ejection reflex of cows (USDA, 2016).With the advancement of milking technology, however, alternative forms of mechanical stimulation have been used.These include rotating brushes (Hopster et al., 2002) and different forms of pulsation (Worstorff et al., 1987) applied to the teat.Worstorff et al. (1987) showed that a reduction in the pulsation chamber vacuum (of 20-22 kPa) combined with a high pulsation frequency (i.e., high-vibration pulsation with 300 pulsation cycles/min) is an effective method to elicit the milk-ejection reflex in dairy cows.These findings were supported by more recent work (Weiss and Bruckmaier, 2005;Watters et al., 2015).Other researchers have compared the effect of reduced pulsation chamber vacuum at a normal pulsation rate (60 cycles/min) and ratio (60:40) on oxytocin concentration and milking performance with that of manual premilking stimulation (Neuheuser et al., 2017).They found that reduced pulsation chamber vacuum with normal pulsation provides adequate stimulation to achieve oxytocin release and milk ejection (Neuheuser et al., 2017).The same research group tested the effect of reduced claw vacuum and B-phase during low milk flow on oxytocin release and milking performance (Neuheuser et al., 2018).They found that oxytocin release and milking performance were similar if milking was performed with premilking stimulation or without premilking stimulation but reduced claw vacuum and B-phase during low milk flow (Neuheuser et al., 2018).These findings indicated a promising method for optimizing the milk harvesting process.Combining automated premilking stimulation (APS) systems with automated application of premilking teat disinfectants can alleviate labor challenges and complement manual premilking stimulation to improve labor efficiency while meeting physiological requirements of individual cows, such as different production levels or milking intervals.However, despite these findings and the technical capability of modern milking parlor systems, as well as their employment in Europe, APS systems have not been adopted in the United States.Further, previous works investigating high-vibration pulsation were as-sociated with a reduction of the pulsation chamber vacuum (Weiss and Bruckmaier, 2005;Watters et al., 2015).Conversely, knowledge about the efficacy of high-vibration pulsation stimulation system without a reduction of the pulsation chamber vacuum has not been investigated by rigorous methods.The objective of this study, therefore, was to investigate the effect of an APS system that employs high-vibration pulsation but no reduction of the pulsation chamber vacuum or claw vacuum on milking performance, teat tissue condition, and udder health.We hypothesized that the APS system would be sufficient to elicit the milk-ejection reflex and harvest the alveolar milk fraction without compromising teat condition or udder health.In addition, we probed the effect of APS on the risk of animal removal from the herd.

MATERIALS AND METHODS
This randomized field trial was conducted at a commercial dairy farm located near Ithaca, New York, from June 10 to September 8, 2021.The study protocol was reviewed and approved by the Institutional Animal Care and Use Committee of Cornell University (protocol no.2020-0068).

Animals and Housing
The lactating herd consisted of approximately 1,600 Holstein cows.These cows were housed year-round in freestall pens with either concrete stalls covered with mattresses and bedded with wastepaper-pulp or dry shavings or deep-bedded stalls with fresh sand.Cows were fed a TMR formulated according to NRC requirements (NRC, 2001).The key herd performance indicators were average milk production (305-d mature equivalent milk production; 12,033 kg), mean test day SCC (213,000 cells/mL), monthly incidence of clinical mastitis (6.0%), 21-d pregnancy rate (26.0%), and culling risk (36.0%).

Milking System
Cows were milked 3 times per day at 0630, 1430, and 2230 h in a 60-stall rotary milking parlor (Madero Premium, Madero Dairy Systems).The vacuum pump (28 kW) was regulated by a variable frequency drive and set to supply a receiver operating vacuum of 44 kPa.The milking unit consisted of the Milking Cluster Classic 300 (GEA Farm Technologies) and a milking liner with a square-shaped barrel (Gea Classic GQ, GEA Farm Technologies).The pulsators (HiFlo, Bou-Matic) were set to a pulsation rate of 60 cycles/min, a ratio of 65:35, and a side-to-side alternating pulsation.

Wieland et al.: AUTOMATED PREMILKING STIMULATION
The average claw vacuum during the peak milk flow period was assessed with a VaDia device (Biocontrol) as the average cyclic vacuum fluctuations (assessed for 10 pulsation cycles, 60 s after the start of the peak milk flow period) over 15 milking observations; this value was 40 kPa.For this purpose, the start of the peak milk flow period was defined according to the manufacturer's manual (VaDia Suite User Manual, version 3.7, Biocontrol) and detected as follows: the average teat-end vacuum (i.e., short milk tube vacuum) in 10-s periods after attachment is monitored.When the average vacuum from 1 period to the next declines less than 0.15 kPa, the midpoint of the first (of the 2) periods is indicated as the start of the peak milk flow period.This automated assessment was visually confirmed by one trained investigator (MW).The parlor was equipped with electronic on-farm milk meters (AfiMilk MPC Milk Meter, Afimilk).The automatic cluster-remover (ACR) settings were as follows: the ACR premilk time (i.e., the minimum amount of time the milking unit is attached) was set to 12 (corresponding to 120 s); the ACR delay (i.e., the cluster-remover milk flow threshold determines the milk flow rate at which milking will be terminated) was set to 6 (2.1 kg/min; June 10-21, 2021), 7 (1.8 kg/min; June 22-30, 2021), and 8 (1.6 kg/ min; July 1 to September 8, 2021); the vacuum decay time controls the delay between the vacuum shut-off to the claw and the beginning of milking unit retraction was set to 0 s; and the autoadjust quick removal was activated.If activated, the autoadjust quick removal initiates removal of the milking unit before the delay time designated in the ACR delay is reached if the following 3 conditions are met: first, the ACR pre-milk time is surpassed; second, the harvested milk yield exceeds the expected milk yield; and third, the milk meter body is less than half full after 50% of the ACR delay time.The milk sweep was initiated 10 s after unit retraction and lasted for 5 s.All milking system settings were assessed and verified by the investigators according to the guidelines outlined by the National Mastitis Council (NMC, 2012) before the start of the study.

Milking Routine
Before the start of the study and during normal operation (i.e., milking of pens not included in the study), the rotational speed of the milking parlor was 8.5 s/ stall, resulting in a rotation time of 510 s.Parlor operation included two 12-h work shifts, each with 4 milking technicians who were assigned to perform the following tasks at 4 different stations: (1) manually forestrip teats and apply premilking teat disinfectant with a dip applicator cup, (2) clean and dry the teat barrel and end of all teats with an individual cloth towel, (3) attach and align the milking unit, and (4) apply postmilking teat disinfectant with a dip applicator cup and monitor milking liner slips, unit fall-offs, or unit kick-offs, realigning or reattaching the milking unit accordingly.For a cow that entered the rotary parlor at stall 1, the positioning of different task stations was stall 3 (station 1), stall 7 (station 2), and stall 14 (station 3).The milking technician at station 4 operated over a wider range between the halfway point and exit of the rotary parlor to ensure timely adjustment of the milking units.This setup resulted in a dip contact time of 34 s, stimulation duration (duration of manual forestripping) of approximately 6 s, and a preparation lag time (i.e., latency from first tactile stimulus to milking unit attachment) of approximately 94 s.

Treatment Allocation
The 3 central components of the treatment allocation were the dairy farm management software program AfiFarm (Afimilk), electronic on-farm milk meters (AfiMilk MPC Milk Meter), and the pulsation system (HiFlo, BouMatic).With AfiFarm, different milking machine settings can be applied for individual cows housed within the same pen.For this purpose, a code entailing the specific settings of the operational parameters was assigned to individual cows according to their treatment allocation for the duration of the study.The APS system of AfiFarm is based on 3 individual parameters.The stimulation duration (STD) measures the maximum length of the stimulation in seconds.The stimulation operation ratio determines the length of the pulsation phases during the stimulation.The stimulation cycles per minute control the number of pulsation cycles per minute during the stimulation.During the pulsation stimulation phase, no reduction of the pulsation chamber vacuum or the claw vacuum was seen, which is different from previously tested APS systems (Weiss and Bruckmaier, 2005;Watters et al., 2015;Neuheuser et al., 2018).Thus, in contrast to previously tested systems (Weiss and Bruckmaier, 2005;Watters et al., 2015), the milking liner was not kept in the closed position to avoid milk harvest and teats were exposed to the unaltered claw vacuum.An additional fixed feature of the APS system is the automated switch from stimulation to milking mode after the first milk meter dump upon milking unit attachment.One milk meterdump corresponds to 200 mL.
We included study cows from 2 lactating pens.All cows housed in these pens at the start of the study were eligible for inclusion and randomly assigned by the first author to treatment and control groups, which were stratified by lactation number, stage of lactation, and average daily milk yield during the previous week, using a random number generated in Microsoft Excel 2019 (Microsoft Corp.).Treatments consisted of a maximum of 80 s (STD = 80 s; APS80) or 99 s (STD = 99 s; APS99) of APS at a stimulation ratio of 25:75 and 300 stimulation cycles per minute.The automated stimulation was initiated immediately after milking unit attachment.The stimulation duration (up to the maximum value) for an individual cow varied according to the duration at which sufficient milk was harvested to initiate the first dump of the milk meter, which then switched the milking unit from stimulation to milking mode.Cows in the control group (CON) received traditional premilking stimulation by means of manual forestripping.To accommodate different premilking udder preparation regimens among groups, we modified the farm's standard milking routine for the study cows.The rotational speed of the milking parlor was set to 12 s/stall.This facilitated the completion of 2 tasks at the same station.Therefore, the positioning of the milking stations was as follows: stall 3 (station 1), stall 6 (station 2), and stalls 11-12 (station 3).Premilking udder preparation for cows in the treatment groups consisted of (1) applying premilking teat disinfectant with a dip applicator cup at station 1, (2) wiping (i.e., drying and cleaning) of teats, and (3) attaching and aligning the milking unit at station 2. Premilking udder preparation for cows in the CON group consisted of (1) applying premilking teat disinfectant with a dip applicator cup and manual forestripping of teats at station 1, (2) wiping (i.e., drying and cleaning) of teats, and (3) attaching and aligning the milking unit at station 3.For cows in the APS80 and APS99 groups, this setup resulted the following premilking preparation: a dip contact time of approximately 36 s, a maximum stimulation duration of 84 s (4 s of teat wiping and 80 s of automated stimulation), and a maximum preparation lag time (i.e., the time period between the first tactile stimulus, which was wiping of teats, and initiation of the milk harvest), of 103 s (4 s of teat wiping and 99 s of automated stimulation), respectively.However, the individual stimulation duration (i.e., time period between attachment of the milking unit and switch to milking mode) could not be extracted from the software program.Therefore, we set out and measured the vacuum dynamics (pulsation chamber vacuum) of the APS system using VaDia device (Biocontrol) and calculated the average stimulation duration from 19 individual cow milking observations.These measurements were taken after completion of the study to provide an idea on the actual stimulation duration that cows in the treatment groups were subjected to.The mean (±SD) stimulation duration from these measurements was 9.2 ± 2.4 s and ranged from 2 to 12 s.For cows in the CON group, the dip contact time was approximately 96 s, the stimulation duration (i.e., time spent on manual forestripping) was approximately 8 s, and the preparation lag time (i.e., time period between the first tactile stimulus, which was manual forestripping, and attachment of the milking unit) was approximately 100 s.To facilitate compliance with the study protocol, cows in the treatment groups were identified with leg bands on both hind legs.Milking technicians were trained on the protocol before the start of the study, and adherence to the protocol was monitored throughout the study period by study authors.Thus, study and farm personnel were not blinded.

Sample Size Calculation
Before the study, we anticipated that approximately 420 cows would be available for enrollment.This sample size was sufficient to detect a minimum difference of 1 kg of milk/milking session in 2 of the 3 groups at an α level of 0.05 with a power of 0.95.This calculation was based on a reported standard deviation of 3.4 kg (Wieland et al., 2017), a presumed correlation between measurements within a cow of 0.5, a total of 264 milking observations (88 d × 3 milking sessions/d), and a repeated-measures ANOVA (G*Power version 3.1.9.7; Faul et al., 2007).

Data Acquisition
Cow Characteristics.We obtained cow characteristics such as lactation number, stage of lactation, SCC at the last test day (3 d before the start of the study), and average daily milk production 1 wk before the start of the study from the dairy management software program (DairyComp 305, Valley Agricultural Software).
Milking Characteristics.We assessed the following milking characteristics during each milking session with the electronic on-farm milk meters (AfiMilk MPC Milk Meter) and recorded them using the dairy farm management software AfiFarm: milk yield (i.e., the yield of milk in kilograms harvested from start of milking to detachment of the milking unit), milking unit-on time (i.e., the latency in seconds from start of milking to detachment of the milking unit), peak milk flow rate (calculated as the maximum milk flow rate between 2 milk meter dumps in kilograms per minute), duration of low milk flow (i.e., the duration in seconds of milk flow rate below 1 kg/min between start of milking and detachment of the milking unit), first 15-s milk flow rate, 15-to-30-s milk flow rate, 30-to-60-s milk flow rate, and 60-to-120-s milk flow rate (i.e., average milk flow rate in kilograms per minute recorded in the first 15 s, from 15 to 30 s, from 30 to 60 s, and from 60 to 120 s of milking, respectively).All the preceding parameters (i.e., milk yield, milking unit-on time, peak milk flow rate, duration of low milk flow, first 15-s milk flow rate, 15-30-s milk flow rate, 30-60-s milk flow rate, and 60-120-s milk flow rate) were measured from the time of unit attachment (i.e., push of the start button) and, for the treatment groups, included the stimulation duration.Reports from each milking session were exported to a text document (.txt file) 3 times per day.For subsequent analyses, we created a new categorical variable (bimodality) and defined it as previously described (Wieland et al., 2020b): bimodality was present if any of the incremental milk flow rates (15-30-s milk flow rate, 30-60-s milk flow rate, or 60-120-s milk flow rate) were lower than any of the preceding incremental milk flow rates (first 15-s milk flow rate, 15-30-s milk flow rate, or 30-60-s milk flow rate, respectively).
Nonlactating Quarter and Teat Tissue Conditions.The presence of a nonlactating quarter and the evaluation of the teat tissue condition were performed by 2 trained investigators (CMG and MW).The presence of a nonlactating quarter was visually assessed during milking 3 d before the start of the study and confirmed to be present if 1 teat cup was not attached to the respective quarter.
Short-term changes to the teat tissue condition induced by machine milking and teat-end callosity were assessed visually and through palpation once per week during the study period according to the scoring systems previously described (Hillerton et al., 2000;Mein et al., 2001).The presences of short-term changes to the teat tissue condition and hyperkeratosis were categorized as previously described (Wieland et al., 2020a).
Milk Sampling and SCC Analysis.Composite milk samples were collected by DHIA service personnel (Dairy One Cooperative Inc., Ithaca, NY) 3 d before the start of the study and on d 26, 54, and 82.Milk samples were analyzed for SCC (cells/mL) with optical fluorescence (Fossomatic FC, Foss; method 978.26,AOAC International, 2023) at Dairy One Cooperative Inc.For subsequent analyses, SCC values were log 10transformed (logSCC).In addition, we calculated the new infection risk for each of the 3 groups for the 3 test days during the study.A new infection was defined present if the previous test day SCC was <200,000 cells/mL and the last test day SCC was ≥200,000 cells/ mL, whereas a new infection was considered absent otherwise.
Clinical Mastitis Detection and Culling.Clinical mastitis detection was performed by trained milking technicians during premilking udder prepara-tion.Because cows in the treatment groups were not forestripped, mastitis detection was based on signs of inflammation [i.e., heat (calor), pain (dolor), redness (rubor), and swelling (tumor)] of the affected mammary gland.Milking personnel were instructed to manually forestrip all quarters of a cow if they observed signs of inflammation of one or more quarters.To mitigate information bias among groups, all cows were manually forestripped once weekly after machine milking by a trained investigator (CMG and MW).A cow was defined as exhibiting clinical mastitis if milk from 1 or more quarters was abnormal with or without signs of local inflammation of the affected quarter as previously described (Erskine et al., 2003).All clinical mastitis events were recorded in Dairy Comp 305.For subsequent analyses, a case was defined as the first case of clinical mastitis that occurred during the study.Culling events were recorded in Dairy Comp 305.
Analytical Approach.We maintained the data in Excel (Microsoft Office Excel 2019; Microsoft Corp.).Cows that were diagnosed with mastitis remained in the study.Data from cows that were lost to follow-up were included in the analyses up until the dry day or the point of removal from the herd.Statistical analyses were performed with the software package SAS (version 9.4, SAS Institute Inc.).
Baseline Characteristics.Chi-squared tests were generated with PROC FREQ to determine differences in lactation number (i.e., first, second, and third and later lactations), stage of lactation (≤100, 101-200, and >200 DIM), nonlactating quarter status, and hyperkeratosis status 3 d before the study among the groups.Differences in logSCC from the test day 3 d before the start of the study and average daily milk yield in the week before the study were assessed with ANOVA using PROC ANOVA.
Milking Characteristics.Data from the first 2 d of the study were excluded, allowing cows to adjust to the differences in milking procedures.In the first step, we screened the data for missing and erroneous values by investigating the distributions of milk yield, milking unit-on time, peak milk flow rate, duration of low milk flow, and the incremental milk flow rates.We removed observations with missing values, outliers, or probable data errors by excluding observations with values of <5 or >55 kg for milk yield, values of <100 or >800 s for milking unit-on time, and values of <0.2 or >10 kg/ min for peak milk flow rate.To evaluate the effect of treatment on milking characteristics (milk yield, milking unit-on time, peak milk flow rate, and duration of low milk flow), 4 separate general linear mixed models were fitted with PROC MIXED.Milk yield, milking unit-on time, peak milk flow rate, and duration of low flow were included as dependent variables.The following steps were similar for all 4 models.To account for the dependence of milking observations between milking sessions and days within a cow, we included a REPEATED statement.We evaluated 3 covariance structures (autoregressive order 1, compound symmetry, and variance components) for modeling the covariance of the repeated measurements and selected the covariance structure that led to the smallest Akaike's information criterion.Treatment was forced into the model as a fixed effect.We considered the following independent variables and initially screened them for inclusion in each model through univariate analysis: lactation number (first, second, or third and later lactations), stage of lactation (≤100, 101-200, or >200 DIM), presence or absence of a nonlactating quarter, and logSCC.All variables with a P-value <0.20 in this step were included in the initial models.Collinearity among the eligible variables was assessed by calculating Spearman correlation coefficients in PROC CORR.A coefficient >|0.60| was considered indicative of collinearity.We performed manual backward elimination until each of the variables had a P-value <0.05.Confounds were assessed by observing regression coefficient changes.Variables that modified regression coefficients by >20% were considered confounding factors.Finally, 2-way interactions between treatment and the remaining variables were assessed individually and retained in the final model if the P-value was <0.05.We used Tukey-Kramer's post hoc test to control for the experimental error rate.For all final models, we assessed the assumptions of homoscedasticity and normality of residuals by inspecting plots of residuals versus the corresponding predicted values and examining quantilequantile residual plots.To satisfy these assumptions, data on the dependent variable duration of low milk flow were log 10 -transformed.The resulting least squares means (LSM) estimates were consequently back transformed and presented as the geometric mean and 95% confidence interval (95% CI).To illustrate the interaction between treatment and milk yield for duration of low milk flow, we computed LSM and the 95% CI for the lower (11 kg) and upper (15 kg) quartiles of milk yield.
To determine differences in bimodality between groups, a generalized linear mixed model with a logit link and a binomial distribution was fitted with PROC GLIMMIX.To account for the clustered structure of the data, cow ID was included as a random effect.Treatment was added to the model as a fixed effect.Lactation number (first, second, or third and later lactations), stage of lactation (≤100, 101-200, or >200 DIM), logSCC from 3 d before the study, presence or absence of a nonlactating quarter, and milk yield were considered independent variables and initially screened for inclusion through univariate analysis.All variables with a P-value <0.20 in this step were included in the initial model.We assessed collinearity among eligible variables by calculating Spearman correlation coefficients in PROC CORR; a coefficient >|0.60| was considered indicative of collinearity.Manual backward elimination was used to achieve the final model.Finally, the 2-way interaction between treatment and the remaining variable was evaluated and retained in the model if the P-value was <0.05.
Teat Tissue Condition.To compare differences in short-term changes to the teat tissue condition and teat-end hyperkeratosis among groups, 2 separate generalized linear mixed models with a logit link and a binomial distribution were fitted with PROC GLIMMIX.We included cow ID as a random effect to account for the clustered nature of the data and modeled the covariance using the compound symmetry covariance structure.Treatment was entered into the models as a fixed effect.We considered lactation number (first, second, or third and later lactations), stage of lactation (≤100, 101-200, or >200 DIM), logSCC from 3 d before the study, and presence or absence of a nonlactating quarter as independent variables and initially screened them for inclusion with univariable analysis.For the dependent variable teat-end hyperkeratosis, the baseline value (i.e., teat-end hyperkeratosis assessed 3 d before the study) was also included.All variables with a P-value <0.20 in univariable analysis were included in the initial models.We assessed collinearity among the eligible variables by calculating Spearman correlation coefficients in PROC CORR; a coefficient >|0.60| was considered indicative of collinearity.Manual backward elimination was used to achieve the final models.Biologically relevant 2-way interactions were evaluated and retained in the final model if the P-value was <0.05.Finally, we calculated adjusted probabilities and 95% CI using the LSMEANS statement.Tukey-Kramer's post hoc test was used to control the experimental error rate when comparing a family of adjusted probabilities.
Somatic Cell Count.To compare logSCC among groups, a general linear mixed model was generated with PROC MIXED.The model was constructed in accordance with the procedure outlined above, except for the 3 following items.First, a REPEATED statement for test day was included to account for clustering of the test day within a cow.Second, in addition to lactation number, stage of lactation, and presence or absence of a nonlactating quarter, the logSCC from the test day 3 d before the start of the study was also included.Third, to calculate the LSMEANS over the Wieland et al.: AUTOMATED PREMILKING STIMULATION time, we included the interaction between treatment and test day.In addition, we calculated the new infection risk among groups using a generalized linear mixed model with a logit link and a binomial distribution with PROC GLIMMIX using the procedure outlined above.
Clinical Mastitis and Culling.Survival curves of the time to clinical mastitis event and time to culling were generated with GraphPad Prism (version 9, GraphPad Software).To explore differences in the risks of clinical mastitis and culling among groups, we constructed 2 separate semiparametric Cox (1972) proportional hazards models with PROC PHREG.The days to occurrence of clinical mastitis and culling were the outcomes of interest and were included as the dependent variables.The independent variable of interest was treatment, which was added to each model as a fixed effect.We considered lactation number, stage of lactation, presence or absence of a nonlactating quarter, and logSCC from the test day 3 d before the start of the study as additional independent variables and initially screened them for inclusion with a univariate analysis.All variables with a P-value <0.20 in the univariate analysis were included in the initial models.We declared significance at P < 0.05.For the analysis of time to mastitis, animals that had no mastitis event by the end of the follow-up and animals that went dry or were removed from the herd before a mastitis event were right censored.For the analysis of time to culling, animals that remained in the herd at the end of the follow-up period were right censored.We assessed each final model for the proportionality assumption with the ASSESS statement.We assumed the model fit if none of the predictor variables violated the Supremum test at a level of P < 0.05.Last, we conducted post hoc power analyses for both outcome variables with JMP (version 14, SAS Institute Inc.).

Description of the Study Population
A total of 427 cows (143 APS80, 141 APS99, and 143 CON cows) were included in the study.The average (mean ± SD) DIM at day of inclusion was 178 ± 99 d, ranging from 9 to 592 d.The average (mean ± SD) daily milk yield during 1 wk before inclusion was 42.4 ± 12.2 kg and ranged from 15.4 to 66.7 kg.The median SCC was 36,500 cells/mL and ranged from 4,000 to 3,494,000 cells/mL.Table 1 shows the baseline characteristics stratified by group.During the study period, 104 cows were dried off, 50 cows were sold, and 10 cows died.Thus, 263 cows completed the study.

Milking Characteristics
We obtained 89,785 cow milking observations over the 90-d study period.The exclusion of the first 2 d (2,454 observations) resulted in 87,331 observations, which were inspected for missing and erroneous values.A total of 3,351 (3.8%) observations were excluded due to missing or erroneous values, resulting in a total of 83,980 individual milking observations available for statistical analyses.Table 2 shows descriptive statistics of the milking characteristics stratified by group.

Teat Tissue Condition
We obtained data from 4,251 cow observations.Of these, 12 values of short-term changes to the teat tissue condition were missing, resulting in a total of 4,239 observations available for analysis.Short-term changes to the teat tissue condition were recorded in 1,515 (35.7%) cow observations [APS80, 591/1,401 (42.2%);APS99, 538/1,425 (37.8%); and CON, 385/1,413 (27.3%)].The final model for the presence or absence of short-term changes to the teat tissue condition included treatment (P = 0.001) and stage of lactation (P = 0.0008; Table 3).The adjusted probabilities of short-term changes to the teat tissue condition among the groups are shown in Figure 3.

Somatic Cell Count
A total of 980 test observations were obtained.Of these, 6 SCC values were missing, resulting in a total of 974 individual test observations available for SCC analyses.The median SCC over the 3 test days was 46,500 cells/mL.The final model included treatment (P = 0.56), stage of lactation (P = 0.001), logSCC 3 d before the start of the study (P < 0.0001), test day (P = 0.005), and the interaction between treatment and test day (P = 0.77).The LSM (95% CI) were 4.74 (4.68-4.81)for the APS80 group, for the APS99 group, for the CON group.A one-unit increase in logSCC 3 d before the start of the study increased the logSCC by 0.61 (0.54-0.68) units.Least squares means for treatment groups over the 3 test days are shown in Figure 5A.
For the calculation of the new infection risk, 795 test day observations were included.We observed a total of 115 (14.5%) new infection over the 3 test days [APS80, 34/273 (12.5%);APS99, 37/258 (14.3%); and CON,44/264 (16.7%)].The final model for the presence or absence of a new infection included treatment (P = 0.66), test day (P = 0.01), and their interaction (P = 0.04).With the CON group as a reference, the OR (95% CI) were 0.78 (0.46-1.33) for cows in the APS80 group and 0.86 (0.50-1.47) for cows in the APS99 group.Figure 5B shows the adjusted probabilities of a new infection stratified by test days.

Clinical Mastitis
We documented a total of 49 (11.5%)mastitis cases [(APS80,17/143 (11.9%);APS99, 12/141 (8.5%); CON, 20/143 (14.0%)].The mean (±SD) days after the start of the study at which the clinical mastitis cases occurred were 39 ± 22 d (APS80,33 ± 24;APS99,44 ± 21;and CON,40 ± 19 d).A total of 378 (88.5%) cow observations were right censored.Among the rightcensored observations, 98 were due to drying (APS80, 28; APS99, 28; and CON, 42) and 34 were due to culling of the animals (APS80, 14; APS99, 15; and CON, 5); 246 cows had no mastitis by the end of the study (APS80,84;APS99,86;and CON,76).Kaplan-Meier survival analysis of time to clinical mastitis is depicted in Figure 6A.The following independent variables were associated with clinical mastitis in univariable analysis and were included in the initial model: stage of lactation (P = 0.10), presence or absence of a nonlactating quarter (P = 0.16), and logSCC (P = 0.10).The final Cox proportional hazards regression model included treatment (P = 0.34) but none of the other initial variables.Compared with cows in the CON group, the hazard risk (HR, 95% CI) of clinical mastitis was 0.85 (0.45-1.63) in the APS80 group and 0.59 (0.29-1.20) in the APS99 group.We assumed model fit based on the Supremum test (P ≥ 0.07).Using an α-level of 0.025 (Bonferroni correction for the comparison of 3 groups), the observed frequency distributions of 8.5 and 14%    in groups APS99 and CON, respectively, and the preexclusion sample sizes of 141 and 143 animals revealed a power of 0.20; 625 animals in each group would be necessary to detect a difference of 5.5% at an α-level of 0.025 and a power of 0.8.

Culling
A total of 60 (14.1%) animals were culled [APS80, 23/143 (16.1%);APS99, 22/141 (15.6%); and CON, 15/143 (10.5%)].The mean ± SD days at which culling occurred were 42 ± 21 d (APS80, 44 ± 21; APS99, 44 ± 22; and CON, 35 ± 18 d).A total of 367 (85.9%) cow observations were right censored.Among the right-censored observations, 104 were due to drying (APS80, 29; APS99, 30; and CON, 45); 263 cows were not culled by the end of the study (APS80, 91; APS99, 89; CON, 83). Figure 6B depicts the Kaplan-Meier survival analysis of the time to culling.Stage of lactation (P = 0.16) and logSCC (P = 0.17) were associated with the risk of culling in the univariate analyses and were included in the initial model.The final Cox proportional hazards regression model included only treatment (P = 0.40).Compared with cows in the CON group, the HR (95% CI) of removal from the herd was 1.53 (0.80-2.94) for the APS80 group and 1.44 (0.75-2.78) for the APS99 group.Based on the Supremum test (P ≥ 0.83), we assumed model fit.Using an α-level of 0.025, the observed frequency distributions of 16.1% and 10.5% in groups APS80 and CON, respectively, and the preexclusion sample sizes of 143 animals in each group led to a power of 0.19; using a power of 0.8, 697 animals in each group would be necessary to detect a difference of 5.6% at an α-level of 0.025.

Milking Characteristics
In this study, we investigated the effect of an APS system that employs high-vibration pulsation but no reduction of the pulsation chamber vacuum or claw vacuum on milking performance in Holstein dairy cows with a thrice daily milking schedule.To our knowledge, this system has not been evaluated by rigorous methods.Our data show no meaningful differences in milk yield and milking unit-on time between cows receiving premilking stimulation by means of the APS system and those that were subjected to the traditional premilking stimulation regimen via manual forestripping of teats.Our results expand the literature (Worstorff et al., 1987;Weiss and Bruckmaier, 2005;Watters et al., 2015) and suggest that the APS system tested here had no negative effects on milk production or milking efficiency.The lack of a measurable differences in milk yield between the treatment and control groups is consistent with the findings of previous works comparing manual and mechanical premilking stimulation regimens (Sagi et al., 1980;Karch et al., 1988;Tuor et al., 2022).Sagi et al. (1980) used 12 Holstein cows and subjected them to 4 different regimens of premilking stimulation: (1) no premilking stimulation, (2) manual premilking stimulation for 60 s, (3) positive pressure pulsation during the first 60 s after attachment of the milking unit, and (4) high-frequency pulsation during the first 60 s after milking unit attachment.The researchers found no differences in milk yield among the groups (Sagi et al., 1980).Karch et al. (1988) found no meaningful differences when comparing 60 s of manual premilking stimulation with 60 s of high-vibration stimulation in a cohort of 37 Brown Swiss dairy cows.In a recent study, Tuor et al. (2022) used 10 Holstein cows to compare 2 manual premilking stimulation regimens including 15 or 5 s of tactile stimulation, respectively, with 2 milking routines consisting of 5 s of tactile stimulation, immediate attachment of the milking unit, and mechanical stimulation by means of reduced pulsation (i.e., pulsation rate, 50 cycles/min; pulsation ratio, 30:70) with or without reduced vacuum.They found no differences in milk yield among the different milking regimens (Tuor et al., 2022).Differences in milking unit-on time between cows receiving manual premilking stimulation and those subjected to mechanical stimulation have been reported in previous studies, with seemingly contradictory results.Watters et al. (2015) employed 30 Holstein cows with a thrice daily milking schedule and subjected them to 5 different regimens of premilking stimulation: (1) no premilking stimulation; (2) manual premilking stimulation (i.e., forestripping and wiping of teats) with a preparation lag time of 30 s; (3) manual premilking stimulation (i.e., forestripping and wiping of teats) with a preparation lag time of 90 s; (4) high-vibration pulsation for 30 s; and (5) high-vibration pulsation for 90 s.The researchers observed differences among groups with the shortest milking duration in cows receiving 90 s of high-vibration pulsation (LSM ± SE: 245 ± 2 s) and the longest milking duration in cows that received no premilking stimulation [262 ± 2 s (Watters et al., 2015)].Others found no differences in milking unit-on time among the groups (Sagi et al., 1980;Tuor et al., 2022).Care must be taken when interpreting the results among these studies due to differences in the definition of milking unit-on time.That is, some research groups defined the machine-on time as the period between attachment of the milking unit and milking unit detachment, including the period of mechanical stimulation (Sagi et al., 1980;Tuor et al., 2022).This is consistent with our study, as we were not able to deduct the actual duration of stimulation from the total milking unit-on time.In contrast, in the study by Watters et al. (2015), the authors deducted the high-vibration pulsation time, and thus, the milking unit-on time did not include the period of mechanical stimulation.In addition to discrepancies in definitions, differences in the study population, stimulation regimen, and milking machine settings may explain the differences among studies.
We observed no meaningful differences in peak milk flow rate among the groups, suggesting that the APS system was able to elicit a similar milk-ejection capacity as the manual premilking stimulation regimen applied in this study.Our findings are consistent with results reported by Tuor et al. (2022) but differ from those of previous studies (Sagi et al., 1980;Karch et al., 1988).We believe that these discrepancies can be mostly attributed to differences in the manual premilking stimulation regimens.In the 2 older studies (Sagi et al., 1980;Karch et al., 1988), the manual premilking stimulation regimen consisted of 60 s of continuous tactile stimulation.In contrast, Tuor et al. (2022) applied stimulation for durations of 5 and 15 s, and cows in our study received approximately 8 s of tactile stimulation.Such a short stimulation duration may have failed to elicit the cows' maximum milk-ejection capacity, leading to a lower peak milk flow rate compared with a premilking stimulation regimen that includes, for example, 60 s of tactile stimulation.However, the application of a premilking stimulation regimen with 60 s of manual stimulation was not practical in our study.Thus, this idea remains a speculation.To a lesser extent, differences among studies may have been due to differences in study population, milking machine settings, and recording systems.
The interaction between treatment and milk yield indicates that differences in the risk of bimodality increased with increasing milk yield among treatment and control groups.Our results are in accordance with those reported by Watters et al. (2015), who documented a higher incidence of bimodality in cows that received 90 s of pulsation stimulation (i.e., 17% bimodal milk flow curves) compared with cows that were subjected to manual forestripping and a preparation lag time of 90 s (7%).As discussed by Watters et al. (2015), the increased frequency of bimodality in the treatment groups could be due to the inherent nature of the APS system, such that a surge of milk generated during the transition from stimulation mode to milking mode creates the peak, and the subsequent decrease results in a seemingly bimodal milk flow curve observed by the milk meter system.Another possible reason for the observed differences between the treatment and control groups could be due to a premature switch from stimulation to milking mode after the first milk meter dump during the stimulation phase.The analysis of the pulsation chamber vacuum dynamics with the VaDia device indicated that the average stimulation time that was truly applied to most cows in both treatment groups was only 9 s.Although, these measurements were obtained after completion of the study, we believe that they closely reflect the situation during the study.These results are also consistent with our own visual observations during the milking sessions throughout the study.This is also supported by the recorded incremental milk flow rates among groups.During the stimulation phase, the milk harvest is halted, or at least minimized, due to the inherently short milking phase (B-phase).This should have resulted in no or little milk harvested during the first 80 and 99 s in the 2 treatment groups and been reflected by decreased values for the incremental milk flow rates for the first 15-s milk flow rate, 15-30-s milk flow rate, 30-60-s milk flow rate, and 60-120-s milk flow rate as shown by Watters et al. (2015).This premature switch resulted in a shortened preparation lag time that is unlikely to provide sufficient time for the oxytocin to circulate, induce the contraction of the myoepithelial cells, and lead to ejection of the alveolar milk into the duct system, resulting in a transiently interrupted milk flow after the harvest of the cisternal milk fraction.We were not able to discriminate between the milk harvested during the stimulation phase and that harvested during the milking phase.Thus, the extent to which these possible explanations contributed to the observed differences, as well as their biological relevance, remain speculative.
We documented differences in the duration of low milk flow among groups.The observed interaction between treatment and milk yield indicates that differences between treatment and control groups increased with increasing milk yield.Due to the inverse relationship between milk flow rate and milking vacuum (Bade et al., 2009;Ambord and Bruckmaier, 2010), the duration of low milk flow increases the vacuum-induced strain on teat tissues, causing congestion and edema.Differences in the duration of low milk flow among groups may therefore explain the observed differences in teat tissue condition between the treatment and control groups in the current study.

Teat Tissue Condition
A novelty of our study was the serial assessment of the teat tissue condition in cows subjected to APS over a 90-d period.Our data demonstrate that cows in the treatment groups had higher risks of short-term changes to the teat tissue condition.Machine-milking-induced short-term changes to the teat tissue condition are associated with an increased risk of new IMI (Zecconi et al., 1996) and diminish animal well-being (Hillerton et al., 2000).Our findings are therefore of particular importance, as they indicate a potential shortcoming of the tested APS system.We attribute the increased risk of short-term changes to the teat tissue condition in the treatment groups to the following circumstances.First, cows in the treatment groups had a higher duration of low milk flow than cows in the CON group.This difference likely led to a higher magnitude of vacuuminduced forces that induced congestion and edema.Second, cows in the treatment groups were not stimulated at the time of milking unit attachment, leading to attachment of the milking unit to teats with a lower diameter and decreased intracisternal pressure due to the lack of milk ejection into the gland and teat cistern.This difference could have resulted in an inferior fit between the teat and the milking liner barrel, compared with that of cows in the CON group.Poor fit between the teat and the milking liner has been associated with increased mouthpiece chamber vacuum (Borkhus and Rønningen, 2003) and, as outlined by Penry et al. (2017), could have resulted in increased congestion and edema in cows in the treatment groups.Third, cows in the treatment groups showed more bimodality, which is associated with increased mouthpiece chamber vacuum (Moore-Foster et al., 2019).
Cows in the treatment groups had higher risk of teatend hyperkeratosis than those in the CON group.These findings are of particular importance because teat-end hyperkeratosis affects teat canal closure, enhances lodging of pathogenic bacteria, and consequently is associated with an increased risk of new IMI (Neijenhuis et al., 2001b;Neijenhuis et al., 2001a).We believe that differences between the treatment and control groups can be attributed to the cumulative effect of short-term changes to the teat tissue condition.In addition, the increased duration of compressive load of the milking liner during the prolonged d-phase during stimulation may have resulted in increased teat-end hyperkeratosis in cows in the APS80 and APS99 groups (Mein et al., 2003).To our knowledge, this is the first study to investigate the effect of APS on short-term changes to the teat tissue condition and teat-end hyperkeratosis over a 90-d study period.Thus, a comparison with other results is not possible.It is possible, that the milking liner used in this study (i.e., square-shaped barrel) had an interactive effect with the APS system influencing the effect on the teat tissue condition.However, we believe that if such an effect existed, it would have been minor.
Taken together, our data show that the APS system tested here had detrimental effects on the teat tissue condition and failed to facilitate a gentle milk harvest.Future studies investigating the efficacy of APS systems should consider the evaluation of both the short-term teat tissue changes to the teat tissue condition and the teat-end hyperkeratosis.

Somatic Cell Count, Clinical Mastitis, and Culling
Our data show that differences in logSCC among groups were likely due to chance.This finding is in accordance with results reported by Watters et al. (2015).As in previous studies (Wieland et al., 2020a), pretrial logSCC was associated with logSCC.Similarly, no meaningful differences were observed for the new infection risk among groups.Differences in the HR of clinical mastitis among groups were likely due to chance.Taken together, these findings indicate that the APS system had no detrimental effect on udder health during the 90-d study period.We can only speculate if an extended study duration would have resulted in a measurable effect on udder health indices, given the effects of the APS system on teat tissue condition.An extension of our study aims was to assess the effect of treatment on the risk of culling.Our results do not support a measurable difference between groups, though we only measured this outcome over 90 d.When interpreting these numbers, one must keep in mind that the sample size in this study was not sufficient to detect differences in clinical mastitis occurrence or culling risk among groups.Thus, the reader should consider the possibilities of Type II errors, as suggested by the post hoc power analyses.

Study Limitations and Future Directions
Our study had some limitations.First, our study was conducted at a single New York dairy with Holstein cows with a thrice daily milking schedule.Thus, our results are likely to reflect implemented changes in operation of commercial dairies managed similarly to this one.However, the generalizability of our results may be limited to similar operations using the same milking parlor equipment.Second, we included only 7 first-lactation animals.Thus, our results apply mostly to second-and third-lactation animals (or higher), and caution is warranted when interpreting our results with first-lactation animals.Third, the duration of the study was limited to 90 d, and the sample size calculation was not based on assessing udder health indices or culling risk.Future studies that investigate the effect of different APS systems on udder health and the risk of removal from the herd over an entire lactation period are warranted.Fourth, cows in the treatment groups were identified with leg bands to facilitate different regimens of premilking preparation in the milking parlor.However, as a result, farm and study personnel were not blinded to groups during monitoring of clinical mastitis or the assessment of teat tissue condition.This lack of blinding may have led to information bias.Fifth, we were not able to discriminate between stimulation and milking mode data.This hampered our ability to make inferences about milk flow parameters.Last, it is likely that omitting the forestripping step for both treatment groups diminished the milking technicians' ability to detect clinical mastitis cases.We attempted to reduce the possibility of information bias through manual stripping of all cows after milking; however, we may have failed to offset it completely.
Opportunities for future research abound.In addition to investigating APS systems over a whole lactation period, future studies could investigate different types of APS systems.These APS system types could include different settings for pulsation rate and ratio, as well as additional modifications, such as a decrease in the pulsation chamber vacuum and the milkline vacuum as previously described (Weiss and Bruckmaier, 2005;Watters et al., 2015;Neuheuser et al., 2017;Tuor et al., 2022).

CONCLUSIONS
We observed no meaningful differences in milk yield and milking unit-on time between cows that were subjected to the APS system and those that were stimulation by manual premilking stimulation.Differences in bimodality and the time spent in low milk flow rate were likely due to a premature switch from the stimulation to the milking mode, leading to a shortened preparation lag time in the treatment groups.The observed differences in short-term changes to the teat tissue condition and teat-end hyperkeratosis suggest that replacing a traditional regimen of premilking stimulation, including manual forestripping of teats, with the APS system tested here, adversely affects teat tissue condition.Future studies investigating APS systems should include the investigation milking performance and udder health over a whole lactation period, as well as the evaluation of short-and long-term changes (e.g., teat-end hyperkeratosis) to the teat tissue condition.

Figure 1 .
Figure 1.Least squares means derived from general linear mixed models showing the effect of 3 different regimens of premilking stimulation on milk yield (A), milking unit-on time (B), peak milk flow rate (C), and duration of low milk flow (D).Cows in the treatment groups (APS80 and APS99) received automated premilking stimulation for a maximum duration of 80 and 99 s, respectively.Control (CON) cows received conventional premilking stimulation by means of manual forestripping.Error bars show the 95% confidence intervals.P-values are shown for the effect of treatment (A-C) and the interaction between treatment and milk yield (D). (D) Least squares means calculated for milk yields (MY) of 11 and 16 kg.Groups with different letters differ at a level of P < 0.05 according to Tukey-Kramer's post hoc test.

Figure 2 .
Figure 2. Adjusted probabilities (Adj.Prob.) of bimodality from the generalized linear mixed model on the effect of 3 different regimens of premilking stimulation over a 90-d period.Cows in the treatment groups (APS80 and APS99) received automated premilking stimulation for a maximum duration of 80 and 99 s, respectively.Control (CON) cows received conventional premilking stimulation by means of manual forestripping.Error bars show the 95% confidence intervals.P-value is shown for the interaction between treatment and milk yield (MY). Adjusted risks were calculated for MY of 11 and 16 kg.Groups with different letters differ at a level of P < 0.05 according to Tukey-Kramer's post hoc test.

Figure 3 .
Figure 3. Adjusted probabilities (Adj.Prob.) of short-term teat tissue changes (STC) induced by machine milking derived from the generalized linear mixed model on the effect of 3 different regimens of premilking stimulation.Cows in the treatment groups (APS80 and APS99) received automated premilking stimulation for a maximum duration of 80 and 99 s, respectively.Control (CON) cows received conventional premilking stimulation by means of manual forestripping.Error bars show the 95% confidence intervals.Groups with different letters differ at a level of P < 0.05 according to Tukey-Kramer's post hoc test.
Figure 4. Adjusted probabilities (Adj.Prob.) of hyperkeratosis (HK) of the teat-end derived from the generalized linear mixed model on the effect of 3 different regimens of premilking stimulation.Cows in the treatment groups (APS80 and APS99) received automated premilking stimulation for a maximum duration of 80 and 99 s, respectively.Control (CON) cows received conventional premilking stimulation by means of manual forestripping.Error bars show the 95% confidence intervals.Groups with different letters differ at a level of P < 0.05 according to Tukey-Kramer's post hoc test.
Figure 5. Least squares means of SCC (log 10 -transformed, logSCC, A) and adjusted probabilities (Adj.Prob.) of the new infection risk (NIR, B) over the 3 test days during the study derived from general linear mixed models on the effect of 3 different regimens of premilking stimulation over a 90-d period on SCC.Cows in the treatment groups (APS80 and APS99) received automated premilking stimulation for a maximum duration of 80 and 99 s, respectively.Control (CON) cows received conventional premilking stimulation by means of manual forestripping.Error bars show the 95% confidence intervals.(A) Mean values and 95% confidence intervals are presented for the pretrial test day (Test 0).P-values are shown for the interaction between treatment and test day.
Figure 6.Kaplan-Meier survival analysis of time to clinical mastitis event (A) and culling (B) for 427 Holstein cows subjected to 3 different regimens of premilking stimulation over a 90-d period.Cows in the treatment groups (APS80 and APS99) received automated premilking stimulation for a maximum duration of 80 and 99 s, respectively.Control (CON) cows received conventional premilking stimulation by means of manual forestripping.

Table 1 .
Wieland et al.:AUTOMATED PREMILKING STIMULATION Baseline characteristics of 427 Holstein cows subjected to 3 different regimens of premilking stimulation over a 90-d period 1 ing unit-on time, peak milk flow rate, duration of low milk flow, and bimodality.The results of the generalized linear mixed model for each outcome variable are demonstrated in Supplemental Tables

Table 2 .
Descriptive statistics of the milking characteristics from 83,980 milking observations of 427 Holstein cows subjected to 3 different regimens of premilking stimulation over a 90-d period 1Bimodality: bimodality was present if any of the incremental milk flow rates (15-30-s milk flow rate, 30-60-s milk flow rate, or 60-120-s milk flow rate) were lower than any of the preceding incremental milk flow rates (first 15-s milk flow rate, 15-30-s milk flow rate, or 30-60-s milk flow rate, respectively). 2 Wieland et al.: AUTOMATED PREMILKING STIMULATION

Table 3 .
Multivariable generalized linear mixed model showing the effect of 3 different regimens of premilking stimulation, milk yield, and their interaction on the risk of machine milking-induced short-term teat tissue changes Wieland et al.: AUTOMATED PREMILKING STIMULATION