- Email: [email protected]
Comparison of three devices for the automated detection of estrus in dairy cows
Theriogenology 82 (2014) 734–741
Contents lists available at ScienceDirect
Theriogenology journal homepage: www.theriojournal.com
Comparison of three devices for the automated detection of estrus in dairy cows Audrey Chanvallon a, *, Stéphanie Coyral-Castel a, Julie Gatien b, Jean-Michel Lamy c, Daniéle Ribaud a, Clément Allain a, Pierre Clément d, Pascal Salvetti b a
Institut de l’Elevage, Paris, France UNCEIA, R&D, Maisons-Alfort, France c Chambre d’Agriculture de Maine-et-Loire, Angers, France d LUNAM Université, Oniris, UMR BioEpAR, Nantes, France b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 25 July 2013 Received in revised form 4 June 2014 Accepted 4 June 2014
Considerable technological advances have been made in the automated detection of estrus in dairy cattle, but few studies have evaluated their relative performance on the same animals or assessed cow-related factors that affect their performance. Our objective was to assess the performance and reliability of three devices commercially available in France for cow estrus detection. The devices were a pedometer (PM; Afitag) and two activity meters (AM1; Heatime-RuminAct, and AM2; HeatPhone). Two algorithms were tested for AM2. We fitted 63 lactating Holstein cows with the three detectors from calving to 90 days after calving. The onset and pattern of cyclicity were monitored from 7 to 90 days postpartum measuring progesterone concentration in milk twice weekly. A total of 211 ovulations were identified. Cyclicity was classified as normal in 60% of cows (38/63). Calculated over the operating period of all the devices (179 periods of estrus), the sensitivities and positive predictive values were, respectively, 71% and 71% for PM, 62% and 84% for AM1, 61% and 67% for the first algorithm of AM2, and 62% and 87% for the second algorithm of AM2. Both activity meters had a lower sensitivity but a higher positive predictive value than the PM (P < 0.05). For all devices, the performance in estrus detection was much poorer at the first postpartum ovulation than at subsequent ovulations (P < 0.05). Lactation rank and milk production affected some devices (P < 0.05). These devices could be used to reinforce visual observations, especially after 50 days postpartum, the minimum recommended delay to insemination. However, their full benefit remains to be verified in different farming systems and taking into account the specific objectives of the dairy farmer. Ó 2014 Elsevier Inc. All rights reserved.
Keywords: Dairy cow Automated estrus detection Activity monitoring
1. Introduction The efficiency and accuracy of estrus detection is one of the most important factors that influence the reproductive performance and profitability of dairy herds that rely on artificial insemination [1,2]. An improvement in the estrus * Corresponding author. Tel.: þ33 2 40 68 28 19; fax þ33 2 40 68 77 78. E-mail address: [email protected] (A. Chanvallon). 0093-691X/$ – see front matter Ó 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.theriogenology.2014.06.010
detection rate of 0.30 to 0.50 would increase profit by V53 per cow per year [3]. The traditional method for estrus detection is visual observation. The accuracy of detection is highly dependent on the intensity of estrus, on the experience of the observer, and the frequency of observations. A cow standing to be mounted is the most specific and accurate sign of estrus [4], but 37% to 54% of detected ovulations are not accompanied by standing estrus in Holstein cows [5–8]. Moreover, the intensity of estrus and
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
its duration have dramatically decreased over the last decades [5,9], making detection more difficult for farmers. In the Holstein breed, the duration of estrus was about 18 to 20 hours in the 1980s but, since the early 2000, it has shortened to only 4 to 8 hours between the first and the last standing mount, or 11 to 14 hours if all signs of estrus are taken into account [7,9,10]. The expression of estrus is influenced mainly by the number of cows in estrus at the same time [4,7]. Many other factors could interfere with the intensity of estrus (for review, see [11]), particularly cow factors (the rank of postpartum ovulation, parity, milk production, lameness) and environmental conditions (nutrition, housing, temperature, and humidity). The rate of successful estrus detection by visual observation varies from 38% to 86% [1,10,12], but it is generally thought that a farmer is able to detect, on average, 50 to 60% of the cows in estrus, depending on the frequency and time of observation, the estrus signs considered, and the experience of the observer. The methods of visual observation have been described, and they include slight signs of sexual behavior, such as sniffing the vagina of other cows and resting a chin on the backs of other cows [5,9]. However, because of the herd size that has increased and the use of skilled manpower that has decreased in contemporary French dairying, the time and expertise available for accurate detection of estrus by visual observation are now compromised [2,13]. The need for this critical but time-consuming task can be avoided altogether by using hormonal induction of estrus, a practice that is now widespread, particularly in the United States [14]. These programs aim to induce synchronous ovulation and thus allowing fixed-time insemination without the need for heat detection. In Europe, the use of hormonal programs to synchronize estrus is not so widespread because of the cost of treatment and because of the reluctance of European consumers to accept products from animals treated with hormones and/or antibiotics [15]. These issues have driven interest in the development of alternative systems that avoid the use of hormones and antibiotics. Particular attention has been paid to the development of inexpensive, reliable, and accurate systems to automatically detect estrus and/or ovulation. They are based on the automated detection of signs of estrus (standing heat and increased ambulatory activity) or ovulations estimated using automated online milk analysis of progesterone (P4) [16–19]. The performance of automated estrus detectors is generally higher than 75% [12,16,19–21], depending on the settings for the threshold and the reference period in the algorithm used to define the estrus. However, only a few studies have directly compared the performance among devices on the same cows, or assessed the cow-related factors that affect their efficiency and accuracy. The aims of our study were as follows: (1) to assess the comparative performance of three automated devices for the detection of estrus; a pedometer (PM) and two activity meters, and (2) to estimate the influence of cow-related factors, such as ovulation rank, lactation rank, postpartum cyclicity, and the various criteria of milk production, on the performance of these three devices. An “estrus” was deemed to be correctly or incorrectly detected by the automated devices on the basis of P4 concentrations in milk.
735
2. Materials and methods 2.1. Animals The experiment was conducted on an experimental farm in France (longitude: 0 220 W; latitude: 47 340 N). Sixty-three Holstein cows were housed in individual stalls during the whole study and fed individually using a total mixed ration based on maize silage and concentrates. The cows had all calved between August 28 and November 8, 2011. Of the 63 cows, 25 were primiparous (40%), 18 were in their second lactation (29%), and 20 were in third or higher lactation (32%). Milking took place twice a day starting at 06.45 and 16.45 hours. On average, peak milk production was 38 6 kg per day, and the minimum content of milk protein calculated for the three first months postcalving (an indicator of negative energy balance) was 29 2 g/kg. At calving, each cow was fitted with three estrus detection devices. The AfiTag PM (AfiMilk, distributed in France by Packo France) was fixed on a back leg. Tags of the two activity meters, Heatime-RuminAct (AM1; SCR Engineers, distributed in France by Créavia SAS) and HeatPhone (AM2; Medria, France), were placed on the same neck collar. 2.2. Description of automated estrus detectors The PM records the number of steps taken by a cow per time unit. The data were transmitted twice a day (morning and evening milking) by radio telemetry to a receiver, sited at the exit of the milking parlor. The data were then automatically forwarded to a database on a central computer. The software created an estrus alert for any cow that had a recent activity that exceeded over 70% the prior 10-day average. The two activity meters detect estrus based on accelerometer technology that records the general activity of the cow in three dimensions. For AM1, data were collected in blocks of 2 hours and transmitted by an infrared connection to an antenna at the exit of the milking parlor and above the drinking troughs. For AM2, data were collected in blocks of 5 minutes and transmitted by radiofrequency to a receiver on the farm, from where they were forwarded every 30 minutes to a central server via the mobile communications network. This system works with a cloud-computing network where data from sensors of all equipped farms are stored. Algorithms and alerts are directly implemented and calculated in this cloud before they are sent to the farmers. Therefore, when the algorithms evolve, it can be validated on a part of the stored data. In this experiment, the manufacturer proposed two different algorithms to be tested (named here AM2 and AM2_2). For both activity meters, an estrus alert was generated when the weighted activity, calculated using the proprietary algorithm developed by the manufacturers, surpassed a defined threshold. 2.3. Recorded data For each cow, the data collected were as follows: date of calving, lactation number, milk production, and milk
736
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
protein content over the first 90 days in milk (DIM), day of installation of the detectors, date and time (morning or afternoon) of PM alerts, date and time of alerts from the activity meters, and malfunction of the devices (e.g., cessation of data transmission, failure of the tags or of the collar on which the tags were attached).
If two alerts from the same device were recorded within 48 hours, only the first alert was kept. If the alerts were spaced by more than 48 hours apart, one of the two was automatically recorded as an FP. In total, five alerts were excluded from the analysis because they could not be classified as TP, FP, or FN.
2.4. Progesterone assay and cyclicity
2.6. Reliability of estrus detection
Milk samples were collected twice a week between 7 and 90 DIM for P4 assay. Each sample was made up by mixing milk in equal proportions from milkings in the afternoon and the following morning. The samples were preserved by adding 8 mg of Bronopol and 0.3 mg of Natamycin (Broad Spectrum Microtabs, Biovalley SA, Marne-La-Vallée, France) and then stored at 4 C. Progesterone was measured by ELISA using the Ovucheck Milk kit from Biovet (AES Chemunex, Bruze, France) following the manufacturer’s instructions, in one replicate for each sample. Intra- and interassay coefficients of variation were 7.2% and 9.9%, respectively. The limit of detection of P4 was 0.5 ng/mL. The concentration of P4 in milk was considered high when it was equal to or higher than 3.5 ng/mL; low when lower than 2.5 ng/ml, and indeterminate when between 2.5 and 3.5 ng/mL. In the latter case, the value was taken as high if the previous and/or following sample was high. The pattern of ovarian cyclicity was determined from the P4 data. The resumption of ovarian cyclicity was estimated as the interval between calving and the day of the first high P4 level, whish is considered to be a valid indicator of luteal phase. The estimated length of a cycle was the interval between a low P4 level preceding a luteal phase and the low P4 level preceding the following luteal phase. On the basis of P4 data, the patterns of cyclicity for each cow were then classified into seven categories using a procedure adapted from those of Opsomer et al. [22] and Kerbrat and Disenhaus [9]: (1) normal, resumption of cyclicity before 50 DIM with a succession of ovarian cycles of 17 to 26 days; (2) prolonged luteal phase, one or more cycles longer than 26 days; (3) short cycle, one or more cycles shorter than 17 days (however, a short cycle immediately following the resumption of ovarian cyclicity was considered physiologically normal); (4) interrupted ovarian cyclicity, one or more periods longer than 12 days of low P4 levels flanked by luteal phases; (5) delayed cyclicity, resumption of cyclicity after 50 DIM; (6) anovulatory pattern, no resumption of cyclicity during the study (i.e. up to 90 DIM); and (7) unclassified, disorderly luteal activity and/or several other abnormalities.
The reliability of each device was defined by their sensitivity (SE, defined as the number of correctly detected ovulations divided by the total number of ovulations that occurred) and by the positive predictive value (PPV, defined as the number of correctly detected ovulations divided by the total number of emitted alerts). Sensitivity and PPV were calculated as percentages using the following equations: SE ¼ TP/(TP þ FN); PPV ¼ TP/(TP þ FP). Further analyses on the basis of these two measures were as follows: (1) the entire study period (SEENT and PPVENT); (2) the operating period of the device (SEOPE and PPVOPE), which excluded periods when the device failed; and (3) the operating period of all three devices (SEOPE3 and PPVOPE3), excluding periods when any device failed.
2.5. Interpretation of estrus alerts Alerts from the three devices were recorded as “true positive” (TP) if it occurred during a period when estrus was expected or “false positive” (FP) if it occurred during a period when estrus was not expected. The absence of an alert during a period when estrus was expected was characterized as a “false-negative” response (FN). An estrus was expected around an ovulation, that is, at a low P4 level just before a luteal phase (high P4) (Fig. 1) [9].
2.7. Statistical analyses Statistical analyses were carried out by the general linear mixed model procedure (GLIMMIX) using a specialized program for statistical analysis (SAS; SAS Institute, Cary, NC, USA). The reliability of estrus detection was compared among devices using the OPE3 data. Risk factors affecting the reliability of estrus detection were analyzed using the OPE data. The factors were initially analyzed individually: rank of ovulation from calving (first ovulation, second ovulation, or more), pattern of cyclicity when the alert was sent out (normal cycle and abnormal cycle), estimated resumption of cyclicity (<20, 20–30, and >30 DIM), interval between calving and alert (50 and >50 DIM), lactation rank (1, 2, and 3), peak milk production (35, 35–40, and 40 kg), and minimum milk protein content (<28 and 28 g/kg). The factors that affected performance in the univariate analysis (P < 0.15) were then included in a multivariate analysis and excluded one by one, beginning by the least significant, finally retaining in the model only those factors with a P value less than or equal to 0.10. However, the interval between calving and alert was excluded from the multivariable analysis because it was confounded with ovulation rankdall cows had ovulated before 50 DIM and 30% of second ovulations had also occurred before 50 DIM. Pairwise comparisons for variables with more than two levels were carried out with the Bonferroni adjustment. Odds ratios (ORs) were determined, with the reference category being the one with the highest performance. When the 95% confidence interval of the OR did not include 1, the estimate of the OR was considered significant at P less than 0.05. Two models were calculated for each device, one for SEOPE and one for PPVOPE. Logistic models were on the basis of the assumption of independence of repeated cow observations over the study period. Only results of the multivariate analysis were presented in the article. Data are presented as mean SEM.
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
737
Fig. 1. Example of a normal profile from the analysis of milk progesterone (A) and definition of the process to describe an estrus alert (B). A true positive alert was sent in an expected estrus period (2); a false negative alert was sent in an anovulatory period (1) or a luteal phase (4); and a false negative was an absence of alert in an expected estrus period (2). The first high sample was considered as the estimated day of resumption of ovarian cyclicity (3). The black line in (A) is the positive threshold of 3.5 ng/mL; the “-” symbol in (B) is a low level of progesterone (<2.5 ng/mL) and the “þ” symbol is a high level of progesterone (3.5 ng/mL).
3. Results 3.1. Description of cyclicity Ovarian cyclicity was estimated to have resumed 28 10 days after calving. Three cows (5%) did not resume ovarian cyclicity during the study period (up to 90 DIM). The percentage of cyclic cows postpartum was 62% (n ¼ 39/ 63) at 30 days and 94% (59/63) at 50 days (Fig. 2). During the whole study, 214 ovulations were identified, in which 28% were first postpartum ovulation (60/214). From the 153 cycles or interrupted cycles, patterns of cyclicity were characterized for each cow: 60% (38/63) were normal, 17% (11/63) were prolonged, 2% (1/63) were short, 3% (2/63) were interrupted, 2% (1/63) were delayed, 5% (3/63) were anovulatory, and 11% (7/63) were unclassified. The estimated length of the normal cycles was 21 2 days.
3.2. Comparative rates of estrus detection For the PM device, two individual malfunctions were noted, but they had no effect on the results because both fell outside any periods when estrus was expected. The AM1 device was unusable for 13 days because of an antenna failure. Furthermore, five tags had to be changed after internal malfunctions. For the AM2 device, two tags had to be changed after an internal malfunction and three because they were lost from a neck collar.
After exclusion of data from a cow without AM2 tags, 211 ovulations were analyzed. Results of the three types of analyses (the entire period, the operating period, and the three-device operating period) are presented in Tables 1 and 2. For the three-device operating period, the sensitivity was higher for PM (71%) than for AM1 (62%; P ¼ 0.02), AM2 (61%; P ¼ 0.005), and AM2_2 (62%; P ¼ 0.02). The positive predictive value of AM1 (84%) and AM2_2 (87%) were higher than those of PM (71%; P ¼ 0.05 and P ¼ 0.007, respectively) and AM2 (67%; P ¼ 0.006 and P ¼ 0.0007, respectively). Generally, the performance of the devices increased as the delay from calving extended (Table 2). Of all ovulations, 80% (168/211) were detected by at least one device and 40% (85/211) by all devices. The number of periods of estrus exclusively detected by each device was 17/ 211 for PM, 6/211 for AM2, and 0 for AM1 and AM2_2. Fiftyone percent (30/59) of the first postpartum ovulations and 91% (138/152) of the following postpartum ovulations were detected by at least one device. Similarly, 12% (7/59) of the first postpartum ovulations and 51% (78/152) of the following postpartum ovulations were detected by all devices. 3.3. Factors affecting the performance of the three devices As can be seen in Table 3, the first postpartum ovulation had a lower sensitivity than for the following ovulations for PM, AM1, and AM2_2 (P < 0.0001) and a lower positive predictive value for AM1, AM2, and AM2_2 (P ¼ 0.002; P ¼ 0.001; P < 0.0001, respectively). Cows in their third lactation had a lower sensitivity than cows in their first lactation for PM (P ¼ 0.019) and a lower predictive positive value for AM1 (P ¼ 0.038). High-producing cows had a lower sensitivity than low-producing cows for AM1 (P ¼ 0.006). Mid-producing dairy cows tended to have a lower sensitivity than low-producing cows for AM2_2 (P ¼ 0.063). Mid-producing dairy cows had a lower positive predictive value compared with low-producing cows for AM2 and AM2_2 (P ¼ 0.046 and P ¼ 0.021). The pattern of cyclicity, the estimated resumption of cyclicity, and the minimum milk protein content had no effect on the ability of any device to detect ovulation. 4. Discussion
Fig. 2. Cumulative percentage of cyclic cows (n ¼ 63) on the study period (from calving to 90 DIM).
The PM had a greater sensitivity but a lower positive predictive value than the two activity meters for the
738
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
Table 1 Sensitivity (SE) and positive predictive value (PPV) of the estrus detection devices. Period
Indicators
Pedometer (PM)
Activity meter 1 (AM1)
Activity meter 2 (AM2)
Activity meter 2 algorithm 2 (AM2_2)
Entire period (ENT)
TP FP FN SEENT (%) PPVENT (%) TP FP FN SEOPE3 (%) [95% CI] PPVOPE3 (%) [95% CI]
154 61 57 73 72 127 53 52 71a[64–77] 71a[63–77]
117 22 94 55 84 111 22 68 62b[55–69] 83b[76–89]
125 63 86 59 66 109 55 70 61b[54–68] 66a[59–73]
128 17 83 61 88 111 17 68 62b[54–69] 87b[80–92]
Operating period of the three devices (OPE3)
Data from the entire study period were analyzed (ENT, from 7 to 90 DIM; n ¼ 211 ovulations), as were data from the operating period of all devices (OPE3; i.e., from 7 to 90 DIM after removing ovulations during a failure of at least one device; n ¼ 179 ovulations). Activity Meter 2 was assessed with two algorithms, AM2 and AM2_2. Abbreviations: FN, false negative; FP, false positive; TP, true positive; [95% CI], 95% confidential interval. a,b Percentage within a row with different superscripts differs (P 0.05).
detection of estrus in dairy cows. We also found that the performance of all three devices was influenced mainly by the number of ovulations from calving (ovulation rank). The results are consistent with the literature for ovarian cyclicity patterns [22–26], cyclicity resumption [27], and the global performance of the three devices (PM [16,18,21], activity meters [18,21,28]). The high rate of FP alerts emitted by the PM in our study (around 29%) is not surprising considering the rates of 17% to 55% reported in the review by Firk et al. [16]. The rate of FP alerts emitted by the activity meters is much more variable among published studies, ranging from a low of 6.5% [18] to high figures of 21% [29] and 28% [30] and our own figures of 12% to 33% being similar. For the HeatPhone activity meter, the second algorithm (AM2_2) clearly improved performance. In any case, false alerts could lead to inseminations during luteal phase and increase the expenses for semen and breeding. Moreover in the present study, 23% to 45% of expected estrus events were not detected and 20% were not detected by any of the devices. This underachievement of the devices can be explained by three factors discussed below: (1) a nonoptimal algorithm, (2) the experimental design, and (3) individual or environmental factors affecting estrus expression. The present results were obtained from observations on one herd, so extrapolations need to be considered with caution. The cows were fed in individual, raised troughs; so repeated, wide head movements would have been
possible, reducing the accuracy of devices attached to neck collars. Moreover, our validation of the estrus alerts was based on ovulations identified using only data from P4 assays in milk and not from video records that allow quantification of estrus expression (duration and intensity). We are cautious about interpretation because FP alerts can be caused by overactivity in the cows induced by their environment and because FN alerts could be caused by silent ovulation. Between 9% and 16% of dairy cows have at least one ovulation of ranks 2 to 4 that are not accompanied by estrus [30], so some FN alerts are almost certainly due to these silent ovulations and the performance of the devices could be underestimated in our study. Finally, we did not study the conception rates and the timing for insemination because of the experimental design (each cow fitted with the three devices), and these aspects should be further investigated [31,32]. The detection of estrus was influenced mainly by ovulation rank. Sensitivity for the first postpartum ovulation varied from 23% to 40% (depending on the device) compared with 78% to 86% for subsequent ovulations. These observations are consistent with those from previous studies [8,16]. It is known that 50% to 80% of first ovulations postpartum are silent [30,33]. We did not have continuous video recording in the study to check which first postpartum ovulations were accompanied by estrus, but it is very likely that the poor performance of all three devices at the first postpartum ovulation was related to low
Table 2 Sensitivity (SE) and positive predictive value (PPV) of the estrus detection devices depending on the interval between calving and alert. Pedometer (PM) Activity meter 1 (AM1) Activity meter 2 (AM2) Activity meter 2 algorithm 2 Interval (AM2_2) between calving <30 [30–50] [50–70] 70 Total <30 [30–50] [50–70] 70 Total <30 [30–50] [50–70] 70 Total <30 [30–50] [50–70] 70 Total and alert (d) TP FP FN SE (%) PPV (%)
20 23 27 43 47
41 13 14 75 76
45 13 9 83 78
48 12 7 87 80
154 61 57 73 72
15 6 32 32 71
32 3 17 65 91
39 8 11 78 83
31 5 10 76 86
117 22 70 63 84
13 12 30 30 52
29 10 22 57 74
40 18 14 74 69
43 23 11 80 65
125 63 77 62 66
10 6 33 23 63
32 6 19 63 84
42 2 12 78 95
44 3 10 81 94
128 17 74 63 88
Analyses were made on the operating period of the studied device (OPE; i.e., after removing ovulations during a failure of the studied device), on 211 ovulations for PM, 187 ovulations for AM1, and 202 ovulations for AM2 and AM2_2. Abbreviations: FN, false negative; FP, false positive; TP, true positive.
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
739
Table 3 Sensitivity (SE) and positive predictive value (PPV) of the estrus devices in relation to ovulation rank, lactation rank, and peak milk production. Risk factor
Pedometer (PM)
Activity meter 1 (AM1)
Activity meter 2 (AM2)
Activity meter 2 algorithm 2 (AM2_2)
SEOPE (%)
SEOPE (%)
PPVOPE (%)
PPVOPE (%)
SEOPE (%)
PPVOPE (%)
24a 0.08 [0.04–0.18]*** 79b
60a 0.18 [0.06–0.51]** 90b
43a 0.29 [0.14–0.61]** 73b
23a 0.08 [0.04–0.18]*** 78b
48a 0.02 [0.00–0.09]*** 98.1b
d d
d d
d d
d
d
d
68a 62b 0.41 [0.17–0.99]* 46 0.76 [0.30–1.90]
62 40 0.40 [0.15–1.04]y 51 0.62 [0.24–1.60]
95a 66b 0.10 [0.01–0.75]* 89 0.40 [0.05–3.30]
Ovulation rank 1 40a 0.11 [0.05–0.22]*** 2 86b Lactation rank 1 77a 2 70 0.71 [0.254–2.00] 3 52b 0.33 [0.12–0.86]* Peak milk production (kg) 35 d 35–40 d 40
d
d d d
68a 50 0.46 [0.16–1.29] 36b 0.26 [0.10–0.72]**
90a 75 0.35 [0.07–1.66] 35b 0.21 [0.05–0.93]* d d d
From the multivariate analysis of the operating period of the device (OPE; i.e. after removing ovulations during a failure of the studied detector), on 211 ovulations for PM, 187 ovulations for AM1, and 202 ovulations for AM2 (Activity Meter 2 was also assessed with a new algorithm, AM2_2). Odds ratio [95% confidence interval] are given in italics. a,b Percentage within a column and per factor differs with P adjust 0.05. yP < 0.10, *P < 0.05, **P < 0.01, ***P < 0.001.
expression of estrus at this time. In our study, the improved performance after 50 DIM is due to the fact that all first ovulations postpartum occurred before 50 days. In practice, it is not advisable to inseminate before 50 DIM so as to ensure the resumption of cyclicity in all cows and complete involution of the uterus [34]. Thus, the first insemination is done generally after 65 DIM [25,35,36] at a time when there are no longer any performance issues. The performance of PM and Heatime-RuminAct activity meter was higher for cows in their first lactation than for cows in their third or higher lactation. This observation is also consistent with previous reports showing that walking activity at estrus decreases with later lactations [37,38], and that estrus is expressed more strongly in primiparous compared with multiparous cows [7,31]. However, in other studies, the duration of estrus and standing activity were not affected by parity [12,39]. The performance of some of the devices was also adversely affected by milk production, and, globally, cows with low milk production showed the best results. This is consistent with the fact that milk yield affects the expression of estrus in dairy cattle, probably due in part to decreased circulating concentrations of estradiol [39,40]. Moreover, high-producing dairy cows have increased nutritional requirements that (1) decrease the time that they spend interacting with their herd mates because of the increased time spent standing while feeding, or lying while ruminating [18] and (2) limit estrus expression, following their survival instinct, in order to avoid conception and therefore the requirement for better nutrition in the later stages of pregnancy [41]. Because of the limitations of our study (e.g., low number of cows from a single herd), further investigations are needed to determine precisely the factors that adversely affect the performance of the three devices. Nevertheless, our results do provide the following: (1) new data to improve the performance of automated estrus devices by including some risk factors in future algorithms and (2)
some useful guidance for dairy farmers wishing to use these tools. The algorithms are now calculated on variation of cow activity, but, even though an increase in activity is concomitant estrus, this is not a very specific behavior (compared with standing mount). One way to reduce this problem could be to regulate the level of alerts using nonsensor data or data from other sensor, for example, animal factors (delay from calving, lactation rank or milk production on the alert day) or environmental conditions not studied here, such as type of housing, temperature, or humidity. In the end, we can speculate whether these automated devices are useful for reducing the constraints to heat detection without decreasing reproductive performance. Our results suggest a qualified “yes” with the proviso that their use should be well considered. First, because of the high risk of silent ovulation, performance before 50 DIM is unreliable, but this is not a serious problem because the recommended time for insemination is 50 days. However, the devices we tested recorded only a part of sexual behaviors [10], so combination of these devices with visual observations, as reported by some authors, might be a better strategy to increase estrus detection, conception rates, and reduce culling due to perceived failure to conceive [18,31]. Second, interpretation of the alerts should be taken into account production data, such as peak milk production, as well as the activity history of the cow, when making a decision to inseminate. Thus, estrus detected by an alert should be accepted as true only if it follows a previous detection (visual or automated) by 18 to 24 days or 36 to 48 days, suggesting an interval of 1 or 2 estrous cycles [42]. Moreover, the purchase of such devices will depend on profitability and must be considered in relation to the farming system and the specific objectives of the breeder. Indeed, the benefit could differ with the breeding system management, including the season of calving, the number of cows in the herd, and whether sexed or
740
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
nonsexed semen is used [21,43]. For example, visual detection of estrus seems to be easier in a seasonal-calving system because estrus is better expressed when several cows are simultaneously on heat [4,20]. In this case, investment in an automated system would require more neck collars, because all cows are at the same stage simultaneously. Another point is finding a balance between sensitivity and predictive values of the device; what’s the most penalizing situation: a FP or a FN alert? The response is not clear and needs to be investigated in relation to semen and breeding expense. Finally, the life style of the farmer needs to be considered: for example, a desire to reduce time-consuming tasks allowing more leisure time or time for others tasks could be part of a decision to invest in an automated device. The cost–benefit ratio of these devices remains to be investigated, with technical, economical, and labor impact, perspectives. On a larger scale, the data collected by the tools could be used to select animals for their potential to express estrus in the context of a continuing decline in the expression and duration of estrus in dairy cows [44]. 4.1. Conclusion Automated devices used for the detection of estrus in dairy cows function by continuously measuring some activity parameters. Three such devices were tested (a PM and two that recorded body activity). All three performed well from the second postpartum ovulation or after day 50 postpartum. The poor performance at the first postpartum ovulation was probably due to the weak expression of estrus or due to silent ovulation, both of which are more common at this stage in high-producing cows. This was not a serious problem in our study because all the first ovulations occurred before the recommended time for insemination (from 50 days postpartum). Acknowledgments The authors thank the staff of the experimental farm Les Trinottières, MC Deloche, from the UNCEIA hormonology service, for progesterone assays, and EVOLUTION and CLASEL for technical support. We also acknowledge Elevage Conseil Loire Anjou for its financial support, Milkline for its financial and technical help, RJ Scaramuzzi for his scientific comments and for English version, and GB Martin for his detailed editing of the article. References [1] Heersche G, Nebel RL. Measuring efficiency and accuracy of detection of estrus. J Dairy Sci 1994;77:2754–61. [2] Gordon P. Oestrus detection in dairy cattle. In Pract 2011;33:542–6. [3] Inchaisri C, Jorritsma R, Vos PLAM, van der Weijden GC, Hogeveen H. Economic consequences of reproductive performance in dairy cattle. Theriogenology 2010;74:835–46. [4] Diskin MG, Sreenan JM. Expression and detection of oestrus in cattle. Reprod Nutr Dev 2000;40:481–91. [5] Van Eerdenburg FJCM, Loeffler HSH, Van Vliet JH. Detection of oestrus in dairy cows: a new approach to an old problem. Vet Q 1996;18:52–4. [6] Lyimo ZC, Nielen M, Ouweltjes W, Kruip TAM, van Eerdenburg F. Relationship among estradiol, cortisol and intensity of estrous behavior in dairy cattle. Theriogenology 2000;53:1783–95.
[7] Roelofs JB, van Eerdenburg FJCM, Soede NM, Kemp B. Various behavioral signs of estrous and their relationship with time of ovulation in dairy cattle. Theriogenology 2005;63:1366–77. [8] Sakaguchi M. Practical aspects of the fertility of dairy cattle. J Reprod Dev 2011;57:17–33. [9] Kerbrat S, Disenhaus C. A proposition for an updated behavioural characterisation of the oestrus period in dairy cows. Appl Anim Behav Sci 2004;87:223–38. [10] Van Vliet JH, Van Eerdenburg FJCM. Sexual activities and oestrus detection in lactating Holstein cows. Appl Anim Behav Sci 1996;50: 57–69. [11] Roelofs J, López-Gatius F, Hunter RHF, van Eerdenburg FJCM, Hanzen C. When is a cow in estrus? Clinical and practical aspects. Theriogenology 2010;74:327–44. [12] At-Taras EE, Spahr SL. Detection and characterization of estrus in dairy cattle with an electronic heatmount detector and an electronic activity tag. J Dairy Sci 2001;84:792–8. [13] Lucy MC. Reproductive loss in high-producing dairy cattle: where will it end? J Dairy Sci 2001;84:1277–93. [14] Caraviello DZ, Weigel KA, Fricke PM, Wiltbank MC, Florent MJ, Cook NB, et al. Survey of management practices on reproductive performance of dairy cattle on large US commercial farms. J Dairy Sci 2006;89:4723–35. [15] Chastant-Maillard S. Is there a future for pharmaceutical management in cow reproduction ? European perspective. 24th World Buiatrics Congress 2006, Nice, France. [16] Firk R, Stamer E, Junge W, Krieter J. Automation of oestrus detection in dairy cows: a review. Livest Prod Sci 2002;75:219–32. [17] Durkin J. Heat detection: trends and opportunities. The first North American Conference on Precision dairy Management 2010, Toronto, Canada. [18] Holman A, Thompson J, Routly JE, Cameron J, Jones DN, GroveWhite D, et al. Comparison of oestrus detection methods in dairy cattle. Vet Rec 2011;169:47. [19] Saint-Dizier M, Chastant-Maillard S. Towards an automated detection of oestrus in dairy cattle. Reprod Domest Anim 2012;47:1056– 61. [20] Roelofs JB, van Eerdenburg F, Soede NM, Kemp B. Pedometer readings for estrous detection and as predictor for time of ovulation in dairy cattle. Theriogenology 2005;64:1690–703. [21] Rutten CJ, Velthuis AGJ, Steeneveld W, Hogeveen H. Sensors to support health management on dairy farms. J Dairy Sci 2013;96: 1928–52. [22] Opsomer G, Gröhn YT, Hertl J, Coryn M, Deluyker H, de Kruif A. Risk factors for post partum ovarian dysfunction in high producing dairy cows in Belgium: a field study. Theriogenology 2000;53:841–57. [23] Lamming GE, Darwash AO. The use of milk progesterone profiles to characterise components of subfertility in milked dairy cows. Anim Reprod Sci 1998;52:175–90. [24] Royal MD, Darwash AO, Flint APE, Webb R, Woolliams JA, Lamming GE. Declining fertility in dairy cattle: changes in traditional and endocrine parameters of fertility. Anim Sci 2000;70:487– 501. [25] Shrestha HK, Nakao T, Higaki T, Suzuki T, Akita M. Resumption of postpartum ovarian cyclicity in high-producing Holstein cows. Theriogenology 2004;61:637–49. [26] Windig JJ, Beerda B, Veerkamp RF. Relationship between milk progesterone profiles and genetic merit for milk production, milking frequency, and feeding regimen in dairy cattle. J Dairy Sci 2008;91: 2874–84. [27] Opsomer G, Coryn M, Deluyker H, de Kruif A. An analysis of ovarian dysfunction in high yielding dairy cows after calving based on progesterone profiles. Reprod Domest Anim 1998;33:193–204. [28] Kamphuis C, DelaRue B, Burke CR, Jago J. Field evaluation of 2 collarmounted activity meters for detecting cows in estrus on a large pasture-grazed dairy farm. J Dairy Sci 2012;95:3045–56. [29] Aungier SPM, Roche JF, Sheehy M, Crowe MA. Effects of management and health on the use of activity monitoring for estrus detection in dairy cows. J Dairy Sci 2012;95:2452–66. [30] Ranasinghe RMSBK, Nakao T, Yamada K, Koike K. Silent ovulation, based on walking activity and milk progesterone concentrations, in Holstein cows housed in a free-stall barn. Theriogenology 2010;73: 942–9. [31] Peralta OA, Pearson RE, Nebel RL. Comparison of three estrus detection systems during summer in a large commercial dairy herd. Anim Reprod Sci 2005;87:59–72. [32] Bar D. Optimal timing of insemination using activity collars. The First North American Conference on Precision Dairy Management 2010, Toronto, Canada.
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741 [33] Kyle SD, Callahan CJ, Allrich RD. Effect of progesterone on the expression of estrus at the first postpartum ovulation in dairy cattle. J Dairy Sci 1992;75:1456–60. [34] Sheldon IM, Lewis GS, LeBlanc S, Gilbert RO. Defining postpartum uterine disease in cattle. Theriogenology 2006;65:1516–30. [35] Berry DP, Buckley F, Dillon P, Evans RD, Rath M, Veerkamp RF. Genetic relationships among body condition score, body weight, milk yield, and fertility in dairy cows. J Dairy Sci 2003;86: 2193–204. [36] Sakaguchi M, Sasamoto Y, Suzuki T, Takahashi Y, Yamada Y. Postpartum ovarian follicular dynamics and estrous activity in lactating dairy cows. J Dairy Sci 2004;87:2114–21. [37] López-Gatius F, Santolaria P, Mundet I, Yániz JL. Walking activity at estrus and subsequent fertility in dairy cows. Theriogenology 2005; 63:1419–29. [38] Yániz JL, Santolaria P, Giribet A, Lopez-Gatius F. Factors affecting walking activity at estrus during postpartum period and subsequent fertility in dairy cows. Theriogenology 2006;66:1943–50.
741
[39] Lopez H, Satter LD, Wiltbank MC. Relationship between level of milk production and estrous behavior of lactating dairy cows. Anim Reprod Sci 2004;81:209–23. [40] Cutullic E, Delaby L, Causeur D, Michel G, Disenhaus C. Hierarchy of factors affecting behavioural signs used for oestrus detection of Holstein and Normande dairy cows in a seasonal calving system. Anim Reprod Sci 2009;113:22–37. [41] Dobson H, Smith RF, Royal MD, Knight CH, Sheldon IM. The highproducing dairy cow and its reproductive performance. Reprod Domest Anim 2007;42:17–23. [42] Philipot JM, Clement P. An innovative tool for monitoring and phenotyping cattle: Heatime-Ruminact. 38th International Committee for Animal Recording 2012, Cork, Ireland. [43] Østergaard S, Friggens NC, Chagunda MGG. Technical and economic effects of an inline progesterone indicator in a dairy herd estimated by stochastic simulation. Theriogenology 2005;64:819–43. [44] Løvendahl P, Chagunda MGG. Short communication: genetic variation in estrus activity traits. J Dairy Sci 2009;92:4683–8.
Contents lists available at ScienceDirect
Theriogenology journal homepage: www.theriojournal.com
Comparison of three devices for the automated detection of estrus in dairy cows Audrey Chanvallon a, *, Stéphanie Coyral-Castel a, Julie Gatien b, Jean-Michel Lamy c, Daniéle Ribaud a, Clément Allain a, Pierre Clément d, Pascal Salvetti b a
Institut de l’Elevage, Paris, France UNCEIA, R&D, Maisons-Alfort, France c Chambre d’Agriculture de Maine-et-Loire, Angers, France d LUNAM Université, Oniris, UMR BioEpAR, Nantes, France b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 25 July 2013 Received in revised form 4 June 2014 Accepted 4 June 2014
Considerable technological advances have been made in the automated detection of estrus in dairy cattle, but few studies have evaluated their relative performance on the same animals or assessed cow-related factors that affect their performance. Our objective was to assess the performance and reliability of three devices commercially available in France for cow estrus detection. The devices were a pedometer (PM; Afitag) and two activity meters (AM1; Heatime-RuminAct, and AM2; HeatPhone). Two algorithms were tested for AM2. We fitted 63 lactating Holstein cows with the three detectors from calving to 90 days after calving. The onset and pattern of cyclicity were monitored from 7 to 90 days postpartum measuring progesterone concentration in milk twice weekly. A total of 211 ovulations were identified. Cyclicity was classified as normal in 60% of cows (38/63). Calculated over the operating period of all the devices (179 periods of estrus), the sensitivities and positive predictive values were, respectively, 71% and 71% for PM, 62% and 84% for AM1, 61% and 67% for the first algorithm of AM2, and 62% and 87% for the second algorithm of AM2. Both activity meters had a lower sensitivity but a higher positive predictive value than the PM (P < 0.05). For all devices, the performance in estrus detection was much poorer at the first postpartum ovulation than at subsequent ovulations (P < 0.05). Lactation rank and milk production affected some devices (P < 0.05). These devices could be used to reinforce visual observations, especially after 50 days postpartum, the minimum recommended delay to insemination. However, their full benefit remains to be verified in different farming systems and taking into account the specific objectives of the dairy farmer. Ó 2014 Elsevier Inc. All rights reserved.
Keywords: Dairy cow Automated estrus detection Activity monitoring
1. Introduction The efficiency and accuracy of estrus detection is one of the most important factors that influence the reproductive performance and profitability of dairy herds that rely on artificial insemination [1,2]. An improvement in the estrus * Corresponding author. Tel.: þ33 2 40 68 28 19; fax þ33 2 40 68 77 78. E-mail address: [email protected] (A. Chanvallon). 0093-691X/$ – see front matter Ó 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.theriogenology.2014.06.010
detection rate of 0.30 to 0.50 would increase profit by V53 per cow per year [3]. The traditional method for estrus detection is visual observation. The accuracy of detection is highly dependent on the intensity of estrus, on the experience of the observer, and the frequency of observations. A cow standing to be mounted is the most specific and accurate sign of estrus [4], but 37% to 54% of detected ovulations are not accompanied by standing estrus in Holstein cows [5–8]. Moreover, the intensity of estrus and
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
its duration have dramatically decreased over the last decades [5,9], making detection more difficult for farmers. In the Holstein breed, the duration of estrus was about 18 to 20 hours in the 1980s but, since the early 2000, it has shortened to only 4 to 8 hours between the first and the last standing mount, or 11 to 14 hours if all signs of estrus are taken into account [7,9,10]. The expression of estrus is influenced mainly by the number of cows in estrus at the same time [4,7]. Many other factors could interfere with the intensity of estrus (for review, see [11]), particularly cow factors (the rank of postpartum ovulation, parity, milk production, lameness) and environmental conditions (nutrition, housing, temperature, and humidity). The rate of successful estrus detection by visual observation varies from 38% to 86% [1,10,12], but it is generally thought that a farmer is able to detect, on average, 50 to 60% of the cows in estrus, depending on the frequency and time of observation, the estrus signs considered, and the experience of the observer. The methods of visual observation have been described, and they include slight signs of sexual behavior, such as sniffing the vagina of other cows and resting a chin on the backs of other cows [5,9]. However, because of the herd size that has increased and the use of skilled manpower that has decreased in contemporary French dairying, the time and expertise available for accurate detection of estrus by visual observation are now compromised [2,13]. The need for this critical but time-consuming task can be avoided altogether by using hormonal induction of estrus, a practice that is now widespread, particularly in the United States [14]. These programs aim to induce synchronous ovulation and thus allowing fixed-time insemination without the need for heat detection. In Europe, the use of hormonal programs to synchronize estrus is not so widespread because of the cost of treatment and because of the reluctance of European consumers to accept products from animals treated with hormones and/or antibiotics [15]. These issues have driven interest in the development of alternative systems that avoid the use of hormones and antibiotics. Particular attention has been paid to the development of inexpensive, reliable, and accurate systems to automatically detect estrus and/or ovulation. They are based on the automated detection of signs of estrus (standing heat and increased ambulatory activity) or ovulations estimated using automated online milk analysis of progesterone (P4) [16–19]. The performance of automated estrus detectors is generally higher than 75% [12,16,19–21], depending on the settings for the threshold and the reference period in the algorithm used to define the estrus. However, only a few studies have directly compared the performance among devices on the same cows, or assessed the cow-related factors that affect their efficiency and accuracy. The aims of our study were as follows: (1) to assess the comparative performance of three automated devices for the detection of estrus; a pedometer (PM) and two activity meters, and (2) to estimate the influence of cow-related factors, such as ovulation rank, lactation rank, postpartum cyclicity, and the various criteria of milk production, on the performance of these three devices. An “estrus” was deemed to be correctly or incorrectly detected by the automated devices on the basis of P4 concentrations in milk.
735
2. Materials and methods 2.1. Animals The experiment was conducted on an experimental farm in France (longitude: 0 220 W; latitude: 47 340 N). Sixty-three Holstein cows were housed in individual stalls during the whole study and fed individually using a total mixed ration based on maize silage and concentrates. The cows had all calved between August 28 and November 8, 2011. Of the 63 cows, 25 were primiparous (40%), 18 were in their second lactation (29%), and 20 were in third or higher lactation (32%). Milking took place twice a day starting at 06.45 and 16.45 hours. On average, peak milk production was 38 6 kg per day, and the minimum content of milk protein calculated for the three first months postcalving (an indicator of negative energy balance) was 29 2 g/kg. At calving, each cow was fitted with three estrus detection devices. The AfiTag PM (AfiMilk, distributed in France by Packo France) was fixed on a back leg. Tags of the two activity meters, Heatime-RuminAct (AM1; SCR Engineers, distributed in France by Créavia SAS) and HeatPhone (AM2; Medria, France), were placed on the same neck collar. 2.2. Description of automated estrus detectors The PM records the number of steps taken by a cow per time unit. The data were transmitted twice a day (morning and evening milking) by radio telemetry to a receiver, sited at the exit of the milking parlor. The data were then automatically forwarded to a database on a central computer. The software created an estrus alert for any cow that had a recent activity that exceeded over 70% the prior 10-day average. The two activity meters detect estrus based on accelerometer technology that records the general activity of the cow in three dimensions. For AM1, data were collected in blocks of 2 hours and transmitted by an infrared connection to an antenna at the exit of the milking parlor and above the drinking troughs. For AM2, data were collected in blocks of 5 minutes and transmitted by radiofrequency to a receiver on the farm, from where they were forwarded every 30 minutes to a central server via the mobile communications network. This system works with a cloud-computing network where data from sensors of all equipped farms are stored. Algorithms and alerts are directly implemented and calculated in this cloud before they are sent to the farmers. Therefore, when the algorithms evolve, it can be validated on a part of the stored data. In this experiment, the manufacturer proposed two different algorithms to be tested (named here AM2 and AM2_2). For both activity meters, an estrus alert was generated when the weighted activity, calculated using the proprietary algorithm developed by the manufacturers, surpassed a defined threshold. 2.3. Recorded data For each cow, the data collected were as follows: date of calving, lactation number, milk production, and milk
736
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
protein content over the first 90 days in milk (DIM), day of installation of the detectors, date and time (morning or afternoon) of PM alerts, date and time of alerts from the activity meters, and malfunction of the devices (e.g., cessation of data transmission, failure of the tags or of the collar on which the tags were attached).
If two alerts from the same device were recorded within 48 hours, only the first alert was kept. If the alerts were spaced by more than 48 hours apart, one of the two was automatically recorded as an FP. In total, five alerts were excluded from the analysis because they could not be classified as TP, FP, or FN.
2.4. Progesterone assay and cyclicity
2.6. Reliability of estrus detection
Milk samples were collected twice a week between 7 and 90 DIM for P4 assay. Each sample was made up by mixing milk in equal proportions from milkings in the afternoon and the following morning. The samples were preserved by adding 8 mg of Bronopol and 0.3 mg of Natamycin (Broad Spectrum Microtabs, Biovalley SA, Marne-La-Vallée, France) and then stored at 4 C. Progesterone was measured by ELISA using the Ovucheck Milk kit from Biovet (AES Chemunex, Bruze, France) following the manufacturer’s instructions, in one replicate for each sample. Intra- and interassay coefficients of variation were 7.2% and 9.9%, respectively. The limit of detection of P4 was 0.5 ng/mL. The concentration of P4 in milk was considered high when it was equal to or higher than 3.5 ng/mL; low when lower than 2.5 ng/ml, and indeterminate when between 2.5 and 3.5 ng/mL. In the latter case, the value was taken as high if the previous and/or following sample was high. The pattern of ovarian cyclicity was determined from the P4 data. The resumption of ovarian cyclicity was estimated as the interval between calving and the day of the first high P4 level, whish is considered to be a valid indicator of luteal phase. The estimated length of a cycle was the interval between a low P4 level preceding a luteal phase and the low P4 level preceding the following luteal phase. On the basis of P4 data, the patterns of cyclicity for each cow were then classified into seven categories using a procedure adapted from those of Opsomer et al. [22] and Kerbrat and Disenhaus [9]: (1) normal, resumption of cyclicity before 50 DIM with a succession of ovarian cycles of 17 to 26 days; (2) prolonged luteal phase, one or more cycles longer than 26 days; (3) short cycle, one or more cycles shorter than 17 days (however, a short cycle immediately following the resumption of ovarian cyclicity was considered physiologically normal); (4) interrupted ovarian cyclicity, one or more periods longer than 12 days of low P4 levels flanked by luteal phases; (5) delayed cyclicity, resumption of cyclicity after 50 DIM; (6) anovulatory pattern, no resumption of cyclicity during the study (i.e. up to 90 DIM); and (7) unclassified, disorderly luteal activity and/or several other abnormalities.
The reliability of each device was defined by their sensitivity (SE, defined as the number of correctly detected ovulations divided by the total number of ovulations that occurred) and by the positive predictive value (PPV, defined as the number of correctly detected ovulations divided by the total number of emitted alerts). Sensitivity and PPV were calculated as percentages using the following equations: SE ¼ TP/(TP þ FN); PPV ¼ TP/(TP þ FP). Further analyses on the basis of these two measures were as follows: (1) the entire study period (SEENT and PPVENT); (2) the operating period of the device (SEOPE and PPVOPE), which excluded periods when the device failed; and (3) the operating period of all three devices (SEOPE3 and PPVOPE3), excluding periods when any device failed.
2.5. Interpretation of estrus alerts Alerts from the three devices were recorded as “true positive” (TP) if it occurred during a period when estrus was expected or “false positive” (FP) if it occurred during a period when estrus was not expected. The absence of an alert during a period when estrus was expected was characterized as a “false-negative” response (FN). An estrus was expected around an ovulation, that is, at a low P4 level just before a luteal phase (high P4) (Fig. 1) [9].
2.7. Statistical analyses Statistical analyses were carried out by the general linear mixed model procedure (GLIMMIX) using a specialized program for statistical analysis (SAS; SAS Institute, Cary, NC, USA). The reliability of estrus detection was compared among devices using the OPE3 data. Risk factors affecting the reliability of estrus detection were analyzed using the OPE data. The factors were initially analyzed individually: rank of ovulation from calving (first ovulation, second ovulation, or more), pattern of cyclicity when the alert was sent out (normal cycle and abnormal cycle), estimated resumption of cyclicity (<20, 20–30, and >30 DIM), interval between calving and alert (50 and >50 DIM), lactation rank (1, 2, and 3), peak milk production (35, 35–40, and 40 kg), and minimum milk protein content (<28 and 28 g/kg). The factors that affected performance in the univariate analysis (P < 0.15) were then included in a multivariate analysis and excluded one by one, beginning by the least significant, finally retaining in the model only those factors with a P value less than or equal to 0.10. However, the interval between calving and alert was excluded from the multivariable analysis because it was confounded with ovulation rankdall cows had ovulated before 50 DIM and 30% of second ovulations had also occurred before 50 DIM. Pairwise comparisons for variables with more than two levels were carried out with the Bonferroni adjustment. Odds ratios (ORs) were determined, with the reference category being the one with the highest performance. When the 95% confidence interval of the OR did not include 1, the estimate of the OR was considered significant at P less than 0.05. Two models were calculated for each device, one for SEOPE and one for PPVOPE. Logistic models were on the basis of the assumption of independence of repeated cow observations over the study period. Only results of the multivariate analysis were presented in the article. Data are presented as mean SEM.
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
737
Fig. 1. Example of a normal profile from the analysis of milk progesterone (A) and definition of the process to describe an estrus alert (B). A true positive alert was sent in an expected estrus period (2); a false negative alert was sent in an anovulatory period (1) or a luteal phase (4); and a false negative was an absence of alert in an expected estrus period (2). The first high sample was considered as the estimated day of resumption of ovarian cyclicity (3). The black line in (A) is the positive threshold of 3.5 ng/mL; the “-” symbol in (B) is a low level of progesterone (<2.5 ng/mL) and the “þ” symbol is a high level of progesterone (3.5 ng/mL).
3. Results 3.1. Description of cyclicity Ovarian cyclicity was estimated to have resumed 28 10 days after calving. Three cows (5%) did not resume ovarian cyclicity during the study period (up to 90 DIM). The percentage of cyclic cows postpartum was 62% (n ¼ 39/ 63) at 30 days and 94% (59/63) at 50 days (Fig. 2). During the whole study, 214 ovulations were identified, in which 28% were first postpartum ovulation (60/214). From the 153 cycles or interrupted cycles, patterns of cyclicity were characterized for each cow: 60% (38/63) were normal, 17% (11/63) were prolonged, 2% (1/63) were short, 3% (2/63) were interrupted, 2% (1/63) were delayed, 5% (3/63) were anovulatory, and 11% (7/63) were unclassified. The estimated length of the normal cycles was 21 2 days.
3.2. Comparative rates of estrus detection For the PM device, two individual malfunctions were noted, but they had no effect on the results because both fell outside any periods when estrus was expected. The AM1 device was unusable for 13 days because of an antenna failure. Furthermore, five tags had to be changed after internal malfunctions. For the AM2 device, two tags had to be changed after an internal malfunction and three because they were lost from a neck collar.
After exclusion of data from a cow without AM2 tags, 211 ovulations were analyzed. Results of the three types of analyses (the entire period, the operating period, and the three-device operating period) are presented in Tables 1 and 2. For the three-device operating period, the sensitivity was higher for PM (71%) than for AM1 (62%; P ¼ 0.02), AM2 (61%; P ¼ 0.005), and AM2_2 (62%; P ¼ 0.02). The positive predictive value of AM1 (84%) and AM2_2 (87%) were higher than those of PM (71%; P ¼ 0.05 and P ¼ 0.007, respectively) and AM2 (67%; P ¼ 0.006 and P ¼ 0.0007, respectively). Generally, the performance of the devices increased as the delay from calving extended (Table 2). Of all ovulations, 80% (168/211) were detected by at least one device and 40% (85/211) by all devices. The number of periods of estrus exclusively detected by each device was 17/ 211 for PM, 6/211 for AM2, and 0 for AM1 and AM2_2. Fiftyone percent (30/59) of the first postpartum ovulations and 91% (138/152) of the following postpartum ovulations were detected by at least one device. Similarly, 12% (7/59) of the first postpartum ovulations and 51% (78/152) of the following postpartum ovulations were detected by all devices. 3.3. Factors affecting the performance of the three devices As can be seen in Table 3, the first postpartum ovulation had a lower sensitivity than for the following ovulations for PM, AM1, and AM2_2 (P < 0.0001) and a lower positive predictive value for AM1, AM2, and AM2_2 (P ¼ 0.002; P ¼ 0.001; P < 0.0001, respectively). Cows in their third lactation had a lower sensitivity than cows in their first lactation for PM (P ¼ 0.019) and a lower predictive positive value for AM1 (P ¼ 0.038). High-producing cows had a lower sensitivity than low-producing cows for AM1 (P ¼ 0.006). Mid-producing dairy cows tended to have a lower sensitivity than low-producing cows for AM2_2 (P ¼ 0.063). Mid-producing dairy cows had a lower positive predictive value compared with low-producing cows for AM2 and AM2_2 (P ¼ 0.046 and P ¼ 0.021). The pattern of cyclicity, the estimated resumption of cyclicity, and the minimum milk protein content had no effect on the ability of any device to detect ovulation. 4. Discussion
Fig. 2. Cumulative percentage of cyclic cows (n ¼ 63) on the study period (from calving to 90 DIM).
The PM had a greater sensitivity but a lower positive predictive value than the two activity meters for the
738
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
Table 1 Sensitivity (SE) and positive predictive value (PPV) of the estrus detection devices. Period
Indicators
Pedometer (PM)
Activity meter 1 (AM1)
Activity meter 2 (AM2)
Activity meter 2 algorithm 2 (AM2_2)
Entire period (ENT)
TP FP FN SEENT (%) PPVENT (%) TP FP FN SEOPE3 (%) [95% CI] PPVOPE3 (%) [95% CI]
154 61 57 73 72 127 53 52 71a[64–77] 71a[63–77]
117 22 94 55 84 111 22 68 62b[55–69] 83b[76–89]
125 63 86 59 66 109 55 70 61b[54–68] 66a[59–73]
128 17 83 61 88 111 17 68 62b[54–69] 87b[80–92]
Operating period of the three devices (OPE3)
Data from the entire study period were analyzed (ENT, from 7 to 90 DIM; n ¼ 211 ovulations), as were data from the operating period of all devices (OPE3; i.e., from 7 to 90 DIM after removing ovulations during a failure of at least one device; n ¼ 179 ovulations). Activity Meter 2 was assessed with two algorithms, AM2 and AM2_2. Abbreviations: FN, false negative; FP, false positive; TP, true positive; [95% CI], 95% confidential interval. a,b Percentage within a row with different superscripts differs (P 0.05).
detection of estrus in dairy cows. We also found that the performance of all three devices was influenced mainly by the number of ovulations from calving (ovulation rank). The results are consistent with the literature for ovarian cyclicity patterns [22–26], cyclicity resumption [27], and the global performance of the three devices (PM [16,18,21], activity meters [18,21,28]). The high rate of FP alerts emitted by the PM in our study (around 29%) is not surprising considering the rates of 17% to 55% reported in the review by Firk et al. [16]. The rate of FP alerts emitted by the activity meters is much more variable among published studies, ranging from a low of 6.5% [18] to high figures of 21% [29] and 28% [30] and our own figures of 12% to 33% being similar. For the HeatPhone activity meter, the second algorithm (AM2_2) clearly improved performance. In any case, false alerts could lead to inseminations during luteal phase and increase the expenses for semen and breeding. Moreover in the present study, 23% to 45% of expected estrus events were not detected and 20% were not detected by any of the devices. This underachievement of the devices can be explained by three factors discussed below: (1) a nonoptimal algorithm, (2) the experimental design, and (3) individual or environmental factors affecting estrus expression. The present results were obtained from observations on one herd, so extrapolations need to be considered with caution. The cows were fed in individual, raised troughs; so repeated, wide head movements would have been
possible, reducing the accuracy of devices attached to neck collars. Moreover, our validation of the estrus alerts was based on ovulations identified using only data from P4 assays in milk and not from video records that allow quantification of estrus expression (duration and intensity). We are cautious about interpretation because FP alerts can be caused by overactivity in the cows induced by their environment and because FN alerts could be caused by silent ovulation. Between 9% and 16% of dairy cows have at least one ovulation of ranks 2 to 4 that are not accompanied by estrus [30], so some FN alerts are almost certainly due to these silent ovulations and the performance of the devices could be underestimated in our study. Finally, we did not study the conception rates and the timing for insemination because of the experimental design (each cow fitted with the three devices), and these aspects should be further investigated [31,32]. The detection of estrus was influenced mainly by ovulation rank. Sensitivity for the first postpartum ovulation varied from 23% to 40% (depending on the device) compared with 78% to 86% for subsequent ovulations. These observations are consistent with those from previous studies [8,16]. It is known that 50% to 80% of first ovulations postpartum are silent [30,33]. We did not have continuous video recording in the study to check which first postpartum ovulations were accompanied by estrus, but it is very likely that the poor performance of all three devices at the first postpartum ovulation was related to low
Table 2 Sensitivity (SE) and positive predictive value (PPV) of the estrus detection devices depending on the interval between calving and alert. Pedometer (PM) Activity meter 1 (AM1) Activity meter 2 (AM2) Activity meter 2 algorithm 2 Interval (AM2_2) between calving <30 [30–50] [50–70] 70 Total <30 [30–50] [50–70] 70 Total <30 [30–50] [50–70] 70 Total <30 [30–50] [50–70] 70 Total and alert (d) TP FP FN SE (%) PPV (%)
20 23 27 43 47
41 13 14 75 76
45 13 9 83 78
48 12 7 87 80
154 61 57 73 72
15 6 32 32 71
32 3 17 65 91
39 8 11 78 83
31 5 10 76 86
117 22 70 63 84
13 12 30 30 52
29 10 22 57 74
40 18 14 74 69
43 23 11 80 65
125 63 77 62 66
10 6 33 23 63
32 6 19 63 84
42 2 12 78 95
44 3 10 81 94
128 17 74 63 88
Analyses were made on the operating period of the studied device (OPE; i.e., after removing ovulations during a failure of the studied device), on 211 ovulations for PM, 187 ovulations for AM1, and 202 ovulations for AM2 and AM2_2. Abbreviations: FN, false negative; FP, false positive; TP, true positive.
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
739
Table 3 Sensitivity (SE) and positive predictive value (PPV) of the estrus devices in relation to ovulation rank, lactation rank, and peak milk production. Risk factor
Pedometer (PM)
Activity meter 1 (AM1)
Activity meter 2 (AM2)
Activity meter 2 algorithm 2 (AM2_2)
SEOPE (%)
SEOPE (%)
PPVOPE (%)
PPVOPE (%)
SEOPE (%)
PPVOPE (%)
24a 0.08 [0.04–0.18]*** 79b
60a 0.18 [0.06–0.51]** 90b
43a 0.29 [0.14–0.61]** 73b
23a 0.08 [0.04–0.18]*** 78b
48a 0.02 [0.00–0.09]*** 98.1b
d d
d d
d d
d
d
d
68a 62b 0.41 [0.17–0.99]* 46 0.76 [0.30–1.90]
62 40 0.40 [0.15–1.04]y 51 0.62 [0.24–1.60]
95a 66b 0.10 [0.01–0.75]* 89 0.40 [0.05–3.30]
Ovulation rank 1 40a 0.11 [0.05–0.22]*** 2 86b Lactation rank 1 77a 2 70 0.71 [0.254–2.00] 3 52b 0.33 [0.12–0.86]* Peak milk production (kg) 35 d 35–40 d 40
d
d d d
68a 50 0.46 [0.16–1.29] 36b 0.26 [0.10–0.72]**
90a 75 0.35 [0.07–1.66] 35b 0.21 [0.05–0.93]* d d d
From the multivariate analysis of the operating period of the device (OPE; i.e. after removing ovulations during a failure of the studied detector), on 211 ovulations for PM, 187 ovulations for AM1, and 202 ovulations for AM2 (Activity Meter 2 was also assessed with a new algorithm, AM2_2). Odds ratio [95% confidence interval] are given in italics. a,b Percentage within a column and per factor differs with P adjust 0.05. yP < 0.10, *P < 0.05, **P < 0.01, ***P < 0.001.
expression of estrus at this time. In our study, the improved performance after 50 DIM is due to the fact that all first ovulations postpartum occurred before 50 days. In practice, it is not advisable to inseminate before 50 DIM so as to ensure the resumption of cyclicity in all cows and complete involution of the uterus [34]. Thus, the first insemination is done generally after 65 DIM [25,35,36] at a time when there are no longer any performance issues. The performance of PM and Heatime-RuminAct activity meter was higher for cows in their first lactation than for cows in their third or higher lactation. This observation is also consistent with previous reports showing that walking activity at estrus decreases with later lactations [37,38], and that estrus is expressed more strongly in primiparous compared with multiparous cows [7,31]. However, in other studies, the duration of estrus and standing activity were not affected by parity [12,39]. The performance of some of the devices was also adversely affected by milk production, and, globally, cows with low milk production showed the best results. This is consistent with the fact that milk yield affects the expression of estrus in dairy cattle, probably due in part to decreased circulating concentrations of estradiol [39,40]. Moreover, high-producing dairy cows have increased nutritional requirements that (1) decrease the time that they spend interacting with their herd mates because of the increased time spent standing while feeding, or lying while ruminating [18] and (2) limit estrus expression, following their survival instinct, in order to avoid conception and therefore the requirement for better nutrition in the later stages of pregnancy [41]. Because of the limitations of our study (e.g., low number of cows from a single herd), further investigations are needed to determine precisely the factors that adversely affect the performance of the three devices. Nevertheless, our results do provide the following: (1) new data to improve the performance of automated estrus devices by including some risk factors in future algorithms and (2)
some useful guidance for dairy farmers wishing to use these tools. The algorithms are now calculated on variation of cow activity, but, even though an increase in activity is concomitant estrus, this is not a very specific behavior (compared with standing mount). One way to reduce this problem could be to regulate the level of alerts using nonsensor data or data from other sensor, for example, animal factors (delay from calving, lactation rank or milk production on the alert day) or environmental conditions not studied here, such as type of housing, temperature, or humidity. In the end, we can speculate whether these automated devices are useful for reducing the constraints to heat detection without decreasing reproductive performance. Our results suggest a qualified “yes” with the proviso that their use should be well considered. First, because of the high risk of silent ovulation, performance before 50 DIM is unreliable, but this is not a serious problem because the recommended time for insemination is 50 days. However, the devices we tested recorded only a part of sexual behaviors [10], so combination of these devices with visual observations, as reported by some authors, might be a better strategy to increase estrus detection, conception rates, and reduce culling due to perceived failure to conceive [18,31]. Second, interpretation of the alerts should be taken into account production data, such as peak milk production, as well as the activity history of the cow, when making a decision to inseminate. Thus, estrus detected by an alert should be accepted as true only if it follows a previous detection (visual or automated) by 18 to 24 days or 36 to 48 days, suggesting an interval of 1 or 2 estrous cycles [42]. Moreover, the purchase of such devices will depend on profitability and must be considered in relation to the farming system and the specific objectives of the breeder. Indeed, the benefit could differ with the breeding system management, including the season of calving, the number of cows in the herd, and whether sexed or
740
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741
nonsexed semen is used [21,43]. For example, visual detection of estrus seems to be easier in a seasonal-calving system because estrus is better expressed when several cows are simultaneously on heat [4,20]. In this case, investment in an automated system would require more neck collars, because all cows are at the same stage simultaneously. Another point is finding a balance between sensitivity and predictive values of the device; what’s the most penalizing situation: a FP or a FN alert? The response is not clear and needs to be investigated in relation to semen and breeding expense. Finally, the life style of the farmer needs to be considered: for example, a desire to reduce time-consuming tasks allowing more leisure time or time for others tasks could be part of a decision to invest in an automated device. The cost–benefit ratio of these devices remains to be investigated, with technical, economical, and labor impact, perspectives. On a larger scale, the data collected by the tools could be used to select animals for their potential to express estrus in the context of a continuing decline in the expression and duration of estrus in dairy cows [44]. 4.1. Conclusion Automated devices used for the detection of estrus in dairy cows function by continuously measuring some activity parameters. Three such devices were tested (a PM and two that recorded body activity). All three performed well from the second postpartum ovulation or after day 50 postpartum. The poor performance at the first postpartum ovulation was probably due to the weak expression of estrus or due to silent ovulation, both of which are more common at this stage in high-producing cows. This was not a serious problem in our study because all the first ovulations occurred before the recommended time for insemination (from 50 days postpartum). Acknowledgments The authors thank the staff of the experimental farm Les Trinottières, MC Deloche, from the UNCEIA hormonology service, for progesterone assays, and EVOLUTION and CLASEL for technical support. We also acknowledge Elevage Conseil Loire Anjou for its financial support, Milkline for its financial and technical help, RJ Scaramuzzi for his scientific comments and for English version, and GB Martin for his detailed editing of the article. References [1] Heersche G, Nebel RL. Measuring efficiency and accuracy of detection of estrus. J Dairy Sci 1994;77:2754–61. [2] Gordon P. Oestrus detection in dairy cattle. In Pract 2011;33:542–6. [3] Inchaisri C, Jorritsma R, Vos PLAM, van der Weijden GC, Hogeveen H. Economic consequences of reproductive performance in dairy cattle. Theriogenology 2010;74:835–46. [4] Diskin MG, Sreenan JM. Expression and detection of oestrus in cattle. Reprod Nutr Dev 2000;40:481–91. [5] Van Eerdenburg FJCM, Loeffler HSH, Van Vliet JH. Detection of oestrus in dairy cows: a new approach to an old problem. Vet Q 1996;18:52–4. [6] Lyimo ZC, Nielen M, Ouweltjes W, Kruip TAM, van Eerdenburg F. Relationship among estradiol, cortisol and intensity of estrous behavior in dairy cattle. Theriogenology 2000;53:1783–95.
[7] Roelofs JB, van Eerdenburg FJCM, Soede NM, Kemp B. Various behavioral signs of estrous and their relationship with time of ovulation in dairy cattle. Theriogenology 2005;63:1366–77. [8] Sakaguchi M. Practical aspects of the fertility of dairy cattle. J Reprod Dev 2011;57:17–33. [9] Kerbrat S, Disenhaus C. A proposition for an updated behavioural characterisation of the oestrus period in dairy cows. Appl Anim Behav Sci 2004;87:223–38. [10] Van Vliet JH, Van Eerdenburg FJCM. Sexual activities and oestrus detection in lactating Holstein cows. Appl Anim Behav Sci 1996;50: 57–69. [11] Roelofs J, López-Gatius F, Hunter RHF, van Eerdenburg FJCM, Hanzen C. When is a cow in estrus? Clinical and practical aspects. Theriogenology 2010;74:327–44. [12] At-Taras EE, Spahr SL. Detection and characterization of estrus in dairy cattle with an electronic heatmount detector and an electronic activity tag. J Dairy Sci 2001;84:792–8. [13] Lucy MC. Reproductive loss in high-producing dairy cattle: where will it end? J Dairy Sci 2001;84:1277–93. [14] Caraviello DZ, Weigel KA, Fricke PM, Wiltbank MC, Florent MJ, Cook NB, et al. Survey of management practices on reproductive performance of dairy cattle on large US commercial farms. J Dairy Sci 2006;89:4723–35. [15] Chastant-Maillard S. Is there a future for pharmaceutical management in cow reproduction ? European perspective. 24th World Buiatrics Congress 2006, Nice, France. [16] Firk R, Stamer E, Junge W, Krieter J. Automation of oestrus detection in dairy cows: a review. Livest Prod Sci 2002;75:219–32. [17] Durkin J. Heat detection: trends and opportunities. The first North American Conference on Precision dairy Management 2010, Toronto, Canada. [18] Holman A, Thompson J, Routly JE, Cameron J, Jones DN, GroveWhite D, et al. Comparison of oestrus detection methods in dairy cattle. Vet Rec 2011;169:47. [19] Saint-Dizier M, Chastant-Maillard S. Towards an automated detection of oestrus in dairy cattle. Reprod Domest Anim 2012;47:1056– 61. [20] Roelofs JB, van Eerdenburg F, Soede NM, Kemp B. Pedometer readings for estrous detection and as predictor for time of ovulation in dairy cattle. Theriogenology 2005;64:1690–703. [21] Rutten CJ, Velthuis AGJ, Steeneveld W, Hogeveen H. Sensors to support health management on dairy farms. J Dairy Sci 2013;96: 1928–52. [22] Opsomer G, Gröhn YT, Hertl J, Coryn M, Deluyker H, de Kruif A. Risk factors for post partum ovarian dysfunction in high producing dairy cows in Belgium: a field study. Theriogenology 2000;53:841–57. [23] Lamming GE, Darwash AO. The use of milk progesterone profiles to characterise components of subfertility in milked dairy cows. Anim Reprod Sci 1998;52:175–90. [24] Royal MD, Darwash AO, Flint APE, Webb R, Woolliams JA, Lamming GE. Declining fertility in dairy cattle: changes in traditional and endocrine parameters of fertility. Anim Sci 2000;70:487– 501. [25] Shrestha HK, Nakao T, Higaki T, Suzuki T, Akita M. Resumption of postpartum ovarian cyclicity in high-producing Holstein cows. Theriogenology 2004;61:637–49. [26] Windig JJ, Beerda B, Veerkamp RF. Relationship between milk progesterone profiles and genetic merit for milk production, milking frequency, and feeding regimen in dairy cattle. J Dairy Sci 2008;91: 2874–84. [27] Opsomer G, Coryn M, Deluyker H, de Kruif A. An analysis of ovarian dysfunction in high yielding dairy cows after calving based on progesterone profiles. Reprod Domest Anim 1998;33:193–204. [28] Kamphuis C, DelaRue B, Burke CR, Jago J. Field evaluation of 2 collarmounted activity meters for detecting cows in estrus on a large pasture-grazed dairy farm. J Dairy Sci 2012;95:3045–56. [29] Aungier SPM, Roche JF, Sheehy M, Crowe MA. Effects of management and health on the use of activity monitoring for estrus detection in dairy cows. J Dairy Sci 2012;95:2452–66. [30] Ranasinghe RMSBK, Nakao T, Yamada K, Koike K. Silent ovulation, based on walking activity and milk progesterone concentrations, in Holstein cows housed in a free-stall barn. Theriogenology 2010;73: 942–9. [31] Peralta OA, Pearson RE, Nebel RL. Comparison of three estrus detection systems during summer in a large commercial dairy herd. Anim Reprod Sci 2005;87:59–72. [32] Bar D. Optimal timing of insemination using activity collars. The First North American Conference on Precision Dairy Management 2010, Toronto, Canada.
A. Chanvallon et al. / Theriogenology 82 (2014) 734–741 [33] Kyle SD, Callahan CJ, Allrich RD. Effect of progesterone on the expression of estrus at the first postpartum ovulation in dairy cattle. J Dairy Sci 1992;75:1456–60. [34] Sheldon IM, Lewis GS, LeBlanc S, Gilbert RO. Defining postpartum uterine disease in cattle. Theriogenology 2006;65:1516–30. [35] Berry DP, Buckley F, Dillon P, Evans RD, Rath M, Veerkamp RF. Genetic relationships among body condition score, body weight, milk yield, and fertility in dairy cows. J Dairy Sci 2003;86: 2193–204. [36] Sakaguchi M, Sasamoto Y, Suzuki T, Takahashi Y, Yamada Y. Postpartum ovarian follicular dynamics and estrous activity in lactating dairy cows. J Dairy Sci 2004;87:2114–21. [37] López-Gatius F, Santolaria P, Mundet I, Yániz JL. Walking activity at estrus and subsequent fertility in dairy cows. Theriogenology 2005; 63:1419–29. [38] Yániz JL, Santolaria P, Giribet A, Lopez-Gatius F. Factors affecting walking activity at estrus during postpartum period and subsequent fertility in dairy cows. Theriogenology 2006;66:1943–50.
741
[39] Lopez H, Satter LD, Wiltbank MC. Relationship between level of milk production and estrous behavior of lactating dairy cows. Anim Reprod Sci 2004;81:209–23. [40] Cutullic E, Delaby L, Causeur D, Michel G, Disenhaus C. Hierarchy of factors affecting behavioural signs used for oestrus detection of Holstein and Normande dairy cows in a seasonal calving system. Anim Reprod Sci 2009;113:22–37. [41] Dobson H, Smith RF, Royal MD, Knight CH, Sheldon IM. The highproducing dairy cow and its reproductive performance. Reprod Domest Anim 2007;42:17–23. [42] Philipot JM, Clement P. An innovative tool for monitoring and phenotyping cattle: Heatime-Ruminact. 38th International Committee for Animal Recording 2012, Cork, Ireland. [43] Østergaard S, Friggens NC, Chagunda MGG. Technical and economic effects of an inline progesterone indicator in a dairy herd estimated by stochastic simulation. Theriogenology 2005;64:819–43. [44] Løvendahl P, Chagunda MGG. Short communication: genetic variation in estrus activity traits. J Dairy Sci 2009;92:4683–8.