J. Dairy Sci.

99:1539–1548
http://dx.doi.org/10.3168/jds.2015-10057
© American Dairy Science Association®, 2016.

Evaluation of calving indicators measured by automated monitoring

devices to predict the onset of calving in Holstein dairy cows
V. Ouellet,* E. Vasseur,† W. Heuwieser,‡ O. Burfeind,‡ X. Maldague,§ and É. Charbonneau*1
*Département de sciences animales, Université Laval, Québec, G1V 0A6, Canada
†Organic Dairy Research Center, University of Guelph, Campus d’Alfred, Ontario, K0B 1A0, Canada
‡Clinic for Animal Reproduction, Faculty of Veterinary Medicine, Freie Universität Berlin, Koenigsweg 65, 14163 Berlin, Germany
§Département de génie électrique et de génie informatique, Université Laval, Québec, G1V 0A6, Canada

ABSTRACT 77%, positive and negative predictive values: 56 and

90%, area under the curve: 0.82). These results indicate
Dystocias are common in dairy cows and often ad- that a device that could simultaneously measure these
versely affect production, reproduction, animal wel- 4 calving indicators could not precisely determine the
fare, labor, and economics within the dairy industry. onset of calving, but the information collected would
An automated device that accurately predicts the assist dairy farmers in monitoring the onset of calving.
onset of calving could potentially minimize the effect Key words: dairy cow, calving indicator, onset of
of dystocias by enabling producers to intervene early. calving, test performance
Although many well-documented indicators can detect
the imminence of calving, research is limited on their
INTRODUCTION
effectiveness to predict calving when measured by au-
tomated devices. The objective of this experiment was Calving is a critical time for both the dam and the
to determine if a decrease in vaginal temperature (VT), calf (Schuenemann, 2012). Difficult births, known as
rumination (RT), and lying time (LT), or an increase in dystocias, are common in dairy cows (Lombard et al.,
lying bouts (LB), as measured by 3 automated devices, 2007). Studies show that dystocia rates in the United
could accurately predict the onset of calving within 24, States range from 28.6 to 51.2% in primiparous cows
12, and 6 h. The combination of these 4 calving indica- and from 10.7 to 29.4% in multiparous cows (Meyer
tors was also evaluated. Forty-two multiparous Holstein et al., 2001; Lombard et al., 2007). Dystocias are as-
cows housed in tie-stalls were fitted with a temperature sociated with increased risk of stillbirth, calf mortal-
logger inserted in the vaginal cavity 7 ± 2 d before ity before 30 d of age, and morbidity (Lombard et al.,
their expected calving date; VT was recorded at 1-min 2007). They also increase the likelihood of trauma on
intervals. An ear-attached sensor recorded rumination the dam (Schuenemann et al., 2011), retained placenta
time every hour based on ear movement while an ac- (Oltenacu et al., 1988), uterine disorders (Sheldon et
celerometer fitted to the right hind leg recorded cow al., 2009), and decreased milk yield (Dematawewa and
position at 1-min intervals. On average, VT were 0.3 Berger, 1997; Rajala and Gröhn, 1998). Furthermore,
± 0.03°C lower, and RT and LT were 41 ± 17 and dystocia is negatively associated with fertility and dam
52 ± 28 min lower, respectively, on the calving day survival (Tenhagen et al., 2007). Prevention of dystocia
compared with the previous 4 d. Cows had 2 ± 1 more in dairy cows should, therefore, be a high priority in
LB on the calving day. Of the 4 indicators, a decrease farm management.
in VT ≥ 0.1°C was best able to predict calving within Predicting the onset of parturition can help preserve
the next 24 h with a sensitivity of 74%, specificity of the integrity of the newborn calf and protect the dam
74%, positive and negative predictive values of 51 and during difficult birth situations by facilitating timely
89%, and area under the curve of 0.80. Combining the human intervention (Shah et al., 2006; Palombi et al.,
indicators enhanced the performance to predict calv- 2013). Moreover, predicting calving time allows care-
ing within the next 24, 12, and 6 h with best overall ful management around the time of parturition, thus
results obtained by combining the 3 devices for predic- minimizing unnecessary pain and distress especially
tion within the next 24 h (sensitivity: 77%, specificity: in situations requiring humane intervention (Miedema
et al., 2011b). External signs such as pelvic ligament
relaxation, udder distension, teat filling, vaginal dis-
Received July 3, 2015.
Accepted October 14, 2015. charge, vulva edema, and behavior changes are often
1
Corresponding author: [email protected] used to predict the onset of calving in dairy cows either

1539
1540 OUELLET ET AL.

manually, visually, or by video observation (Berglund MATERIALS AND METHODS

et al., 1987; Streyl et al., 2011). Such assessments are
subjective and time consuming, and the signs them- The study was conducted on a commercial dairy
selves vary widely among dairy cows. Furthermore, farm (Saint-Anselme, QC, Canada), which had a milk-
whereas the number of farms is decreasing in North ing herd of 108 Holstein cows producing, on average,
America and Europe, the number of cows per farm is 10,390 kg/cow per yr. All experimental procedures
increasing. Hence, less time is accorded to individual were approved by the Animal Care Committee from
cow supervision around parturition time, especially in Université Laval, Quebec, Canada.
small operation farms. Therefore, an automated device
that could accurately predict the onset of calving would Animals, Housing, and Feeding
be valuable to minimize the effects of dystocia on dairy
cows. A total of 42 multiparous Holstein cows (average ±
Changes in cow behavior and physiology observed SD; parity: 1.9 ± 1.2; gestation length: 281 ± 3 d; calving
on the day of calving have been well documented. interval: 408 ± 52 d) housed in freestalls were enrolled
Schirmann et al. (2013) report that cows spent, on from November 2013 to June 2014. About 21 d before
average, 63 ± 30 min less time ruminating on the day their expected calving date, the cows were moved to 1
of calving. Miedema et al. (2011a) and Jensen (2012) of the 12 tie-stalls reserved for calving (mean dimension
report an increase of lying bouts (LB) on the day of 2.4 m long × 1.3 m wide; chain length 0.60 to 0.63 m).
calving compared with a control period during gesta- All 12 tie-stalls were fitted with a water bed mattress
tion, whereas a decrease of 1 h in daily duration of (DCC waterbeds HQ, Reedsburg, WI) covered with a
lying time (LT) was observed. A decrease in body tem- thin layer of sawdust. A TMR was provided once daily
perature before the onset of parturition has also been (at 0800 h) consisting of 54% corn silage, 21% alfalfa
reported for dairy cows (Burfeind et al., 2011; Streyl et hay, and 26% concentrate and minerals on a DM basis.
al., 2011). Burfeind et al. (2011) report that a decrease Feed was pushed up 4 times per day. Water was freely
in vaginal temperature (VT) of ≥0.3°C over 24 h as available from water bowls (1 per 2 neighboring cows).
measured by a temperature logger can predict calving
within 24 h with a sensitivity (Se) ranging from 62 to Experimental Measurements
71% and with a specificity (Sp) ranging from 81 to
87%. Rumination time (RT), VT, LB, and LT show Cows were fitted with 3 automated devices to mea-
measurable changes that are consistent between indi- sure 4 calving indicators (i.e., VT, RT, LB, and LT).
viduals. Therefore, RT, VT, LB, and LT are considered Vaginal temperature was recorded continuously every
to be useful calving indicators that can help predict the minute using a microprocessor-controlled temperature
onset of parturition. data logger (Minilog II-t, Vemco Ltd., Halifax, Canada)
Various automated devices that record RT and the as validated by Vickers et al. (2010). The temperature
number of LB and LT are available commercially for data loggers were attached to a modified, controlled, in-
producers. To our knowledge, the test performance of ternal drug-release device without progesterone (CIDR,
those calving indicators measured by automated moni- InterAg, Hamilton, New Zealand) and inserted into the
toring devices has not been determined. Therefore, the vaginal cavity 6 ± 2 d before the predicted calving
objective of this study was to determine the perfor- date, as described by Burfeind et al. (2011). During the
mance of 3 automated devices to predict the onset of process of calving, the temperature logger was expelled
calving based on measuring 4 calving indicators (de- from the vaginal cavity and collected by the farm staff,
crease of VT, RT, and LT; increase of LB). Specifically, resulting in a sudden decrease in measured tempera-
we set out (1) to investigate the test performance of ture, as described by Burfeind et al. (2011). The time of
the calving indicators during 3 time periods (within the complete expulsion of the logger was established as the
next 24, 12, and 6 h) before calving, and (2) to evaluate calving time for each cow. The temperature data were
the test performance using combined calving indica- downloaded after calving.
tors. We hypothesized that, of the 4 calving indicators, Rumination time was measured continuously every
a decrease in VT, recorded by a temperature logger, hour using a 3-dimensional accelerometer (SensOor;
would obtain the highest test performance to predict Agis Automatisering BV, Harmelen, the Netherlands)
calving, with the best predictive value for a calving designed to be attached to the ear identification tag of
prediction within the next 24 h, and that combining the the cows as validated by Bikker et al. (2014). Rumi-
indicators would improve the predictive performance of nation data were sent through a wireless connection,
the individual indicators. via routers and coordinators, to an on-farm computer

Journal of Dairy Science Vol. 99 No. 2, 2016

 PREDICTION OF CALVING IN DAIRY COWS 1541

and was available through a web-based application, the assumption of variance homogeneity. Pearson cor-
as described by Bikker et al. (2014). The cows were relations between THI and VT, RT, LB, and LT were
equipped with the ear-attached sensor 7 ± 1 d before calculated using PROC CORR of SAS.
their predicted calving date. Measurements were taken Differences in RT, LB, and LT between the day of
starting at 6 d before the anticipated calving date until calving assigned d 0 and the 4 d antepartum assigned
calving. Rumination data were downloaded from the d −4, −3, −2, and −1 were determined using PROC
CowManager SensoOr system every day until the time GLIMMIX of SAS with the cow as a random effect.
of calving. For VT, PROC MIXED was used with a model that
The position of the cow (lying or standing) was assumed heterogeneity of variance to account for the
recorded continuously every minute using an Onset normal modification in the amplitude of temperature
Pendant G data logger (Onset Computer Corpora- for the baseline compared with the day of calving. To
tion, Bourne, MA) as validated for measuring the ly- explore the approximate time of change for all indica-
ing behavior of dairy cows by O’Driscoll et al. (2008). tors, differences between the last twenty 6-h periods
Cows were fitted with the device 6 ± 2 d before their before calving were also calculated using PROC GLIM-
expected calving date. The data loggers were wrapped MIX of SAS with cows as random effect and time of
in VetWrap cohesive bandage (3M Products, St. Paul, day (morning, afternoon, evening, and night) set as a
MN) to provide cushioning and were placed on the right covariate with 4 levels. Morning was defined as the pe-
hind leg of the cow. The data were downloaded after riod from 0600 to 1200 h, afternoon from 1200 to 1800
calving using Onset HOBOware Software (Onset Com- h, evening from 1800 to 0000 h, and night from 0000
puter Corporation) and exported to Microsoft Excel to 0600 h.
(Microsoft Corp., Redmond, WA). Lying time and the Differences between each indicator for a particular
number of LB were computed using Excel macros with 6-h period, and the equivalent 6-h period 24 h previ-
LB defined as a period of lying for at least 2 consecu- ously, were calculated for the last 120 h before calving.
tive min (Jensen, 2012) separated by periods of walking Receiving operating characteristic (ROC) analyses
or standing (Miedema et al., 2011a). for the differences were then conducted using PROC
The ambient temperature (AT, °C) and the relative LOGISTIC of SAS to determine the diagnostic perfor-
humidity (RH, %) of the barn were measured continu- mance and cut-off points of a decrease in VT, RT, and
ously every minute throughout the research project us- LT, and an increase in LB. Cut-off points were defined
ing a temperature and RH data logger (HOBO U23 Pro as the threshold calculated for each indicator optimiz-
v2, Onset Computer Corporation) secured on the barn’s ing both Se and Sp for predicting calving. The continu-
ceiling about 1 m above the cows. The temperature- ous variable was the difference in VT, RT, LB, or LT,
humidity index (THI) was calculated using the equa- whereas the classification variable was the occurrence
tion reported by Kendall et al. (2008): THI = (1.8 × of calving within 24, 12, or 6 h. Because indicators were
AT + 32) − [(0.55 – 0.0055 × RH) × (1.8 × AT – 26)]. summarized in 6-h periods, 4, 2, and 1 positive events,
defined as the occurrence of calving within the 24, 12,
Statistical Analysis or 6 h, existed per cow, respectively. A prediction of
calving was considered when a decrease in VT, RT, LT,
Data were analyzed with SAS 9.3 (2011, SAS Insti- or an increase in LB, were observed and were greater or
tute Inc., Cary, NC). During the study, VT, LB, and equal to the cut-off point.
LT were measured every minute by their respective Test characteristics (Se, Sp, and predictive values)
automated device. For further analysis, hourly means for predicting the onset of calving for each cut-off point
were calculated for each cow independently. Vaginal and the 95% confidence interval were then calculated
temperatures below 38.0°C were considered to be arti- using PROC FREQ of SAS. Sensitivity was defined as
facts due to movement of the temperature logger and the proportion of positive events (occurrence of calv-
were excluded from the data set as described by Bur- ing within 24, 12, or 6 h) correctly predicted by the
feind et al. (2011). Rumination time was already cal- test (calving correctly predicted/total calving events).
culated per hour by the CowManager SensoOr system Specificity was defined as the proportion of negative
and expressed as a percentage of behavior per hour for events (absence of calving within 24, 12, or 6 h) cor-
each cow. The percentage was then transformed (i.e., rectly diagnosed as being negative by the test (absence
divided by 100 and multiplied by 60) to have the RT of calving correctly predicted/total of absence of calv-
in minutes per hour. Each calving indicator was sum- ing). The positive predictive value was defined as the
marized per day to obtain 1 value per cow per day and proportion of events with a positive prediction of calv-
in 6-h periods to obtain 4 values per cow per day. The ing that resulted in a calving within the expected time
number of LB was square-root transformed to meet interval (calving correctly predicted/total of calving
Journal of Dairy Science Vol. 99 No. 2, 2016
1542 OUELLET ET AL.

predicted). The negative predictive value was defined ings occurring in the morning, 8 in the afternoon, 9 in
as the proportion of events with a negative prediction the evening, and 10 in the night. Only 2 cows calved
for calving that were correctly diagnosed negative by on their predicted date, whereas 14 cows calved before
the test (absence of calving correctly predicted/total of (average ± SD; days before the predicted calving date:
absence of calving predicted). 3.0 ± 1.5 d) and 16 cows calved after (average ± SD;
After individual evaluation of each calving indica- days after the predicted calving date: 3.1 ± 1.7 d) their
tor, it was possible to measure the test characteristics predicted date.
of combinations because the indicators were not cor-
related. This analysis was done using a multivariate Differences Between Days
logistic regression in SAS, and was followed by PROC
LOGISTIC and PROC FREQ to compute the area Cows exhibited distinctive changes in the 4 calv-
under the curve and to evaluate performance of each ing indicators within the last 24 h before parturition
combination. Observations for a specific time preceding compared with the 4 d precalving (Table 1). Mean VT
calving with missing data from 1 of the indicators were was lower (P < 0.05) on the day of calving compared
excluded from this analysis. with 1, 2, 3, and 4 d before calving. Vaginal tempera-
ture recorded on the 4 d before the day of calving did
RESULTS AND DISCUSSION not significantly differ. An average decrease (mean ±
SE) of 0.3 ± 0.03°C (P < 0.05) was observed on the
During the study, average daily ambient temperature day of calving compared with 4 d before parturition.
(±SD) was 13.6 ± 2.6°C, whereas THI was 56.7 ± 3.9. This agrees with the findings of Burfeind et al. (2011)
Vaginal temperature and THI (r = 0.06; P < 0.01), RT and Streyl et al. (2011) who also measured a decrease
and THI (r = 0.06; P < 0.01), and LT and THI (r = of 0.3°C in VT the day of calving compared with the
0.10; P < 0.01) were correlated. The number of LB and preceding days. The similarity between the findings
THI were not correlated (P > 0.05). Considering the confirms that the amplitude of VT variations before
low or lack of correlation observed between THI and calving is relatively constant in Holstein dairy cows.
all the indicators, this measure was not used for fur- Similarly, RT was lower (P < 0.05) on the calving
ther analysis. No significant correlation was observed day compared with the 4 d precalving (Table 1). No
between the 4 indicators (P > 0.05). significant difference was observed between the 4 d be-
Ten cows were excluded from the analysis due to fore parturition. Cows spent, on average (mean ± SE),
technical problems with 1 of the 3 automated devices 41 ± 17 min/24 h (P < 0.05) less time ruminating on
(e.g., 8 cows lost their temperature logger before calv- the calving day compared with the 4 d before calving,
ing, and 2 cows did not have RT data due to technical which is comparable but lower than the decrease of 63 ±
problems with the rumination sensor). Therefore, 32 30 min/24 h observed by Schirmann et al. (2013). The
multiparous cows were included in the final analysis discrepancy in the results could partly be related to the
with data collected from the 3 devices. In the experi- different devices used to measure the RT. Schirmann
ment, 39,691 of the 230,400 VT measures (17%) were et al. (2013) used a rumination collar based on an
below 38°C and excluded from further analysis. A total acoustic measure, whereas a rumination sensor based
of 66 h were excluded from the RT analysis (0.02%) on ear movements was used in our study. Moreover, the
due to loss of signal from the system. Calvings were cows in Schirmann et al. (2013) were checked multiple
distributed irregularly throughout the day with 5 calv- times for relaxation of tail ligament, vulval discharge,

Table 1. Daily vaginal temperature (mean ± SE), daily rumination time (mean ± SE), daily number of lying bouts (mean ± SE), and daily
lying time (mean ± SE) on the 4 d before and the day of parturition for dairy cows (n = 32 multiparous cows)

Days relative to calving

Indicator1 −4 −3 −2 −1 0 P
a a a a b
VT (°C) 38.7 ± 0.03 38.8 ± 0.03 38.8 ± 0.03 38.7 ± 0.03 38.5 ± 0.03 <0.001
RT (min/24 h) 664.5 ± 21.8a 657.4 ± 22.2a 655.3 ± 22.2a 653.0 ± 22.3a 617.0 ± 22.6b <0.001
LB2 (bouts/24 h) 8.9 ± 0.9c 9.8 ± 1.0bc 10.4 ± 0.9abc 11.3 ± 1.1ab 11.8 ± 1.1a <0.001
LT (min/24 h) 802.4 ± 41.4a 734.2 ± 39.6ab 760.9 ± 39.0ab 774.4 ± 39.0ab 712.9 ± 38.3b <0.001
a–c
Means within a row with different superscripts differ (P < 0.05).
1
Indicators: VT = vaginal temperature; RT = rumination time; LB = lying bouts; LT = lying time.
2
Values were square root transformed to meet variance homogeneity assumption. Back-transformed values are shown in the table.

Journal of Dairy Science Vol. 99 No. 2, 2016

 PREDICTION OF CALVING IN DAIRY COWS 1543

and milk letdown. They were moved to a calving pen the periods from 120 to 24 h before calving and tended
when calving was considered imminent, which was, on to be lower than VT measured during 24 to 18 h before
average, less than 4 h before the expulsion of the calf. calving (Figure 1A). Burfeind et al. (2011), in 3 ex-
Moving the cows at that time could have contributed periments, also observed that VT reached a minimum
to the decrease in RT, resulting in the greater overall of respectively 18, 13, and 15 h before the onset of
decrease observed in their trial. In our study, fewer calving. The time of day (morning, afternoon, evening,
changes in the environment occurred because cows were and night) had an effect (P < 0.001) on VT during the
moved to a tie-stall reserved for calving 21 d before last 120 h before calving (Figure 2A). Burfeind et al.
their expected calving, where they later calved. (2011) observed that VT exhibited a diurnal rhythm
The daily number of LB was also influenced by calv- throughout the experiment with a minimum reached
ing time, which corroborates earlier findings (Huzzey in the night (0000 to 0600 h) and a maximum reached
and Von Keyserlingk, 2005; Miedema et al., 2011a; during the evening (1800 to 0000 h).
Jensen, 2012). On a daily basis, our results indicate Rumination time reached a minimum in the last 6-h
that the number of LB started to increase from d −3 period before parturition (Figure 1B). In the last 24 h
before parturition (Table 1). A maximum of LB per before calving, the RT declined in the last 6-h period
day was reached on the calving day. On average (mean before calving compared with the 12 to 6 h before calv-
± SE), 2 ± 1 more LB (P < 0.05) were found on the ing and was lower (P < 0.05) than the periods 24 to
day of calving compared with the 4 d precalving. Our 12 h before the onset of parturition. Cows spent, on
result is lower than that measured in 2 other studies average (mean ± SE), 162.8 ± 8.0 min/6 h ruminating
that observed 7.8 and 7 more LB during the last 24 h during the periods from 120 to 6 h before calving and
before parturition compared with 24 h and 4 d precalv- 131.6 ± 7.5 min/6 h in the last 6 h before calving.
ing, respectively (Miedema et al., 2011a; Jensen, 2012). This result agrees with previous studies where RT was
The cause of variation in the increase of LB observed significantly reduced by 25.6 min/6 h in the final 6 h
on the calving day between the studies might be due to before calving and by 10 min/2 h in the last 4 h ante-
the different housing systems. Miedema et al. (2011a) partum (Büchel and Sundrum, 2014; Pahl et al., 2014).
group-housed their cows in a large straw-bedded barn, The time of day (morning, afternoon, evening, night)
and Jensen (2012) kept their cows in individual calv- had a significant effect (P < 0.01) on RT with cows
ing pens, also bedded with deep straw; the cows in ruminating less during the morning and afternoon, and
our study were kept in a tie-stall with a thin layer of more in the evening and at night (Figure 2B).
sawdust. Cows in a tie-stall are more restrained in their The number of LB reached a maximum in the last 6
movements, which could explain the smaller increase h before parturition (Figure 1C). In the last 24 h, the
in LB observed on the day of calving compared with number of LB was higher (P < 0.05) during the period
the days before. Furthermore, the differences observed 6 to 0 h before calving than the period 24 to 18 h. The
between the studies can be partly explained by the increased number of LB observed in the last 6 h before
definitions of the calving time. Miedema et al. (2011a) the onset of parturition is well documented (Miedema
and Jensen (2012) defined calving time as when the calf et al., 2011a; Jensen, 2012) and reflects the increased
was fully expelled, whereas our calving time was set to degree of restlessness and the growing discomfort of the
when the temperature logger was fully expelled. cow with the imminence of calving. The time of day
Daily LT was, on average, lower (P < 0.05) on the (morning, afternoon, evening, night) had no effect (P
day of calving compared with d −4 before calving (Ta- > 0.05) on the number of LB (Figure 2C).
ble 1). Variation for this indicator was progressive and When all the experimental periods were compared,
reached a nadir on the day of calving. The amplitude no difference (P > 0.05) in LT across periods was
of variation in daily LT (mean ± SE; 52 ± 28 min) be- observed (Figure 1D). However, LT was numerically
tween the day of calving compared with the 4 d before lower in the interval of 12 to 6 h before parturition.
parturition was similar to previous studies. Miedema et This result agrees with previous studies that did not
al. (2011a) and Jensen (2012) also measured a decrease identify any difference in LT duration between the
in LT of about an hour on calving day compared with four 6-h periods before calving (Miedema et al., 2011a)
the control period. and between the last twelve 2-h periods before calving
(Jensen, 2012). The time of day (morning, afternoon,
evening, night) had an effect (P < 0.001) on the LT
Difference Between 6-h Periods with most cows resting more during the night (0000 to
0600 h) and less during the afternoon (1200 to 1800 h),
Vaginal temperatures were lower (P < 0.05) during the morning and evening periods having intermediary
the last three 6-h periods (18 to 0 h) compared with results (Figure 2D).
Journal of Dairy Science Vol. 99 No. 2, 2016
1544 OUELLET ET AL.

Figure 2. Mean vaginal temperature (A; SEM = 0.02), mean ru-

Figure 1. Mean vaginal temperature (A; SEM = 0.08), mean ru- mination time (B; SEM = 3.68), mean number of lying bouts (C;
mination time (B; SEM = 13.20), mean number of lying bouts (C; SEM = 0.14), and mean lying time (D; SEM = 6.59) during morning,
SEM = 0.32), and mean lying time (D; SEM = 14.05) in the last 120 afternoon, evening, and night in the last 120 h before parturition in
h before parturition in multiparous dairy cows (n = 32). multiparous dairy cows (n = 32).

Journal of Dairy Science Vol. 99 No. 2, 2016

 PREDICTION OF CALVING IN DAIRY COWS 1545

Test Performance of Calving Indicators Vaginal temperature results are comparable with the
findings of Burfeind et al. (2011); they measured the
Changes observed in VT, RT, LB, and LT in rela- test performance to predict calving events within 24
tion to the onset of parturition in dairy cows are well h of an hourly decrease in VT compared with 24 h
documented. However, to our knowledge, the only test earlier in 3 experiments. In their experiments, the tests
characteristics of a decrease of VT measured over 24 h obtained a Se ranging from 55 to 76%, a Sp ranging
by a temperature logger as an indicator to predict calv- from 71 to 92%, a +PV ranging from 42 to 70%, a –PV
ing within the next 24 h is currently available (Burfeind ranging from 86 to 92%, and an AUC ranging from 0.77
et al., 2011). to 0.84 with best results achieved when a decrease of
Sensitivity, Sp, +PV, and –PV are necessary factors ≥0.3°C was measured. Our data showed optimal test
to evaluate the validity of a predictive test (Burfeind performance to predict calving within the next 24 h
et al., 2011). Among all indicators, a decrease in VT when a decrease of ≥0.1°C was measured over 24 h.
measured by the temperature logger showed the high- This discrepancy in optimal difference is likely related
est predictive value for calving within the next 24, 12, to the calculation of the cut-off points. In our study,
and 6 h (Table 2). Variation in VT also obtained the a single cut-off point allowed the optimization of both
greatest area under the curve (AUC). The AUC is a Se and Sp, which were calculated, whereas Burfeind et
useful tool to assess the diagnostic accuracy of a test al. (2011) calculated several temperature decreases to
and to compare the performance of more than one test measure their test characteristics without attempting
for the same outcome (Bewick et al., 2004). The AUC to optimize both Se and Sp.
indicates the ability of the test to discriminate cows The number of LB and RT showed important changes
that will calve and the cows that will not calve within during the last 6-h period before the onset of calving,
the next 24, 12, or 6 h. Therefore, a test that would be whereas LT reached a minimum 12 to 6 h before calv-
able to differentiate the 2 populations perfectly would ing, and VT were lower during the 18 to 12 h before
have a AUC of 1, whereas a predictor that is not able calving. Therefore, it appears that the test performance
to categorize the 2 populations at all would have an for different prediction times is associated with the
AUC of less than 0.5 (Bewick et al., 2004; Burfeind et period during which the changes in the indicators are
al., 2011). Several scales are available for AUC inter- most important. Moreover, a decrease in RT and LT
pretation; in general, a test with an AUC ≤0.75 is not and an increase in LB showed lower performances to
clinically useful (Fan et al., 2006). predict calving with lower Se, Sp, +PV, –PV, and AUC

Table 2. Test performance (95% confidence interval in parentheses) of optimal cut-off point of decreases in vaginal temperature, rumination
time, lying time, and increase in lying bouts measured over 6-h period and compared with the same period 24 h earlier as a predictor of
parturition within 24, 12, and 6 h1

Calving indicator
Prediction
2
time Test performance VT RT LB LT
3
24 h Cut-off point 0.1°C/6 h 3.6 min/6 h 0 bout/6 h 8 min/6 h
Se (%) 74 (65–82) 51 (42–60) 67 (59–75) 56 (47–65)
Sp (%) 74 (69–79) 51 (46–57) 27 (22–32) 57 (51–62)
+PV (%) 51 (43–59) 27 (21–33) 25 (20–30) 32 (26–39)
−PV (%) 89 (85–92) 75 (69–80) 69 (61–77) 78 (72–83)
AUC 0.80 0.54 0.52 0.58
12 h Cut-off point 0.2°C/6 h 5.4 min/6 h 1 bout/6 h 11 min/6 h
Se (%) 69 (56–80) 52 (39–65) 39 (27–53) 57 (44–70)
Sp (%) 69 (64–74) 55 (49–59) 63 (58–67) 57 (52–62)
+PV (%) 26 (20–34) 15 (11–21) 14 (9–20) 17 (12–23)
−PV (%) 93 (90–96) 88 (83–92) 87 (82–91) 90 (85–93)
AUC 0.74 0.60 0.53 0.56
6h Cut-off point 0.2°C/6 h 12.0 min/6 h 1 bout/6 h 3 min/6 h
Se (%) 68 (43–83) 63 (44–79) 53 (34–69) 48 (30–67)
Sp (%) 67 (62–71) 63 (58–67) 63 (59–68) 47 (42–52)
+PV (%) 13 (8–19) 11 (7–17) 9 (5–15) 6 (4–10)
−PV (%) 97 (94–98) 95 (93–98) 95 (91–97) 93 (88–96)
AUC 0.68 0.67 0.60 0.52
1
Calving indicators: VT = vaginal temperature; RT = rumination time; LB = lying bouts; LT = lying time.
2
Test performance: Se = sensitivity; Sp = specificity; +PV = positive predictive value; −PV = negative predictive value; AUC = area under
the curve.
3
Cut-off point = threshold calculated for each indicator optimizing both Se and Sp for predicting calving.

Journal of Dairy Science Vol. 99 No. 2, 2016

1546 OUELLET ET AL.

compared with VT. Those indicators exhibited larger used in the different combinations. Therefore, the ma-
variation between the cows (Table 1) making it harder jor improvement for the combinations including VT or
to calculate a common optimal cut-off point for all the LT is for a prediction within the next 6 h when they
cows. The constant decrease in VT of 0.3°C the day are combined with RT or LB; the major improvement
before calving measured in our study and in 2 other including LB or RT is for a prediction within the next
studies (Burfeind et al., 2011; Streyl et al., 2011) may 6 h when they are combined with VT. The LT had a
indicate less variation between cows for this calving in- low effect on the test performance of the combinations
dicator. We speculate that physiological changes might of indicators. The low effect of LT in combinations can
be more stable among cows than behavioral ones, mak- be related to the lack of specific evolution of this be-
ing them more accurate for predicting calving time. havior related to the imminence of calving observed in
Likewise, Matsas et al. (1992) suggest that a reduction this study when comparing the 6-h period results. A
in progesterone is the most accurate measure to predict combination with VT generally improved the results of
calving within the next 12 h in dairy cows. This mea- all the other parameters taken individually.
sure is also a physiological indicator which supports our The combination of VT, RT, LB, and LT obtained
hypothesis. the best test performance to predict calving within the
Each indicator showed low +PV, especially for a next 24 and 12 h, whereas the combination of VT, RT,
prediction of calving within the next 6 h (Table 2). and LB obtained the best results for predicting calving
Sensitivity and Sp are test characteristics that are not within the next 6 h (Table 3). This result emphasizes
affected by the prevalence of positive events (occur- that the test performance for different prediction times
rence of calving). However, the predictive values are is associated with the period during which the changes
affected by the Se, the Sp, and by the prevalence of are most important. The best results for a combina-
positive events (Bewick et al., 2004). Therefore, when tion that does not include VT, which requires vaginal
the prevalence of positive events is low, the +PV will installation, is obtained by combining RT and LB for a
be automatically low irrespective of the Se and the Sp, prediction in the next 6 h. Such a combination could be
whereas –PV will be high (Bewick et al., 2004). In our advantageous because the devices used to measure RT
study, all calving indicators were summarized in 6-h and LB require minimal setup and are less invasive for
periods, and 4, 2, and 1 positive events were defined as the cows than those required for VT. Despite improved
the occurrence of calving within the next 24, 12, and 6 test performance through combining the parameters,
h, respectively. This approach explains why lower +PV +PV remained low, whereas –PV were high for all the
and higher –PV were measured for a prediction of calv- combinations. A device that would be able to measure
ing within the next 6 h compared with the prediction a combination of the 4 indicators would not be able to
of calving within the next 12 and 24 h. Moreover, a test accurately predict the onset of calving but could be
with higher Se and Sp will automatically have higher a useful tool to assist calving management on dairy
predictive values. Therefore, the predictive values are a farms. More work is needed to determine if the uti-
useful tool to evaluate performance of a given test but lization of the tested devices can result in beneficial
cannot be used, in our study, to compare the different interventions that justify the cost. Future work should
indicators. investigate whether other commercially available de-
vices that monitor the tested indicators (i.e., infrared
Test Performance of Different Combinations cameras) could improve test performance in predicting
of Calving Indicators the onset of calving. Also, whether devices that moni-
tor other calving indicators (i.e., tail raise) can predict
Test performance of all the possible combinations the onset of calving.
of the 4 calving indicators were conducted (Table 3). Finally, of note, the temperature logger and the ac-
Because false positive events were high when predicting celerometers used in this study are not practical for
individual parameters, the combinations chosen were commercial use because both of them require data
inclusive (i.e., a prediction of calving was made when downloading. Furthermore, the automated devices
all the indicators in its composition were conclusive). used in our study were chosen because they have been
Combining the devices to allow simultaneous consider- validated in previous studies and thus are valuable
ation of the calving indictors enhanced the performance for research purposes. However, a temperature logger
to predict calving within the next 24, 12, or 6 h com- that requires inserting a device into the cow’s vaginal
pared with when indicators were used independently cavity may not be popular with producers due to the
(Table 3). This result indicates that combining calving installation required and the potential for infection and
indicators can merge the strength of each indicator irritation.

Journal of Dairy Science Vol. 99 No. 2, 2016

 Table 3. Test performance (95% CI in parentheses) of different combinations of calving indicators as a predictor of parturition within 24, 12, and 6 h1

Calving indicator combinations

Prediction Test VT, RT, VT, RT, VT, RT, VT, LB, RT, LB,
time performance2 LB, LT LB LT LT LT VT, LT VT, RT VT, LB RT, LB RT, LT LB, LT
24 h Se (%) 77 75 77 76 57 76 75 74 75 54 58
(68–85) (66–83) (68–85) (67–83) (47–66) (67–83) (66–82) (65–82) (64–84) (45–64) (49–67)
Sp (%) 77 76 77 76 57 76 75 74 30 54 58
(72–82) (71–81) (72–82) (71–81) (51–62) (71–80) (70–80) (69–79) (24–37) (49–60) (53–64)
+PV (%) 56 54 56 54 33 54 52 51 29 31 34
(48–64) (46–62) (48–64) (46–62) (27–40) (46–61) (44–60) (43–59) (24–37) (25–38) (27–41)
−PV (%) 90 89 90 89 78 89 89 89 74 76 79
(85–93) (85–93) (86–93) (85–93) (72–83) (85–93) (84–92) (84–92) (63–83) (70–81) (73–84)
AUC 0.82 0.81 0.82 0.81 0.61 0.81 0.81 0.80 0.57 0.58 0.60
12 h Se (%) 70 70 70 67 57 69 70 67 55 55 52
(57–81) (57–81) (57–81) (54–79) (43–69) (56–80) (57–81) (54–79) (42–68) (42–68) (39–65)
Sp (%) 72 71 70 67 57 69 70 67 57 55 54
(67–77) (66–76) (65–75) (62–72) (51–62) (69–74) (65–75) (63–72) (52–69) (50–60) (49–59)
+PV (%) 30 29 29 25 18 26 28 25 17 17 15
(22–38) (22–37) (21–37) (21–37) (13–24) (20–34) (21–36) (18–32) (12–23) (12–23) (11–21)
−PV (%) 93 93 93 93 89 93 93 93 89 88 88
(90–96) (90–96) (90–96) (89–95) (84–93) (90–96) (90–96) (89–96) (84–92) (83–92) (83–92)
AUC 0.77 0.77 0.76 0.75 0.62 0.73 0.76 0.75 0.62 0.61 0.56
6h Se (%) 68 71 65 68 71 61 68 68 71 61 58
(49–83) (52–86) (45–81) (40–83) (52–86) (42–78) (49–83) (49–83) (52–86) (42–78) (39–75)
PREDICTION OF CALVING IN DAIRY COWS

Sp (%) 68 71 65 68 71 61 67 70 71 63 61
(63–72) (66–76) (60–70) (63–72) (66–75) (56–66) (62–71) (65–74) (66–75) (58–68) (56–65)
+PV (%) 15 17 13 14 16 11 14 14 16 12 10
(9–21) (11–24) (8–19) (9–20) (10–23) (7–16) (9–20) (9–21) (10–23) (7–18) (6–15)
−PV (%) 96 97 96 97 97 95 96 97 97 95 95
(93–98) (94–99) (93–98) (94–98) (94–99) (92–98) (93–98) (94–98) (94–99) (92–98) (92–97)
AUC 0.78 0.78 0.75 0.71 0.73 0.68 0.75 0.70 0.73 0.69 0.61
1
Calving indicators: VT = vaginal temperature; RT = rumination time; LB = lying bouts; LT = lying time.
2
Test performance: Se = sensitivity; Sp = specificity; +PV = positive predictive value; −PV = negative predictive value; AUC = area under the curve.

Journal of Dairy Science Vol. 99 No. 2, 2016

1547
1548 OUELLET ET AL.

CONCLUSIONS Lombard, J. E., F. B. Garry, S. M. Tomlinson, and L. P. Garber. 2007.

Impacts of dystocia on health and survival of dairy calves. J. Dairy
Sci. 90:1751–1760.
Vaginal temperature, RT, LT, and the number of LB Matsas, D. J., R. L. Nebel, and K. D. Pelzer. 1992. Evaluation of an
showed clear changes associated with the onset of par- on-farm blood progesterone test for predicting the day of parturi-
turition. The technologies used in this study were able tion in cattle. Theriogenology 37:859–868.
Meyer, C. L., P. J. Berger, K. J. Koehler, J. R. Thompson, and C. G.
to capture those changes. Combining these indicators Sattler. 2001. Phenotypic trends in incidence of stillbirth for Hol-
improved the performance to predict calving within stein in the United States. J. Dairy Sci. 84:515–523.
the next 24, 12, and 6 h. Nonetheless, the measured Miedema, H. M., M. S. Cockram, C. M. Dwyer, and A. I. Macrae.
2011a. Behavioural predictors of the start of a normal and dys-
improvement was not sufficient to precisely predict tocic calving in dairy cows and heifers. Appl. Anim. Behav. Sci.
calving. However, these automated devices could be a 132:14–19.
useful tool to assist calving management on commercial Miedema, H. M., M. S. Cockram, C. M. Dwyer, and A. I. Macrae.
2011b. Changes in the behaviour of dairy cows during the 24 h
dairy farms. before normal calving compared with behaviour during late preg-
nancy. Appl. Anim. Behav. Sci. 131:8–14.
O’Driscoll, K., L. Boyle, and A. Hanlon. 2008. A brief note on the
ACKNOWLEDGMENTS validation of a system for recording lying behaviour in dairy cows.
Appl. Anim. Behav. Sci. 111:195–200.
This study was funded through les Fonds de recherche Oltenacu, P. A., A. Frick, and B. Lindhe. 1988. Use of statistical mod-
du Québec-Nature et technologies (FRQNT) (Québec, eling and decision analysis to estimate financial losses due to dys-
tocia and other disease in Swedish cattle. Pages 353–355 in Proc.
QCc, Canada). The first author received a scholarship 5th Int. Symp. Vet. Epidemiol. Econ., Copenhagen, Denmark.
from the Canadian Dairy Commission in collaboration Pahl, C., E. Hartung, A. Grothmann, K. Mahlkow-Nerge, and A.
with Novalait Inc. (Québec, QC, Canada). Apprecia- Haeussermann. 2014. Rumination activity of dairy cows in the 24
hours before and after calving. J. Dairy Sci. 97:6935–6941.
tion is extended to the farm staff and owner of the com- Palombi, C., M. Paolucci, G. Stradaioli, M. Corubolo, P. B. Pascolo,
mercial dairy farm where this experiment took place and M. Monaci. 2013. Evaluation of remote monitoring of parturi-
for their help and hospitality during the project. The tion in dairy cattle as a new tool for calving management. BMC
Vet. Res. 9:191.
authors also want to extend their appreciation to Linda Rajala, P. J., and Y. T. Gröhn. 1998. Effects of dystocia, retained
Saucier for her contribution on this project. placenta, and metritis on milk yield in dairy cows. J. Dairy Sci.
81:3172–3181.
Schirmann, K., N. Chapinal, D. M. Weary, L. Vickers, and M. A.
REFERENCES G. von Keyserlingk. 2013. Short communication: Rumination and
feeding behavior before and after calving in dairy cows. J. Dairy
Berglund, B., J. Philipsson, and O. Danell. 1987. External signs of Sci. 96:7088–7092.
preparation of calving and course of parturition in Swedish dairy Schuenemann, G. M. 2012. Calving Management in Dairy Herds: Tim-
cattle breeds. Anim. Reprod. Sci. 15:61–79. ing of Intervention and Stillbirth. The Ohio State University Ex-
Bewick, V., L. Cheek, and J. Ball. 2004. Statistics review 13: Receiver tension, Columbus.
operating characteristic curves. Crit. Care 8:508–512. Schuenemann, G. M., I. Nieto, S. Bas, K. N. Galvao, and J. Work-
Bikker, J. P., H. van Laar, P. Rump, J. Doorenbos, K. van Meurs, G. man. 2011. Assessment of calving progress and reference times for
M. Griffioen, and J. Dijkstra. 2014. Technical note: Evaluation of obstetric intervention during dystocia in Holstein cows. J. Dairy
an ear-attached movement sensor to record cow feeding behavior Sci. 94:5494–5501.
and activity. J. Dairy Sci. 97:2974–2979. Shah, K. D., T. Nakao, and H. Kubota. 2006. Plasma estrone sul-
Büchel, S., and A. Sundrum. 2014. Short communication: Decrease in phate (E1S) and estradiol-17-β (E2β) profiles during pregnancy
rumination time as an indicator of the onset of calving. J. Dairy and their relationship with the relaxation of sacrosciatic ligament,
Sci. 97:3120–3127. and prediction of calving time in Holstein-Friesian cattle. Anim.
Burfeind, O., V. S. Suthar, R. Voigtsberger, S. Bonk, and W. Heu- Reprod. Sci. 95:38–53.
wieser. 2011. Validity of prepartum changes in vaginal and rec- Sheldon, I. M., J. Cronin, L. Goetze, G. Donofrio, and H.-J. Schuberth.
tal temperature to predict calving in dairy cows. J. Dairy Sci. 2009. Defining postpartum uterine disease and the mechanisms of
94:5053–5061. infection and immunity in the female reproductive tract in cattle.
Dematawewa, C. M. B., and P. J. Berger. 1997. Effect of dystocia on Biol. Reprod. 81:1025–1032.
yield, fertility, and cow losses and an economic evaluation of dys- Streyl, D., C. Sauter-Louis, A. Braunert, D. Lange, F. Weber, and H.
tocia scores for Holsteins. J. Dairy Sci. 80:754–761. Zerbe. 2011. Establishment of a standard operating procedure fort
Fan, J., S. Upadhye, and A. Worster. 2006. Understanding receiver predicting the time of calving in cattle. J. Vet. Sci. 12:177–185.
operating characteristic (ROC) curves. CJEM 8:19–20. Tenhagen, B. A., A. Helmbold, and W. Heuwieser. 2007. Effect of
Huzzey, J. M., and M. A. G. Von Keyserlingk. 2005. Changes in feed- various degrees of dystocia in dairy cattle on calf viability, milk
ing, drinking and standing behavior of dairy cows during the tran- production fertility and culling. J. Vet. Med. A Physiol. Pathol.
sition period. J. Dairy Sci. 88:2454–2461. Clin. Med. 54:98–102.
Jensen, M. B. 2012. Behaviour around the time of calving in dairy Vickers, L. A., O. Burfeind, M. A. G. Von Keyserlingk, D. M. Veira, D.
cows. Appl. Anim. Behav. Sci. 139:195–202. M. Weary, and W. Heuwieser. 2010. Technical note: Comparison of
Kendall, P. E., C. B. Tucker, D. E. Dalley, D. A. Clark, and J. R. rectal and vaginal temperatures in lactating dairy cows. J. Dairy
Webster. 2008. Milking frequency affects the circadian body tem- Sci. 93:5246–5251.
perature rhythm in dairy cows. Livest. Sci. 117:130–138.

Journal of Dairy Science Vol. 99 No. 2, 2016