Accounting for pregnancy diagnosis in predicting days open.

Abstract

The system for estimating days open for cows with no subsequent lactation was examined to determine if estimates should vary depending on pregnancy diagnosis. Pregnancy diagnosis information was unavailable when the original prediction system was developed, but collection was begun in 2002. New prediction equations were estimated from nearly 1.1 million Holstein lactations for 20-d intervals from 110 to 250 days in milk (DIM). Use of pregnancy diagnosis improved accuracy compared with the original system. The improvement was particularly evident for lactations of cows confirmed to be open in the 130-to-149 DIM interval, where predicted days open increased by > 96 d. For lactations of cows with a confirmed pregnancy, predicted days open decreased by 18 d for the same interval. Prediction errors decreased with increasing DIM. Jersey lactations averaged fewer days open, but in most cases Holstein solutions provided adequate predictions. Specific adjustments were generated for Jersey lactations with no breedings reported. Those adjustments reduced the predicted days open averaged across parity by an amount that increased from 9 to 27 d with DIM interval. The new prediction equations were implemented for November 2004 evaluations for daughter pregnancy rate.

Full text

Page 1

J. Dairy Sci. 88:1873–1877
 American Dairy Science Association, 2005.
Accounting for Pregnancy Diagnosis in Predicting Days Open
G. R. Wiggans and R. C. Goodling Jr.*
Animal Improvement Programs Laboratory, Agricultural Research Service,
USDA, Beltsville, MD 20705-2350
ABSTRACT
The system for estimating days open for cows with
no subsequent lactation was examined to determine if
estimates should vary depending on pregnancy diagno-
sis. Pregnancy diagnosis information was unavailable
when the original prediction system was developed, but
collection was begun in 2002. New prediction equations
were estimated from nearly 1.1 million Holstein lacta-
tions for 20-d intervals from 110 to 250 days in milk
(DIM). Use of pregnancy diagnosis improved accuracy
compared with the original system. The improvement
was particularly evident for lactations of cows con-
firmedto beopen inthe 130-to-149DIMinterval, where
predicted days open increased by >96 d. For lactations
of cows with a confirmed pregnancy, predicted days
opendecreasedby18dforthesameinterval.Prediction
errorsdecreasedwithincreasingDIM.Jerseylactations
averaged fewer days open, but in most cases Holstein
solutions provided adequate predictions. Specific ad-
justments were generated for Jersey lactations with
no breedings reported. Those adjustments reduced the
predicted days open averaged across parity by an
amountthatincreasedfrom9to27dwithDIMinterval.
The new prediction equations were implemented for
November 2004 evaluations for daughter pregnancy
rate.
(Key words: pregnancy confirmation, days open)
Abbreviation key: DO = days open, DPR = daughter
pregnancy rate, PD = pregnancy diagnosis.
INTRODUCTION
In February 2003, the Animal Improvement Pro-
grams Laboratory implemented an evaluation for fe-
male fertility called daughter pregnancy rate (DPR;
VanRaden etal., 2004). Theevaluation is basedon days
open (DO) and includes a system for estimating DO
Received November 4, 2004.
Accepted January 18, 2005.
Corresponding author: G. R. Wiggans; e-mail: wiggans@aipl.
arsusda.gov.
*Current address: Penn State Cooperative Extension, 2120 corn-
wall Road, Lebanon, PA 17042 (rcg133@psu.edu).
1873
developed by Kuhn et al. (2004), which allows inclusion
of records before DO can be confirmed by a subsequent
calving. Data available for developing that system did
not include pregnancy diagnosis (PD) because collec-
tion of that information did not start until 2002. With
more than 2 yr of data now available, predictions could
be developed specifically for cows confirmed to be preg-
nant or to be open.
Pregnancy diagnosis is a common management prac-
tice. Fricke (2002) reports that ultrasound imaging can
provide accurate information as early as 30 d after in-
semination.Historically,rectalpalpationhasbeenused
at ≥45 d.
For cows confirmed to be pregnant, actual DO may
be greater than DO at last breeding because the cow
became pregnant from a later unreported breeding, the
PD was wrong, or the cow aborted after PD. Cows con-
firmed to be open are expected to have longer DO than
cows with the same DIM at last breeding and no con-
firmationbecausemanyunconfirmedcowsmaybepreg-
nant, but few of the cows confirmed to be open are
expected to be pregnant. A cow bred after having been
confirmed to be open would revert to unknown preg-
nancy status. The purpose of this study was to deter-
mine if information on PD improves prediction of DO
and if so, modify the prediction of DO to use PD.
MATERIALS AND METHODS
Data
Breeding records supplied as part of the lactation
recordoraspartofarecentimplementationofcollection
of reproduction information were extracted for parities
1 through 5 for calvings from October 2001 through
March 2003. The upper limit on date was chosen to
allow for a subsequent calving to verify DO. Lactations
designated as “do not breed” were excluded because
such cows do not provide useful information for predic-
tion of DO, particularly if the designation is early in
lactation. Because of national reporting of pregnancy
confirmation,collectioncouldberestrictedtoherdswith
a high level of reporting. To be included, herds were
required to have a test on or after October 1, 2001, ≥365
d between the first and last test, ≥7 tests during the
365 d following the first test. To eliminate small herds,

Page 2

WIGGANS AND GOODLING JR.
1874
records from a herd were used starting with the first
test date for which a herd had ≥50 cows in milk.
The DPR evaluation imposes a ceiling of 250 d on
DOtolimittheeffect oftheskewnessofthedistribution
(VanRaden et al., 2004). For a lactation to be used in
the study, there must have been a subsequent lactation
to verify DO or evidence that DO was ≥250. Such evi-
dence included being culled for reproductive reasons or
a breeding or confirmed-open diagnosis at ≥250 DIM.
Estimated breeding date was calculated as subsequent
calving date minus gestation interval (290 d for Brown
Swiss and 280 d for other breeds). Breedings where
the date was >18 d after estimated breeding date were
excluded to eliminate breedings to pregnant cows.
Herds were eliminated if less than 50% of cows had
a PD reported. Requiring herds to have a high level of
reporting of PD ensured exclusion of herds for which
only problem breeders were checked. Herds with <10%
or >75% of breedings that resulted in conception were
excluded to eliminate herds with selective reporting.
Seventy-fivepercentofthecowsinaherdwererequired
to have a breeding reported. Records for 1,095,629 Hol-
stein and Red-and-White and 76,802 Jersey lactations
were included in the analysis after imposing the edits.
Model
The model of Kuhn et al. (2004) was used:
y = intercept + parity + CE + b1age + b2age2
+ b3DOL+ e,
where y = DO (breeding date − calving date), CE =
calving ease score (1 through 5), age = calving age in
years (e.g., 2.5 yr), DOL= DO at last breeding before
the end of the interval (may be a breeding in a prior
interval; the term was dropped from model if the cow
had not been breed yet), b = regression coefficient for
effect, and e = residual. That model, in full or part, was
applied to 56 data sets, which were defined by seven
20-d intervals starting at 110 DIM, the presence or
absence of calving ease information, and 4 classes for
breeding and PD information (no breeding, pregnancy
status unknown, confirmed to be pregnant, and con-
firmed to be open). A particular breeding with a diagno-
sis contributed to the unknown-status group until 45
d after breeding when the diagnosis was assumed to
have occurred. The end of the DIM interval was used
forthis determination.The actualdate ofdiagnosis was
missing for a majority of data.
Cows with subsequent heats reported were included
in the group diagnosed to be open for that breeding.
When more than one PD or indication followed a breed-
ing, the last one was used. In addition to analysis of
Journal of Dairy Science Vol. 88, No. 5, 2005
Holsteindata,applicabilityofHolsteinresultstoJersey
data was investigated. Holstein results were assumed
to apply to other breeds because mean DO values were
similar to Holstein values.
To determine the characteristics of PD reporting, a
separate study of data supplied by AgSource (Verona,
WI) was conducted. AgSource was the only center that
reportedthedateofdiagnosis.Forcowscalvingbetween
December 2003 and December 2004, with diagnosis be-
tweenMayandDecember2004,themeantimebetween
breeding and a diagnosis of pregnant was 45 d, and
where a diagnosis of not pregnant occurred, 40 d. Of
the 122,974 diagnoses, 39% were not pregnant and 61%
were pregnant. Data from Dairy Records Management
Systems (Raleigh, NC) had only 15% not pregnant of
140,865 diagnoses from the same period. This lower
value probably results from only the last diagnosis in
atestintervalbeingreportedwithanapproximatedate.
The other processing centers did not report the date
of diagnosis.
Genetic Correlations
Genetic correlations were estimated among predic-
tions for 7 DIM intervals and actual DO using REML
andasiremodelasinKuhnetal.(2004).Themodelwas
y = hysp + s + e
where y = vector of 7 predicted DO and actual DO;
hysp = herd-year-season-parity effect, with seasons
starting in January, March, June, September, and No-
vember; s = sire effect; and e = residual. Relationships
through sire and maternal grandsire were considered.
Comparison with Original Prediction
Original prediction equations (Kuhn et al., 2004)
were applied to data used for estimating regressions.
Prediction errors and standard deviations were calcu-
lated. The same values also were calculated for the new
prediction equations. Because the same data were used
for the estimation, the prediction errors for the new
prediction equations were expected to average 0.
RESULTS AND DISCUSSION
Regressions
Eight sets of regression equations were estimated for
each of the 7 DIM intervals. Results are reported for
only the second and last intervals. Results for intervals
not displayed followed the trend established by this
range. The first interval was not displayed because re-
sults from that interval were not implemented. Mean

Page 3

PREGNANCY DIAGNOSIS IN PREDICTING DAYS OPEN
1875
Table 1. Number of lactations by category of information, mean prediction errors (predicted − actual) of
previous prediction equation, and SD of prediction error for 2 DIM intervals.
Information
SD of prediction error
Mean
prediction
error
Calving
ease
Lactations
(no.)DIMBreedingDiagnosis NewPreviousCorrelation
130 to 149No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
Pregnant
Pregnant
Open
Open
No
No
No
No
Pregnant
Pregnant
Open
Open
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
46,037
15,411
393,659
223,218
235,574
151,211
19,405
11,114
9201
2812
204,878
107,624
455,757
277,433
24,839
13,085
4.2
1.8
43.0
39.6
55.9
55.0
27.7
26.2
59.5
55.8
63.3
48.1
32.3
29.0
18.2
16.3
54.6
52.1
43.3
40.0
56.5
55.5
28.0
26.6
60.6
57.4
64.6
49.1
32.6
29.3
18.3
16.4
59.1
57.7
0.12
0.15
0.39
0.41
0.50
0.51
0.15
0.14
0.20
0.20
0.90
0.92
0.92
0.93
0.41
0.41
−17.2
−17.1
18.4
17.2
−96.4
−99.7
32.8
16.3
−9.3
−7.8
3.9
4.0
−79.2
−80.3
230 to 249
prediction errors and standard deviations are in Table
1 and solutions are in Tables 2 and 3. The percentage
of lactations without a breeding decreased from 5.6 to
1.1overthe 100dbetweenreported periods;percentage
of lactations with PD, both pregnant and open, in-
creased from 38.1 to 70.4. The percentage of PD that
were not pregnant was lower than found in recent data.
This may reflect that as DIM increases, a cow is rebred
and then may be diagnosed pregnant. A cow may also
be diagnosed not pregnant because the diagnosis was
too early to detect the pregnancy. In this study, the
last diagnosis in an interval was used. All 1,095,629
Holstein and Red-and-White lactations contributed to
each interval, because each interval used all data avail-
able during and before that interval.
Table 2. Coefficients for prediction of days open for 2 DIM intervals and the 8 combinations of information.
Information
Parity
Calving
ease
Days
open1
DIM BreedingDiagnosis12345Age Age2
130 to 149 No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
Pregnant
Pregnant
Open
Open
No
No
No
No
Pregnant
Pregnant
Open
Open
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
195.7
196.4
32.5
40.6
3.8
5.6
151.6
171.2
166.1
200.7
3.5
1.0
1.0
2.5
142.8
154.0
189.7
189.0
17.1
28.5
2.2
5.3
144.3
167.8
160.1
202.0
−5.8
−7.0
−0.3
2.0
142.3
152.2
183.2
183.4
−1.4
10.0
−0.6
3.4
135.2
162.3
155.1
195.5
−14.3
−15.5
−1.6
1.0
137.8
151.1
176.9
179.7
−21.2
−9.4
−4.1
0.6
123.8
154.9
149.8
198.0
−23.0
−23.4
−3.3
−0.1
132.0
148.8
167.8
173.5
−42.2
−30.8
−8.0
−2.7
113.0
145.1
137.9
194.8
−30.7
−32.0
−4.9
−1.9
127.1
147.0
5.28
11.41
10.49
7.25
0.45
−0.71
2.18
0.38
18.27
15.88
8.16
8.36
0.69
−0.22
−0.56
3.45
0.36
−0.39
0.91
1.21
0.31
0.35
0.92
0.83
−0.63
−0.99
0.02
−0.06
0.10
0.16
0.70
0.03
0.90
0.96
0.98
0.99
0.49
0.42
230 to 249
0.97
0.99
1.00
1.00
0.54
0.50
1Days open at last breeding.
Journal of Dairy Science Vol. 88, No. 5, 2005
Mean prediction errors in Table 1 resulted from
applying the equations of Kuhn et al. (2004). Largest
mean errors were for the confirmed-open groups in the
130-to-149 DIM interval. The correlations also show
the value of a successful last breeding in predicting
DO. Underprediction of DO was >96 d. As expected,
confirmed-pregnantgroupshadpredictedDOthatwere
too high by about 18 d. Records for cows with unknown
pregnancy status (those records that remained after
removing records of cows confirmed to be pregnant and
the relatively few cows confirmed to be open) had a
corresponding underprediction. Those mean prediction
errors tended to be smaller in the last DIM group. As
expected, mean prediction errors were small for lacta-
tions without breedings in the early interval, which

Page 4

WIGGANS AND GOODLING JR.
1876
Table 3. Adjustments to prediction of days open for calving ease score differences from 5 by calving ease
score for 2 DIM intervals and 4 combinations of information and predicted days open for calving ease score
5 with last breeding at 100 DIM in parity 5 at 6.3 yr of age.
Information Calving ease score
Predicted
days openDIMBreedingDiagnosis1234
130 to 149No
Yes
Yes
Yes
No
Yes
Yes
Yes
No
No
Pregnant
Open
No
No
Pregnant
Open
−10.2
−10.7
−1.1
−9.6
−14.4
−2.8
−0.9
−11.0
−7.9
−8.1
−0.7
−7.4
−7.8
−1.9
−0.7
−8.6
−6.2
−8.1
−1.2
−3.0
−9.6
−2.4
−0.8
−1.9
−0.4
−2.8
−0.8
−0.6
3.3
−1.0
−0.5
−2.2
230.2
169.2
116.2
227.0
255.8
128.1
113.1
224.7
230 to 249
indicatedthatthenewpredictionsweresimilartothose
of Kuhn et al. (2004). Mean prediction errors for lacta-
tions of cows with unreported breedings were greater
in the last interval, but the counts were much smaller.
The standard deviations of prediction errors were
generallysmallerforthenewpredictions,butthatcould
be because the same data were used to estimate the
predictions as were used for the comparison. As with
the mean prediction errors, the greatest benefit was for
cows confirmed tobe open. Even for the230-to-249 DIM
group, DO of the last breeding was not final DO for
3.5%ofthelactations,which explainswhythestandard
deviation of prediction errors was not zero. Fifteen per-
cent of cows confirmed to be open were actually preg-
nant from their last breeding, and 52% of the lactations
of cows with unknown pregnancy status had final DO
different from DO reported at last breeding.
The solutions in Table 2 show that the values are
similar for equations with and without calving ease
data. The regression coefficients on DO for the con-
firmed pregnant and unknown pregnancy status equa-
tions are >0.9, indicating that DO at last breeding com-
prises most of the estimate. For the confirmed open
Table 4. Predicted days open without calving ease scores for last breeding at 110 DIM for parities 1 and
5 with and without breeding and confirmation data by DIM.
Days open
Age
(yr)
No
breeding
No
confirmation
Confirmed
pregnant
Confirmed
open ParityDIM
1 2.0140
160
180
200
220
240
140
160
180
200
220
240
207.7
215.5
219.2
218.4
212.8
200.1
215.7
223.0
227.8
230.1
230.1
228.5
156.2
151.9
145.7
138.9
132.7
127.1
159.9
154.9
148.2
141.1
134.5
129.2
114.1
114.0
113.8
113.5
113.0
112.7
115.5
115.5
114.9
114.5
113.9
113.2
213.6
215.5
215.1
212.6
208.6
203.7
217.9
220.2
220.4
218.0
215.0
210.8
56.3
Journal of Dairy Science Vol. 88, No. 5, 2005
equations, the coefficients are around 0.5, indicating
that failed breedings provide some information on the
eventual DO. Table 3 gives the solutions for calving
ease and generally shows that predicted DO increases
with calving ease score, particularly at early DIM when
there is no PD. However, at late DIM, predicted DO
changes little with changes in calving ease score when
PD is not known. This indicates an interaction between
CE scores and the 2 information categories for no PD
and pregnant. The predicted DO for calving ease score
5, parity 5, and breeding at 110 DIM also is given to
permit comparison with results in Table 4.
To illustrate the effect of PD information on predic-
tions, Table 4 shows predictions for 110 DIM at last
breeding and 2 ages for the 4 information categories
for breeding and pregnancy confirmation. Without a
breeding, predicted DO was >200 d. With a breeding,
predicted DO decreased toward the last breeding date
at 110-DIM with increasing DIM. For cows confirmed
to be pregnant, predicted DO was never >115.5 d. The
benefit of the pregnancy confirmation compared with
an unknown pregnancy status declined with increasing
DIM from >40 d to 15 d. Predicted DO for cows con-

Page 5

PREGNANCY DIAGNOSIS IN PREDICTING DAYS OPEN
1877
Table 5. Reduction in predicted days open for Jersey cows without
a breeding by parity for 2 DIM intervals.
Parity
DIM12345
130 to 149
230 to 249
8 11
34
10
32
98
2527 17
firmed open also declined with increasing DIM. That
decline appeared to reflect the abnormal situation pre-
sented in the example where an open cow is not rebred
and not designated as “do not breed.” The comparison
of parities 1 and 5 showed slightly higher predictions
for the later lactation, particularly when no breeding
wasreportedorthebreedingwasconfirmedtobeunsuc-
cessful.
Application to Jersey Data
Solutions estimated from Holstein data were applied
to Jersey data and were found to have similar accuracy
and prediction error except for records without breed-
ings. For that group, adjustments by parity (1 to 5) and
DIM interval (6 intervals starting at 130 DIM) were
calculated (Table 5). When averaged across parity,
mean prediction error increased from 9 to 27 d with
the DIM intervals. The values in Table 5 are applied
to reduce predicted DO.
Genetic Correlations
Correlations and heritabilities are in Table 6 for the
7 DIM intervals and final DO. The results are nearly
the same as those of Kuhn et al. (2004). The heritabilit-
iesareslightlylower,whichpossiblyreflectstheshorter
period included in the data.
Table 6. Heritability of predicted days open and its correlations with
days open from completed lactations by DIM.
Correlations
DIM
(20-d interval)PhenotypicGenotypicHeritability1
110
130
150
170
190
210
230
Completed
0.615
0.709
0.778
0.828
0.866
0.895
0.918
1.000
0.964
0.979
0.986
0.990
0.995
0.997
0.997
1.000
0.0326
0.0318
0.0333
0.0337
0.0343
0.0342
0.0346
0.0364
1Standard errors are approximately 0.0029.
Journal of Dairy Science Vol. 88, No. 5, 2005
CONCLUSIONS
InformationfromPDimprovestheaccuracyofpredic-
tion of DO. The largest improvement was for cows diag-
nosed to be open where DO was previously underpre-
dicted by >96 d for the 130-to-149 DIM interval. A
smaller improvement was observed for the much larger
number of cows confirmed to be pregnant. The 3.5% of
cows for which the confirmed last breeding was not the
final DO demonstrates the value of applying a predic-
tion process instead of equating a pregnancy confirma-
tion to having a subsequent calving. Although coeffi-
cients for prediction equations were estimated for the
110-to-129 DIM interval, there was not a sufficient im-
provement in accuracy to lower the threshold for pre-
dicting DO to include them; therefore, the 130-d re-
quirement established by Kuhn et al. (2004) was re-
tained. The prediction system developed in this study
was implemented for the November 2004 DPR evalua-
tion. Expansion factors adjust the variance of incom-
plete records to meet the expectations of the model,
whichassumes thatsuch recordshave thesame genetic
variance but more error variance than completed re-
cords. Because incorporation of PD into the prediction
system increased the variance of the DO in early lacta-
tion, the expansion factors in the evaluation system
were reduced proportionally. Records are weighted to
reflect their accuracy. Although, as shown in Table 1,
accuracy in predicting DO differs by PD and presence
of a breeding, these factors were not considered in the
weights because they are correlated with the value of
DO. At a given DIM, open cows have higher DO than
confirmed pregnant cows. Weights were based only on
DIM for simplicity and to avoid bias.
ACKNOWLEDGMENTS
TheauthorsthankM.T.Kuhn,AnimalImprovement
Programs Laboratory (Beltsville, MD), for his advice
on the project and review of the manuscript; L. L. M.
Thornton, Animal Improvement Programs Laboratory,
for assistance in manuscript preparation; and G. E.
Shook, University of Wisconsin, and 2 anonymous re-
viewers for suggestions on the manuscript.
REFERENCES
Fricke, P.M. 2002. Scanning thefuture − ultrasonographyas a repro-
ductive management tool for dairy cattle. J. Dairy Sci.
85:1918–1926.
Kuhn, M. T., P. M. VanRaden, and J. L. Hutchison. 2004. Use of
early lactation days open records for genetic evaluation of cow
fertility. J. Dairy Sci. 87:2277–2284.
VanRaden, P. M., A. H. Sanders, M. E. Tooker, R. H. Miller, H. D.
Norman, and G. R. Wiggans . 2004. Development of a national
genetic evaluation for cow fertility. J. Dairy Sci. 87:2285–2292.