Livestock Science 189 (2016) 8–16

Contents lists available at ScienceDirect

Livestock Science
journal homepage: www.elsevier.com/locate/livsci

Causal relationships between clinical mastitis events, milk yields and

lactation persistency in US Holsteins
K. Dhakal a,n, F. Tiezzi b, J.S. Clay c, C. Maltecca b
a
Bayer, Crop Science Division, 407 Davis Drive, Morrisville, NC 27560, USA
b
Department of Animal Science, North Carolina State University, Raleigh, NC 27695, USA
c
Dairy Records Management Systems, Raleigh, NC 27603, USA

art ic l e i nf o a b s t r a c t

Article history: Complex relationships exist between udder susceptibility to mastitis and milk production traits. Iden-
Received 17 August 2015 tifying causal association between these traits could help to disentangle these complex relationships. The
Received in revised form main objective of the study was to use producer-recorded health data to examine the causal relationship
11 April 2016
between mastitis events, milk yield and lactation persistency. A total of 48,058 first lactation cows,
Accepted 22 April 2016
daughters of 2213 Holstein bulls and raised across 207 herds were analyzed using structural equation
models. Traits included in the dataset were mastitis events and average test day milk yields recorded in
Keywords: three different periods: period 1 (5–60 DIM), period 2 (61–120 DIM) and period 3 (121–180 DIM). In
Causal effects addition, lactation persistency was also included. A subset including 28,867 daughters of 1809 Holstein
Mastitis sires having both first and second lactation across 201 herds was further investigated. In these datasets,
Milk yield
mastitis events were defined on a lactation basis as binary trait; either a cow was assigned a score of 1
Structural equation models
(had a mastitis event in that lactation) or a score of 0 (healthy) for that particular lactation, regardless of
the time of occurrence. Total milk yield from first and second lactation were also included in the analyses.
We estimated negative structural coefficient (
0.032) between clinical mastitis and test day milk pro-
duction in early lactation period suggesting that mastitis results in a direct decline in milk production in
early lactation. We nonetheless elicited little impact of mastitis on test day milk production of mid and
late lactation periods, and on milk yield lactation persistency. Likewise the positive estimate of the
structural coefficient (0.123) from mastitis event in first lactation to second lactation suggests an in-
creased risk of mastitis in second lactation if a case of mastitis occurs in the primiparous cow. Heritability
estimates obtained from the structural equation models were low for mastitis (ranged 0.04 to 0.07), and
negative genetic correlations were found between mastitis events and milk yield. The study illustrates
how mastitis events and production are causally linked. Through the use of structural equation models
we elicited the causal effect among mastitis and production traits that evolve over the course of cow life.
& 2016 Elsevier B.V. All rights reserved.

1. Introduction mastitis directly impact the profitability of dairy enterprises

(Huijps et al., 2008). The average cost of a clinical mastitis has been
Mastitis is a mammary inflammation and is one of the most previously estimated at $179 with $115 from milk lost per case,
economically impacting health events in the dairy cattle industry. $14 due to increased mortality loss, and $50 from treatment costs
The losses due to mastitis are costly (Koeck et al., 2012b), mostly (Bar et al., 2008). The antagonistic relationship between health
due to veterinary and treatment costs (Hinrichs et al., 2005), dis- disorders and milk yield in dairy cows is generally accepted (Rauw
carded milk (Shim et al., 2004), and reduced milk production (Bar et al., 1998). In the past 50 years, there has been an intense se-
et al., 2008), but also due to increased risk of culling (Hertl et al., lection for yield traits. This has resulted in an increased dete-
2011), and increased reproductive problems (Moore et al., 1991). rioration of dairy health (Miglior et al., 2005). These problems in
Moreover, replacement costs and increased labor cost due to dairy cattle have pointed towards genetic selection for increased
disease resistance and several researchers in the past decade have
suggested inclusion of clinical mastitis in the overall breeding goal
Abbreviations: LP, lactation persistency of milk yield; LMAST, liability to mastitis; of Holstein dairy cattle (Kadarmideen and Pryce, 2001; Ødegård
MY, milk yield; SEM, structural equation models; TD, test-day; TMY, total milk
yield
et al., 2003). Direct selection for mastitis resistance has been so far
n
Corresponding author. fully implemented in Nordic cattle (Heringstad et al., 2003; Phi-
E-mail address: [email protected] (K. Dhakal). lipsson and Lindhé, 2003) and national genetic and genomic

http://dx.doi.org/10.1016/j.livsci.2016.04.015
1871-1413/& 2016 Elsevier B.V. All rights reserved.
 K. Dhakal et al. / Livestock Science 189 (2016) 8–16 9

evaluations for clinical mastitis has been started in Canada (Jam- days, period 2 included 61–120 days, and period 3 included 121–
rozik et al., 2013) and France (Govignon-Gion et al., 2012). Simi- 180 days similarly to the procedure adopted by Wu et al. (2008).
larly, routine genetic evaluation of mastitis is done in Austria Single MY TD records were assigned to each period as the closest
(Fuerst et al., 2011; Koeck et al., 2015). in time to the midpoint of each segment and will be hereafter
Several researchers in the past have used mixed models to referred to as MY1, MY2, and MY3, respectively. Cows were as-
compute genetic correlations between mastitis and milk yield. signed a value of 0 (healthy) or 1 (mastitis) in each period. Only
Recently, Pfeiffer et al. (2015) described genetic relationships be- mastitis records that were prior and temporally closer to the as-
tween functional longevity and mastitis as well as other direct signed TD for each period were considered. This definition implies
health traits. Results from those studies mostly revealed the un- that pre-existing mastitis events would affect the MY of the fol-
favorable genetic correlations between mastitis events and milk lowing TD. Lactation persistency of MY (LP), a measure describing
yield traits. Yet correlations do not imply causation and there is the shape of the lactation curve after peak milk yield, was calcu-
still a lack of knowledge about the cause and effect between these lated for each cow using BESTPRED software (Cole and VanRaden,
traits, which could be addressed using structural equation models 2007).
(SEM). In the context of animal breeding, Gianola and Sorensen A subset (First & Second-Lactation) dataset was formed to
(2004) extended multivariate mixed model theory to infer re- identify causal effects from mastitis in first lactation to second
cursive relationships between phenotypes by accounting for pos- lactation and from mastitis to total milk yield for first and second
sible feedback situations. Several papers that have been published lactation. The First & Second-Lactation dataset included 28,867
in the realm of animal breeding over the past few years used daughters from 1809 sires having first and second lactation across
structural equation models to infer causal relationships between 201 herds. Only cows showing records for both lactations were
health traits (Wu et al., 2008; Heringstad et al., 2009; Dhakal et al., included in the dataset. To reduce complexity and improve the
2015). Wu et al. (2008) used a dataset of Norwegian Red cows to interpretation of the results mastitis events were in this case de-
study the causal effect between mastitis and milk yield. To the best fined on a lactation basis as binary trait; a cow was assigned either
of our knowledge, no study has been conducted to infer causal a score of 1 (had a mastitis event in that lactation) or a 0 (healthy)
relationships between mastitis events and milk yield in the US for that particular lactation, regardless of the time of occurrence.
Holstein cattle population. Similarly, there is a knowledge gap Total milk yield (305-day milk yield) from first and second lacta-
regarding the causal effect of mastitis occurring in first lactation tion were also included in the analysis as calculated from the
and mastitis events occurring in later lactations. Thus, the objec- BESTPRED software (Cole and VanRaden, 2007).
tive of the current study was to elicit direct causal phenotypic
effects and genetic relationships among mastitis events and pro- 2.2. Statistical analysis
duction traits (milk yield and lactation persistency of milk yield) in
US Holsteins using recursive models. Recursive Gaussian-threshold sire models were used for the
statistical analyses. The threshold model assumed an underlying
continuous variable, liability (li), for binary mastitis events that
2. Materials and methods defines the observed binary variable into a value of 1 if liability is
larger than a fixed threshold and 0 otherwise. Two different series
2.1. Data of analysis (Lactation first (LAC1) and Lactation first and second
(LAC12)) were defined for the purpose of identifying causal re-
Health information records were made available from Dairy lationships between mastitis events and production measures. The
Records Management Systems (Raleigh, NC) from US dairy farms LAC1 series of analyses employed a SEM to find recursive re-
from 1996 through June 2013. Holstein cows with mastitis records lationships between mastitis events, TD milk yields, and LP in first
in first and second parity were retained for the analyses which lactation. Four analyses were performed, which are as follows:
included calving records from 1996 to 2012. Health data quality
edits were applied as described in detail in Parker Gaddis et al.  LAC1. A: This analysis included two traits: liability to mastitis in
(2012) with some slight modifications for herd edits. In order to the first period (LMAST1) and MY1. The direct recursive effect
avoid herds that over or under reported mastitis events, maximum was assumed from LMAST1 to MY1.
and minimum constraints were applied to the dataset. The max-  LAC1. B: This analysis included MY1, liability to mastitis in the
imum constraint was imposed by excluding records when re- second period (LMAST2) and MY2. The direct recursive effect
porting frequency of herds were greater than two standard de- was assumed from MY1 to LMAST2 and from LMAST2 to MY2.
viations above the mean reporting frequency of mastitis event.  LAC1. C: This analysis included MY2, liability to mastitis in the
Similarly, minimum constraint was imposed by selecting records third period (LMAST3) and MY3. The direct recursive effect was
from herd with at least one reported incidence of the mastitis assumed from MY2 to LMAST3 and from LMAST3 to MY3.
event and herds consisting of at least 5 cows. In addition, only  LAC1. D: Liability to mastitis of each period and LP were in-
records of cows having lactation length less than or equal to 400 cluded in this analysis. The direct recursive effects were as-
days in milk (DIM) were included. After applying data quality sumed from mastitis of each period (LMAST1, LMAST2, and
edits, a dataset was formed to identify causal effects between LMAST3) to lactation persistency of milk yield.
mastitis and production measures (test-day milk yields and lac-
tation persistency) in first parity US Holsteins. This dataset in- Lastly, the LAC12 analysis (using First & Second-Lactation da-
cluded 48,058 first parity daughters of 2213 Holstein sires across taset) assumed direct recursive effects from liability to first lac-
207 herds and will be referred to as First-Lactation dataset. tation mastitis (LM1) to liability to second lactation mastitis (LM2).
The First-Lactation dataset included test-day (TD) records for Direct recursive effects were also considered from LM1 to total
milk yield (MY) and lactation persistency of MY in addition to milk yield of first parity (TMY1) and also to total milk yield of
mastitis events. Test Day records from 5 to 180 days after calving second parity (TMY2). An indirect recursive effect was also as-
were included and cows with missing TD records were removed sumed from LM1 to LM2 and TMY2. A direct recursive effect
from the original dataset in the process of forming the First-Lac- measures how much TMY1, LM2, and TMY2 would be affected by
tation dataset. Days in milk up to 180 days after calving were di- changes in LM1. An indirect recursive effect measures how much
vided into 3 lactation periods such that period 1 included 5–60 LM2, and TMY2 would be affected by changes in LM1 through the
10 K. Dhakal et al. / Livestock Science 189 (2016) 8–16

mediating effect of TMY1. For instance, the indirect recursive effect the residual covariances with unit residual variances was an in-
from LM1 to LM2 can be calculated as the product of structural verse-Wishart distribution. Transformation of the estimated cov-
coefficients (LM1-LM2) ¼(LM1-TMY1)  (TMY1-LM2). The ariance matrices for the SEM in multiple trait model scale was
overall causal effect on LM2 can be calculated as (LM1-LM2) þ performed as:
(LM1-TMY1)  (TMY1-LM2) (Lopez de Maturana et al., 2009;
G*n = (I − Λ)−n1G n (I − Λ)′n− 1
Shipley, 2002).
The models used in LAC1 and LAC12 analyses can be sum- H*n = (I − Λ)−n1Hn (I − Λ)′n− 1
marized as follows: R *n = (I − Λ)−n1 R n (I − Λ)′n− 1
⎧ y = Xb1 + Zh h1 + Zs s1 + e1
LAC1. A: ⎨
⎪
1 where the index n indicates the models used in first and second
series of analyses, and G, H, R, and Λ were as defined above.
⎪
⎩ y2 = λ21y1 + Xb2 + Zh h2 + Zs s2 + e2
Heritabilities and genetic correlations were then calculated in the
usual manner from (co)variance components in G*n, H*n, and R*n.
⎧ y = Xb1 + Zh h1 + Zs s1 + e1 Data analyses were conducted in Bayesian framework using the
⎪ 1 SIR-BAYES package (Wu et al., 2008) in which all Bayesian models
LAC1. B/LAC1. C: ⎨ y2 = λ21y1 + Xb2 + Zh h2 + Zs s2 + e2
⎪ were implemented via Markov chain Monte Carlo (MCMC) sam-
⎩ y3 = λ32 y2 + Xb3 + Zh h3 + Zs s3 + e3 pling. For each model, 100,000 iterations were generated and the
first 20,000 iterations were discarded as burn-in. Posterior sam-
ples from each chain were thinned every 25 iterations after burn-
⎧ y = Xb1 + Zh h1 + Zs s1 + e1
⎪ 1 in and retained for analysis. Posterior distributions of parameters
⎪ y2 = Xb2 + Zh h2 + Zs s2 + e2 of interest were inferred based on posterior samples after burn-in.
LAC1. D: ⎨ Markov chain convergence was assessed by visual inspection of
⎪ y3 = Xb3 + Zh h3 + Zs s3 + e3
⎪ trace plots. Additional diagnostic tests such as Geweke's con-
⎩ y4 = λ 41y1 + λ 42 y2 + λ 43 y3 + Xb4 + Zh h4 + Zs s4 + e4
vergence statistic (Geweke, 1992) was obtained to confirm con-
vergence through R (http://cran.r-project.org) with the CODA
package (Plummer et al., 2012).
⎧ y = Xb1 + Zh h1 + Zs s1 + e1
⎪ 1 Transformation of lambda coefficients estimates from liability
⎪ y2 = λ21y1 + Xb2 + Zh h2 + Zs s2 + e2 to observable scale was done following Wu et al. (2008). For ex-
LAC12: ⎨
⎪ y3 = λ31y1 + λ32 y2 + Xb3 + Zh h3 + Zs s3 + e3 ample, the difference in mean peak milk yield between sick
⎪ (1) cows due to MAST1 and healthy (0) cows can be calculated as
⎩ y4 = λ 41y1 + λ 42 y2 + λ 43 y3 + Xb4 + Zh h4 + Zs s4 + e4
– –
Δ ≈ λ( l1 – l0)
where, the y1 and y2 are vectors reporting LMAST1 and MY1 re-
spectively in LAC1.A analysis. The y1, y2, and y3 are the vectors where l¯1 and l¯0 are averages of augmented liabilities for sick cows
reporting MY1, LMAST2, and MY2 respectively in LAC1.B analysis. due to MAST1 and healthy cows, respectively.
Similarly, the y1, y2, and y3 are vectors reporting MY2, LMAST3,
and MY3 respectively in LAC1.C analysis. In the case of LAC1.D
analysis, the y1, y2, y3, and y4 of are vectors reporting LMAST1, 3. Results and discussion
LMAST2, LMAST3, and LP respectively. In equations reported above
for LAC12 analysis, the y1, y2, y3, and y4 are vectors reporting LM1, The incidence of mastitis events in First-Lactation dataset were
TMY1, LM2, and TMY2 respectively. The λij is the structural coef- 6.58%, 4.13%, and 3.90% for lactation periods 5–60, 61–120, and
ficients describing the rate of change for trait i with respect to trait 121–180 DIM respectively (Table 1). The TD MYs decreased over
j, b is a vector of systematic effects including the effect of year- the three lactation periods. The mean (standard deviation) of TD
season of calving, h is a vector of herd effects, s is a vector of sire of MY was 34.66 (7.18) kg, 34.28 (7.92) kg, and 28.55 (7.32) kg at
cow effects, and e is a vector of residuals; X, Zh, and Zs are the lactation periods 1, 2, and 3, respectively (Table 2). The mean
corresponding incidence matrices. In matrix form, the general (standard deviation) of LP was 0.38 (0.97) LP units. The mean of LP
model was: in our study was lower than that reported by Appuhamy et al.
y = (Λ ⊗ I) y + Xb + Zh h + Z s s + e (2009). In their study, they reported a mean (standard deviation)
0.53 (1.19) LP units for first parity cows. Incidences of mastitis
where, Λ are lower triangular matrices with 1 on diagonal, λij on events in the First & Second-Lactation dataset were 10.87% and
off-diagonals representing the recursive effects from j to i, and 14.05% for lactations 1 and 2, respectively (Table 1). Incidences of
0 everywhere else.
Multivariate normal prior distributions were assigned to Table 1
structural coefficients as N(1λ0, I τ2), where hyperparameters Descriptive statistics for mastitis events.
were λ0 ¼ 0 and τ2 ¼10,000. Elements of b were assigned normal
Traitsa Number of records Mastitis event frequency (%)
prior distributions, with mean 0 and variance 10,000. Sire effects
were assigned a multivariate normal prior distribution s  N(0, G ⊗ Healthy Diseased
A), where G is the sire covariance matrix for the traits involved and
A is the matrix of additive genetic relationships among bulls. The MAST1 44,896 3162 6.58
MAST2 46,073 1985 4.13
prior distribution of herd effects was h  N(0, H ⊗ I), where H is MAST3 46,184 1874 3.90
the herd (co)variance matrix and I is an identity matrix. In- M1 25,729 3138 10.87
dependent inverse-Wishart prior distributions were used for H M2 24,811 4056 14.05
and G, the covariance matrices of h and s, respectively. In order to a
MAST1 is the mastitis event occurring in 5–60 DIM in first lactation; MAST2 is
achieve identifiability, residual variances of threshold traits were
the mastitis event occurring in 61–120 DIM in first lactation; MAST3 is the mastitis
fixed to 1. Furthermore, all residual covariances were forced to be event occurring in 121–180 DIM in first lactation; M1 is the mastitis event occurring
equal to 0. In this case, the prior distribution of the R matrix fixing in first lactation; and M2 is the mastitis event occurring in second lactation.
 K. Dhakal et al. / Livestock Science 189 (2016) 8–16 11

Table 2 mastitis events in this study were slightly higher than those re-
Descriptive statistics of production measures. ported by Parker Gaddis et al. (2014) in a similar dataset (9.53%
and 10.24% in parities 1 and 2, respectively). The mean (standard
Traitsa Number of records Mean SD Minimum Maximum
deviation) of TMY1 in parity 1 and TMY2 in parity 2 were 9851.28
MY1 (kg) 48,058 34.66 7.18 6.50 81.22 (2034.32) kg, and 10,800.60 (1900.98) kg respectively (Table 2).
MY2 (kg) 48,058 34.28 7.92 9.64 79.34
MY3 (kg) 48,058 28.55 7.32 8.32 72.09
LP (units) 48,058 0.38 0.97
3.12 4.86
3.1. Recursive effects
TMY1 (kg) 28,867 9851.28 2034.32 2938.00 17,944.00
TMY2 (kg) 28,867 10,800.60 1900.98 3094.00 17,470.00 Posterior distribution of recursive effects from liability to
a
mastitis to TD milk yields of three lactation periods from the LAC1
MY1 is the test-day milk yield of first period (5–60 DIM) of first lactation;
series of analyses (LAC1.A, LAC1.B, LAC1.C analysis using First-
MY2 is the test-day milk yield of second period (61–120 DIM) of first lactation; MY3
is the test-day milk yield of third period (121–180 DIM) of first lactation; LP is the Lactation dataset) are shown in Fig. 1(A) and that of TD milk yields
lactation persistency of milk yield in first lactation; TMY1 is the total milk yield of to liability to mastitis in the following lactation period are shown
first lactation; TMY2 is the total milk of second lactation. in Fig. 1(B); the posterior mean, standard deviation (SD) and 95%

Fig. 1. Recursive effects between milk yields and liability to mastitis in three different periods of First-Lactation (5–60, 61–120, and 121–180 days in milk): (A) Recursive
effects from liability to mastitis (LMAST1, LMAST2, LMAST3) to milk yields (MY1, MY2, MY3) of corresponding period of First-Lactation; (B) Recursive effects from milk yields
(MY1 and MY2) to liability to mastitis (LMAST2 and LMAST3) of corresponding period of First-Lactation; (C) Recursive effects from liability to mastitis (LMAST1, LMAST2,
LMAST3) to lactation persistency of milk yield (LP).
12 K. Dhakal et al. / Livestock Science 189 (2016) 8–16

Table 3 Jamrozik et al., 2010; Wu et al., 2007) and found that increases in
Posterior mean, standard deviation (SD), and 95% highest posterior density interval SCS decreased the milk yield production. These results were also
(95% HPD) of causal relationships between liability to mastitis, milk yields, and
reflected in our study because SCS can be considered as an in-
lactation persistency.
dicator of udder infection (Jamrozik et al., 2010) and high SCS is
Traitsa Recursive effects often associated with clinical mastitis cases (Shook and Schutz,
1994). High genetic correlations between SCS and mastitis ranging
Liability scale Observable scale from 0.63 to 0.85 (Heringstad et al., 2006; Koeck et al., 2012a,
Mean SD 95% HPD Mean
2012b) have been reported in literature.
Based on the recursive effects from TD milk yields to liability to
LAC1.A mastitis in the following lactation period (e.g. MY1-LMAST2), it
LMAST1-MY1
0.032 0.010 [
0.051;
0.013]
0.24 kg was concluded that a weak positive relationship might exist,
nonetheless zero was included in the 95% HPD credible region
LAC1.B (Table 3). This may be an indication of the fact that the increase in
MY1-LMAST2 0.005 0.008 [
0.011; 0.021] TD milk production in lactation period 1 and 2 had no effect on
LMAST2-MY2
0.004 0.008 [
0.021; 0.010]
0.09 kg
occurrence of mastitis events. It is otherwise possible that our data
were insufficient to estimate the true recursive effects between TD
LAC1.C milk yields and liability to mastitis.
MY2-LMAST3 0.001 0.008 [
0.015; 0.017]
Posterior distribution of recursive effects from liability to
LMAST3-MY3
0.003 0.009 [
0.020; 0.016]
0.08 kg
mastitis of lactation periods 1, 2 and, 3 to LP are shown in Fig. 1(C);
the posterior mean, standard deviation (SD) and 95% HPD are
LAC1.D shown in Table 3. All the 95% HPD for the structural coefficients
LMAST1-LP
0.002 0.005 [
0.013; 0.008]
0.01 LP units
LMAST2-LP
0.003 0.005 [
0.015; 0.007]
0.01 LP units
obtained in this analysis included zero in the credible interval.
LMAST3-LP 0.006 0.005 [
0.005; 0.016] 0.02 LP units Posterior means of structural coefficients indicate a weak effect of
liability to mastitis of each period to LP. An increase of 1-unit of
LAC12
liability to mastitis decreased LP by 0.002 (LMAST1-LP) and 0.003
LM1-TMY1
0.031 0.021 [
0.087; 0.028]
0.24 kg (LMAST2-LP) LP units and increased LP by 0.006 (LMAST3-LP)
LM1-LM2 0.123 0.033 [0.058; 0.186] LP units, thus indicating that a mastitis event (LMAST1 and
LM1-TMY2 0.018 0.039 [
0.056; 0.096] 0.09 kg LMAST2) happening in a cow's early lactation period would mar-
TMY1-LM2
0.005 0.015 [
0.035; 0.022]
ginally affect persistency while late mastitis (LMAST3) would
TMY1-TMY2 0.740 0.009 [0.722; 0.755]
LM2-TMY2
0.119 0.04 [0.038;
0.195]
0.83 kg slightly increase persistency. Mastitis events occurring at the be-
ginning of the lactation might compromise later production more
a
Liability to mastitis in first parity divided into three periods (LMAST1, than those occurring possibly in part because in late lactation cows
LMAST2, LMAST3); test-day milk yields (MY1, MY2, MY3); Lactation persistency of may have enough energy reserves to utilize slowly and efficiently
milk yield denoted as LP; liability to mastitis in lactation 1 and 2 denoted as LM1
and LM2. Total amount of milk yield in first and second lactation denoted as TMY1
to maintain their production (Ferris et al., 1985).
and TMY2. The posterior distribution of direct recursive effects from LM1 to
TMY1, LM2, and TMY2 are shown in Fig. 2(A); from TMY1 to LM2
and TMY2 are shown in Fig. 2(B); from LM2 to TMY2 is shown in
posterior density interval (95% HPD) are shown in Table 3. The Fig. 2(C), and the posterior mean, SD, and 95% HPD are shown in
graphical representation of causal structure assumed and the Table 3 (Fig. 3). The causal structure assumed and the posterior
posterior means obtained for LAC1 series are shown in Fig. 3. means obtained for LAC12 series are shown in Fig. 4. The results
The recursive effects from liability to mastitis to milk yields show how the direct recursive effect from LM1 to TMY1 is negli-
were between
0.032 and
0.003. Among the recursive effects gible. The direct recursive effect from LM1 to LM2 had a positive
from liability from mastitis to milk yields, only the recursive effect posterior mean of approximately 0.123 liability unit increase of LM2
from LMAST1 to MY1 did not include zero in 95% HPD credible for a 1-unit increase of LM1. The indirect effect of LM1 to LM2
interval. Posterior means of structural coefficients indicate a ne- through the mediating effect of TMY1 was likewise weak. The
gative recursive effect of liability to mastitis to TD milk yields. An overall causal effect of LM1 to LM2 was positive (0.124 liability unity
increase in 1-unit of liability to mastitis decreased TD milk yields increase of LM2 for a 1-unit increase of LM1). A cow with a mastitis
by 0.032, 0.004, and 0.003 kg per day, in lactation period 1, 2, and infection in first parity would have an increased risk of incurring in
3, respectively. The decrease in TD milk yield was higher in period mastitis in second parity due to the direct causal effect of the first
1 but similar for period 2 and 3. This may be interpreted as cows event on the second. The direct recursive effect of LM1 to TMY2 was
affected by mastitis during the early lactation period might have positive with approximately 0.018 kg per day increase of TMY2 for a
acquired some immunity against mastitis causing pathogens, 1-unit increase of LM1. The indirect recursive effect of LM1 to TMY2
thereby reducing the effect of mastitis in milk yields during suc- was negligible. Thus cows with mastitis events in first lactation
cessive lactation periods. Wu et al. (2008) reported similar de- would produce a slightly higher amount of milk in second lactation
crease in TD milk yields in Norwegian Red cows with clinical compared to cows which were assumed healthy in this analysis. It
mastitis. In the observable scale, the difference in mean MY1, MY2 should be noted here that an increase in milk yield in second lac-
and MY3 between the sick cows due to mastitis and healthy cows tation for the cows having mastitis problem in first lactation could
in our study were
0.24 kg,
0.09 kg and
0.08 kg per day for be due to other unidentified management effects in second lacta-
lactation periods 1, 2, and 3, respectively. Based on results, an in- tion such as additional care or better nutrition, or the cows that
creased liability to mastitis slightly reduces milk yield at the fol- were assumed healthy (no mastitis event) in second lactation may
lowing TD. This result is in agreement with a study by Wu et al. have other health events affecting milk yield not accounted in the
(2008), in which they reported the presence of a causal relation- current analysis. The direct recursive effect of LM2 to TMY2 was
ship between mastitis incidence and milk yield production. Esti- negative with approximately 0.119 kg per day decrease of TMY2 for
mates of structural coefficients obtained were also similar. Several a 1-unit increase of LM2.
other authors have reported the causal relationship between so- The direct causal effect of TMY1 on LM2 had negligible effect.
matic cell score (SCS) and milk yield (de los Campos et al., 2006; The direct recursive effect of TMY1 on TMY2 had a positive
 K. Dhakal et al. / Livestock Science 189 (2016) 8–16 13

Fig. 2. Recursive effects between liability to mastitis events between parity 1, 2, total milk yield of first parity (TMY1) and second parity (TMY2): (A) Recursive effects from
liability to mastitis in parity 1 (LM1) to liability to mastitis in parity 2 (LM2), LM1 to TMY1 and LM1 to TMY2; (B) Recursive effects from TMY1 to LM2, TMY1 to TMY2;
(C) Recursive effects from LM2 to TMY2.

recursive effect of approximately 0.74 kg per day increase of TMY2 were similar to those estimated by Wu et al. (2008). The posterior
for a 1-unit increase of TMY1. mean of heritability for LP was approximately 0.14 and fell within
the range of previous estimates of heritability for LP. Gengler (1996)
3.2. Heritabilities and genetic correlations reported heritability for LP of 0.14 for Holstein cows. Cole and
VanRaden (2006) estimated the heritability of LP equal to 0.10.
Posterior mean, SD, and 95% HPD heritabilities of mastitis events Posterior mean, SD, and 95% HPD of heritabilities for mastitis events
and milk yields of all three lactation periods of first lactation and LP in first and second lactation and TMY are shown in Table 4. The
are shown in Table 4. The posterior mean of heritabilities for posterior mean of heritabilities for LM1 and LM2 were 0.04. These
LMAST1 (0.04), LMAST2 (0.07) and LMAST3 (0.04) were slightly estimates of heritabilities were lower than that reported by Zwald
lower than the study done by Wu et al. (2008) using recursive et al. (2006) using producer-recorded data where heritability of
Gaussian threshold model for Norwegian Red cows. Heritabilities in liability to mastitis for first parity was 0.12 and for second parity
that study of liability to clinical mastitis (LCM) for three periods of was 0.10. Heritability estimates of liability to mastitis in the present
lactation were 0.067 (LCM1), 0.094 (LCM2) and 0.079 (LCM3), re- study were in agreement with Parker Gaddis et al. (2014) where
spectively. The lower heritabilities of mastitis in this study could be they reported a heritability of 0.06 for mastitis in first parity cows
due to the fact that no selection was performed in respect to health- and 0.03 for mastitis in later parity cows using producer-recorded
related traits in the US Holstein population, and also may be due to dataset. The posterior mean of heritability for TMY1 and TMY2
the use of farmer recorded data in estimating heritability. Lower were 0.22 and 0.18 respectively. Carlen et al. (2004) reported her-
estimate of heritability of mastitis (0.003) was also reported in the itability estimate of 305-day milk yield of first and second lactation
study of farmer observed health data of Austrian Fleckvieh cows by of 0.34 and 0.25 respectively, in the study done in Swedish Holstein
Koeck et al. (2015). The posterior mean of heritabilities for MYs cows. Similar heritability estimates ranging from 0.19 to 0.25 of
ranged from 0.12 to 0.24. These estimates of heritabilities for MYs 305-day milk yield across three parities in small and large herds of
14 K. Dhakal et al. / Livestock Science 189 (2016) 8–16

Table 4
Posterior mean, standard deviation (SD), and 95% highest posterior density interval
(95% HPD) of heritability of liability to mastitis, milk yields (MY), and lactation
persistency, and total milk yield of first lactation (TMY1) and second lactation
(TMY2).

Traitsb Heritabilitya

Mean SD 95% HPD

c
LMAST1 0.04 0.01 [0.020; 0.070]
LMAST2c 0.07 0.02 [0.034; 0.104]
LMAST3d 0.04 0.01 [0.017; 0.058]
MY1 0.12 0.01 [0.091; 0.141]
MY2b 0.13 0.02 [0.104; 0.161]
MY3 0.24 0.02 [0.203; 0.281]
LP 0.14 0.02 [0.114; 0.172]
LM1 0.04 0.01 [0.018; 0.065]
LM2 0.04 0.02 [0.013; 0.075]
TMY1 0.22 0.01 [0.198; 0.242]
TMY2 0.18 0.01 [0.145; 0.204]

4σs2
a
Heritability computed as h2= , where σs2 is the sire additive genetic
σs2 +σ 2 +σe2
h
variance, σ h2 is the herd environmental variance, σe2 is residual variance.
b
Liability to mastitis in first lactation divided into three periods (LMAST1,
LMAST2, LMAST3); test-day milk yields (MY1, MY2, MY3); Lactation persistency of
milk yield denoted as LP; liability to mastitis in lactation 1 and 2 denoted as LM1
and LM2. Total amount of milk yield in first and second lactation denoted as TMY1
and TMY2 respectively.
c
Heritability of liabilities of mastitis were reported from LAC1.D analysis.
d
Heritability of MY2 was reported from LAC 1.B analysis.

Table 5
Posterior mean, standard deviation (SD), and 95% highest posterior density interval
(95% HPD) of genetic correlation between liability to mastitis, milk yields, and
Fig. 3. Acyclic graphical representation of causal structure assumed and posterior
lactation persistency.
means obtained for causal effects between milk yields and liability to mastitis in
three different periods of First–Lactation (5–60, 61–120, and 121–180 days in milk):
Traitsa Genetic Correlation
(A) Causal structure assumed in first period between liability to mastitis (LMAST1)
and milk yield (MY1); (B) Causal structure assumed between milk yield (MY1) of
Mean SD 95% HPD
first period, liability to mastitis (LMAST2) and milk yield (MY2) of second period;
(C) Causal structure assumed between milk yield (MY2) of second period, liability
LMAST1 and MY1 0.21 0.20 [
0.191; 0.612]
to mastitis (LMAST3) and milk yield (MY3) of third period; (D) Causal structure
LMAST2 and MY1
0.14 0.20 [
0.517; 0.260]
assumed between liability to mastitis (LMAST1, LMAST2, LMAST3) to lactation
LMAST2 and MY2
0.15 0.19 [
0.532; 0.237]
persistency of milk yield (LP).
LMAST3 and MY2 0.44 0.20 [0.084; 0.816]
LMAST3 and MY3 0.43 0.20 [0.046; 0.780]
LMAST1 and LMAST2 0.58 0.15 [0.268; 0.814]
LMAST1 and LMAST3 0.45 0.16 [0.134; 0.745]
LMAST2 and LMAST3 0.59 0.13 [0.342; 0.820]
LMAST1 and LP 0.008 0.17 [
0.316; 0.355]
LMAST2 and LP 0.003 0.15 [
0.291; 0.289]
LMAST3 and LP 0.20 0.15 [
0.068; 0.502]
LM1 and LM2 0.48 0.23 [
0.149; 0.526]
LM1 and TMY1
0.004 0.20 [
0.042; 0.036]
LM1 and TMY2
0.01 0.02 [
0.060; 0.035]
TMY1 and LM2 0.006 0.02 [
0.031; 0.043]
TMY1 and TMY2 0.59 0.10 [0.286; 0.785]
LM2 and TMY2
0.001 0.02 [
0.038; 0.041]

a
Liability to mastitis in first lactation divided into three periods (LMAST1,
LMAST2, LMAST3); test-day milk yields (MY1, MY2, MY3); Lactation persistency of
milk yield denoted as LP; liability to mastitis in lactation 1 and 2 denoted as LM1
and LM2. Total amount of milk yield in first and second lactation denoted as TMY1
and TMY2 respectively.
Fig. 4. Acyclic graphical representation of causal structure assumed and posterior
means obtained between liability to mastitis events between parity 1 (LM1), parity estimates of genetic correlations are moderate and were lower
2 (LM2), total milk yield of first parity (TMY1) and second parity (TMY2).
than those obtained by Wu et al. (2008). The genetic correlations
between LMAST1 and MY1, LMAST2 and MY1, LMAST2 and MY2,
Holstein cows in US was reported by Tsuruta et al. (2015). LMAST3 and MY2, and LMAST3 and MY3 were 0.21,
0.14,
0.15,
Posterior means, SD and 95% HPD genetic correlations between
0.44, and 0.43, respectively. Among these posterior means only
mastitis and between mastitis event and milk yields and between
genetic correlations between LMAST3 and MY2 and LMAST3 and
mastitis event and LP are shown in Table 5. Posterior means of
MY3 were well defined and others include zero in their 95% HPD
genetic correlations among mastitis events in three lactation per-
iods of first lactation were all positive. Genetic correlations be- credible interval. The genetic correlations between LMAST1 and LP,
tween LMAST1 and LMAST2, LMAST1 and LMAST3, and LMAST2 LMAST2 and LP, and LMAST3 and LP were 0.008, 0.003, and 0.20,
and LMAST3 were 0.58, 0.45, and 0.59, respectively. These respectively, and included zero in the 95% HPD credible interval.
 K. Dhakal et al. / Livestock Science 189 (2016) 8–16 15

Posterior mean, SD, and 95% HPD genetic correlations between describing relationships between somatic cell score and milk yield in first-
mastitis events of first and second lactation, TMY1 and TMY2 were lactation dairy cows. J. Dairy Sci. 89, 4445–4455.
de Maturana, E.L., Wu, X.L., Gianola, D., Weigel, K.A., Rosa, G.J.M., 2009. Exploring
shown in Table 5. The genetic correlation between LM1 and LM2 biological relationships between calving traits in primiparous cattle with a
was 0.48. The genetic correlation between TMY1 and TMY2 was Bayesian recursive model. Genetics 181, 277–287.
0.59. The genetic correlations of mastitis events with total milk Dhakal, K., Tiezzi, F., Clay, J.S., Maltecca, C., 2015. Inferring causal relationships
between reproductive and metabolic health disorders and production traits in
yields were negligible. first-lactation US Holsteins using recursive models. J. Dairy Sci. 98, 2713–2726.
Ferris, T.A., IMao, I.L., Anderson, C.R., 1985. Selection for lactation curve and milk
yield in cattle. J. Dairy Sci. 68, 1438–1448.
Fuerst, C., Koeck, A., Egger-Danner, C., Fuerst-Waltl, B., 2011. Routine genetic eva-
4. Conclusion luation for direct health traits in Austria and Germany. Interbull Bull. 44,
210–215.
Causal relationships between mastitis events along with pro- Gengler, N., 1996. Persistency of lactation yields: a review. Proc. Int. Workshop on
Genetic Improvement of Functional Traits in Cattle. Gembloux, Belgium. In-
duction traits such as milk yields and lactation persistency can
terbull Bull., vol. 12, pp. 87–96.
help us to identify the real biological pathway of disease process Geweke, J., 1992. Evaluating the accuracy of sampling-based approaches to the
and its consequences in production traits. The causal relationship calculation of posterior moments. In: Bernardo, J.M., Berger, J., Dawid, A.P.,
between mastitis events and milk yields showed that with an in- Smith, A.F.M., ed. Oxford University Press, Oxford, UK, Pages 169–193 in
Bayesian Statistics 4.
crease in mastitis events there would be a decline in milk pro- Gianola, D., Sorensen, D., 2004. Quantitative genetic models for describing si-
duction. There is little to no impact of mastitis events on lactation multaneous and recursive relationships between phenotypes. Genetics 167,
persistency of milk yield. The causal relationship among mastitis 1407–1424.
Govignon-Gion, A., Dassonneville, R., Baloche, G., Ducrocq, V., 2012. Genetic eva-
events in first and second lactation found in this study indicate luation of mastitis in dairy cattle in France. Interbull Bulletin. (Available):
that having a mastitis event in first lactation is likely to increase 〈https://journal.interbull.org/index.php/ib/article/view/1276/1314〉 (accessed
the risk of a mastitis event in second lactation. Based on the causal 26.05.15).
Heringstad, B., Klemetsdal, G., Steine, T., 2003. Selection responses for clinical
relationships between clinical mastitis events and production mastitis and protein yield in two Norwegian dairy cattle selection experiments.
traits, economic loss from clinical mastitis events can be mitigated J. Dairy Sci. 86, 2990–2999.
by addressing proper disease management strategies such as Heringstad, B., Gianola, D., Chang, Y.M., Ødegård, J., Klemetsdal, G., 2006. Genetic
associations between clinical mastitis and somatic cell score in early first-lac-
providing proper vaccination to boost immunity, proper treatment tation cows. J. Dairy Sci. 89, 2236–2244.
of infected cows, feeding improved feed stuffs, having better nu- Heringstad, B., Wu, X.L., Gianola, D., 2009. Inferring relationships between health
tritional standards to cope with disease, etc. Greater insight into and fertility in Norwegian red cows using recursive models. J. Dairy Sci. 92,
1778–1784.
relationships between mastitis and production traits could be
Hertl, J., Schukken, Y., Bar, D., Bennett, G., González, R., Rauch, B., Welcome, F.,
achieved by incorporating other factors in a recursive model such Tauer, L., Gröhn, Y., 2011. The effect of recurrent episodes of clinical mastitis
as risk factors of mastitis, herd demographics, housing conditions, caused by gram-positive and gram-negative bacteria and other organisms on
and feeding procedures. mortality and culling in Holstein dairy cows. J. Dairy Sci. 94, 4863–4877.
Hinrichs, D., Stamer, E., Junge, W., Kalm, E., 2005. Genetic analyses of mastitis data
using animal threshold models and genetic correlation with production traits. J.
Dairy Sci. 88, 2260–2268.
Conflict of interest statement Huijps, K., Lam, T.J., Hogeveen, H., 2008. Costs of mastitis: facts and perception. J.
Dairy Res. 75, 113–120.
Jamrozik, J., Koeck, A., Miglior, F., Kistemaker, G., Schenkel, F., Kelton, D., Van
The authors declare that there is no conflict of interest asso- Doormaal., B., 2013. Genetic and genomic evaluation of mastitis resistance in
ciated with this publication and there has been no significant fi- Canada. Interbull Bulletin. (Available): 〈https://journal.interbull.org/index.php/
ib/article/view/1291/1365〉 (accessed 26.05.15).
nancial support for this work that could have influenced its Jamrozik, J., Bohmanova, J., Schaeffer, L., 2010. Relationships between milk yield
outcome. and somatic cell score in Canadian Holsteins from simultaneous and recursive
random regression models. J. Dairy Sci. 93, 1216.
Kadarmideen, H., Pryce, J., 2001. Genetic and economic relationships between so-
matic cell count and clinical mastitis and their use in selection for mastitis
Author's contribution resistance in dairy cattle. Anim. Sci. 73, 19–28.
Koeck, A., Fuerst, C., Egger-Danner, C., 2015. Farmer-observed health data around
calving –genetic parameters and association with veterinarian diagnoses in
C. Maltecca, F. Tiezzi, and K. Dhakal designed the research. J.S.
Austrian Fleckvieh cows. J. Dairy Sci. 98, 2753–2758.
Clay provided the data for research. K.Dhakal performed the sta- Koeck, A., Miglior, F., Kelton, D., Schenkel, F., 2012a. Alternative somatic cell count
tistical analyses and wrote the first draft of the manuscript. F. traits to improve mastitis resistance in Canadian Holsteins. J. Dairy Sci. 95,
432–439.
Tiezzi contributed in statistical analyses. C. Maltecca, F. Tiezzi, and
Koeck, A., Miglior, F., Kelton, D., Schenkel, F., 2012b. Short communication: genetic
K. Dhakal contributed to the interpretation and discussion of the parameters for mastitis and its predictors in Canadian Holsteins. J. Dairy Sci. 95,
results. All authors approve the final draft of manuscript for 7363–7366.
publication. Miglior, F., Muir, B., Van Doormaal, B., 2005. Selection indices in Holstein cattle of
various countries. J. Dairy Sci. 88, 1255–1263.
Moore, D., Cullor, J., Bondurant, R., Sischo, W., 1991. Preliminary field evidence for
the association of clinical mastitis with altered interestrus intervals in dairy
References cattle. Theriogenology 36, 257–265.
Ødegård, J., Klemetsdal, G., Heringstad, B., 2003. Variance components and genetic
trend for somatic cell count in Norwegian Cattle. Livest. Prod. Sci. 79, 135–144.
Appuhamy, J., Cassell, B., Cole, J., 2009. Phenotypic and genetic relationships of Parker Gaddis, K., Cole, J., Clay, J., Maltecca, C., 2014. Genomic selection for pro-
common health disorders with milk and fat yield persistencies from producer- ducer-recorded health event data in US dairy cattle. J. Dairy Sci. 97, 3190–3199.
recorded health data and test-day yields. J. Dairy Sci. 92, 1785–1795. Parker Gaddis, K., Cole, J., Clay, J., Maltecca, C., 2012. Incidence validation and re-
Bar, D., Tauer, L., Bennett, G., Gonzalez, R., Hertl, J., Schukken, Y., Schulte, H., Wel- lationship analysis of producer-recorded health event data from on-farm
come, F., Gröhn, Y., 2008. The cost of generic clinical mastitis in dairy cows as computer systems in the United States. J. Dairy Sci. 95, 5422–5435.
estimated by using dynamic programming. J. Dairy Sci. 91, 2205–2214. Pfeiffer, C., Fuerst, C., Ducrocq, V., Fuerst-Waltl, B., 2015. Short communication:
Carlén, E., Strandberg, E., Roth, A., 2004. Genetic parameters for clinical mastitis, genetic relationships between functional longevity and direct health traits in
somatic cell score, and production in the first three lactations of Swedish Austrian Fleckvieh cattle. J. Dairy Sci. 98, 7380–7383.
Holstein cows. J. Dairy Sci. 87, 3062–3070. Philipsson, J., Lindhé, B., 2003. Experiences of including reproduction and health
Cole, J., VanRaden, P., 2007. A Manual for Use of BESTPRED: A program for esti- traits in Scandinavian dairy cattle breeding programmes. Livest. Prod. Sci. 83,
mation of lactation yield and persistency using best prediction. (Available): 99–112.
〈http://aipl.arsusda.gov/software/brestpred/〉 (accessed 14.11.13). Plummer, M., Best, N., Cowles, K., Vines, K., Sarkar, D., Almond, R., 2012. Package
Cole, J., VanRaden, P., 2006. Genetic evaluation and best prediction of lactation ‘coda’. (Available): 〈http://cran.r-project.org/web/packages/coda/coda.pdf〉 (ac-
persistency. J. Dairy Sci. 89, 2722–2728. cessed 12.01.14).
de los Campos, G., Gianola, D., Heringstad, B., 2006. A structural equation model for Rauw, W., Kanis, E., Noordhuizen-Stassen, E., Grommers, F., 1998. Undesirable side
16 K. Dhakal et al. / Livestock Science 189 (2016) 8–16

effects of selection for high production efficiency in farm animals: a review. interactions on culling rates and 305-day milk yield of Holstein cows in 3 US
Livest. Prod. Sci. 56, 15–33. regions. J. Dairy Sci. 98, 5796–5805.
Shim, E., Shanks, R., Morin, D., 2004. Milk loss and treatment costs associated with Wu, X.L., Heringstad, B., Chang, Y.M., de los Campos, G., Gianola, D., 2007. Inferring
two treatment protocols for clinical mastitis in dairy cows. J. Dairy Sci. 87, relationships between somatic cell score and milk yield using simultaneous
2702–2708. and recursive models. J. Dairy Sci. 90, 3508–3521.
Shipley, B., 2002. Cause and Correlation in Biology: A User's Guide to Path Analysis, Wu, X.L., Heringstad, B., Gianola, D., 2008. Exploration of lagged relationships be-
Structural Equations and Causal Inference. Cambridge University Press, United tween mastitis and milk yield in dairy cows using a Bayesian structural equa-
Kingdom. tion Gaussian-threshold model. Genet. Sel. Evol. 40, 333–358.
Shook, G., Schutz, M., 1994. Selection on somatic cell score to improve resistance to Zwald, N.R., Weigel, K.A., Chang, Y.M., Welper, R.D., Clay, J.S., 2006. Genetic analysis
mastitis in the United States. J. Dairy Sci. 77, 648–658. of clinical mastitis data from on-farm management software using threshold
Tsuruta, S., Lourenco, D.A.L., Misztal, I., Lawlor, T.J., 2015. Genotype by environment models. J. Dairy Sci. 89, 330–336.