Risk factors for delayed milk ejection in Holstein dairy cows milked 3 times per day

Delayed milk ejection, manifested most often as bi-modal milk flow, occurs when the cisternal milk fraction is removed before the alveolar milk reaches the gland cistern. It is thought to be a consequence of not meeting cows’ physiological needs, due to insufficient premilking teat stimulation, inadequate timing of milking unit attachment, or both. It has been associated with decreased milking efficiency, reduced milk yield, and impaired teat and udder health. Traditionally, portable electronic milk meters have been used to assess the presence of delayed milk ejection in dairy cows. By contrast, incremental milk flow rates from on-farm milk meters and their suitability as a measure to assess delayed milk ejection have not been studied by rigorous methods. The objectives were (1) to describe a protocol for identification of cows with chronically delayed milk ejection (CDME) and (2) to investigate risk factors for CDME using incremental milk flow rates obtained from automated on-farm milk meters. In a retrospective case control study, milk flow data from a 4,300-cow dairy with a thrice-daily milking schedule were obtained over a 1-wk period. Incremental milk flow rates (0–15 s, 15–30 s, 30–60 s, and 60–120 s) were used to identify cows with delayed milk ejection. Cases of CDME were defined as presence of delayed milk ejection at all 21 milking observations. Cows that had no delayed milk ejection at any of the same 21 milking observations were included as controls. A total of 171 cases and 393 controls were included in the study based on these criteria. A logistic regression model was used to evaluate associations of the following risk factors with CDME: parity (1, 2, ≥3), stage of lactation (<100, 101–200, >200 DIM), presence of a nonlactating quarter, milk somatic cell count, average daily milk production, and health and management events. Parity and CDME were associated such that compared with cows in their third or greater lactation, the odds (95% confidence


INTRODUCTION
The analysis of milk flow curves provides valuable information for improving milking efficiency and enhancing udder health by adjusting milking equipment and milking protocols to the physiological requirements of the cow (Sandrucci et al., 2007).The 4 main phases are (1) incline, (2) plateau, (3) decline, and (4) overmilking phase, as described by Tančin et al. (2006).The shape of the milk flow curve during the first 2 min of milking has been related to the effectiveness of premilking udder preparation (Kaskous and Bruckmaier, 2011).Tactile teat stimulation during premilking udder preparation is essential to activate the milk ejection reflex that results in expulsion of the alveolar milk fraction into the gland cistern (Bruckmaier and Blum, 1998).The milk flow curve from a milking event of a cow receiving appropriate tactile stimulation in combination with proper timing of milking unit attachment (i.e., preparation lag time) is therefore characterized by a short (i.e., approximately 15 s) and steep incline progressing uninterruptedly into the plateau phase (Wallace et al., 2003).By contrast, insufficient stimulation or inadequate preparation lag time before unit attachment, or both, are known to result in delayed milk ejection (Bruckmaier and Blum, 1998).
Delayed milk ejection manifests most often as bimodal milk flow.A milk flow curve is defined as bimodal when an increasing milk flow rate is followed by a decreasing flow rate during the first 2 min of the milking observation (Tančin et al., 2006) and is due to removal of the cisternal milk fraction before the alveolar milk reaches the gland cistern (Bruckmaier and Blum, 1998).Bimodal milk flow has been associated with decreased milking efficiency (Bruckmaier and Blum, 1996), reduced milk yield (Erskine et al., 2019), and impaired teat and udder health (Bruckmaier and Blum, 1996;Bruckmaier and Hilger, 2001;Zecconi et al., 2018).Additionally, delayed milk ejection can occur in the absence of a bimodal milk flow (Moore-Foster et al., 2019).Knight et al. (1994) demonstrated that only small volumes of cisternal milk are present until 4 h after milking and that the cisternal fraction only starts filling when the "elastic limit" volume (i.e., the pressure level within the alveolar compartment when milk is forced into the cisternal compartment) has been reached.In cows with short milking intervals, this can lead to small amounts of cisternal milk and, in turn, result in delayed milk ejection without obvious bimodality.
Traditionally, milk flow dynamics have been assessed with portable electronic milk meters that measure continuous milk flow.Among the portable milk meters, the Lactocorder (WMB AG) has gained significant popularity among milk quality consultants and researchers worldwide over the last decades.Data from the Lactocorder have been established as measures of milking machine settings (Wallace et al., 2003), premilking udder preparation (Schukken et al., 2005), and genetic evaluations of milk flow traits of individual cows (Sandrucci et al., 2007;Tamburini et al., 2010;Gray et al., 2011), including bimodality (Samoré et al., 2011).However, over the last decades, on-farm milk meter technology has advanced significantly, and the ability to record individual cow milk flow characteristics more accurately has increased.In specific, several electronic on-farm milk meters from different manufacturers are now able to calculate and record incremental milk flow rates from 0 to 15, 15 to 30, 30 to 60, and 60 to 120 s.This offers unique opportunities to collate these automatically recorded incremental milk flow rates from an individual cow milking observation and identify cows with delayed milk ejection.Despite these recent advancements in on-farm milk meter technology, incremental milk flow rates have not been investigated by rigorous methods.The objectives of this study were (1) to describe a protocol to identify cows with chronically delayed milk ejection (CDME) and (2) to investigate risk factors of CDME using incremental milk flow rates obtained with on-farm milk meters.

Animals and Housing
This retrospective case control study was conducted in November 2019 at a commercial dairy farm located near Ithaca, New York.Because this retrospective study was based on the analysis of data that were collected from routine parlor and herd diagnostics, this study was exempt from animal care and use approval.During the study period, approximately 4,300 lactating Holstein cows were housed in freestall pens, bedded with manure solids, and fed a TMR formulated to meet the requirements outlined by the National Research Council (NRC, 2001).Herd data were maintained in a dairy management software program (DairyComp 305, Valley Agricultural Software).The farm used DHIA services including the individual cow SCC option.The rolling herd key performance indicators were projected 305-d mature equivalent milk production, 13,765 kg; bulk tank SCC, 230,000 cells/mL; monthly clinical mastitis incidence, 7.8%; 21-d pregnancy rate, 29.0%; and culling rate, 36.9%.

Milking System
Cows were milked 3 times daily at 0100, 0900, and 1700 h in a 100-stall parallel rotary parlor (RP3100HD, DeLaval International AB).The vacuum pump (22.4 kW; 30 HP) was regulated by a variable frequency drive and set to supply a receiver operator vacuum of 44 kPa (13.0 inHg).The milking unit was composed of the cluster MC70 (DeLaval International AB) and a milking liner with a round barrel shape (LS-01 NC, DeLaval International AB).The pulsators (EP100, DeLaval International AB) were set to a pulsation rate of 60 cycles/min, a ratio of 65:35, and a side-to-side alternating pulsation.The average claw vacuum during the peak milk flow period was calculated from 10 milking observations using a digital vacuum recorder (VPR100, DeLaval International AB) according to the guidelines outlined by the National Mastitis Council (NMC, 2012) and was 36 kPa.The automatic cluster removers were set to a cluster remover milk flow thresh- old of 1.6 kg/min, a 1-s delay, and a vacuum decay time of 1.5 s.The milk sweep was initiated 3.0 s after unit retraction and lasted for 8 s.Milking system settings and milking characteristics were monitored with a dairy farm management software program (DelPro 5.1, DeLaval International AB).

Milking Routine
The rotational speed of the milking parlor was 5.3 s/stall (rotation time: 530 s), resulting in a theoretical throughput of 679 cows/h.Four teat spray robots (TSR, DeLaval International AB), 2 at the parlor entrance and 2 at the parlor exit, were installed for pre-and postmilking teat dip application.The parlor was operated by two 12-h work shifts, each with 4 milking technicians who were assigned to perform the following tasks at 4 different stations: (1) wipe the teat barrel of all teats from lactating quarters with an individual clean cloth towel, (2) wipe the teat end with an individual clean cloth towel, (3) attach and align the milking unit, and (4) monitor milking liner slips, unit fall-offs, and unit kick-offs, and realign or reattach the milking unit accordingly.Assuming a cow entered the rotary parlor in stall 1, the positioning of the pre-dip teat spray robots and milking stations, respectively, were as follows: robot 1, stall 3; robot 2, stall 5; milking station 1, stall 10; milking station 2, stall 11; and milking station 3, stall 22.The milking technician at station 4 was positioned at the halfway point of the rotary and operated within a wider range to ensure timely adjustment of the milking units.This setup resulted in a dip contact time of 27 to 37 s, duration of tactile stimulation (calculated as the sum of wiping duration of stations 1 and 2) of approximately 6 s, and a preparation lag time (i.e., time from first tactile stimulus to milking unit attachment) of approximately 64 s.

Data Acquisition
Cow Characteristics.Cow-level information (i.e., parity, DIM, SCC, average daily milk yield, and presence or absence of a nonlactating quarter) as well as health and management events (i.e., displaced abomasum, hoof trimming, ketosis, lameness, mastitis, metritis, pen movement, retained placenta, and vaccination) were obtained from the dairy management software program (DairyComp 305).
Milk Flow Characteristics.The following milk flow characteristics were assessed at each milking with electronic on-farm flow-through milk meters using nearinfrared technology (MM27BC, DeLaval International AB) and recorded with the dairy farm management software (DelPro, DeLaval International AB): first 15 s milk flow rate (i.e., average milk flow rate recorded in the first 15 s of milking, kg/min, 15S), milk flow rate 15 to 30 s (average milk flow rate recorded between the first 15-30 s of milking, kg/min, 30S), milk flow rate 30 to 60 s (average milk flow rate recorded between the first 30-60 s of milking, kg/min, 60S), milk flow rate 60 to 120 s (average milk flow rate recorded between the first 60-120 s of milking, kg/min, 120S), 2-min milk yield (amount of milk harvested within the first 2 min of milking, kg), milk yield (yield of milk harvested from start of milking to detachment of the milking unit, kg), milking unit-on time (time recorded from start of milking to detachment of the milking unit, s), average milk flow rate (calculated as milk yield/milking unit-on time, kg/min), peak milk flow rate (calculated as the maximum 60-s average milk flow, kg/min), and time spent in low milk flow rate (time spent below milk flow rate of 1 kg/min between start of milking and detachment of the milking unit, s).Reports for each milking session were exported to a comma-separated values (csv) file once daily.

Study Design
Risk Period and Study Population.The risk period was defined as 1 wk (September 22-28, 2019).The source population from which cases and controls were selected consisted of all lactating cows during the study period.
Case Definitions.For subsequent analysis, a new binary variable (i.e., delayed milk ejection) was created and defined as follows: a delayed milk ejection was present if any of the incremental milk flow rates 30S, 60S, or 120S were lower than any of the previous ones (15S, 30S, and 60S) or if the average milk flow rate during the first 60 s of milking remained below 1 kg/min (i.e., 15S <1 kg/min, 30S <1 kg/min, and 60S <1 kg/min); whereas delayed milk ejection was absent otherwise.The first condition of this case definition was intended to represent bimodality.We employed this algorithm previously and found it a valid tool to evaluate different premilking stimulation regimens (Wieland et al., 2020b).We selected the second condition (i.e., 15S <1 kg/min, 30S <1 kg/min, and 60S <1 kg/min) to identify delayed milk ejection without obvious bimodality.The time interval (0-60 s of milking) was chosen in accordance with Erskine et al. (2019), who suggested that the time interval between 30 and 60 s after the start of milking is a critical indicator of milking performance.We based the threshold value of 1 kg/min on the definition for the milking characteristic "time spent in low milk flow rate," defined as time spent below 1 kg/min milk flow rate, which has been used in previous studies as an outcome variable to assess the effects of differ-ent automatic cluster remover settings (Wieland et al., 2020a) and different premilking stimulation regimens (Wieland et al., 2020b).Cases of CDME were defined as presence of delayed milk ejection at all 21 milking observations during the risk period.
Controls.All cows with a complete record and absence of delayed milk ejection events for all 21 milking observations were included in the study as control cows.
Exclusion Criteria.Cows missing 1 or more milking observations were excluded.

Identification of Cows with Chronically Delayed Milk Ejection.
In a first step, we merged all 7 reports containing milk flow characteristics for each day together with data of cow characteristics from DairyComp 305 in a single Excel file.Subsequently, we screened the data for missing and erroneous values by investigating distributions of 15S, 30S, 60S, 120S, 2-min milk yield, milk yield, milking unit-on time, average milk flow rate, peak milk flow rate, and time spent in low milk flow rate.Observations with a missing value in any of the aforementioned items, or a value of 0 in 2-min milk yield, milk yield, milking unit-on time, or average milk flow rate, were excluded.Next, we calculated the frequency distribution of delayed milk ejection for each individual cow, to identify cases and controls according to definitions outlined previously.Finally, data of case and control cows were extracted into a separate file.
Risk Factors for Chronically Delayed Milk Ejection.Risk factors of interest included recorded events for ketosis (i.e., blood BHB ≥1.2 mmol/L), metritis [presence of fetid uterine discharge and rectal temperature >39.7°C (>103.5°F)],milk fever (presence of weakness, cold ears, or recumbency within 48 h after parturition), retained placenta (presence of fetal membranes ≥24 h after parturition), lameness [locomotion scores 3-5 according to Sprecher et al. (1997)], clinical mastitis (abnormal milk from 1 or more quarters with or without signs of local inflammation according to Erskine et al., 2003), hoof trimming, pen movement, and vaccination, which were diagnosed based on the farm protocols and extracted from the on-farm records.To investigate risk factors for CDME, we constructed a multivariable logistic regression model in 3 subsequent steps.In step 1, we explored health and management events that occurred 2 wk before (wk −2), 1 wk before (wk −1), and during the week of the risk period (wk 0), depending on their presumed influence on milk flow characteristics in the following or the same week, respectively.That is, we assumed that a health event such as ketosis, metritis, milk fever, retained placenta, or lameness could have had a sustained influence on milk ejection of a cow before and during its diagnosis, as the day of diagnosis may lag the beginning of its influence on milk flow.Conversely, management events with known onset and no possible prior effect on milk flow, such as hoof trimming, pen movement, or vaccination, were considered only 2 and 1 wk before the risk period, but not during wk 0. Because cows diagnosed with clinical mastitis were milked separately at the time of diagnosis with a system that bypassed the milk meters, leading to incomplete data, clinical mastitis was considered only at wk −2 and −1.
In step 2, we screened all baseline characteristics and health and management events for inclusion into the multivariable model through univariable analysis.For this purpose, univariable logistic regression models were fitted using PROC LOGISTIC.Presence or absence of CDME was the dependent variable, and baseline characteristics and health and management events were the independent variables and were tested one at a time.Tables 1 and 2 list baseline characteristics, health and management events, and the resulting P-values from univariable analyses.
In step 3, we fitted a multivariable logistic regression model using PROC LOGISTIC.All variables with a Pvalue <0.20 in univariable analysis were considered in the initial model.Collinearity among eligible variables was assessed by calculating Spearman correlation coefficients in PROC CORR.We considered that a coefficient of >|0.60| indicated collinearity.Manual stepwise backward elimination was performed until each of the remaining variables had a P-value ≤0.05, to establish the final model.Finally, biologically relevant 2-way interactions between the remaining variables were tested, and those with a P-value <0.05 were retained to fit the final model.We used deviance and Pearson goodnessof-fit statistics to assess the final model fit and the presence of overdispersion.

Description of Study Population
We obtained 89,487 milking observations from 4,361 cows over the 1-wk risk period.A total of 5,970 (6.7%) milking observations were excluded due to erroneous or missing values, resulting in 83,517 observations from 4,176 cows.Complete records of 21 milking observations were available from 2,937 out of 4,176 cows (70.3%).We identified 171 cases of CDME and 393 controls, contributing 11,844 cow milking observations.Case and control cows were in their first (175, 31%), second (122, 22%), and third or greater lactation (267, 47%), and between 10 and 749 DIM (mean ± SD, 158 ± 107) on the day of inclusion into the study.A total of 76 (13.5%) cows had 1 nonlactating quarter.The mean (±SD) milk yield per milking session calculated from the 11,844 milking observations was 14.3 ± 4.4 kg.The mean (±SD) values for milking characteristics were 3.3 ± 0.8 kg/min for average milk flow rate and 257 ± 56 s for milking unit-on time.Bimodality (i.e., any of 30S, 60S, or 120S lower than any of 15S, 30S, and 60S) was observed in 3,536 out of 11,844 (29.9%) observations, among which 207 (1.7%) had an average milk flow rate during the first 60 s of milking <1 kg/min.We documented 55 of 11,844 (0.5%) milking observations that had an average milk flow rate <1 kg/min during the first 60 s of milking without bimodality.This resulted in a total of 3,591 of 11,844 (30.3%) observations with delayed milk ejection.The median milk SCC from the last DHIA test day on September 4, 2019, was 51,000 cells/mL.Tables 1-3 depict frequency distribution of health and management events, baseline characteristics, and descriptive statistics of case and control cows, respectively.Figure 1 illustrates the frequency distribution 15S, 30S, 60S, and 120S of case and control cows.

Identification of Cows with Chronically Delayed Milk Ejection
Here we provide a protocol for objective identification of cows with CDME using the incremental milk flow rates obtained from automated electronic on-farm milk meters.The idea for this project arose through our extension service activities with dairy producers who ask for help identifying cows whose maximum physiological milk ejection capacity is not elicited with their current milking routine.Being able to automatically detect CDME events would allow producers to target cows that may benefit most from an enhanced premilking stimulation regimen.Because changes in premilking stimulation require more labor and incur higher operat-ing costs, equipment and labor resources could be used most efficiently.
Our proposed case definition of delayed milk ejection comprises 2 conditions.Based on the principle outlined by Tančin et al. (2006), bimodality, the first condition for delayed milk ejection, was fulfilled when any of the incremental milk flow rates 30S, 60S, or 120S were lower than any of the previous ones.The second condition represented delayed milk ejection without obvious bimodality (i.e., a delayed milk ejection was present if 15S <1 kg/min, 30S <1 kg/min, and 60S <1 kg/min) and was based on the idea from Moore-Foster et al. ( 2019) that delayed milk ejection can also occur in the absence of a bimodal milk flow when only small amounts of cisternal milk are available.Our eligibility criteria (i.e., number of observed milking observations included for each cow) and case definition for CDME (i.e., number of milking observations with delayed milk ejection per total number of milking observations observed) were conservative.Our reasoning was that we wanted to examine the most extreme scenarios, allowing for the best separation between cases and controls.Both the eligibility criteria as well as the threshold value can be adjusted to classify a cow at risk of CDME.This would allow dairy operators to achieve individual farm goals and adjust the case definition according to the capacity of their milking parlor and personnel.

Risk Factors for Chronically Delayed Milk Ejection
Our second objective was to investigate risk factors for CDME in the study herd.For this purpose, we used data from 1 dairy operation.Our results show that parity was associated with CDME such that, compared with cows in the third or greater lactation, the odds of CDME were greater for cows in first and second lactations.We believe that the observed variability can be attributed to differences in familiarity with the milking environment among cows of different parities and its relationship to milk ejection.Previous research has showed that cows milked in unfamiliar surroundings are subject to central inhibition of milk ejection due to no or reduced release of oxytocin (Bruckmaier et al., 1993).We speculate that, in the current study, cows in first and second lactation were less familiar with the milking center environment compared with cows in parity 3 or greater.This could have led to reduced oxytocin release and inhibition of the milk ejection reflex, which in turn may help explain the increased risk of CDME in these groups of animals.In a recent study investigating the effect of 2 different premilking stimulation regimens on milking characteristics, we found that second-lactation animals had greater odds of bimodality compared with cows in third or greater lactation (Wieland et al., 2020b).Conversely, first-lactation cows had lower odds of bimodality in that study (Wieland et al., 2020b).Sandrucci et al. (2007) analyzed a total of 2,486 milk flow curves from cows in 82 Holstein-Friesian dairy herds in Lombardy, Italy, using a portable milk meter (Lactocorder, WMB AG).They found that the percentage of bimodality was not different (P = 0.43) between first-lactation animals (34.7%) and cows that were in second and greater lactation (33.1%).We believe that differences in study herds and diagnostic techniques help explain the discrepancies between results of the current study and those reported previously.Differences in the case definitions for bimodality and CDME as detected with the Lactocorder and the on-farm milk meters, respectively, may be another reason.
Stage of lactation was associated with CDME such that late lactation cows (>200 DIM) had higher odds of CDME compared with cows in early (≤100 DIM) and mid-lactation (101-200 DIM).These results are similar to those in the aforementioned study by Sandrucci et al. (2007), in which the authors found that 27.3% of cows <150 DIM and 40.6% of cows >150 DIM had a bimodal milk flow curve.Our results are in contrast to those reported in a recent study (Erskine et al., 2019) in which researchers recorded milking data from 663 Holstein cows on a single 3,600-cow dairy in Michigan with a thrice-daily milking schedule, using digital vacuum recorders to investigate the relationship between bimodal milk flow and milk yield.They observed no evidence that the probability of delayed milk ejection differed significantly within different categories of stage of lactation (Erskine et al., 2019).Discrepancies in study population, management factors inherent to the study herds, diagnostic techniques, and thresholds used to assign stage of lactation categories could be variables that account for differences between studies.The novelty of our study was the assessment of bimodality over a 1-wk period.This likely strengthened our observations, as it allowed the outcome variable to be based on serial assessment rather than a single milking observation.
Average daily milk yield was associated with lower odds of CDME such that a 1-kg increase decreased the odds of CDME by 11%.This finding supports the results reported by other investigators (Bruckmaier and Hilger, 2001;Kaskous and Bruckmaier, 2011).Bruckmaier and Hilger (2001) used 18 cows to investigate the relationship between udder filling and milk ejection.The authors calculated the degree of udder filling as a function of the actual milk yield and maximum storage capacity.The maximum storage capacity of the mammary gland was estimated as 50% of the average daily milk yield in the second month of the current lactation.They found that the delay until start of milk ejection increased with decreasing udder filling.Kaskous and Bruckmaier (2011) conducted 2 subsequent experiments in 7 and 21 cows, respectively, to study the suitability of different premilking stimulation regimens in cows with different degrees of udder filling that were milked twice daily.The investigators confirmed the findings by Bruckmaier and Hilger (2001) and concluded that the delay in milk ejection due to udder fill needs to be accommodated by stimulation time and preparation lag time.As outlined by Samoré et al. (2011), delayed milk ejection can be attributed to 2 separate but interrelated conditions: (1) delayed ejection of the alveolar milk fraction into the cistern (Bruckmaier and Hilger, 2001) or (2) a progressively decreasing cisternal milk fraction over the course of lactation, as demonstrated by Caja et al. (2004).Accordingly, an early lactation cow with a large amount of cisternal milk can exhibit a bimodal milk flow curve if the degree of tactile stimula-tion is not sufficient to elicit the milk ejection reflex or the timing of unit attachment is improper.A late lactation cow with a small amount of cisternal milk will not show a bimodal milk flow curve if premilking stimulation intensity and timing of unit attachment accommodate the individual physiological requirements of the cow.Evidence also suggests that, under certain circumstances, delayed milk ejection can cause reduced milk yield.The underlying concept has been recently described by Erskine et al. (2019) and seconded by work from our group (Wieland et al., 2021).However, the reader should keep in mind that a cause-effect relationship cannot be inferred from this observational study.
In addition to variables that have been previously reported, we investigated possible risk factors for delayed milk ejection that have not been investigated by rigorous methods, to the best of our knowledge.We found that presence of lameness at wk 0 and a vaccination event at wk −1 increased the odds of CDME.These results are of particular importance because they expand the current knowledge about risk factors that possibly influence milk ejection in dairy cows.We believe that the observed associations can be attributed to the negative effects of lameness and vaccination on milk production.Several researchers have reported a negative effect of lameness on milk production (Warnick et al., 2001;Green et al., 2002;King et al., 2017).Similarly, a post-vaccinal milk drop has been observed by several authors (Bosch et al., 1997;Scott et al., 2001;Bergeron and Elsener, 2008).Under certain circumstances, milk production is linked with the relative degree of udder filling, which in turn has been shown to be positively associated with milk ejection, such that higher udder filling facilitates milk ejection (Wellnitz et al., 1999).Conversely, lower milk production or decreased relative degree of udder filling may result in delayed milk ejection.In a recent study (Wieland et al., 2020b), we found that milk yield was associated with bimodality such that a 1-kg increase in milk yield per milking session decreased the odds of bimodality [OR (95% CI): 0.81 (0.79-0.83)].Another reason that may help explain the observed association between lameness and CDME is a possible interrelationship between lameness, inflammation, and milk ejection.There is consensus that lameness is associated with inflammation (Whay and Shearer, 2017), which in turn may negatively influence milk flow dynamics through impairment of the milk ejection reflex.Our theory is supported by results reported by Zecconi et al. (2018), who showed that application of ketoprofen decreased the frequency of bimodality in cows with chronic mastitis.The authors attributed the observed effect to a decrease in the inflammatory response with the application of ketoprofen (Zecconi et al., 2018).However, because the study addressed inflammation of the mammary gland rather than inflammation of the limb, our explanation remains speculative.
Our final multivariable logistic regression model yielded a robust adjusted coefficient of determination (R 2 adj = 0.54).The adjusted coefficient of determination may be interpreted such that the model explains 54% of the variability.We believe that some of the unexplained variability can be attributed to unobserved variations in premilking stimulation among cows that are likely to occur on dairy operations between milking shifts and over a 7-d period.
Measures to manage cows at risk of CDME include changes in milking routine and parlor settings.Modern parlor technologies facilitate application of different automated premilking stimulation regimens to individual cows, using inherent features such as the pulsator system to stimulate milk ejection.A complementation of the protocol described herein with current parlor technologies would therefore offer unique opportunities to adjust milking protocols to the physiological requirements of individual cows.In the current study population, where manual forestripping during premilking stimulation was omitted during the study period, managing cows with CDME would likely require an increase in both the duration of tactile stimulation and the preparation lag time.According to Kaskous and Bruckmaier (2011), 15 s of stimulation followed by a latency period of 45 s (i.e., preparation lag time of 60 s) is necessary to achieve adequate milk letdown in midlactation animals.Conversely, stimulation time must be increased to 30 s with a subsequent latency period of 60 s (i.e., preparation lag time of 90 s) for cows with very low udder filling (Kaskous and Bruckmaier, 2011).This is in accordance with results of a recent study from our own group, where we tested the effects of 2 different premilking stimulation regimens on milking performance and teat tissue condition (Wieland et al., 2020b).We found that cows whose teats were forestripped and wiped for a total duration of 16 s had lower odds of bimodality and teat tissue changes after machine milking compared with cows whose teats were wiped for 7 s.Evidence of the effect of preparation lag time on the risk of bimodality has been provided by Watters et al. (2015).The researchers showed that cows who received no premilking stimulation, or premilking stimulation and a short preparation lag time of 30 s, had a higher incidence of bimodality compared with animals that were prestimulated and had a 90 s of preparation lag time (Watters et al., 2015).

Study Limitations and Future Directions
Although our findings are encouraging to further investigate risk factors for CDME, our study had several limitations.First, our study was carried out on only 1 dairy farm.The external validity of our study is therefore limited.A peculiarity of this study herd was likely that the manual forestripping step during premilking stimulation was omitted during the study period.The reader should keep this in mind when comparing frequency distributions of other study populations with the one described herein.Second, some mention should be made of our case definition for delayed milk ejection in the current study.Our choice was based on previous reports from groups with extensive experience in the analyses of milk flow dynamics (Tančin et al., 2006;Moore-Foster et al., 2019).However, although we were able to identify previously reported risk factors for delayed milk ejection, the biological relevance of the applied case definition has yet to be confirmed by additional studies.Future research should therefore focus on identifying the most discriminatory case definition.Decreased milk yield, as suggested by Erskine et al. (2019), changes to the teat tissue condition after machine milking, as well as a comparison with other portable milk flow meters (e.g., Lactocorder) could be variables that help optimize the case definition for delayed milk ejection using incremental milk flow rates from electronic on-farm milk meters.Last, as indicated by the wide 95% CI, the uncertainties of the magnitude were relatively high for the risk factors lameness and vaccination events.This could be due to the low number of observations in our data set.Although the observed associations are biologically coherent, the observed associations should be interpreted with caution and confirmed by future studies.To increase external validity, future studies should be designed as multicenter studies, representing a variety of dairy operations with different parlor systems and milking routine regimens.

CONCLUSIONS
The protocol described herein can serve to identify cows whose physiological requirements for adequate milk ejection are not met with the current milking routine in a dairy herd.This would help dairy producers target cows that benefit most from a more laborious and cost-intensive premilking stimulation regimen, allowing producers to use their parlor and labor resources most efficiently.Incremental milk flow rates from electronic on-farm milk meters may be a useful tool to monitor milk flow dynamics on dairy farms and study potential risk factors, including health and management events that are associated with delayed milk ejection.Future research is warranted to determine a biologically relevant threshold for delayed milk ejection using incremental milk flow rates.
Wieland et al.: DELAYED MILK EJECTION Data were managed in Microsoft Excel (2019 version, Microsoft Corp.) and JMP (version 14, SAS Institute Inc.).Statistical analyses were performed with the software package SAS (version 9.4, SAS Institute Inc.).Figures were created in GraphPad Prism (version 7, GraphPad Software).Descriptive Statistics.Boxplots were generated with PROC UNIVARIATE to visually assess normality of continuous variables.Descriptive statistics were generated with PROC MEANS and PROC FREQ.

Table 1 .
Wieland et al.:DELAYED MILK EJECTION Frequency distribution of fresh cow disease events (i.e., 1Percentages are provided in parentheses.P-values derived from univariable logistic regression analyses testing the association between the respective health or management event and the presence or absence of chronically delayed milk ejection.

Table 2 .
Baseline characteristics of Holstein dairy cows with (cases, n = 171) and without (controls, n = 393) chronically delayed milk ejection 1 1P-values derived from univariable logistic regression analyses testing the association between the respective baseline characteristic and the presence or absence of chronically delayed milk ejection.Percentages are provided in parentheses; may not add up to 100 due to rounding error.LogSCC and average daily milk yield values are mean ± SD.

Table 3 .
Wieland et al.:DELAYED MILK EJECTION Descriptive statistics of milk yield (kg/milking session) and milk flow characteristics of Holstein dairy cows with (cases, n = 171) and without (controls, n = 393) chronically delayed milk ejection 1

Table 4 .
Wieland et al.:DELAYED MILK EJECTION Results of multivariable logistic regression model showing the association of parity, stage of lactation, average daily milk yield, presence of a lameness event during the risk period (wk 0), and presence of a vaccination 1 wk before the risk period (wk −1) with chronically delayed milk ejection 1 3 aOR = adjusted odds ratio.