If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
To determine whether individual cow milking vacuum (within the short milk tube and the liner mouthpiece) could be substituted for milk flow technology to identify delayed (bimodal) milk ejection, and the possible relationship between bimodal milk flow and milk yield, we recorded milking data from 663 Holstein cows on a 3,600-cow Michigan dairy that milked 3 times per day. Overall, delayed milk ejection occurred in 45.6% of the milkings, and 98% of the cows with delayed milk ejection also had bimodal flow. Multivariable analysis revealed that milk yield during each individual cow milking was positively associated with increasing lactation number but negatively associated with increasing days in milk and delayed milk ejection. As the time between unit attachment and the estimated milk letdown (the lag period) increased, milk yield decreased; relative to a lag of <30 s, milk yield decreased by 1.8 and 3.1 kg for lags of 30–59 and ≥60 s, respectively. The final multivariate model had an adjusted coefficient of determination of 0.27. The negative association between delayed milk ejection and decreased milk yield in this study suggested that milking vacuum parameters from individual cows could serve as a useful tool to qualitatively estimate milk flow within a herd and that this information may be used to enhance herd productivity.
Analysis of milk flow dynamics can improve udder health and milking efficiency by highlighting opportunities to improve milking protocols and equipment function that align with the physiology of the cow (
Best combination of pre-stimulation and latency period duration before cluster attachment for efficient oxytocin release and milk ejection in cows with low to high udder-filling levels..
). Conversely, insufficient stimulation before milking results in delayed milk ejection (DME) and bimodal milk flow during the incline phase, which is associated with poor milking efficiency, impaired teat health, and possibly reduced milk yield (
). Bimodal milk flow periods are defined as those when milk flow is interrupted after removal of cisternal milk but before alveolar milk reaches the gland cistern (
The most commonly used instruments to assess milking dynamics are those that measure continuous milk flow, either portable or incorporated into the milking system. However, use of instruments that record vacuum in the milking unit might serve to estimate key changes in the phases of milk flow dynamics during individual cow milkings. For example, mouthpiece chamber (MPC) vacuum (VMPC) is strongly and negatively correlated with milk flow and positively associated with teat congestion (
). Delayed milk ejection results in poor contact of the teat barrel with the wall of the liner during the transient period of zero flow, resulting in an increase in VMPC as milking vacuum penetrates into the MPC and subsequently causes teat congestion and poor milk flow (
VaDia digital vacuum recorders (Biocontrol, Rakkestad, Norway) measure simultaneous vacuum events during milking in 4 channels while attached to the cluster. One investigator can attach and monitor numerous recorders in a milking parlor simultaneously, allowing collection of data from large numbers of cows at different positions within milking strings. This approach was reported by
in a study of 64 dairy herds that found an association between an increased risk of DME and decreased teat stimulation before milking. Although digital vacuum recorders are becoming more commonly used by veterinarians and milk quality consultants to help assess milking performance in dairy herds and can reliably identify key time points of change relative to milking curves, little is known of the relationship between the dynamic of milking vacuum (as a proxy measure for milk flow) and milk yield, especially in large commercial dairy herds. Additionally, it is unclear whether digital vacuum recorders can serve as a possible proxy for milk flow meters to estimate this relationship.
The purpose of this study was to determine whether milking vacuum, measured in the short milk tube (SMT) and MPC with digital recorders, could serve to (1) identify key time points of phase changes in milk flow, (2) quantitatively determine the lag period (time between cluster attachment and start of the incline phase) related to DME, and (3) determine the association of DME on milk yield in a Michigan dairy herd.
MATERIALS AND METHODS
Study Herd, Cows, and Data Collection
Data for this study were collected from a commercial dairy herd with approximately 3,600 cows that milked 3 times per day in a double-30 herringbone parlor, with a low-line milking plant in a basement level below the parlor. The vacuum was supplied by a variable speed pump (18.7 kW; 25 HP) and vacuum levels were 45.7 kPa (13.5 inHg) at the pump and 44.9 kPa in the milking vacuum line above the sanitary trap. The farm has separate barns and a double-20 herringbone milking parlor for recently fresh cows (<35 DIM, approximately), cows with discarded milk (including cows in the hospital pen), and those with clinical mastitis, lame cows, and cows that are within about 30 to 40 d of drying off. Thus, our data excluded those cows that were milked in the double-20 parlor. The herd was on monthly DHI test to collect SCC, milk fat, and milk protein data, but milk weights were recorded at each milking with the use of an infrared-based, continuous flow meter system (MM27BC, DeLaval, Kansas City, MO). The DHI SCC data were used to determine the most recent test-date linear SCC score (LS) for each cow on the trial.
During the 6-mo period from October 2016 through March 2017, and excluding DHI test dates, we evaluated individual cow vacuum events during milking by attaching VaDia recorders on the first, third, fifth, sixth, eighth, tenth, eleventh, and fifteenth milking clusters on the west side of the parlor. The vacuum analysis was performed during a single milking from 836 Holstein cows on 8 separate days during the 6-mo trial period and only during the second (afternoon) shift of 3 milking shifts each day. Thus, milking groups of cows and milking personnel were consistent throughout the trial. However, 123 cows were found to have been recorded more than once; a priori at the start of the trial, we decided to include only the first test from each cow so as to have each cow give equal weight to the analysis. Additionally, we excluded 31 cows that were milking with only 3 quarters, 18 cows that were determined (based on either observation or from analysis) to have a milking cluster fall off and reattached during the milking event, and 1 cow that was determined to have the onset of clinical mastitis during our parlor evaluation. Thus, our final data set included 663 cows.
Digital vacuum (reported as kPa) was recorded by inserting 2.4-mm (i.d.) silicon tubing into the following positions on the milking cluster to record vacuum: (1) a rear quarter liner MPC, (2) a front quarter liner MPC, (3) the SMT, and (4) a short pulsation tube. Milking periods for each cow were continuously recorded from the time milking clusters were attached until removed, either automatically by cluster detachers or manually by employees. The automatic detachers were set to remove the clusters at a flow of 1.1 kg/min (2.5 lb/min) with a 3-s delay and vacuum decay time of 3.5 s. The start of milking delay was set at 120 s. All vacuum recordings were downloaded and then reviewed using the VaDia Suite software (Biocontrol). We determined key events such as unit attachment, start of milk flow, and cluster removal for each cow by manual evaluation rather than the default settings of the VaDia Suite software. For consistency, one investigator (RJE) conducted the parlor evaluations, which included visually evaluating the milking protocols, recording milk weights (kg), identifying the cow and stall that each cow occupied during milking, and interpreting the digital vacuum recordings for all cows on the study. The lag period between cluster attachment (start of milking time) and the start of the incline phase of milk flow (recorded in seconds) was then calculated and termed “let-down time” (LDT), as previously described and demonstrated in example plots (
). Briefly, for each cow, the start of the incline phase was marked when the vacuum level in at least 1 of the 2 MPC channels decreased to <13.5 kPa (4 inHg) and the vacuum in the SMT (VSMT) decreased from maximum and fluctuated by ≥3.4 kPa (1 inHg). We selected the VMPC cut-point based on extensive milking system experience of one author (RST), our previous research to determine the prevalence of DME in dairy herds (
), and the round geometry of the liner (DeLaval LS-01 R12) that is designed to reduce liner slips in a system with abundant vacuum reserve. Bimodal milk flow was defined when VMPC and VSMT decreased after the start of milking but then markedly increased, indicating decreased milk flow.
To better elucidate the vacuum differences between cows that were determined to have DME versus normal milk ejection, we manually elongated the recordings and used the “average” function of the software to determine average VMPC for both the front and back liner and VSMT during the 60-s period after cluster attachment. For VMPC, we reported the mean (kPa) of the 2 liners (front and rear).
In addition to milking time vacuum analysis, we observed milking procedure and protocols to determine the parlor routine, the time of teat stimulation, and the interval between first teat stimulation and cluster attachment (latency period).
Statistical Analysis
For each cow, we entered identification; milk weight (kg); the VaDia analysis parameters LDT (s), milking time (s), VMPC and VSMT during the first 60 s of milking (kPa), and overmilking time (s); previous DHI LS; lactation; and DIM into Excel (Microsoft Corp., Redmond, WA) and then imported the data into Stata 15.2 (Stata Corp., College Station, TX) for data management and descriptive and inferential statistical analyses. The frequency distribution of estimated LDT in Figure 1 showed a bimodal distribution. There was a break between cows with an LDT <30 s and cows with an LDT ≥30s, which was also observed on scatterplots for individual animals (not shown). This is similar to the distribution reported by
. For this reason, we chose to treat LDT as a categorical variable: cows with LDT ≥30s were determined to have DME compared with cows with LDT <30s. Among cows with LDT ≥30s, we additionally calculated the proportions of cows with LDT for the following categories: 30 to 59 s and ≥60s.
Figure 1Distribution of time from cluster attachment to let-down time (start of sustained milk flow, also termed “lag period”) as measured by digital vacuum recorders for 663 dairy cows on a Michigan dairy.
For bivariable and multivariable analyses, explanatory variables proposed to be associated with milk yield were categorized because they were bimodal (LDT) or right skewed [lactation category (LACT), DIM, and LS]. Categories for LDT are described above; LACT was divided into 1, 2, and 3+; DIM into <150, 150–199, 200–249, and ≥250; and LS into <4 and ≥4. The DIM categories were designed, in part, to represent a similar peak milk category (<150 DIM) for first-lactation and multiparous cows.
The possible association of LDT on milk yield was investigated using multivariable linear regression adjusting for LACT, DIM, and LS. Bivariable associations between milk yield and LDT, LACT, DIM, and LS with P < 0.2 were eligible for use in the multivariable model. For the multivariable model, first-order interaction terms were created to account for potential effect modification by model variables. Using the Wald test in the contrast command in Stata, we used a manual backward elimination procedure to build the final multivariable model until only significant (P < 0.05) variables and interactions were retained. The residual distribution was assessed visually for normality and homoscedastic variance. Least squares means (LSM) of milk yield for the 3 LDT categories and other variables when adjusted for all other variables in the model were estimated using the margins command and treating all factor variables as balanced (“as balanced” option in the margins command).
RESULTS AND DISCUSSION
Each cow was prepared for milking by one employee, who performed 3 passes in 5-cow territories. Eight 5-cow milking strings were observed to attain duplicate observations for each of the 4 employees milking in the parlor. Teats were dipped with a germicide during the first visit to each cow, and then stripped (mean duration 3.6 ± 0.2 s) in the second pass. During the third pass, teats were dried (mean duration 6.4 ± 0.3 s) with a separate cloth towel, and milking clusters were immediately attached after drying each cow. The average time between loads (30-cow row) of cows in the parlor was 11.5 min, or 5.2 turns of the parlor per hour, which was higher than the mean for 64 Michigan dairy herds (4.4 turns per hour;
). This rate was facilitated by a crowd gate in the holding pen and rapid exit system, which minimized employee interactions with cows. Nonetheless, the 3 employees dedicated to premilking preparation for each string of cows (a fourth disinfected teats after milking) were pressed to maintain milking groups on schedule.
Descriptive statistics for milk yield, variables potentially associated with milk yield, and vacuum parameters are presented in Table 1. The distribution of duration of milking time (total time cluster was attached) was slightly right skewed, and the median time of attachment was 279 s and ranged from 148 to 682 s (mean = 291 ± 72 s; Table 1). The proportion of cows that completed milking in ≤360 s was 85.5% (95% CI: 82.6–88.1%). Only 62 cows (8.8%; 95% CI: 6.7–11.2%) were estimated to have been overmilked by more than 30 s. However, as the VaDia units only measured vacuum in 2 of the 4 MPC for each cluster, and considerable variation can exist between quarters in milk flow rate (
), this could be considered a limitation of the study. Nonetheless, we believe the high proportion of cows that were milked in under 6 min coupled with the low proportion of cows that were overmilked reflects the drive in this herd to maximize parlor throughput, as discussed earlier relative to the low time of teat stimulation.
Table 1Descriptive statistics of digital vacuum determined let-down time, lactation, DIM, linear SCS at previous DHI test date average, duration of milking cluster attachment (unit-on time), average mouthpiece chamber vacuum (VMPC; mean of a rear and front quarter) and short milk tube vacuum (VSMT) during the first 60 s after cluster attachment for 663 cows on a Michigan dairy
Bimodal milk ejection, indicated by an initial decrease in VSMT and VMPC after cluster attachment followed by an increase, was not observed on manual inspection of VaDia recordings for any of the 361 cows that had an LDT of <30 s. Conversely, 296/302 (98.0%) of cows with a LDT ≥30 s demonstrated bimodal milk ejection, and the remaining 6 cows demonstrated high MPC vacuum and low within-pulsation-cycle VSMT fluctuations (indicative of low milk flow) despite a lack of bimodality. Thus, in our study herd, we estimated that 302/663 cows (45.6%; 95% CI: 41.7–49.4%) had DME. This is higher than previous studies in Michigan (25.0%;
). However, it should be noted that the Italian studies measured actual milk flow, not vacuum, to determine DME, and all of the previous studies had a more robust representation of herd populations due to inclusion of >50 herds in each.
This higher proportion of DME in our study herd relative to previous studies may be related to the observed milking protocols. Stimulation during the second pass (stripping) was <3 s and total stimulation (including drying) averaged less than 10 s.
reported that the optimal duration of prestimulation to reduce delayed milk ejection was 90 s in udders containing small amounts of milk but only 20 s in well-filled udders. In mid-lactation cows, stimulation of at least 15 s before milking is ideal for optimal milk let-down if a latency period between first stimulation and unit attachment of at least 45 s is included (
Best combination of pre-stimulation and latency period duration before cluster attachment for efficient oxytocin release and milk ejection in cows with low to high udder-filling levels..
). In our opinion, the short time of stimulation may have reflected the need for employees to maintain parlor throughput. Previous studies found that a greater number of clusters per operator (
reported that the frequencies of bimodal milking among cows that had clusters attached immediately, attached after dipping and forestripping followed by a 30-s latency period, or attached after dipping and forestripping followed by a 90-s latency period, were 21, 14, and 7%, respectively. For our study herd, the mean latency period from stripping to unit attachment was 86 ± 3 s, which is considered acceptable for most routines, and suggested that the key risk for DME was inadequate stimulation.
Results of the bivariable analyses of milk yield and previously described explanatory variables are shown in Table 2; SCC, LACT, DIM, and LDT were all found to meet the requirements for inclusion in the multivariable analysis.
Table 2Bivariable analyses estimating associations between milk yield (kg) and let-down time (s), lactation, DIM, and linear SCS for 663 cows on a Michigan dairy
Our final multivariable model, which included the categorical variables LDT (P < 0.001), DIM (P < 0.001), and LACT (P < 0.018), and LACT × DIM (P < 0.001) and LACT × LDT (P < 0.05) interactions, revealed a negative association on milk yield as LDT increased (Table 3). This agrees with 2 Italian field studies that found an association between better udder preparation and greater milk yield and less bimodal milking (
). Interestingly, in our study, there was a “dose effect”; that is, milk yield decreased with each 30-s increment in the time interval from unit attachment to start of milk flow. Adjusting for DIM, LACT, and their interaction, cows with LDT of 30 to 59 s and ≥60 s yielded 1.8 and 3.1 kg less milk (Table 3), respectively, than cows with LDT <30 s.
Table 3Multivariable analysis model of milk yield (kg) for 663 cows on a Michigan dairy
With respect to vacuum dynamics of the milk cluster, we believe the 30 to 60 s after the start of milking is perhaps the most critical indicator of milking performance. When we compared average VMPC and VSMT for the first 60 s of milking using the nonparametric Dunn's test of multiple comparisons (rank sums), we found that as LDT increased (P < 0.001), so did VMPC and VSMT (Table 4). Thus, a strong positive relationship exists between DME and these measures of vacuum during the early phase of the milking curve that mirrored the “dose effect” between LDT and milk yield.
Table 4Comparisons of average mouthpiece chamber vacuum (VMPC; mean of rear and front quarter) and short milk tube vacuum (VSMT) during the first 60 s after cluster attachment by digital vacuum determined let-down time for 663 cows on a Michigan dairy
Values within a column with different superscripts are significantly different at P < 0.001 using Dunn's nonparametric test of multiple comparisons.
0.16
42.8–43.4
a–c Values within a column with different superscripts are significantly different at P < 0.001 using Dunn's nonparametric test of multiple comparisons.
1 As measured by digital recorders after teatcup attachment.
The mechanism for the association between DME and decreased milk yield, and the fact that milk loss increases with longer DME, may be explained by the higher VMPC in cows with DME than in cows with normal milk ejection. Early research noted discoloration from congestion and circular rings at the base of the teat associated with high VMPC (
). When milk flow is low or nonexistent, the teat barrel becomes thinner from the lack of positive pressure within the teat cistern, which results in a poor seal between the teat and liner wall (
). This decreases the depth of teat penetration into the barrel of the liner, which in turn increases leakage of milking vacuum into the MPC from the teat end, thus increasing VMPC (
). An increased duration of high VMPC, as occurs with longer DME, will likely cause the liner mouthpiece to have a tighter seal at the base of the teat, constricting the cricoid (annular) ring and impeding milk flow between the gland and teat cistern (
). Elevated VMPC may further congest the teat barrel, which decreases the cross-sectional area of the teat canal during the milking phase of pulsation (
). Thus, a transient decrease in milk flow rate leads to repositioning of the teat within the liner and subsequent constriction on the junction between the gland and teat cistern and, on average, decreased milk harvested during an individual milking.
Earlier reports found that milking without prestimulation caused prolonged milking times and reduced milk flow rates but did not consistently affect total milk yield (
), there were no significant differences in milk yield, milk unit attachment time, or milk flow for cows that were forestripped compared with those that were not. In a New York study, cows that received no premilking stimulation, or prestimulation and lag times of only 30 s, were more likely to have bimodal milking compared with cows that were prestimulated and had 90 s of lag time (
). Both the Wisconsin and New York studies had relatively small numbers of cows that received all preparation treatments in a crossover design. Our study included considerably more cows but they were evaluated only once. Additionally, in the Wisconsin study, cows were milked twice a day and the influence of milk pressure on milk ejection likely differed compared with that of cows milking 3 times per day, as in our study herd. This is supported by reports that found higher intramammary milk pressure is likely in cows that are in early lactation and milked less frequently (
Because prestimulation is more important in cows with reduced udder filling in late lactation, as described above, we speculated that late-lactation cows might have accounted for most of the DME events observed in this study. However, an interaction between LDT and DIM was not significant (P = 0.125) when tested in the final model. Hence, there is no evidence in our data that the probability of DME differed significantly within DIM categories (Table 5). For example, the 56% of cows <150 DIM without DME (LDT category <30 s) was essentially equivalent to cows in other DIM categories (45 to 60%; Table 5). Similarly, the 32% of cows <150 DIM with a DME of 30 to 59s did not differ from cows in other categories (28 to 33%). Least squares means estimates of milk yield for LDT categories did not change when the LDT × DIM interaction term was added to the final model. This is supported by
, who reported that milk ejection is, in large part, a function of udder filling (as a percentage of storage capacity) after the start of teat stimulation. Although there is a slight trend toward more rapid milk ejection in earlier compared with later lactation cows, decreases in time to milk ejection were more profound than stage of lactation following longer (12 h) milking intervals compared with shorter (8 h) milking intervals (
). Thus, as mentioned earlier, the frequency of milking in our study herd may be a more important explanatory variable for the prevalence of DME, along with the short stimulation time, than stage of lactation. Our results could have been affected by the thresholds we used to assign stage of lactation categories, however.
Table 5Cross-tabulation of let-down time (s) and DIM categories
As expected, our model also showed that milk yield decreased with increasing DIM and increased with increasing LACT (Table 3). The minor decrease in estimated LSM milk yield in the DIM category 200–249 and no decrease in the DIM category ≥250 compared with the 150–199 category might be explained by selective culling of cows in later DIM because of low production or reproductive infertility.
Our model had a fairly robust adjusted coefficient of determination (R2adj = 0.27).
estimated that the heritability for bimodal milk ejection was 0.43, indicating a genetic correlation with milk flow traits. We did not include genetic variability in our model, although unlike the previous studies linking milk yield and bimodal milking (
), our data were collected from one herd where cows were housed and milked in the same facilities. Thus, we did not have to account for herd-to-herd variation.
One of our objectives was to investigate whether milking unit vacuum could serve as a proxy for milk flow to identify key points in the milk curve of a cow, in this case LDT. Although we had previously applied this concept in a large field study (
Malmo, J., and G. Mein. 2015. A new tool for milking-time investigations: Using the VaDia and interpreting results. In Proceedings of the Countdown Symposium, Melbourne, Australia. Dairy Australia, Melbourne, Australia.
previously demonstrated a strong relationship between higher milk flow and lower milking unit vacuum when simultaneously recording milk flow (as measured by LactoCorder) and milking unit vacuum (as measured by VaDia). It is important to note that we did not use the LDT time point on our vacuum plot for each cow to determine maximum milk flow but as an estimate of the start of the incline phase of the milking curve. The use of vacuum as a measure of key time points in the milking curve was previously supported by
Malmo, J., and G. Mein. 2015. A new tool for milking-time investigations: Using the VaDia and interpreting results. In Proceedings of the Countdown Symposium, Melbourne, Australia. Dairy Australia, Melbourne, Australia.
, who found that estimated LDT, when measured either by actual milk flow or milking unit vacuum, were highly correlated (R2 = 0.81).
Additionally, our association between DME and reduced milk yield was based solely on one milking for each cow. We did not continue to monitor cows in subsequent milkings for either bimodality or milk yield. Thus, to better understand the biological relevance of our study and the effect of DME on longer-term milk yield losses, a longitudinal study that monitored several milkings for each cow would be required. Nonetheless, that LDT was found to remain in our final model, along with such variables as DIM and LACT that are known to affect milk yield, suggests that this line of research may be useful.
CONCLUSIONS
Milk flow analysis by use of VSMT and VMPC served as a useful indicator for DME, bimodality, and estimating LDT in a herd milking 3 times per day. Additionally, milk yield in the parlor was negatively associated with increased LDT. We believe this information is related to premilking behaviors and thus could impel farm management to help train and educate employees to improve milking protocols, performance, and ultimately the milking experience of the cows. Further research to better understand the association of DME and milk yield on a longer-term basis would be beneficial.
ACKNOWLEDGMENTS
This project was supported by Agriculture and Food Research Initiative Competitive Grant no. 2013-68004-20439 from the USDA National Institute of Food and Agriculture (Washington, DC).
REFERENCES
Borkhus M.
Rønningen O.
Factors affecting mouthpiece chamber vacuum in machine milking..
Best combination of pre-stimulation and latency period duration before cluster attachment for efficient oxytocin release and milk ejection in cows with low to high udder-filling levels..
Malmo, J., and G. Mein. 2015. A new tool for milking-time investigations: Using the VaDia and interpreting results. In Proceedings of the Countdown Symposium, Melbourne, Australia. Dairy Australia, Melbourne, Australia.
30389-3&id=gr1.jpg){kind=link}