Application of Syndromic Surveillance on Routinely Collected Cattle

Preventive Veterinary Medicine 124 (2016) 15–24

Contents lists available at ScienceDirect

Preventive Veterinary Medicine

j ournal homepage: www.elsevier.com/locate/prevetmed

Application of syndromic surveillance on routinely collected cattle

reproduction and milk production data for the early detection of

outbreaks of Bluetongue and Schmallenberg viruses

a,∗ a b,c

Anouk Veldhuis , Henriëtte Brouwer-Middelesch , Alexis Marceau ,

b,c d,e b,c d

Aurélien Madouasse , Yves Van der Stede , Christine Fourichon , Sarah Welby ,

a a,f

Paul Wever , Gerdien van Schaik

GD Animal Health, PO Box 9, 7400 AA Deventer, The Netherlands

INRA, UMR1300, Biologie, Epidémiologie et Analyse de Risque en santé animale, F-44307 Nantes, France

LUNAM Université, Oniris, F-44307 Nantes, France

CODA-CERVA, Groeselenberg 99, 1180 Brussels, Belgium

Ghent University, Laboratory of Veterinary Immunology, Merelbeke, Belgium

Utrecht University, Department of Farm Animal Health, Utrecht, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: This study aimed to evaluate the use of routinely collected reproductive and milk production data for the

Received 20 April 2015

early detection of emerging vector-borne diseases in cattle in the Netherlands and the Flanders region of

Received in revised form 2 December 2015

Belgium (i.e., the northern part of Belgium). Prospective space-time cluster analyses on residuals from a

Accepted 11 December 2015

model on milk production were carried out to detect clusters of reduced milk yield. A CUSUM algorithm

was used to detect temporal aberrations in model residuals of reproductive performance models on two

Keywords:

indicators of gestation length. The Bluetongue serotype-8 (BTV-8) epidemics of 2006 and 2007 and the

Early-warning

Schmallenberg virus (SBV) epidemic of 2011 were used as case studies to evaluate the sensitivity and

Syndromic surveillance

Vector-borne timeliness of these methods. The methods investigated in this study did not result in a more timely

Cattle detection of BTV-8 and SBV in the Netherlands and BTV-8 in Belgium given the surveillance systems

in place when these viruses emerged. This could be due to (i) the large geographical units used in the

analyses (country, region and province level), and (ii) the high level of sensitivity of the surveillance

systems in place when these viruses emerged. Nevertheless, it might be worthwhile to use a syndromic

surveillance system based on non-speciﬁc animal health data in real-time alongside regular surveillance,

to increase the sense of urgency and to provide valuable quantitative information for decision makers in

the initial phase of an emerging disease outbreak.

1. Introduction emerging diseases is of major importance to minimize their impact

on animal welfare, animal trade and the costs associated with an

Vector-borne emerging or re-emerging diseases can spread over outbreak. Two vector-borne diseases have emerged and spread

large areas having major consequences for the livestock industry. In throughout the north-western part of Europe in the last decade:

cattle, vector-borne diseases can be difﬁcult to detect at the early Bluetongue virus serotype-8 (BTV-8) and the Schmallenberg virus

onset of the disease when they result in non-speciﬁc symptoms, (SBV). BTV-8 emerged in August 2006 and re-emerged in July 2007

such as fever, loss of appetite and drop in milk production. They may and affected animal welfare and caused important economic losses

therefore go unnoticed as such symptoms can be misinterpreted as (Wilson and Mellor, 2009). At the end of the vector-active season

the result of common endemic diseases, environmental conditions in 2006, the southern provinces and some central provinces of the

or reduced feed quality. However, early detection of vector-borne Netherlands were affected by BTV-8 (Van Schaik et al., 2008). In

Belgium, BTV-8 had affected all provinces by January 2007, with a

gradient of decreasing seropositivity toward the south and the west

of the country (Méroc et al., 2008). SBV emerged in Europe in the

∗

Corresponding author. Fax: +31 0570660354. late summer of 2011, causing diarrhea and drop in milk production

E-mail address: [email protected] (A. Veldhuis).

http://dx.doi.org/10.1016/j.prevetmed.2015.12.006

16 A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24

in adult cattle (Muskens et al., 2012) and congenital malforma- 2. Materials and methods

tions in newborn ruminants (Hoffmann et al., 2012). SBV spread

throughout the Netherlands and Belgium in one vector-active sea- 2.1. Data collection and validation

son leading to high herd prevalences (Méroc et al., 2013; Veldhuis

et al., 2013). Given their vector-borne nature, the spread of viruses 2.1.1. Milk production data

like BTV and SBV is not limited by territorial borders and therefore Milk recording is the process of measuring milk yield and

difﬁcult to prevent. In addition, regarding SBV, control measures composition of (all) individual lactating cows on a regular basis,

such as animal movement restrictions are expected to have little usually monthly, as a decision-support tool for farmers and the

effect on the spread of the disease (Gubbins et al., 2014). cattle breeding programme. Milk recording data collected in the

Surveillance systems for early warning of emerging animal dis- Netherlands and the Flanders region of Belgium were obtained from

eases are often based on passive surveillance that consists of clinical the cattle improvement company CRV. CRV carries out milk record-

diagnosis in the ﬁeld and on the reporting of suspect cases. Elbers ing on a daily basis (week days only), covering 80–85% of Dutch and

et al. (2008) concluded that BTV-associated clinical signs in cat- approximately 48% of Belgian dairy herds each month. Geograph-

tle (and sheep) are important in the recognition of BTV, but as a ical location data were obtained from GD Animal Health and from

diagnostic tool they result in a maximal sensitivity of 67% and speci- the SANITEL (National animal identiﬁcation and registration system

ﬁcity of 72% in cattle (based on a combination of speciﬁc clinical in Belgium) database for Dutch and Flemish dairy herds, respec-

signs). The performance of passive (clinical) surveillance systems tively. Unfortunately, in Flemish data 76% of the milk production

for emerging diseases with non-speciﬁc clinical signs is likely to be records were not associated with herd location data. Therefore, no

lower, in particular if the disease is unknown or if animal health space-time analyses could be carried out on such data. Milk pro-

observers are unfamiliar with the disease. The efﬁciency of such duction data from Flanders are therefore not further described in

a system depends on the awareness of the farmer and/or the vet- this manuscript.

erinarian and the willingness to report a clinical suspicion to the Milk recording data from cows from 1 to 305 days in milk from

authorities. Therefore, the implementation of a surveillance system the Netherlands were provided from January 1, 2003 to March

based on non-speciﬁc herd production data may be more objective 31, 2012 and included an anonymous unique farm identiﬁcation

and may, in addition, support clinical observations from the ﬁeld. number, location, date of milk recording (i.e., test-day), cow iden-

This type of surveillance is known as syndromic surveillance. Syn- tiﬁcation numbers and milk production per cow per test-day (kg).

dromic surveillance systems are generally based on health-related Milk production data were aggregated at herd and test-day level to

information (often clinical signs) or information obtained by pas- obtain a mean milk production per cow per herd per test-day (in

sive surveillance that might precede (or may substitute for) formal kg). Records with 0 kg milk were removed from the dataset (0.3%).

diagnosis (Hoinville et al., 2013). As a result, timeliness is an asset of Records with a mean milk production below 10 kg per cow or above

syndromic surveillance. Also, syndromic surveillance is not usually 50 kg per cow were removed from the dataset (0.2%), because of

focused on a speciﬁc hazard providing an ability to detect several possible recording errors. Herd location was unknown for 0.02% of

diseases of interest with similar symptoms. These characteristics the data. These data were excluded from further analyses.

make syndromic surveillance most applicable for early-warning

surveillance (Hoinville et al., 2013). In veterinary health, the appli- 2.1.2. Reproduction data

cation of syndromic surveillance is frequently based on clinical data Routinely collected data on reproductive events of cows in the

from practitioners and laboratory data, although a number of other Netherlands and Flanders, such as artiﬁcial inseminations (AI) and

data sources are being explored (Dórea et al., 2011; Perrin et al., dates of calving were supplied by CRV. Each dataset contained

2010). Two recent examples in the ﬁeld of cattle health surveil- (anonymous) unique herd numbers, location at postal district and

lance illustrate the potential of non-speciﬁc herd productivity data province level (data from the Netherlands only), cow identiﬁcation

for veterinary syndromic surveillance. Madouasse et al. (2014) eval- numbers, parity numbers, insemination dates and calving dates.

uated milk yield from milk recording in dairy cattle as an indicator A maximum of 10 inseminations per parity was set to exclude

to be included in an emerging disease surveillance system. Marceau repeated inseminations carried out for embryo transfer procedures.

et al. (2013, 2014) described the use of routinely recorded repro- Data from nulliparous heifers were excluded. Also, data from cat-

ductive events as indicators of disease emergence in dairy cattle. tle with parity >8 were excluded, in agreement with Marceau et al.

Both studies used data to retrospectively detect the BTV-8 outbreak (2014). Data from the Dutch National Identiﬁcation and Registra-

in France in 2007–2008, i.e., the years when these performances tion database were used to obtain each animal’s date of arrival in

had been shown to be impaired by the epidemics (Nusinovici et al., the herd and date of removal (data from the Netherlands only). For

2012). In the Netherlands, the BTV-8 and the SBV epidemics had a the Netherlands, data were available for the period April 1, 2006

negative effect on reproductive performance of cattle and milk pro- to December 31, 2011. This period included the BTV epidemics in

duction (Santman-Berends et al., 2010a; Santman-Berends et al., 2006 and 2007 and the SBV epidemic in 2011 (Table 1). For Flan-

2011; Veldhuis et al., 2014). However, the clinical signs induced ders, data were available for the period January 1, 2005– June 30,

by these viruses differed as well as their transmission parameters 2011, including the BTV epidemic.

(Gubbins et al., 2014). Nevertheless, they provide an opportunity

to validate the potential of syndromic surveillance on routinely 2.2. Data analysis

collected production data for early detection purposes. This paper

describes the application of the methods described by Marceau 2.2.1. Milk production data

et al. (2013, 2014) and Madouasse et al. (2014) on routinely col- To capture ﬂuctuations in the seasonal herd-level milk produc-

lected reproductive and milk production data from the Netherlands tion baseline, a linear mixed model as described by Madouasse

and the northern region of Belgium (Flanders). The aim of this study et al. (2014) was used. Brieﬂy, herd-test-day milk production was

was to evaluate whether these non-speciﬁc data sources can be modeled as a function of the day of the year at which milk record-

used to deﬁne indicators for the early detection of emerging vector- ing occurred. Seasonal differences were accounted for using linear

borne diseases in cattle in the Netherlands and Flanders. By doing splines, by splitting data in eight time periods deﬁned by knots

so, the external validity of the approaches developed by Marceau placed at days 60, 90, 120, 150, 200, 250 and 300 of the year. Each

et al. (2013, 2014) and Madouasse et al. (2014) is assessed. of the eight change points was included as both ﬁxed and random

effects in the model. The coefﬁcients were the estimated milk pro-

A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24 17

Table 1

Dates of ﬁrst notiﬁcation of BTV-8 and SBV suspicions in cattle in the Netherlands and Belgium.

Epidemic The Netherlands Belgium

First report Source of information First report Source of information

BTV-8 August 14, 2006 Reporting of clinical signs/suspicion ∼June 1, 2006 Reporting of clinical signs/suspicion

BTV-8 (re-emergence) July 20, 2007 Sentinel study July 30, 2007 Reporting of clinical signs/suspicion

SBV August 25, 2011 Reporting of clinical signs/suspicion December 4, 2011 Reporting of clinical signs/suspicion

Not conﬁrmed by PCR testing.

Table 2

Dates used to deﬁne the baseline and prediction periods for the detection of low milk production clusters in cattle in the Netherlands.

BTV-8 January 1, 2005–December 31, 2005 January 1, 2006–December 31, 2007

Model Baseline period Prediction period

SBV January 1, 2009–December 31, 2010 January 1, 2011–March 31, 2012

Table 3

ductions in kg at these change points. The outbreak of BTV-8 in

Boundaries of the intervals used to deﬁne the premature calving and short gestation

2006 (from August 14th to December 15th) and 2007 (from July

indicators in the Netherlands and Belgium. Both indicators were based on the inter-

20th to December 15th) and the outbreak of SBV in 2011 (start-

val between the last known AI date and calving date. The boundaries were based on

ing around August 25th but conﬁrmed on December 8th, 2011) the observed percentiles of the interval distributions for both countries during the

were used in separate models to test the sensitivity and timeli- non-epidemic periods.

ness of the milk production indicator for the detection of these

Indicator Gestation length boundaries (days)

epidemics. Model parameters were estimated from data from years

Premature calving: (P1 of the frequency distribution—40 days) to P1

without epidemics (‘baseline period’) and were then used to pre-

The Netherlands 226–266

dict milk production for the years in which the BTV-8 and SBV

Belgium 225–265

epidemics occurred (‘prediction period’). The dates used to deﬁne a

Short gestation: P1 to P25 of the frequency distribution

each period are provided in Table 2. Differences between observed

The Netherlands 267–278

and predicted milk production per herd-test-day were determined Belgium 266–277

for the prediction periods, aggregated by postal district and sent a

P1: 1% percentile. P25: 25% percentile.

to SaTScan (Kulldorff, 2009) to identify space-time clusters of

low milk production. To clarify, the Netherlands is comprised of

2 Table 4

90 postal districts (2-digit level) with an average size of 387 km .

Dates used to deﬁne the baseline and prediction periods for the construction of time

Prospective analyses were run on 5-week moving windows using

series of short gestation and premature calving rates of cattle in the Netherlands and

a normal probability model. In SaTScan, the differences between Belgium.

observed and predicted productions were weighted by the square

Model Baseline period Prediction period

root of the number of cows recorded. The maximum spatial clus-

The January 1, 2009–December April 1, 2006–December 31,

ter size was set at 10% of the population at risk. This relatively low

Netherlands 31, 2010 2007 (BTV-8 ) + January 1,

cluster size was chosen as the spatial scan statistic was used for

2011–March 31, 2012 (SBV)

early detection of vector-borne diseases. The temporal cluster size Belgium January 1, 2005–December January 1, 2006–December 31,

was set at four weeks, with one week used as background. A circular 31, 2005 + January 1, 2007 (BTV-8)

2009–June 30, 2011

window shape was chosen. We scanned for ‘low mean’ clusters, i.e.,

lower observed milk production than would have been expected. Because of limiting data availability, prediction could only be done to detect the

re-emergence of BTV-8 in 2007.

For each window, a likelihood-ratio test statistic was calculated and

As SBV emerged in the late summer of 2011, no predictions could be done to

the window with the maximum value was the most likely cluster.

detect the emergence of SBV in Belgium.

A p-value was assigned to each tested window using Monte Carlo

hypothesis testing (999 simulations). Clusters of low milk produc-

tion were deﬁned as windows with a p-value ≤ 0.05. Clusters of low mal gestation lengths due to abortion or erroneous recording (i.e.,

milk production that were identiﬁed outside the prediction periods lower than 260 days and greater than 310 days) were (temporar-

were considered false alerts. ily) removed from the dataset before P1 and P25 were calculated, to

capture parameters describing the normal symmetric distribution

2.2.2. Reproduction data of gestational length. For that same reason, data recorded over a

Two indicators of reproductive events were built for dairy cows period without any major epidemics (‘baseline period’) were used

(i.e., parity > 0 at AI) as described in Marceau et al. (2014). Brieﬂy, to obtain P1 and P25. An overview of the baseline and prediction

the indicators ‘premature calving’ and ‘short gestation’ were con- periods for both countries is provided in Table 4.

structed based on gestation length, i.e., the interval between the Daily rates for each indicator were calculated per country as the

last recorded AI before calving and calving date. These indicators number of indicator events on a given day related to the popula-

were chosen according to their relation to particular reproduction at risk on that day. For the Netherlands, daily indicator rates

tive disorders such as abortion or stillbirth (leading to premature were also calculated per province, with similar parameter settings

calving) or health disturbances at the end of gestation (potentially as at country level, to assess a possible increase in sensitivity by

inducing calving a few days earlier than expected). Indicator rates applying a smaller geographical unit. This exercise could not be

for the Netherlands and Flanders could not be combined in one carried out for Flanders, because data were available for the whole

model due to differences in geographical units between the two region of Flanders without identiﬁcation of provinces. A 7-day mov-

countries. Table 3 displays an overview of the indicators and corre- ing average of the daily rate was calculated for each day to reduce

sponding time intervals. Interval boundaries of the indicators were the variability observed on a daily basis due to a low size of the

chosen according to the 1% and 25% percentiles (P1 and P25) of population at risk. Time series of the moving average daily rates

the observed frequency distribution of gestation lengths. Abnor- of premature calving and short gestation were modeled using a

18 A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24

harmonic linear regression model. Details about the model can be −0.04 kg in 2006, +0.20 kg in 2007, +0.03 kg in 2011 and +0.12 kg in

found in Marceau et al. (2014). Brieﬂy, the model contained an the ﬁrst three months of 2012.

intercept and two sine/cosine harmonics as predictors, to take into In 2006, two signiﬁcant low milk production clusters were found

account annual seasonality of reproductive events. The number of in the southern part of the Netherlands (Table 5; cluster A and B).

harmonics which best ﬁtted the observed data according to the This part of the Netherlands was affected by BTV-8 in 2006 (Van

AIC criterion was chosen. The models were ﬁtted on the baseline Schaik et al., 2008). However, these alerts were 2–3 months later

period that corresponded to a period without any major epidemics than the first official notifications of the BTV-8 emergence (Table 5).

(Table 4). Then, expected indicator rates for premature calving and No signiﬁcant clusters of low milk production were found in 2007

short gestation were estimated for the corresponding prediction when BTV-8 re-emerged in the Netherlands and spread further

period. Our model could not be used to detect the BTV-8 epidemic toward the north of the country. In the period from January 1,

of 2006 in the Netherlands as data on reproductive events from the 2011 to March 31, 2012, ﬁve signiﬁcant low milk production clus-

Netherlands were available from April 1st 2006 onwards and BTV- ters were found (Table 5; cluster C–G). Cluster C was found in a

8 emerged in north-western Europe in August 2006 (OIE, 2013). period in which it was unlikely that SBV had emerged yet. The sec-

As SBV was most likely introduced in the late summer of 2011 (De ond SBV cluster (D) was found in the period in which numerous

Regge et al., 2012; Veldhuis et al., 2013), Belgian data (available up reports of sudden drops in milk production and diarrhea in dairy

to June 30, 2011) could not be used to detect the SBV epidemic. cattle were made by veterinarians (starting around August 25th,

To detect anomalies in the daily rate of indicator events, a 2011). Cluster E was found in the period during which SBV quickly

one-sided positive CUSUM function was used to calculate the spread among Dutch dairy cattle. Clusters F and G were found late

cumulative sum of differences between observed and predicted 2011/early 2012, a period in which no drop in milk production was

daily indicator rates, as described by Marceau et al. (2013) (Eq. (1)): observed at country level (Fig. 1). Nine signiﬁcant clusters for low

milk production were found outside the prediction periods (six in

Cusumt = max 0, Cusumt−1 + Yt − Yˆt − k (1) 2009 and three in 2010) and were located in four different areas.

These were considered as false alerts in the context of detecting

where t is the time unit in days, Yt is the observed indicator rate BTV-8 or SBV.

at day t, Yˆt is the predicted indicator rate at day t and k is a ref-

erence value to explain the variation of the mean of the baseline 3.2. Reproduction data

period. At t0, the CUSUM is set at 0. Once Yt − Yˆt (i.e., the residual)

exceeds k, a change in the CUSUM value has been found. If then, 3.2.1. The Netherlands

for example, the reference value is not exceeded the next day, the In the Netherlands, a mean gestation length of 281.4 days was

CUSUM value is set at 0 again. The level of k was set at the 95% observed for cows during the baseline period. In that period, the

percentile value of the residuals in the baseline period. Also, we mean daily premature calving rate was 0.00065%. In the prediction

adapted the CUSUM function to a moving time lag of 14 days. By period, the mean daily premature calving rate was 0.00068%. The

doing so, impact of past anomalies was restricted to 14 days. Once mean daily short gestation rate was 0.025% and 0.024% in the base-

all the CUSUM values were calculated for each dataset, a threshold line period and prediction period, respectively. In the summer of

value (h) was applied as decision limit to trigger an alert. Different 2007, when BTV-8 re-emerged in the Netherlands, the daily rate

values of h were applied, exploring the optimal balance between of premature calving in cows was elevated leading to an increase

the algorithm’s sensitivity, timeliness and speciﬁcity. Values of h in the cumulative sum of differences between observed and pre-

were chosen as a function of k : 3k, 5k, 7 k and 10k . The ﬁrst day dicted rates (Fig. 2). Applying the detection threshold h resulted in

on which a CUSUM value exceeds the decision limit was set as the a ﬁrst alert on August 31, 2007 (Table 6). This is a delay of 42 days

day of the alert. When the h-value was set at 3k, more alerts were compared to the detection of BTV-8 by the surveillance system in

generated compared to 5k, leading to a greater false positive rate. place at the time. No increase in premature calving rates could be

In this paper, alerts based on a threshold value of 5k are reported. observed in the late summer of 2011, when SBV emerged. Daily

Alerts outside the prediction periods were considered false alerts. rates of short gestations were elevated during the two epidemics

(BTV-8 and SBV) in the Netherlands (Fig. 2). This resulted in alerts on

the 9th of September 2007 and August 29, 2011 (Table 6). One alert

3. Results was generated outside the prediction period with the short gesta-

tion indicator (in April 2009) and one with the premature calving

3.1. Milk production data indicator (in March 2008). These alerts were considered false alerts

in the context of detecting BTV-8 or SBV.

3.1.1. The Netherlands Premature calving and short gestation rates were also cal-

The number of milk recording dairy herds in the Netherlands culated per province to assess a possible increase in timeliness

decreased from 19,016 in 2005 to 16,983 in 2011. During that by applying a smaller geographical unit. Daily premature calv-

period, milk production and composition was measured on average ing rates were elevated in 7 out of 12 provinces during the BTV

10.2 times per dairy herd per year. The mean number of lactating period (Apr. 1, 2006–Dec. 31, 2007) and resulted in 15 alerts (all

cows per herd was 63.1 for the period 2005–2011. It increased from in 2007). The ﬁrst alert was obtained in the western province of

55.4 cows in 2005 to 69.7 cows in 2011. Overall, the mean num- Zuid-Holland on April 6th, 2007, which was considered a false

ber of tested cows per week was 221,337 in the period 2005–2011 positive alert as there is no knowledge of any major disease out-

with a minimum of 91,486 cows per week. The observed mean breaks at that time. The subsequent alerts were on August 19th

daily milk production per cow decreased from 25.7 kg in 2005 and (province of Limburg), August 28th (Noord-Brabant and Gelder-

2006 to 25.5 kg in 2007 and 25.3 kg in 2008. From 2009, the mean land), September 1st (Zuid-Holland), September 10th (Utrecht)

daily milk yield per cow remained constant in time with 25.8 kg. and September 16th (Overijssel) (Fig. 3). Based on the short ges-

Based on observed milk production in 2005, milk production was tation rate per province, 11 elevations were detected in 9 out of

predicted for the years 2006 and 2007. In addition, observed milk 12 provinces in the prediction period associated with BTV. The

productions in 2009 and 2010 were used to predict milk produc- ﬁrst alert was found in the province of Noord-Brabant on August

tions for 2011 and the ﬁrst three months of 2012 (Fig. 1). The mean 16th, 2007 (Fig. 3). Further alerts in August were found in the

difference between observed and predicted milk productions were provinces Limburg (August 20th) and Zuid-Holland (August 29th).

A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24 19

Fig. 1. Country-level mean 7-day moving average milk production per cow per test-day in the Netherlands in the period January 1st 2005–December 31st 2007 (left) and

January 1st 2009–March 31st 2012 (right), with observed milk production (grey solid line), ﬁtted milk production (solid black line) and predicted milk production (black

dashed line). Grey vertical bars represent detection dates of the BTV-8 epidemics (left) and SBV epidemic (right) by surveillance systems in place at the time.

Table 5

Overview of signiﬁcant (P-value < 0.05) detected clusters by SaTScan of decreased milk production in the Netherlands in 2006–2007 (BTV period) and 2011–2012 (SBV

period), with time period, radius (km), P-value, geographical location and delay in days between actual ﬁrst report of the epidemic by surveillance components in place and

the ﬁrst day of the space/time window of each cluster.

Model Cluster Time period Cluster radius (km) P-value Location Detection delay (days)

BTV- A October 7–13, 2006 17.2 0.032 South 54

8 B November 11–17, 2006 31.6 0.026 South 89

SBV C April 22–May 19, 2011 14.6 0.011 Centre −125

D August 5–September 1, 2011 31.4 0.002 Centre −20

E September 23–October 20, 2011 0.0 0.026 South-East 29

F December 16, 2011–January 14, 2012 0.0 0.028 West 113

G January 8–February 4, 2012 15.7 0.008 West 136

A cluster with a radius of 0 km indicates it comprised one (two-digit) postal code area.

Fig. 2. Country-level mean 7-day moving rate of premature calvings (left) and short gestations (right) in cows in the Netherlands between January 1, 2007 and December 31,

2011, with observed rates (solid light grey line), ﬁtted and predicted rates (solid black and solid dashed line) and CUSUM values (solid dark grey line). Threshold h is presented

by a horizontal dashed line. Grey vertical bars represent ﬁrst detection dates of BTV-8 and SBV by surveillance systems in place at the time (2007 and 2011, respectively).

Table 6

Detection dates of elevations in the daily rate of reproductive indicators in the Netherlands and Belgium and the corresponding delay in days between actual ﬁrst report of

the BTV-8 and SBV epidemics by surveillance components in place and dates of elevations.

The Netherlands Belgium

Model Indicator Detection date Detection delay Detection date Detection delay

BTV-8 Premature calving n.a. n.a. No alerts –

(2006) Short gestation n.a n.a. No alerts –

BTV-8 Premature calving August 31, 2007 42 August 17, 2007 18

(2007) September 27, 2007 59

Short gestation September 9, 2007 51 September 1, 2007 33

SBV Premature calving No alerts – n.a. n.a.

Short gestation August 29, 2011 4 n.a n.a.

20 A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24

Fig. 3. Location of alerts between April 2006–December 2007 (A, B) and January 2011–March 2012 (C, D) based on the premature calving indicator and short gestation

indicator at province level for the Netherlands. Chronology of alerts is indicated with decreasing color intensity (with white indicating no alert). Province names are indicated

in C: FR, Friesland; GR, Groningen; DR, Drenthe; FL, Flevoland; OV, Overijssel; GE, Gelderland; LB, Limburg; NB, Noord-Brabant; ZE, Zeeland; NH, Noord-Holland; UT, Utrecht;

ZH, Zuid-Holland.

In September 2007, alerts were found in the provinces Gelderland province, a total of 14 elevations were detected in 12 out of 12

(September 1st), Utrecht (September 8th), Overijssel (September provinces in the prediction period associated with SBV (Fig. 3).

10th) and Noord-Holland (September 23rd). On October 30, 2007, The ﬁrst alert that could be associated with the SBV outbreak

a ﬁnal alert was found in the province of Drenthe. Irrespective of was found in the province Overijssel on August 29th, 2011. A sec-

indicator, no alerts were obtained in 2007 from the most northern ond alert was found in Gelderland, on August 30th, 2011. Then,

provinces of the Netherlands (Friesland and Groningen) and the alerts were found in Utrecht (August 31st), Flevoland (September

province of Flevoland. 1st), Noord-Brabant (September 4th), Friesland (September 5th),

In the prediction period associated with the SBV epidemic (Jan. Noord-Holland (September 5th), Drenthe (September 6th), Zuid-

1, 2011–Mar. 31, 2012) premature calving rates were elevated Holland (September 6th), Groningen (September 10th) and Zeeland

twice and were obtained from 2 out of 12 provinces (Fig. 3). The (October 17th).

ﬁrst alert that could be associated with the SBV outbreak was Over the complete study period, the mean number of false alerts

found in the province Noord-Brabant on September 13th, 2011. per province (i.e., alerts outside prediction periods) was 1.8 for

A second alert was found in Limburg, on September 16th, 2011. the premature calving indicator (range: 0–8) and 1.4 for the short

No other alerts were found in 2011 or 2012 based on province- gestation indicator (range: 0–3). In general, the timeliness of the

level premature calving rates. Based on short gestation rates per algorithm to detect elevations in short gestations and premature

A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24 21

calvings that could be associated with the BTV-8 and SBV epidemics production may have gradually increased to become detectable

increased slightly by reducing the geographical unit to province months after the first notification. Yet, no significant drop in milk

level. More speciﬁcally, the timeliness of detection of an elevation production was detected during the re-emergence of BTV-8 in the

in reproductive indicators improved in particular for the detection Netherlands (2007), which is somewhat surprising as the mean

of BTV-8 in 2007 (from August 31st to August 19th, 2007). within-herd seroprevalence in cattle herds in the southern and cen-

tral part of the Netherlands increased rapidly in the summer of 2007

3.2.2. Flanders (Santman-Berends et al., 2010b). Nonetheless, timeliness and sen-

During the baseline periods, a mean gestation length of 280.9 sitivity of detection may be improved by accounting for more of

days was observed in cows in Flanders. In that period, the mean the residual variability in our models. It is likely that a large part

daily premature calving rate was 0.00066%. In the prediction period, of the unexplained variability in milk production is caused by fac-

the mean daily premature calving rate was 0.00062% in cows. The tors such as climate, feed quality and feed price, which were not

mean daily short gestation rate during the baseline period was taken into account in this study. Also, milk production records were

0.025% and 0.022% in the prediction period. aggregated at herd level prior to analysis; a possible drop in milk

Premature calving rates were elevated in the summer of 2007, production on individual cow level caused by BTV-8 is probably

corresponding to the re-emergence of BTV-8 in Belgium, but could lost in the normal variation in milk production on herd level. In

not be observed in the summer of 2006 when BTV-8 was ﬁrst addition, a relatively short baseline of one year (2005) was used

detected in Belgium (Fig. 4). Applying the threshold to the CUSUM to predict milk production for 2006 and 2007, which was perhaps

values resulted in alerts on August 17, 2007 and September 27, 2007 not long enough to capture seasonal and environmental ﬂuctua-

(Table 6), which is 17 and 58 days later than the ﬁrst conﬁrmed case tions in milk production. Another explanation could lie in the fact

following the surveillance that was in place (July 31, 2007). A third that milk production data based on monthly milk recording were

alert was obtained on November 18, 2010, which was considered used in this study, providing monthly observations on herd level.

an aspeciﬁc alert due to the absence of a BTV-8 epidemic, yet a local As an alternative, the use of bulk milk collection data (resulting in

outbreak of Brucellosis occurred at that time in Belgium (OIE, 2014). approximately 3 observations per herd per week) might increase

The pattern of the rate of short gestations in Flanders is comparable the timeliness and sensitivity of the syndromic surveillance system

to the pattern of premature calvings (Fig. 4), apart from that only proposed in this study, due to a larger daily (or weekly) coverage.

one alert was obtained, on September 1st, 2007 (Table 6). This is A signiﬁcant drop in milk production was found in the

33 days after the initial detection by the surveillance systems in Netherlands at the start of the SBV outbreak. The cluster was found

place at the time. No alert was generated in July or August 2006 on September 1st 2011 in the eastern part of the Netherlands, seven

that might indicate the start of the epidemic in Belgium. days after the day that farmers started to report a severe drop in

milk production in dairy cattle in that area (Muskens et al., 2012).

4. Discussion However, our analyses also resulted in a number of clusters in early

2011 and early 2012 that are not likely to be associated with SBV or

The objective of this study was to evaluate whether routinely another emerging disease outbreak, indicating a lack of speciﬁcity.

collected cattle production data can be used to build indica- This may be improved if the model used to predict the expected

tors for early detection of emerging vector-borne diseases in the milk production were to be extended with variables explaining

Netherlands and the northern part of Belgium (Flanders). The BTV- climatological factors and feed prices.

8 epidemics of 2006 and 2007 and the SBV epidemic of 2011 were When Madouasse et al. (2013) evaluated milk yield as an indica-

used as case studies. In the last decade, several applications of syn- tor for syndromic surveillance through simulation, the timeliness

dromic surveillance have been described to enhance surveillance of detection depended mostly on how easily the disease spread

of both public and veterinary health. Most of these applications between and within herds. This is in agreement with our ﬁnd-

in veterinary public health were based on clinical data from prac- ings: the analysis of milk production data may trigger an alert in a

titioners and laboratory data, although a number of other data surveillance system even when the impact of the disease on milk

sources are being explored (Perrin et al., 2010; Dórea et al., 2011; production is limited, provided that the disease spreads fast (as was

Dupuy et al., 2015). This work demonstrates that data collected for the case with SBV, but not with BTV-8).

purposes other than surveillance can supplement, yet not replace,

traditional targeted and passive surveillance systems. Milk yield 4.2. Reproductive performance as indicator

from milk recording and artiﬁcial insemination data show different

advantages and drawbacks in this regard. The gestation-based reproductive indicators used in this study

have the potential to add value to existing passive surveillance

4.1. Milk production records as indicator strategies in the Netherlands and Belgium (Flanders) to detect

emerging diseases in cattle similar to SBV but not BTV-8. The lat-

Spatiotemporal cluster analyses using milk production records ter could be due to differences in epidemiological characteristics

were able to detect the SBV epidemic in a timely manner but not between SBV and BTV-8. BTV-8 had a higher impact on reproductive

the BTV-8 epidemic. The ﬁrst signiﬁcant cluster of low milk produc- performance (and milk production) than SBV (Santman-Berends

tion during the 2006 BTV-8 outbreak came two to three months et al., 2010a; Santman-Berends et al., 2011; Veldhuis et al., 2014).

later than the detection of the epidemic by current surveillance The speed at which SBV spread geographically however, was quite

components. This is consistent with the ﬁndings of Madouasse different from BTV-8. At the end of the vector-active season in

et al. (2014), who detected a signiﬁcant drop in milk production 2006, the southern provinces and some central provinces of the

seven weeks after the ﬁrst notiﬁcation of BTV-8 based on clini- Netherlands were affected by BTV-8 (Van Schaik et al., 2008). In

cal signs in the less dense French dairy population. Although these Belgium, BTV-8 had affected all provinces by January 2007, with a

alarms occurred late after the emergence, the clusters identiﬁed gradient decreasing seropositivity toward the south and the west

are likely consequences of BTV-8 infections. A major reason for of the country (Méroc et al., 2008). SBV, on the contrary, spread

this long delay could be that clinical signs associated with BTV- throughout the Netherlands and Belgium in one vector-active sea-

8 are obvious resulting in the ﬁrst cases being notiﬁed at a time son leading to high herd prevalences (Méroc et al., 2013; Veldhuis

when a small proportion of the cattle population was affected. As et al., 2013). In addition, the fact that no aberrations in reproductive

the diseased progressed through the population, the effect on milk indicators were observed in Belgium (Flanders) in July or August

22 A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24

Fig. 4. Country-level mean 7-day moving rate of premature calvings (left) and short gestations (right) in cows in Flanders in the period January 1, 2006–June 30, 2011, with

observed rates (solid light grey line), ﬁtted and predicted rates (solid black and solid dashed line) and CUSUM values (solid dark grey line). Threshold h is presented by a

horizontal dashed line. Grey vertical bars represent ﬁrst detection dates of BTV-8 by surveillance systems in place at the time (2006 and 2007).

2006 – indicating the start of the BTV-8 epidemic – may be due to tion dates and birth registrations. Insemination dates were only

the large geographical unit that was used for the analyses of Belgian available for herds that were participating in the national cattle

data. The unit of analysis was the Flanders region (i.e., the North of breeding scheme. An alternative reproductive performance indica-

Belgium) while the ﬁrst cases in 2006 were limited to the Walloon tor that is worth exploring is the rate of abortions, as viruses such as

region in the South of the country. BTV-8 and SBV are known to cause abortions. For example, Bronner

In 2007, a certain level of BTV-8 awareness was present after et al. (2015) used AI data to calculate the weekly incidence rate of

the primary outbreaks in 2006. Additional (temporary) surveil- midterm abortions in French dairy herds. They found that the mean

lance strategies led to the conﬁrmation of BTV-8 re-emergence in number of BTV-8 cases over a départment-speciﬁc time interval

the Netherlands on July 26th, 2007 and in Belgium on July 31st, was associated with an increase in the weekly incidence rate of

2007. The enhanced surveillance in place in 2007 might explain mid-term abortions in 47% and 71% of the départments for heifers

why the timeliness and added value of the reproductive indica- and parous cows, respectively. Marceau et al. (2014) also included

tors to detect BTV-8 in 2007 seem low. If the outbreak had been a ‘very late return-to-service’ indicator (indicative for reproductive

completely unexpected, there wouldn’t be a sentinel study or disorders in the early stage of gestation) in their study but they con-

mandatory reporting of clinical suspicions, which would have likely cluded that indicators based on gestation length are particularly

resulted in a greater added value of the reproductive indicators for interesting for early detection due to the limited delay between

the detection of the BTV-8 outbreak in the Netherlands and Belgium disease effect and the indicator of this effect.

in 2007. In the Netherlands, BTV-8 eventually spread toward the The system’s timeliness of detecting elevations in short gesta-

center and the north of the Netherlands in 2007, although preva- tions and premature calvings that could be associated with the

lences remained low in the Northern provinces by December 2007 BTV-8 and SBV epidemics increased slightly by reducing the size

(Santman-Berends et al., 2010b). The detection of BTV-8 in 2007 of the geographical unit (e.g., from country to province level). An

based on province-level short gestation rates and early calving rates additional reduction in size of geographical unit, for example to

followed a similar pattern from south to north, but lacked alerts in municipality or postal district level, might lead to an additional

the most Northern provinces. At province level, the sensitivity of increase in sensitivity and timeliness. This could particularly bene-

the indicators to detect the BTV-8 re-emergence in the Netherlands ﬁt the detection of diseases with a moderate speed of geographical

in 2007 was 50% (i.e., 6 out of 12 provinces) and 75% for premature spread, such as BTV-8. However, considering the rare incidence

calvings and short gestations, respectively. of premature calvings or short gestations at herd level, a further

A ﬁrst alert indicating the outbreak of SBV in the Netherlands decrease in geographical unit might have to be combined with an

was obtained on August 29th, 2011, in the province of Overijssel. increase in time unit to assure a sufﬁcient demographic coverage.

This alert was based on short gestation rates. The alert was four

days later than the ﬁrst suspicion of SBV obtained by the passive

4.3. Cross-border application

surveillance systems in place in the Netherlands, originating from

the same province on August 25th, 2011. The high speed of SBV’s

Intuitively, it seems interesting to combine data from multiple

geographical spread was shown by four consecutive alerts of ele-

countries into one surveillance system for vector-borne diseases.

vations in short gestation rates in four neighboring provinces of the

Within this study it was not feasible to combine data from Belgium

Netherlands. At province level, the sensitivity of the indicators to

and the Netherlands to due to differences in data availability (geo-

detect the emergence of SBV in 2011 was 17% and 100% for pre-

graphical and temporal). Moreover, it is questionable to which

mature calvings and short gestations, respectively. Thus, the short

extent it is desirable to apply syndromic surveillance in a cross-

gestation indicator seems most promising as an early indicator

border manner. First of all, intrinsic differences in data exist due

of emerging viruses affecting reproductive performance in cattle,

to differences in data collection and availability. Secondly, cat-

which is in agreement with ﬁndings of Marceau et al. (2014). The

tle populations differ between countries, for example breed in

biological relation between a short gestation and infection is not

composition, herd management practices and production circum-

known but can be due to the fever following infection that induces

stances, which cannot be accounted for from the available data.

parturition earlier. Alternatively, the viruses we used as case stud-

Even between regions within a country this may be an issue.

ies might have a direct effect on the gestation process in cattle.

In addition, increasing the geographical area to be monitored by

In our study, we calculated gestation lengths based on insemina-

aggregating data over countries might dilute the effect of a local dis-

A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24 23

ease outbreak, hampering the system’s sensitivity. Thus, it seems the investigated cases it would not have led to an increased sen-

sensible to maintain some level of spatial segregation when apply- sitivity. Nevertheless, it might be worthwhile to use a syndromic

ing one syndromic surveillance system covering multiple countries. surveillance system based on non-speciﬁc animal health data in

By doing so, baselines can be estimated separately for each coun- real-time alongside regular surveillance, to provide valuable quan-

try, region or province and data property issues might be avoided. titative information for decision makers in the initial phase of an

As an alternative, it would be of great value to share experiences emerging disease outbreak.

and choices regarding the data and the statistical models to be used

and share and follow-up results of the analyses on national datasets

Acknowledgements

between countries (e.g., model residuals).

This study was funded by the Dutch Ministry of Economic Affairs

4.4. Added value to conventional surveillance (EZ) and the Belgian Federal Agency for Safety of the Food Chain

(FASFC) in the framework of the European project EMIDA ERA-NET

The methods we investigated did not result in a more timely Early Detection Data.

detection of BTV-8 and SBV in the Netherlands and BTV-8 in

Belgium given the surveillance systems in place when these viruses

References

emerged. However, this might be different for other countries and

different between emerging diseases. Drawbacks of a syndromic

Bronner, A., Morignat, E., Hénaux, V., Madouasse, A., Gay, E., Calavas, D., 2015.

surveillance system as proposed in this study are the need for data

Devising an indicator to detect mid-term abortions in dairy cattle: a ﬁrst step

with sufﬁcient demographic and geographic coverage that are gath- towards syndromic surveillance of abortive diseases. PLoS One 10 (3), http://

dx.doi.org/10.1371/journal.pone.0119012.

ered at frequent time intervals, which might be difﬁcult to obtain

De Regge, N., Deblauwe, I., De Deken, R., Vantieghem, P., Madder, M., Geysen, D.,

within a short time period, and the occurrence of false-positive

Smeets, F., Losson, B., van den Berg, T., Cay, A.B., 2012. Detection of

alerts (in particular with milk production data). Schmallenberg virus in different Culicoides spp. by real-time RT-PCR.

Transbound. Emerg. Dis. 59, 471–475.

The magnitude of the added value of a syndromic surveillance

Dórea, F.C., Sanchez, J., Revie, C.W., 2011. Veterinary syndromic surveillance:

system as proposed in this study depends on (i) the availability and

current initiatives and potential for development. Prev. Vet. Med. 101 (1–2),

performance of conventional passive surveillance systems, (ii) the 1–17.

availability and demographic coverage of data usable for syndromic Dupuy, C., Morignat, E., Dórea, F., Ducrot, C., Calavas, D., Gay, E., 2015. Pilot

simulation study using meat inspection data for syndromic surveillance: use of

surveillance, (iii) the costs of the follow-up of signals from the

whole carcass condemnation of adult cattle to assess the performance of

syndromic surveillance system and (iv) the characteristics of the

several algorithms for outbreak detection. Epidemiol. Infect., 1–11, http://dx.

emerging disease. In countries without a sensitive passive surveil- doi.org/10.1017/S09502688140 03495, FirstView.

Elbers, A.R.W., Backx, A., Ekker, H.M., van der Spek, A.N., van Rijn, P.A., 2008.

lance system, the added value of syndromic surveillance will be

Performance of clinical signs to detect bluetongue virus serotype 8 outbreaks

greater, provided that data that allow the monitoring of the at risk

in cattle and sheep during the 2006-epidemic in The Netherlands. Vet.

population are available quickly, at a small cost. A follow-up pro- Microbiol. 129, 156–162.

Gubbins, S., Turner, J., Baylis, M., Van der Stede, Y., van Schaik, G., Cortinas

cedure to deal rapidly and effectively with the alerts must be in

Abrahantes, J., Wilson, A., 2014. Inferences about the transmission of

place. When the system’s speciﬁcity is imperfect, it is challenging

Schmallenberg virus within and between farms. Prev. Vet. Med. 116, 380–390.

to keep the costs of such a follow-up procedure at an acceptable Hoffmann, B., Scheuch, M., Höper, D., Jungblut, R., Holsteg, M., Schirrmeier, H.,

Eschbaumer, M., Goller, K.V., Wernike, K., Fischer, M., Breithaupt, A.,

level. A solution for this issue could be the application of a sys-

Mettenleier, T.C., Beer, M., 2012. Novel orthobunyavirus in cattle, Europe, 2011.

tem on multiple data sources simultaneously, after which in-depth

Emerg. Infect. Dis. 18 (3), 469–472.

investigation of alerts could be limited to combined alerts only. Hoinville, L.J., Alban, L., Drewe, J.A., Gibbens, J.C., Gustafson, L., Häsler, B.,

Saegerman, C., Salman, M., Stärk, K.D.C., 2013. Proposed terms and concepts for

Finally, regarding disease characteristics, the examples of BTV-8

describing and evaluating animal-health surveillance systems. Prev. Vet. Med.

and SBV are good illustrations of added values that differ between

112, 1–12.

two pathogens. With BTV-8, the clinical signs were obvious and led Kulldorff, M. and Information Management Services, Inc., 2009. SaTScan v8.0:

to a quick notiﬁcation and identiﬁcation of the etiological agent, Software for the spatial and space-time scan statistics. http://www.satscan.org.

Madouasse, A., Marceau, A., Lehebel, A., Brouwer-Middelesch, H., van Schaik, G.,

when a small proportion of the population was affected. In this case,

Van der Stede, Y., Fourichon, C., 2013. Evaluation of a continuous indicator for

the performance of syndromic surveillance insufﬁcient. The situa-

syndromic surveillance through simulation: application to vector borne

tion was different with SBV, for which the immediate clinical signs disease emergence detection in cattle using milk yield. PLoS One 8, e73726.

Madouasse, A., Marceau, A., Lehébel, A., Brouwer-Middelesch, H., van Schaik, G.,

were mild and non-speciﬁc. Using syndromic surveillance, signals

Van der Stede, Y., Fourichon, C., 2014. Use of monthly collected milk yields for

of reduced reproductive performance and milk production in the

the detection of the emergence of the 2007 French BTV epizootic. Prev. Vet.

Netherlands would have been generated within a week after the Med. 113, 484–491.

Marceau, A., Madouasse, A., Lehébel, A., Lesufﬂeur, T., Van der Stede, Y., Fourichon,

ﬁrst signal following passive surveillance. Therefore, even for coun-

C., 2013. Syndromic surveillance in dairy cattle: development of indicators and

tries with sensitive passive surveillance such as the Netherlands

methods based on reproduction data for early detection of emerging diseases.

and Belgium (Welby et al., 2013), it is likely that the alerts from Proceedings of the Annual Meeting of the Society for Veterinary Epidemiology

and Preventive Medicine, Madrid, 115–124.

multiple indicators would have generated an increased sense of

Marceau, A., Madouasse, A., Lehébel, A., van Schaik, G., Veldhuis, A., Van der Stede,

urgency in the initial stage of an epidemic, potentially triggering

Y., Fourichon, C., 2014. Can routinely recorded reproductive events be used as

authorities to initiate in depth investigation. indicators of disease emergence in dairy cattle?: an evaluation of ﬁve

indicators during the emergence of bluetongue in France in 2007–2008. J.

Dairy Sci. 97, 1–16.

Méroc, E., Faes, C., Herr, C., Staubach, C., Verheyden, B., Vanbinst, T.,

5. Conclusion

Vandenbussche, F., Hooyberghs, J., Aerts, M., De Clercq, K., Mintiens, K., 2008.

Establishing the spread of bluetongue virus at the end of the 2006 epidemic in

Syndromic surveillance based on milk production or reproduc- Belgium. Vet. Microbiol. 131, 133–144.

tive performance data may trigger an alert in the early phase of Méroc, E., Poskin, A., Van Loo, H., Quinet, C., Van Driessche, E., Delooz, L.,

Behaeghel, I., Riocreux, F., Hooyberghs, J., De Regge, N., Caij, A.B., Van den Berg,

an emerging disease outbreak, provided that the impact on these

T., Van der Stede, Y., 2013. Large-scale cross-sectional serological survey of

parameters and/or spread of the disease are similar to SBV and

Schmallenberg virus in Belgian cattle at the end of the ﬁrst vector season.

potentially also to those of BTV-8. Despite the issue of timeliness, Transbound. Emerg. Dis. 60, 4–8.

Muskens, J., Smolenaars, A.J.G., van der Poel, W.H.M., Mars, M.H., van Wuijckhuise,

a syndromic surveillance system as proposed in this study could

L., Holzhauer, M., van Weering, H., Kock, P., 2012. Diarree en productiedaling

have an added value in the Netherlands and Belgium, yet as com-

op Nederlandse melkveebedrijven door het Schmallenbergvirus. Tijdschr.

plement to traditional/regular passive surveillance systems as in Diergeneeskd. 137, 112–115.

24 A. Veldhuis et al. / Preventive Veterinary Medicine 124 (2016) 15–24

Nusinovici, S., Seegers, H., Joly, A., Beaudeau, F., Fourichon, C., 2012. Quantiﬁcation Santman-Berends, I.M.G.A., Hage, J.J., Lam, T.J.G.M., Sampimon, O.C., van Schaik, G.,

and at-risk period of decreased fertility associated with exposure to 2011. The effect of bluetongue virus serotype 8 on milk production and

bluetongue virus serotype 8 in naïve dairy herds. J. Dairy Sci. 95, 3008–3020. somatic cell count in Dutch dairy cows in 2008. J. Dairy Sci. 94 (3), 1347–1354.

OIE, World Organization for Animal Health, 2013. World animal health Van Schaik, G., Berends, I.M.G.A., van Langen, H., Elbers, A.R.W., Vellema, P., 2008.

information database interface. Summary of Immediate Notiﬁcations and Seroprevalence of bluetongue serotype 8 in cattle in the Netherlands in spring

Follow-Ups—2006. Available at; http://www.oie.int/wahis 2/public/wahid. 2007, and its consequences. Vet. Rec. 163, 441–444.

php/Diseaseinformation/Immsummary (accessed on 01.07.13.). Veldhuis, A.M.B., van Schaik, G., Vellema, P., Elbers, A.R.W., Bouwstra, R., van der

OIE, World Organization for Animal Health, 2014. World animal health information Heijden, H.M.J.F., Mars, M.H., 2013. Schmallenberg virus epi-demic in the

database interface. Report of an immediate notiﬁcation of Brucellosis (Brucella Netherlands: spatiotemporal introduction in 2011 and seroprevalence in

abortus) in Belgium—2010. Available at; http://www.oie.int/wahis 2/public/ ruminants. Prev. Vet. Med 112, 35–47.

wahid.php/Reviewreport/Review?reportid=10017 (accessed on 25.10.14.). Veldhuis, A.M.B., Santman-Berends, I.M.G.A., Gethmann, J.M., Mars, M.H., van

Perrin, J.B., Ducrot, C., Vinard, J.L., Morignat, E., Gaufﬁer, A., Calavas, D., Hendrikx, Wuyckhuise, L., Vellema, P., Holsteg, M., Höreth-Böntgen, D., Conraths, F.J., van

P., 2010. Using the National Cattle Register to estimate the excess mortality Schaik, G., 2014. Schmallenberg virus epidemic: impact on milk production,

during an epidemic: application to an outbreak of bluetongue serotype 8. reproductive performance and mortality in dairy cattle in the Netherlands and

Epidemics 2, 207–214. Kleve district, Germany. Prev. Vet. Med. 116, 412–422.

Santman-Berends, I.M.G.A., Hage, J.J., van Rijn, P.A., Stegeman, J.A., van Schaik, G., Welby, S., Méroc, E., Faes, C., De Clercq, K., Hooyberghs, J., Mintiens, K., Van der

2010a. Bluetongue virus serotype 8 (BTV-8) infection reduces fertility of Dutch Stede, Y., 2013. Bluetongue surveillance system in Belgium: a stochastic

dairy cattle and is vertically transmitted to offspring. Theriogenology 74 (8), evaluation of its risk-based approach effectiveness. Prev. Vet. Med. 112, 48–57.

1377–1384. Wilson, A.J., Mellor, P.S., 2009. Bluetongue in Europe: past, present and future.

Santman-Berends, I.M.G.A., Bartels, C.J.M., van Schaik, G., Stegeman, A.A., Vellema, Philos. Trans. R. Soc. B Biol. Sci. 364, 2669–2681.

P., 2010b. The increase in seroprevalence of bluetongue virus (BTV) serotype 8

infections and associated risk factors in Dutch dairy herds, in 2007. Vet.

Microbiol. 142, 268–275.