Preventive Veterinary Medicine 96 (2010) 170–178

Contents lists available at ScienceDirect

Preventive Veterinary Medicine

journa l homepage: www.e lsev ier .com/ locate /prevetmed

Risk factors for abortion in dairy cows from commercial Holstein dairyherds in the Tehran region

Nima Rafati a, Hassan Mehrabani-Yeganeha,∗, Timothy E. Hansonb

a Department of Animal Science, University of Tehran, Karaj, 4111, Iranb Department of Statistics, University of South Carolina, Columbia, SC 29208, United States

a r t i c l e i n f o

Article history:Received 10 July 2008Received in revised form 25 April 2010Accepted 14 May 2010

Keywords:Bovine abortionIranLogistic regressionSurvival analysis

a b s t r a c t

In last decade, pregnancy loss in dairy cattle has had an upward trend bringing difficultiesfor breeders: the annual cost is estimated around 396 billion Rials (i.e. around 40 millionUS$) for the Iranian dairy industry. The present study was conducted to determine theinfluence of maternal factors on abortion and to predict the probability of abortion as wellas the effect of these factors on the fetal lifetime in Holstein dairy cattle. Data from 44,629established pregnancies that included 14,226 heifers and 30,403 pregnancies from 12,265parous cows in nine industrial dairy herds around Tehran were used. Overall, 4871 preg-nancies of parous cows resulted in abortion. Prediction of the probability of abortion (PPA)was estimated by a logistic regression model. Survival analysis was performed using anaccelerated failure time (AFT) model assuming a multi-modal hazard function. Effectivefactors included age of dam at conception, gravidity, open days, number of previous abor-tion(s), abortion before/after 60 days of gestation in previous conception, herd and seasonof insemination. The PPA decreased with increasing open days, increasing gravidity andno previous abortion. In addition, the PPA was greater for cows which had been insemi-nated during summer versus winter. However, the difference between autumn and spring

was not significant. Overall, 25 sires out of 695 from which sperm was collected for arti-ficial insemination (AI) had significantly higher risk of abortion, with odds ratios rangingbetween 1.44 and 4.73 compared to the average. The survival probability increased slightlyduring gestation as gravidity increased for cows that had a previous abortion. Cows thathad aborted before 60 days of gestation in previous conception tended to abort later in their next conceptions.

1. Introduction

Abortion in dairy cows brings about economic, breed-ing and productive damages. The cost of abortion variesaccording to such effective factors as the time of gestation,

milk production, days in milk, the time of inseminationafter parturition, the cost of nutrition, sperm costs, insem-ination time, and labor costs, which differ from regionto region. The calculated economic loss per animal for

∗ Corresponding author. Tel.: +98 261 2248082; fax: +98 261 2246752.E-mail address: hmehrbani@ut.ac.ir (H. Mehrabani-Yeganeh).

0167-5877/$ – see front matter © 2010 Elsevier B.V. All rights reserved.doi:10.1016/j.prevetmed.2010.05.008

© 2010 Elsevier B.V. All rights reserved.

abortion varies between 810,000 and 12,760,000 Rials (i.e.82–1302 US$) in the Tehran region (Samia-Kalantari etal., 2008). Therefore, controlling abortion and preventingthis huge amount of economic loss are vital for breed-ers in Iran. With respect to breeding efficiency, abortioncauses a larger calving interval that hampers achieving the-oretical genetic response. On the other hand, late abortionincreases premature culling and inflicts replacement costs.

The worldwide reported rate of abortion varies from 8%to 14% depending on the stage of gestation when preg-nancies are diagnosed (Thurmond et al., 2005). In Iran,abortion frequency varies. For example, whereas abor-tion frequency in herds with leptosprisis background is

terinary

aa2

actpBrBaabatbd(eniusn1Smctrut1

hi1H2dbcippsItcs

2

2

trfv

N. Rafati et al. / Preventive Ve

round 30% (Hassanpour et al., 2007), the frequency ofbortion from non-infectious causes are between 10% and0% (Samia-Kalantari et al., 2008).

Despite ongoing research investigations to diagnosend evaluate the epidemiology of infectious factors thatause abortion, breeders are still confronted with abor-ion continuously. In Iran, like other countries, importantutative infectious agents that cause abortion includerucella abortus, Listeria monocytogenes, Leptospira inter-ogans (Hassanpour et al., 2007; Sakhaee et al., 2007),ovine virus diarrhea (BVD) (Kargar Moakhar et al., 2001)nd infectious bovine rhinotracheitis (IBR) (Kargar Moakharnd Hemmatzadeh, 2004). Recently, Neospora caninum haseen identified as one of the most important causes ofbortion in Iran (Razmi et al., 2007). However, most ofhese infectious factors have been brought under controly appropriate vaccination. Moreover, successful etiologyiagnosis of abortion shows less than half of the fetal deathGeoffrey et al., 1992; Markusfeld-Nir, 1997; Thurmondt al., 2005). Non-infectious factors include genetic andon-genetic disorders that have been reported in some

nvestigations. Heat stress, production stress, and othernfavorable conditions including seasonal effect and sea-on changes, especially summer are the most importanton-genetic factors (Labernia et al., 1996; Markusfeld-Nir,997; Hansen, 2002; Lopez-Gatious et al., 2002; Bitarafani and Amanloo, 2007). Genetic disorders include chro-osomal and single gene disorders. Contrary to human

hromosomal disorders, which result in about 50% abor-ion, in farm animals the rate is lower, but a sterile calf canesult (Geoffrey et al., 1992). Mutation in codon 405 of theridine monophosphate synthase (UMPS) gene is the puta-ive example of single gene disorders (Fries and Ruvinsky,999).

Different non-infectious maternal and paternal factorsave been reported for fetal death; for example cow par-

ty (Lee and Hwa Kim, 2007), sire effect (Markusfeld-Nir,997), age at conception (Thurmond et al., 1990a, 2005;anson et al., 2003), and abortion history (Hanson et al.,003; Thurmond et al., 2005). We examined these pre-ictors of the probability of abortion (PPA) that can helpreeders to decide if they want to inseminate or cull theirows. Additionally we examined times of elevated risk dur-ng gestation, beneficial for herd management during theseeriods. In an attempt to disseminate this research intoractice, the first and second authors have developed aoftware package and are providing it to herd managers inran. This helps dairy cows to be inseminated according toheir PPA given maternal, paternal (sperm), and seasonalharacteristics. Furthermore, cows receive more supervi-ion during times of elevated hazard, during gestation.

. Materials and methods

.1. Data

The data used to conduct this study were collected overhe past 10 years from industrial herds around the Tehranegion. Tehran, the capital city of Iran, is located at theoot of Alborz mountain range with 1200 m (3900 ft) ele-ation and has mild weather in general but is cooler in the

Medicine 96 (2010) 170–178 171

hilly north. All herds that have been investigated in thisstudy are located in areas with mild weather, are among thelargest herds in Iran, and run with relatively modern man-agement. A private consulting firm under supervision of achief veterinarian provides consultation regarding areas ofmanagement, collects data, and does quality assurance inan advisement capacity.

2.2. Herds and cows studied

Pregnant cows from nine commercial, well-managed(receiving regular veterinary service, heat synchronizationand cow performance monitoring), Holstein dairy herdsin Tehran registered for their artificial insemination (AI)time and parturition/abortion time between 1997 and2006 were used in this investigation. The herd size variesbetween ∼700 and ∼6000 cows and the average milk pro-duction in this data set is approximately 32 kg/day. Theherds are vaccinated for Blackleg, FMD, and Anthrax. Youngstocks are vaccinated by B. abortus strain 19 and recentlyby Rb51. Additionally, herds were tested for Brucellosis ona quarterly basis. The herds were selected randomly frommore than 100 herds which used a computerized regis-tration system and were located in Tehran province. Cowswith at least one previous confirmed pregnancy were con-sidered which included 30,403 pregnancies of 12,265 cows.This restriction is related to factors included in the model,since all of them are measurable when cows have experi-enced at least one pregnancy.

Pregnancy diagnosis was done between 40 and 50 daysand then repeated between 160 and 170 days of gestationby rectal palpation. Only cows that had a non-pregnantinterval between 1 and 12 months were considered for thisstudy, to omit some registration errors.

2.3. Case definition

Pregnancy loss is divided into embryo loss and fetalloss. Fetal loss is considered if any of the following eventsoccurred between 42 and 260 days of gestation (after preg-nancy diagnosis): (1) an expelled fetus or fetal membrane,(2) the cow was observed in estrus and examination ver-ified that there was an abortion, (3) the cow was foundnon-pregnant on a follow-up examination. The reason forconsidering this period is that around 42 days of gestationthe placenta would be completed and fetus is dependenton its mother through placenta.

The model is similar to Hanson et al. (2003), includ-ing maternal factors that are associated with fetal loss:age of dam at conception (AGE), the interval betweencalving/abortion and next conception (NPI), number ofprevious abortion(s) (NPA), the outcome of previous preg-nancy as to whether abortion occurred before/after 60days of gestation (E/L), and the number of previous con-

firmed pregnancies or gravidity (GR). As well, four seasonsof insemination and herd effect were considered as fixedeffects in the logistic model. The covariates AGE, NPI, NPAand GR were centered at zero and standardized to unitvariance.

terinary

172 N. Rafati et al. / Preventive Ve

2.4. Model and statistical analysis

The risk of abortion was estimated through logisticregression incorporating random cow effects. Since seasonof insemination can affect the pregnancy loss, here we addthe season effect too:

logit(Pijk) = ˇ0 + ˇ1NPIijk + ˇ2GRijk + ˇ3(NPI × GR)ijk

+ ˇ4AGEijk + ˇ5NPAijk + ˇ6Eijk + ˇ7Lijk

+ ˇ8(AGE × NPA)ijk + hi + cij + sijk (1)

where Pijk is the probability of abortion during the kth preg-nancy of the jth cow in herd i. hi is a fixed herd effect, cijis a random, mean-zero Gaussian distributed cow effect,and sijk is season of insemination including autumn, winter,spring and summer. The dummy variable Eijk = 1 denotesabortion <60 days in the most recent previous pregnancy,and 0 otherwise; Lijk = 1 denotes abortion >60 days in themost recent previous pregnancy, and 0 otherwise.

The cutpoint of 60 days was chosen according to Hansonet al. (2003) who noted that “An inhospitable uterine envi-ronment, such as that commonly seen in dairy cows withresidual endometritis or incomplete or damaged placenta,is expected to manifest as an initial phase of elevated riskof death between 30 and 60 days.” Finally, the probabilityof abortion with covariate vector x in herd i was estimatedby pi = exp(x′ˇ + hi)/{1 + exp(x′ˇ + hi)} in which the randomcow effect was set to zero.

The fetal lifetime is the time to abortion or birth if noabortion has occurred. Since times of birth are censoredat 260 days of gestation, one can study the distribution ofabortion time. Survival analysis was performed using anaccelerated failure time (AFT) model:

ln(Tijk) = ˛1NPIijk + ˛2GRijk + ˛3Eijk + ˛4Lijk

+ hi + cij + Uijk (2)

where Tijk is the time-to-abortion (days) (TTA) for kth preg-nancy of jth cow in herd i. Uijk ∼ S0(t) is the baseline survivalfunction, which was modeled as a mixture of three normaldistributions. This model allows for multi-modal hazardand density functions, motivated and validated for bovinetime-to-abortion modeling by Hanson et al. (2003) andThurmond et al. (2005):

S0(t) = 1 −3∑

m=1

�m˚(

t − �m

�m

)(3)

where Ф(x) denotes standard normal distribution function,�m and �m (for m = 1, 2, 3) are the mean and stan-dard deviation of normal component m and �m denotesthe probability of fetal death according to componentN(�m, �m). All model variables are assumed independentexcept the ordering �1 < �2 < �3 is enforced for identifia-bility. Prior distributions on the component means were�1 ∼ N(4.1, 100), �2 ∼ N(4.78, 100), and �3 ∼ N(5.36, 100),

centered at the natural logarithm of 60.3, 119, and 212.7days of gestation, and precision (1/�2

m) prior distributionswere Gamma (10−6, 10−6). ˇ0 is the base log-odds of abor-tion for typical cow with average covariates. FollowingThurmond et al. (2005), we centered the prior for ˇ0 at

Medicine 96 (2010) 170–178

−2 to yield a best guess for the overall abortion rate at12% (logit(0.12) ≈ −2). The weights were assumed (�1, �2,�3) ∼ Dirichlet(1, 1, 1). We performed a sensitivity anal-ysis by considering several sets of flat, noninformativepriors in additional model fits; there were no appreciablechanges in regression coefficients, indicating robustness toprior specification. The overall survival function and hazardfunctions based on the logistic regression and AFT modelsare described in Appendix A.

The logistic model was fitted in WINBUGS (Version1.4.2) and inferences based on 25,000 samples from twochains after a 2500 burn-in period. Other logistic regressionanalyses were performed using the LOGISTIC and GENMODprocedures in SAS 9.1. The AFT model was fit in WINBUGSand inferences based on 50,000 samples from one chainafter a 5000 sample burn-in. Convergence of the Markovchain was assessed by trace plots and the Gelman–Rubinconvergence statistic (Spiegelhalter et al., 2007). The cer-tainty of estimated coefficients was assessed by posteriorprobability (POPR). POPR values vary between 0 and 1 andshow the proportion of samples that are positive (coeffi-cient > 0) or negative (coefficient < 0). Receiver operatingcharacteristic (ROC) curves represent the fraction of truepositive rate (TRP) versus the fraction of false positive rate(FPR), useful for assessing the diagnostic accuracy of thelogistic regression.

There are unknown paternal factors that can result inpregnancy loss. Thus, the odds ratios for the bull semenresponsible for 30,403 pregnancies were calculated. Oddsratios of sires versus population mean were calculatedusing the Mantel–Haenszel method by SAS 9.1.

3. Results

Of the 44,629 pregnancies considered, 30,403 pregnan-cies were from 12,265 parous cows of which 4871 resultedin abortions (16.02%). The proportion of abortion amongthe 14,226 heifers was 9.24%. Age of conception (AGE)ranged between 1.5 and 15.1 years, with an average of4.21 years. The median gravidity was 2, varying from 1 to13. The number of cows within gravidity categories 1, 2,3, 4, ≥5 are 9593 (31.55%), 7459 (24.53%), 5308 (17.46%),3450 (11.35%), and 4593 (15.11%), respectively. Open days(NPI) ranged between 30 and 365 days, with a medianof 102 days. The median of NPI in successful pregnancieswas 102 days while in pregnancies that resulted in abor-tion was 97 days. Maximum number of previous abortions(NPA) was 4, and 8.25% of all pregnancies had an abor-tion in the previous conception that included 12% abortion<60 days in gestation (E = 1) and 88% >60 days of gestation(L = 1).

3.1. Prediction of the probability of abortion

The regression coefficient summaries and their POPRare given in Table 1. Herd and seasonal effects were also

significant. Goodness of fit for this model was assessed bythe Hosmer–Lemeshow statistic (P = 0.2939) in SAS PROCLOGISTIC.

The probability of abortion increased in with the cow’sage (POPR = 1), shorter intervals between calving to next

N. Rafati et al. / Preventive Veterinary Medicine 96 (2010) 170–178 173

Table 1Logistic regression model with estimated coefficients’ mean, S.D.a and 95% probability interval (95% PI). The POPRb used in order to show that estimatedcoefficients were <0 or >0 in studied herds of Tehran region.

Coefficient Mean S.D. (95% PI) POPR

2.5% 97.5%

Intercept −1.935 0.1229 −2.179 −1.687 1 (<0)NPIc (days) −0.1534 0.021 −0.1945 −0.112 1 (<0)GRd −0.7564 0.1158 −0.9836 −0.5251 1 (<0)AGE (days) 0.7813 0.1126 0.5555 1.003 1 (>0)NPAe 0.3746 0.0266 0.3226 0.4265 1 (>0)Ef −0.4457 0.1613 −0.7674 −0.1331 0.99 (<0)Lg 0.3901 0.0699 0.2534 0.5275 1 (>0)NPI × GR −0.0335 0.0159 −0.649 −0.0023 0.98 (<0)AGE × NPA −0.0474 0.0136 −0.0744 −0.0209 0.99 (<0)

a S.D.: standard deviation.b POPR: posterior probability, indicates the proportion of Monte Carlo samples which the estimated coefficient was <0 and >0.c NPI: non-pregnant interval.d GR: gravidity.

c>aAioA

(tOd

TAo

e NPA: number of previous abortion(s).f E: aborted ≤60 days of gestation in previous pregnancy.g L: aborted >60 days of gestation in previous pregnancy.

onception (POPR = 1), lower gravidity (POPR = 1), abortion60 days in gestation of previous pregnancy (POPR = 1),nd more abortions in previous pregnancies (POPR = 1).dditionally, the risk of abortion for cows which were

nseminated in summer is significantly higher than forther seasons. The interactions between NPI and GR, andGE and NPA were significant (Table 1).

Fig. 1 gives the receiver operating characteristic curve

ROC) from SAS and WINBUGS analyses, which indicateshe concordance of observed and predicted abortions.verall, PPA ranged between 0.08 and 0.56. The model pre-icted approximately 62% of abortions.

able 2ccelerated fetal time (AFT) estimated coefficients’ mean, S.D.a The POPRb used inf Tehran region.

Coefficient Mean S.D.

NPIc 0.0009 0.004GRd −0.0084 0.004Ee −0.0699 0.042Lf 0.0274 0.013�1

g 0.3274 0.019�2 0.5307 0.026�3 0.1419 0.012�1

i 4.181 0.021�2 4.847 0.021�3 5.355 0.018�1

−2j 32.66 3.618�2

−2 16.67 2.157�3

−2 178.8 51.180

a S.D.: standard deviation.b POPR: posterior probability, indicates the proportion of Monte Carlo samplesc NPI: non-pregnant interval.d GR: gravidity.e E: aborted ≤60 days of gestation in previous pregnancy.f L: aborted >60 days of gestation in previous pregnancy.g �: the probability of fetal death in each normal distribution.h NA: not applicable.i Means of normal distributions.j The inverse of normal distributions variances.

3.2. Survival analysis

The survival model included GR, NPI, E and L effects. NPIwas inconclusive (POPR = 0.59) and GR shows significance(POPR = 0.98) (Table 2).

Previous abortion elevates the hazard of fetal deathduring gestation. However, for a given NPA, fetal sur-vival increased in cows with higher gravidity and those

who experienced abortion before 60 days of gestationin previous pregnancy. However, fetal survival decreasedif previous pregnancy was aborted after 60 days ofgestation.

order to show that estimated coefficients were <0 or >0 in studied herds

(95% PI) POPR

2.5% 97.5%

−0.0070 0.0090 0.59 (>0)−0.0146 −0.0005 0.98 (<0)−0.1527 0.0145 0.94 (<0)0.0010 0.0547 0.97 (>0)0.2882 0.3660 NAh

0.4778 0.5818 NA0.1188 0.1627 NA4.137 4.222 NA4.802 4.886 NA5.317 5.389 NA26.35 40.51 NA13.11 21.51 NA106.6 302.7 NA

which the estimated coefficient was <0 and >0.

174 N. Rafati et al. / Preventive Veterinary Medicine 96 (2010) 170–178

eiver op

Fig. 1. The comparison of two software estimation by rec

Survival probabilities and hazards for a typical cow with120 days open, three previous pregnancies, aged 4.5 yearsat conception, and inseminated in winter, are presented inFig. 2 for different abortion histories. As is shown in Fig. 2,there are three modes during gestation, which have higherrisk of abortion. The hazard reaches its peak on 64 and 122days, and a slight peak on 210 days.

4. Discussion

4.1. Open days

The results show that a higher number of days openlowers the risk of abortion and this is in accordance withother researchers’ findings (Hanson et al., 2003; Thurmondet al., 2005). Uterine involution in cows due to previ-

Table 3Predicted probability of abortion (PPA) and 95% probability interval (95% PI) for cgravidity, season of insemination, number of previous abortions, days open).

NPAa = 0, season = summer AGE = 4.5

Days open GRb = 2PPA (95% PI)

GR = 3PPA (95% PI)

50 0.17 (0.14–0.20) [n = 4]c 0.13 (0.11–0.1150 0.15 (0.13–0.18) [n = 92] 0.12 (0.09–0.1

Season = winter GR = 2PPA (95% PI)

GR = 3PPA (95% PI)

50 0.15 (0.12–0.18) [n = 17] 0.12 (0.1–0.14150 0.13 (0.11–0.16) [n = 220] 0.10 (0.08–0.1

a Number of previous abortion.b Gravidity.c Sample size.

Table 4Odds ratio of abortion, with 95% confidence interval, for different parities, calcula

Parity Odds ratio 95co

3 versus 2 1.052a 0.4 versus 2 1.119b 1.5 versus 2 1.143b 1.

a Non-significant (P < 0.1).b Significant (P < 0.05).

erating characteristic curve (ROC) for WINBUGS and SAS.

ous pregnancy’s outcome (successful parturition, dystocia,retained placenta and abortion) is different. Inhospitableuterine environment seen in dairy cows as a result of com-mon problems such as endometritis and retained placentaincrease the risk of abortion (Hanson et al., 2003). Addition-ally, in the first two to three months of production cows arefaced with negative energy balance, especially those withhigh milk yield (Ghorbani and Asadi-Alamoti, 2004). Theylose weight during this period which can cause a reduc-tion in body condition score with a corresponding negativecorrelation with progesterone secretion, harmful to fetal

survival (Spicer et al., 1990; Grimard et al., 2006). As a casein point, the relative risk of abortion for a cow within a gra-vidity level (GR = 3), age of insemination (AGE = 5.5 year),season of insemination (summer), and days open (NPI = 50versus 150 days) is 1.2. This indicates a 20% increase in

ows in nine herds of Tehran region, with the combination of factors (age,

AGE = 5.5

GR = 3PPA (95% PI)

GR = 4PPA (95% PI)

6) [n = 302] 0.18 (0.14–0.21) [n = 11] 0.14 (0.11–0.18) [n = 180]4) [n = 255] 0.15 (0.13–0.18) [n = 67] 0.12 (0.09–0.14) [n = 167]

GR = 3PPA (95% PI)

GR = 4PPA (95% PI)

) [n = 509] 0.16 (0.13–0.19) [n = 33] 0.12 (0.10–0.14) [n = 31]2) [n = 416] 0.14 (0.11–0.16) [n = 172] 0.10 (0.08–0.12) [n = 261]

ted by the Mantel–Haenszel method.

% Waldnfidence interval

968 1.143021 1.227051 1.242

N.R

afatietal./Preventive

Veterinary

Medicine

96(2010)

170–178175

Fig. 2. Fetal survival and hazard of fetal death for differing numbers of previous abortions (0, 1 and 2) in a typical cow which was inseminated at age of 4.5 years in winter, 120 days open and three previous confirmedpregnancies. Solid lines correspond to cows with NPA = 0. Cows with one previous abortion (NPA = 1) are shown in A and C, and cows with two previous (NPA = 2) abortions are shown in B and D. The dashed line(—) correspond to cows that aborted before 60 days of gestation in previous pregnancy; the dashed-dotted (- · -) correspond to cows that did not abort the previous pregnancy; and dashed-dotted-dotted-dash(- · · -) correspond to cows that aborted after 60 days of gestation in previous pregnancy.

176 N. Rafati et al. / Preventive Veterinary Medicine 96 (2010) 170–178

Table 5The effect of the outcome of previous pregnancy on fetal survival and PPA in 4.5-year-old cow with three previous pregnancy, 120 non-pregnant interval,and inseminated in winter at next conception.

AGE (year) NPIa (days) GRb NPAc Ed Le Season 10th percentile PPAf

4.5 120 3 0 0 0 Winter 229 9.874.5 120 3 1 1 0 Winter 164 12.714.5 120 3 1 0 0 Winter 122 18.524.5 120 3 1 0 1 Winter 103 25.144.5 120 3 2 1 0 Winter 102 23.204.5 120 3 2 0 0 Winter 82 32.054.5 120 3 2 0 1 Winter 74 41.064.5 120 2 1 0 0 Winter 99 25.824.5 120 3 1 0 0 Winter 122 18.524.5 120 4 1 0 0 Winter 179 12.92

a NPI: non-pregnant interval.b GR: gravidity.

c NPA: number of previous abortion(s).d E: aborted ≤60 days of gestation in previous pregnancy.e L: aborted >60 days of gestation in previous pregnancy.f PPA: prediction of the probability of abortion.

the risk of abortion if inseminated shortly after parturition(Table 3).

In the survival analysis, open days did not show anysignificant effect on survival probability, also noted inThurmond et al. (2005).

4.2. Age and parity

In spite of culling cows with high risk of abortion, thePPA increases with age. According to Dairy Herd Improve-ment (DHI) rules, abortions after 152 days of gestationwere advanced to a new lactation number (Thurmond andPicanso, 1990b). The difference between second and thirdlactation was inconclusive; however, in higher lactationsthe risk of abortion is larger (Table 4). Markusfeld-Nir(1997) reported higher pregnancy losses in parous cows,and suggested that varied levels of risk in different paritiesmight reflect different degrees of immunity to the infect-ing agents. The elevated risk of abortion at higher ages incattle is in similar to humans, which is dependent on parity(Skjærven and Klungsøyr, 2007). Although the risk of abor-tion in older cows is higher, the PPA decreased at a given agewith increasing gravidity; see the AGE × GR effect Table 3.

4.3. Gravidity

Studies show that gravidity is an effective factor onpregnancy wastage (Thurmond et al., 1990a, 2005; Hansonet al., 2003). As the gravidity of a cow increases there is ahigher chance of maintaining the fetus until the end of ges-tation. For a given age and number of previous abortionsthe PPA decreased with higher gravidity. For illustration,the PPA of a 4.5-year-old cow with 120 days open, insem-inated in winter, without any abortion in last gestationwas calculated. Her PPA decreased by about 13% as gra-vidity increased from 2 to 4. Additionally, the survivabilityof fetus is augmented in higher gravidity, similar to what

Thurmond et al. (2005). For instance, the 10th percentileof fetal survival rises from 99 to 179 days in gestation byincreasing the gravidity from 2 to 4 (Table 5 – last threerows). Although there is not any known mechanism for thisassociation, cows that have successfully conceived at the

time of pregnancy examination have higher likelihood tomaintain their pregnancy successfully, possibly due to par-tial immunity that cows attain against infectious agents asa result of previous pregnancies (Thurmond et al., 2005).The policy of culling a cow according to her reproductiveperformance can influence the GR effect on pregnancy suc-cess. Breeders try to retain fecund cows and consequentlycows with higher GR can be overly represented.

4.4. Previous abortion(s)

One of the most influential factors that provide theground for miscarriage is the previous abortion(s) that acow has experienced. The next abortion can be due to aninfectious disease. Nesopora caninum, for instance, is oneof the most putative infectious agents that cause habit-ual abortion in cows; repeated abortions associated withprevious abortion are termed a recurrence risk (Hafez andHafez, 2000; Razmi et al., 2007; Youssefi et al., 2009). Habit-ual abortion related to an intrinsic problem of the cow istermed heterogeneous risk (Thurmond et al., 1990a).

The outcome of previous pregnancies is categorizedinto successful pregnancy, abortion <60 days and abortion>60 days of gestation. In contrast with Thurmond et al.(2005), effects due to previous abortion in last pregnancybefore/after 60 days of gestation significantly (POPR = 0.94and POPR = 0.97, respectively) affect TTA. Early abortion canbe a protective mechanism against the infectious agent,providing immunity for next pregnancy. Furthermore,early abortion could be due to false-positive pregnancydiagnosis (Markusfeld-Nir, 1997). The estimated effect for≤60 days is qualitatively different for Tehranian cows,significantly delaying the time to abortion in later preg-nancies; this is also at odds with Thurmond et al. (2005).Moreover, results show that this type of abortion increasesPPA, slightly (Table 5).

Herd 7 among all herds had higher abortion incidence.

Therefore, the history of this herd was studied in greaterdetail. The NPA of this herd was overwhelmingly higherthan other herds (Fig. 3). This can be related to inade-quate reproductive management. As well, milk productionof this herd was higher than average of milk production in

N. Rafati et al. / Preventive Veterinary Medicine 96 (2010) 170–178 177

sidered due to lack of cows representing all NPA ≥ 3 values across herds.

ap

fwdcin

9fa(dtms

4

ttpftqaha(i

stfG

TAi

a

Table 7Number of bulls which had higher significant odds ratio than mean(˛ = 0.05) with the number of insemination and the abortion percentagewith 95% probability interval (95% PI).

Bull (95% PI) #Inseminations Abortion (%) Odds ratio

S1 59 47 4.73 (2.84–7.9)S2 66 36 3 (1.81–4.95)S3 65 32 2.5 (1.49–4.21)S4 139 30 2.27 (1.58–3.26)

Fig. 3. NPA summery in herds. Only NPA = 1 and NPA = 2 were con

ll herds and culling decisions were based mainly on milkroduction rather than reproduction.

In modeling the PPA, the estimated regression effectsor gravidity, age, the number of previous abortions, andhether a previous abortion occurred before or after 60ays were 40–80% greater in magnitude for Tehranianows. Moreover, the age by number of previous abortionsnteraction is significant among Tehranian cows, whereasot for the Californian herds.

Tehranian herds have elevated hazards between 50 and0 days, with a second, less drastic elevation of risk in theollowing 100–150 days. This is consistent with Forar etl. (1996), while Thurmond et al. (2005) and Hanson et al.2003) reported highest risk in second trimester aroundays 130 and 145. This suggests different etiology withinhe two populations, and thus points to different manage-

ent strategies in terms of the timing of more focusedurveillance.

.5. Seasonal effects

A higher frequency of abortion was observed in cowshat were inseminated during summer. In contrast, cowshat were inseminated during winter had the smallestrobability of abortion (Table 6). The odds ratio of abortionor winter to summer was 0.814 (0.742–0.894). However,here was no significant difference between abortion fre-uencies of cows that were inseminated during autumnnd spring. In contrast, Labernia et al. (1996) reportedigher rate of abortion for cows that were inseminated inutumn. It is worth mentioning that Lopez-Gatius et al.2004) did not report any significant effect of season ofnsemination on pregnancy loss.

In summer the risk of abortion is higher than other sea-

ons due mainly to heat stress. Other reasons that elevatehe probability of abortion in this season include glandularunction, ratio and infectious agent (Hafez and Hafez, 2000;horbani and Asadi-Alamoti, 2004).

able 6bortion proportion in different seasons of insemination and the compar-

son of them.

Winter Spring Summer Autumn

Abortion (%) 14.35a,b,c 16.72a 17.06a,b 16.36a

,b,cMeans with different superscripts differ (P < 0.05).

S5 64 30 2.21 (1.29–3.79)

4.6. The sire effect

Semen from 695 bulls were responsible for the 30,403pregnancies; the average abortion proportion was 16.02%.Of these, 83 bulls had higher odds ratio than aver-age and only 25 of them showed significance (˛ = 0.05)(Table 7). Likewise, Markusfeld-Nir (1997) reported signifi-cantly higher than average frequency of abortion by semenfrom certain bulls.

5. Conclusion

Repeated measures methodology of Thurmond et al.(2005) was applied to Tehranian herds, modeling both thePPA and the TTA given various maternal risk factors, withsire and seasonal effects included in the model, which werefound to significantly affect abortion. The strength of theeffects and their significance are different across Tehranianand Californian herds, with more pronounced effects, andgreater predictive ability seen in the Tehranian population.The shapes of the abortion hazards are also quite different.The highest risk of abortion befell Tehranian cows in theirfirst trimester; thus, Tehranian dairy herd managers canallocate more resources towards surveillance during theinitial 1–2 months of pregnancy. Since the risk of abortionis higher in the first trimester, we suggest using other, morerecent pregnancy diagnosis methods. In addition, consid-ering the herd effect is useful for ranking herds accordingto their reproduction performance and elucidating the role

of management in controlling the malfunction of repro-duction. Also, for Tehranian herds to have a successfulpregnancy we suggest longer open days.

terinary

178 N. Rafati et al. / Preventive Ve

Acknowledgements

The authors wish to thank Dr. Baniasadi for generouslyproviding the data used in this study, to Dr. Eskandari fromthe Department of Statistics, Allameh Tabatabaie Univer-sity, Tehran, Iran, for providing some helpful commentsand directions. Furthermore, authors thank two anony-mous referees for their helpful notes to improve the paperon earlier version.

Appendix A.

The general fetal lifetime distribution model is given by:

fL(t|�, x) = p(A|�, x)fA(t|�, x) + {1 − p(A|�, x)}fD(t|�, x) (A.1)

where p(A|�, x) is the probability of abortion which is cal-culated using logistic regression model and fA and fD arefetal lifetime density functions referred to aborted fetusesand to surviving fetuses, respectively. The covariate vectorx and variable vector � are not necessarily the same vectorsin each of the two model components. The survival functionis

SL(t|�, x) = p(A|�, x)SA(t|�, x) + {1 − p(A|�, x)}SD(t|�, x) (A.2)

where SA and SD are survival functions associated with fAand fD. For pregnancies that did not result in abortion by260 days, SD(t|�, x) = 1. Therefore, the survival and hazardfunctions for the 260-day fetal period are as follows:

SL(t|�, x) = p(A|�, x)SA(t|�, x) + {1 − p(A|�, x)} (A.3)

and

hL(t|�, x) = p(A|�, x)fA(t|�, x)SL(t|�, x)

(A.4)

Appendix B. Supplementary data

Supplementary data associated with this article can befound, in the online version, at doi:10.1016/j.prevetmed.2010.05.008.

References

Bitaraf Sani, M., Amanloo, H., 2007. Heat stress effect on open days inHolstein dairy cattle in Yazd province, Iran. In: 3rd Congress of AnimalScience, Mashhad, Iran.

Forar, A.L., Gay, J.M., Hancock, D.D., Gay, C.C., 1996. Fetal loss frequency inten Holstein dairy herds. Theriogenology 45, 1505–1513.

Fries, R., Ruvinsky, A., 1999. The Genetics of Cattle, first ed. CABI Publishing,New York, pp. 59–60.

Geoffrey, H.A., Noakes, D.E., Pearson, H., Parkinson, T.J., 1992. VeterinaryReproduction & Obstetrics, sixth ed. Saunders Publishing.

Ghorbani, G., Asadi-Alamoti, A., 2004. Developed Management of DairyCattle, first ed. Industrial Unit of Jahad, Isfahan, Iran, pp. 92–100.

Grimard, B., Freret, S., Chevallier, A., Pinto, A., Ponsart, C., Humblot, P., 2006.Genetic and environmental factors influencing first service conception

Medicine 96 (2010) 170–178

rate and late embryonic/foetal mortality in low fertility dairy herds.Anim. Rep. Sci. 91 (1–2), 31–44.

Hafez, E.S.E., Hafez, B., 2000. Reproduction in Farm Animals, seventh ed.Wiley Blackwell.

Hansen, P.J., 2002. Embryonic mortality in cattle from the embryo’sprospective. J. Anim. Sci. 80 (Suppl. 2), E33–E44.

Hanson, T., Bedrick, E.J., Johnson, W.O., Thurmond, M.C., 2003. A mix-ture model for bovine abortion and foetal survival. Stat. Med. 22,1725–1739.

Hassanpour, A., Fartashvand, M., Abdollahpour, Gh., Mogaddam, Gh.,Nadalian, A., Sattari, Gh.S., 2007. Seroprevalence of leptospiral infec-tion in dairy herds in Tabriz – Iran. Pajohesh-va-Sazandegi 74,67–77.

Kargar Moakhar, R., Bokaie, S., Akhavizadegan, M.A., Charkhkar, S.,Meshkot, M., 2001. Seroepidemiological survey for antibodies againstinfectious bovine rhinotracheitis and bovine herpes 4 viruses amongcattle in different provinces of Iran. J. Razi Inst. 52, 93–100.

Kargar Moakhar, R., Hemmatzadeh, F., 2004. A survey for detectingpestivirus antigen in persistently infected cattle around Tehran. J.Pajohesh-va-Sazandegi 63, 21–25.

Labernia, J., Lopez-Gatius, F., Santolaria, P., Lopez-Bejar, M., Rutllant, J.,1996. Influence of management factors on pregnancy attrition in dairycattle. Theriogenology 45, 1247–1253.

Lee, J., Hwa Kim, I., 2007. Pregnancy loss in dairy cows: the contributingfactors, effect on reproductive performance and the economic impact.J. Vet. Sci. 8 (3), 283–288.

Lopez-Gatious, F., Santolaria, P., Yaniz, J.L., lopez-Bezar, M., 2002. Factorsaffecting pregnancy loss from gestation day 38 to 90 in lactating dairycows from a single herd. Theriogenology 57, 1251–1261.

Lopez-Gatius, F., Pabon, M., Almeria, S., 2004. Neospora caninum infectiondoes not affect early pregnancy in dairy cattle. Theriogenology 62,606–613.

Markusfeld-Nir, O., 1997. Epidemiology of bovine abortions in Israeli dairyherds. Prev. Vet. Med. 31, 245–255.

Razmi, G.R., Maleki, M., Farzaneh, N., Talebkhan Garoussi, M., Fallah, A.H.,2007. First report of Neospora caninum-associated bovine abortion inMashhad area, Iran. Parasitol. Res., 755–757.

Sakhaee, E., Abdollahpour, Gh.R., Bolourchi, M., Hasani Tabatabayi, A.M.,Sattari Tabrizi, S., 2007. Serologic and bacteriologic diagnosis of bovineleptospirosis in Tehran suburb dairy farms. Iran. J. Vet. Res. Univ. Shi-raz 8 (4), 325–332.

Samia-Kalantari, A., Mehrabani-Yeganeh, H., Moradi, M., 2008. Evaluationof Economic Losses Due to Abortion and Decision Making Processfor Culling Aborted Cow. Animal Science Department, University ofTehran, Iran.

Skjærven, R., Klungsøyr, K., 2007. Selective fertility the examplesof perinatal death and preeclampsia. Norsk Epidemiol. 17 (2),175–180.

Spicer, L.J., Tucker, W.B., Adams, G.D., 1990. Insulin-like growth factor-Iin dairy cows: relationships among energy balance, body condi-tion score, ovarian activity, and estrous behavior. J. Dairy Sci. 73,929–937.

Spiegelhalter, D., Thomas, A., Best, N., Gilks, W., 2007. WINBUGS: win-dows Bayesian inference using Gibbs sampling, version 1.4.2. MRCBiostatistics Unit, Cambridge. Available at: www.mrc-bsu.cam.ac.uk/bugs/winbugs/.

Thurmond, M.C., Picanso, J., Jameson, C., 1990a. Considerations for use ofdescriptive epidemiology to investigate foetal loss in dairy cows. J.Am. Vet. Med. Assoc. 197, 1305–1312.

Thurmond, M.C., Picanso, J., 1990b. A surveillance system for bovine abor-tion. Prev. Vet. Med. 8, 41–53.

Thurmond, M.C., Branscum, A.J., Johnson, W.O., Bedrick, E.J., Hanson, T.E.,

2005. Predicting the probability of abortion in dairy cows: a hierarchi-cal Bayesian logistic-survival model using sequential pregnancy data.Prev. Vet. Med. 68, 223–239.

Youssefi, M.R., Arabkhazaeli, F., Tabar Molla Hassan, A., 2009. Seropreva-lence of Neospora caninum infection in rural and industrial cattle inNorthern Iran. Iran. J. Parasitol. 4 (1), 15–18.