Use of register data to assess the association between use ofantimicrobials and outbreak of Postweaning MultisystemicWasting Syndrome (PMWS) in Danish pig herds

Hakan Vigre a,*, Ian R. Dohoo b, Henrik Stryhn b, Vibeke Frøkjær Jensen a

a The National Veterinary Institute, Technical University of Denmark, Bulowsvej 27, DK-1790 Copenhagen C, Denmarkb Department of Health Management, Atlantic Veterinary College, University of Prince Edward Island, 550 University Avenue, Charlottetown, PEI, Canada C1A 4P3

Preventive Veterinary Medicine 93 (2010) 98–109

A R T I C L E I N F O

Article history:

Received 20 October 2008

Received in revised form 9 October 2009

Accepted 16 October 2009

Keywords:

Pig

PMWS

Antimicrobial usage

Register data

Mixed models

A B S T R A C T

In 2001, the first case of Postweaning Multisystemic Wasting Syndrome (PMWS) was

reported in the Danish pig population. During subsequent years, the number of affected

farms increased exponentially. The aim of this study was to determine how this increase

influenced the use of antimicrobials between 2002 and 2004. We used national register

data of herd characteristics, antimicrobial usage and disease occurrence. The analysis

included data on antimicrobial usage in 3371 pig herds with weaners and 7434 pig herds

with finishers, which accounted for 56 and 82% of the national amount of antimicrobials

prescribed to weaners (prescribed by 347 practitioners) and finishers (prescribed by 522

practitioners), respectively.

The estimation of the effect of PMWS was done by comparing the amount of

antimicrobials (measured as Animal Defined Daily Doses (ADDkg) used per pig-day at risk

each month in each herd) used in herds before and after an outbreak of PMWS, and by

comparing the amount of antimicrobials used in herds experiencing PMWS with the

amount of antimicrobials used in herds not experiencing PMWS. The effects were

estimated in a three-level (veterinarian/herd/study-month) linear mixed regression

model with an autoregressive correlation of order 1 (AR1).

We found that after a herd had experienced an outbreak of PMWS, the antimicrobial

usage in weaners was increased for a year. During the first 3 months post outbreak the

usage increased by 22%, followed by an increase of 7% during the next 4th to 12th month

when compared to the pre-outbreak usage. There was a significant variation between

herds in this effect. Additionally, in herds experiencing an outbreak of PMWS, the usage of

antimicrobials before the outbreak was 37 and 19% higher in herds with weaners and

finishers, respectively, compared to herds not experiencing PMWS.

Generalisation of the results to the entire Danish pig population indicated that the

increase of PMWS infected herds from almost zero to about 20% during a 4-year period

resulted in a national increase of 4–5% in antimicrobials usage in weaners. The effect of

PMWS on usage of antimicrobials in finishers was unclear.

Contents lists available at ScienceDirect

Preventive Veterinary Medicine

journal homepage: www.elsevier.com/locate/prevetmed

� 2009 Elsevier B.V. All rights reserved.

1. Introduction

Use of antimicrobials in food-producing animals hasbecome a major concern in relation to the risk of

* Corresponding author. Tel.: +45 40 60 13 03; fax: +45 35 88 70 01.

E-mail address: hvig@food.dtu.dk (H. Vigre).

0167-5877/$ – see front matter � 2009 Elsevier B.V. All rights reserved.

doi:10.1016/j.prevetmed.2009.10.010

emergence and dissemination of resistant bacteria. As aresponse to this, the Danish government established amonitoring-system for surveillance of the veterinary use ofdrugs in production animals – VetStat (Stege et al., 2003).From 2001, VetStat has collected detailed data onantimicrobial usage in animals at the herd level. Nationalsummary statistics describing the amount and type of

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109 99

antimicrobials used at the species level in Denmark havebeen published (DANMAP, 2004).

According to the legislation in Denmark, use ofantimicrobials in animals is only allowed for treatmentof diagnosed infectious diseases and not for prophylaxis orgrowth promotion. Therefore, it can be assumed that thevariability with which infectious diseases manifest them-selves in animal production is the primary cause of trendsin antimicrobial prescription. Between 2001 and 2004 thenational use of antimicrobials in pigs increased by 22%(DANMAP, 2004), whereas the corresponding increase inthe size of the pig population was only 6% (StatisticsDenmark, 2004). The increased use of antimicrobials perpig produced suggests that the occurrence of diseasestreated with antimicrobials increased during this period.

In October 2001, the first case of PostweaningMultisystemic Wasting Syndrome (PMWS) was reportedin the Danish pig population. During subsequent years,the number of affected farms increased exponentially(Vigre et al., 2005). During the same period, the quantityof antimicrobials (kg active compound) used for pigs inDenmark increased by 5% in 2002 compared to 2001(DANMAP, 2002), and by 10% in 2003 compared to 2002(DANMAP, 2003). Based on a direct comparison at thenational-level between the usage of antimicrobials andoccurrence of PMWS, it has been speculated that theincreased amount of antimicrobials used in the industrymay be caused by the therapeutic treatment ofsecondary bacterial infections following outbreaks ofPMWS.

PMWS was first described in Canada in 1996 (Clark,1996). Since then, reports of PMWS have been made inalmost every pig producing country around the world.PMWS targets the immune system in weaners, resulting inslower growth rates (wasting), chronic disease andimmune dysfunction. The syndrome is associated with aporcine circovirus, but causality remains unclear andprobably involves other factors. Poor response to anti-microbial therapy has been reported (Madec et al., 2000;Trujano et al., 2001; Segales et al., 2004). However,experience with PMWS in the United Kingdom (Anon.,2002) and Denmark (DANMAP, 2003) suggests that farmsaffected with PMWS experience a great increase inincidence of secondary bacterial infections and conse-quently, the use of antimicrobials in farms with PMWS issubstantially increased.

The aim of this study was to determine if the increasedoccurrence of pig herds experiencing an outbreak of PMWSin the Danish pig population contributed to the increaseduse of antimicrobials at both the herd- and national-level.The study was based on national register data.

2. Materials and methods

2.1. Data sources

The study period in this project was from 1 January2002 until 31 December 2004. Information on use ofantimicrobials, number of pigs in different age groups,producer, herd health status, type of herd and occurrenceof the disease syndrome PMWS was extracted from

relevant national databases. Summary data about theentire pig population in Denmark, used as nationalreference data, was obtained from Statistics Denmark(2004).

2.1.1. Use of antimicrobials

Information on use of antimicrobials was obtained fromthe VetStat database (Stege et al., 2003). VetStat includedinformation on all antimicrobials prescribed for use inanimals in Denmark. Each prescription is represented by arecord, including information on day of purchase, drug(product identity and quantity), recipient (farm, animalspecies, age group (sows and piglets; weaners; finishers))and disease category (respiratory, gastro-intestinal; jointsand limbs; reproduction; udder; metabolic; other) andprescribing veterinarian. For each record, the amount ofthe commercial antimicrobial product was converted intoa unit measuring how many kilograms of pig that could betreated per day when the drug was used. The dosagerecommended by the pharmaceutical companies – AnimalDefined Daily Doses (ADDkg) (Jensen et al., 2004) – wasused for the calculations.

2.1.2. Number of pigs

Information on the number of pigs in different agegroups in each farm was obtained from the Danish centralhusbandry register (CHR). A farm is defined as apopulation of pigs kept at the same geographical locality.The reporting principle in the CHR is, that each producershould annually report the number of: (i) sows and gilts;(ii) weaners (7–30 kg living weight) and finishers (>30 kgliving weight) at each farm. In addition, significantchanges in the number of pigs or the closing down ofthe farm should be reported immediately. Each record inthe register is equivalent to an update performed by thepig producer. However, there has been no systematicfollow up on whether producers actually perform theannual reporting.

According to the register, pigs were produced at 21,236farms at some time during the study period, and thenumber of records for these pig farms was 29,366. Thirty-three percent of the records represented updates per-formed after the study start, whereas 67% of the recordsrepresented updates performed before study start. Inspec-tion of the dates for when updates had been performedrevealed that for most of the farms there had been morethan 1 year between the updates.

When the register was implemented in 1998, onlythe number of sows and finishers were recorded. SinceJuly 2000, reporting the number of weaners to theregister became mandatory. However, we have empiricalknowledge telling us that the recording of weaners inthe CHR has a considerable lag in time – for some farmsseveral years.

According to the register, there were 7622 sow-farmsat the study start. The absolute majority of Danish sow-farms also rear weaners. According to the CHR, at studystart only 1834 sow-farms also had the number ofweaners recorded. Between 1 January 2002 and 30 June2005 a large number of farms updated their number ofweaners, and at 30 June 2005, 4499 sow-farms had

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109100

number of weaners recorded in the register whereas only3123 did not. To increase the amount of data on weanersavailable for the statistical analysis, an algorithm forimputation of the number of weaners was utilised. Thealgorithm was based on the number of sows recorded inthe register. For each farm, the newest record includingboth the number of sows and weaners were comparedwith the second newest. If the number of sows were equal,we assigned the number of weaners in the newest recordto the second newest record. Subsequently, within eachfarm the second most recent record was compared withthe third most recent record and so on. As a result of theimputation, 20% of the records with a number of weanersincluded an imputed value of the number of weaners. Ifthe number of sows differed between two adjacentrecords, no imputation of number of weaners wasperformed.

2.1.3. Pig producer

In the CHR, each record includes information on the pigproducer of each farm. The pigs owned by a producer (theherd) can be kept at many geographical localities (farms).For the sake of simplicity, all animals owned by theproducer will be referred to as a herd throughout the studyeven though some herds are kept at many farms.

2.1.4. Occurrence of PMWS

Information on the occurrence of the disease syn-drome PMWS was obtained from the database in thenational reference laboratory situated at the NationalVeterinary Institute, Technical University of Denmark.The laboratory is the only laboratory in Denmark makingthis diagnosis. In this study the diagnosis of PMWS wasmade at the herd level according to the definition ofPMWS that has been agreed upon by the committee ofthe EU-consortium of the Control of Porcine CircovirusDiseases (Bækbo, 2005). A herd was considered positivefor PMWS when pigs showed postweaning wasting andlesions indicative of PMWS (i.e.: (i) lymphocyte deple-tion together with histiocytic infiltration and/or inclu-sion bodies and/or giant cells and (ii) presence of PCV2 inmoderate to massive quantities) were present inlymphoid tissue of one or more autopsied pigs.

2.1.5. Herd type and health status

Information on herd type (breeding and multiplying,conventional production or outdoor/organic) wasextracted from the CHR. The Danish SPF (specific pathogenfree) Company provided data concerning health status foreach herd. If a herd was situated on several farms withdifferent health statuses, the herd was classified to thelowest status in the analysis.

2.1.6. Study-month

In the study we used the variable ‘study-month’ torepresent time. To adjust for Christmas and New Yearrelated decrease in prescription of medicine and anincrease immediately after, we defined study-month 1as the period from 10 of January until 9 of February 2002and so on. The periods between 1 and 9 of January 2002and from 10 of December 2004 until the end of the study

period were excluded from the data analysis, resulting in atotal of 35 study-months.

Calendar month (January–December; 1–12) wasdefined according to the study-month, i.e. study-month1 (10 of January until 9 of February) is equivalent tocalendar month 1 (January) and so on.

2.2. Definition of dependent variable

The dependent variable in our study was the amountof antimicrobials (ADDkg) used per pig-day in a herdeach month the herd was included in the study. Thevalue of the dependent variable was estimated forweaners and finishers separately. Due to inconsistentand large variation between veterinarians in recordingdisease category in VetStat, all records with prescribedantimicrobials to weaners and finishers, respectively,were utilised in the study. This means that allantimicrobials prescribed for several concurrent dis-eases in a herd at the same time were included in thecalculation of the dependent variable. The dependentvariable was derived for weaners and finishers sepa-rately in four steps.

1. F

or every day, the amount of antimicrobials used withina herd (# ADDkgherd, day), was estimated as the averagedaily use of antimicrobials within the herd during thatprescription period:

AADkgherd;date ¼ADDkgherd; prescribed

daysherd; prescri ption period(1)

where #ADDkgherd, prescribed = amount of antimicrobials

measured as ADDkg prescribed to the herd at the initial

day of the prescription period; #daysherd, prescription

period = number of days between the day of the actual

prescription and the day of the subsequent prescrip-

tion.It was assumed that all antimicrobials prescribed at

one event were used at a constant amount every dayuntil the event of a subsequent prescription for the agegroup. If number of days within the prescription periodwas >90 then the prescription period was censored to90 days, based on the assumption that a longerprescription period in the register presumably wascaused by a missing registration between two succes-sive registrations or a missing registration or theclosing of the herd. Because of the censoring strategy,values of #ADDkgherd, date were missing for some periods(missing until next date new amounts of antimicrobialswere prescribed to the herd). Also, the value of#ADDkgherd, date could never be zero due to how it isderived.

2. T

he #ADDkgherd, day was summarized for each month.

AADkgherd;month ¼Xn

date¼1

ðAADkgherd;dayÞ (2)

The number of pig-days at risk for antimicrobial

3. treatment in a herd was estimated for each day, andwas equal to the actual number of pigs recorded in theCHR on the actual day. The number of pig-days at risk for

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109 101

antimicrobial treatment every month was obtained bysummarizing the daily values.

pig� daysherd;month ¼Xn

date¼1

ð pigsherd;dayÞ (3)

where #pigsherd, day is the number of pigs according to

the CHR in that herd on any given day.

4. T he amount of antimicrobials used per pig-day (ADDkg/

pigday) within a herd each month (the dependentvariable) was estimated as:

ADDkgpigday

� �herd;month

¼ADDkgherd;month

pig� daysherd;month(4)

2.3. Editing and organising data used in the statistical

analysis

Data were organised, edited and summarized using thesoftware SAS1 version 8.2. The evaluation of the potentialfor selection bias was based on descriptive statistics andcomparisons. Descriptive statistics were compared tonational statistics (e.g. distribution of herds into differentproduction types and health status; herd size; number ofprescriptions per herds; percentage of the national useantimicrobials included in the study compared to percen-tage of the national pig population included in the study)and expert opinions (e.g. number of locations per herd;number of herds per veterinarian; length of prescriptionperiod). Data for number of pigs in the herd andregistration of prescription in the herd were cross-tabulated to reveal mismatches such as antimicrobialsbeing prescribed to herds without pigs or herds havingreared pigs without use of any antimicrobials. Thedistribution of both errors in registrations of prescriptionand missing registration of pigs were compared betweenherds with and without the diagnose PMWS.

According to the hierarchy of usage of antimicrobials(the veterinarian is prescribing antimicrobials to the herd,the herd is using the antimicrobials over time), the datawere organised in a three-level nested hierarchy: study-month within herd within veterinarian.

During the study period, many herds changed veter-inarian, which caused cross-classification of herds atveterinarian-level – more than one veterinarian could bethe responsible veterinarian for a herd during the studyperiod. To avoid cross-classification in the data analysis,only the time period between the first prescription by theveterinarian that issued the majority of the prescriptionsduring the study period and the first prescription by thesubsequent veterinarian were included in the analysis.

If a farm was part of two different herds during thestudy (caused by change of producer) we assumed that theuse of antimicrobials in these two herds were non-correlated.

2.4. Data analysis

The distributions of the amount of antimicrobials usedper pig-day within a herd each month (dependentvariable) were right skewed for both weaners and

finishers, and to improve the normality of the responsevariable it was log-transformed both for weaners andfinishers. Logarithmic transformed data were approxi-mately normally distributed.

Due to the repeated measures within herd (giving riseto serial dependence and autocorrelated residuals) and thehierarchical structure where study-months are clusteredin herds and herds are clustered in veterinarians (givingrise to correlation of observations within the samesubgroup), data were analysed using linear mixed-effect(multilevel) models including serial correlation structures(see Section 2.4.1). The models were fitted by MaximumLikelihood estimation (ML) using the lme function of theNLME (Nonlinear Mixed-Effects Models) library (Pinheiroand Bates, 2000) in the software package R version 2.5.1.

The modelling strategy was to initially build ahierarchical model with an appropriate correlationstructure for the serial dependency. Akaike InformationCriterion (AIC) was used to compare the fit of the modelswith different correlation structures and different hier-archical levels (herd/study-month versus veterinarian/herd/study-month). After refining the initial model(Model I), the unconditional fixed effects of the indepen-dent variables were added to Model I separately. This wasdone to investigate the significance of different functionalforms before the effects were added to the multivariablemodels.

The multivariable models were built in stages, succes-sively adding to Model I the fixed effects of confoundingvariables at different levels (Models II–IV), and finally theeffect of PMWS (Models V–VI). Model II added the study-month level variables ‘study month’ and ‘season’. Model IIIadded the herd-level variables ‘herd size’, ‘production type’and ‘health status’. Model IV added the veterinarian-levelvariable ‘number of herds per veterinarian’. In Model V thePMWS related variables are added. A description of howthe effect of outbreak PMWS was parameterized in themodel is given in Section 2.4.2. Finally, in Model VI randomeffects for outbreak of PMWS were evaluated at theveterinarian and herd levels respectively, by adding theseeffects to the model assuming independence between therandom effects at the respective levels. The significance ofthe models was assessed using Likelihood Ratio tests, andthe change in model fit was assessed using the quantitativechange in AIC.

Subsequently, a set of more complex candidate modelswere estimated and compared to Model VI.

To evaluate whether the effect of PMWS was influencedby other variables, the interaction terms between ‘out-break of PMWS’ and the confounding variables at herdlevel (‘herd size’, ‘production type’ and ‘health status’)were assessed.

In Model VI we assumed no covariance between therandom intercept and random slope at the herd level. Toevaluate this assumption, Model VI was re-estimatedallowing for this covariance.

Approximate 95% confidence intervals on the randomand fixed effects were obtained using the approximatedistributions for the ML estimates. Prediction from themodels was obtained by adding together the populationpredictions (based on only the fixed effect estimates) and

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109102

the estimated contributions of the random effects to thepredictions. The resulting values are the Best LinearUnbiased Estimates (BLUEs).

2.4.1. Estimation of random effects of independent variables

and residuals

The effects of veterinarians and herd (nested withinveterinarians) were estimated as normal distributedrandom effects (veterinarian-level and herd-level) of theintercept, assuming that both the veterinarians and theherd represent a random sample from the total population.In addition, in Model VI, we estimated normal distributedrandom slopes for the effects of ‘outbreak of PMWS’ at theveterinarian and herd levels, respectively.

At the lowest level (study-month) we assumed that thewithin group error was identically distributed randomvariables with mean zero. Because of the repeated structureof the study-month within herd, models with differentcorrelation structures between the monthly measures wereestimated: no correlation, autoregressive of order 1 (AR(1))and autoregressive of order 1 with varying moving-averagecorrelation (ARMA(1,q)), where q varied between 1 and 5.The adequacy of the correlation structures was evaluated inplots of the empirical autocorrelation function correspond-ing to the normalized residuals of the models.

The proportion of the total variance which resides ateach level provides an estimate of the intra-class correla-tion coefficient. However, this approach can only beapplied in models in which the only random effects arethose associated with the intercept. Therefore, the intra-class correlation coefficients were estimated using theestimates obtained in Model V.

To assess the adequacy of the assumption that therandom effects were normally distributed, normal plots ofstudy-month-level, herd-level as well as veterinarian-levelstandardised residuals from models were constructed. Theassumption of constant variance (homoscedasticity) wasevaluated by fitting models in which the residual varianceswere allowed to vary depending on whether the herd was aPMWS-herd, or experienced a PMWS-outbreak and byherd size, respectively and comparing these models withmodels assuming constant variance.

A Box–Cox transformed version of the response wassubstituted by log-transformed response variable in thefinal model to see if any further improvement in thenormality of the residuals could be obtained.

2.4.2. Estimation of fixed effects of independent variables

The continuous confounding variables ‘season’, ‘long-term effect’, ‘herd size’ and ‘number of herds per veterinar-ian’ were added to the mixed-effect model in a functionalrelationship with the dependent variable that agreed withvisual impression in scatterplots of the dependent variableagainst each confounding variable separately. The effect ofseason was represented by a sine–cosine curve based oncalendar months. The long-term effect was estimated as alinear effect of study-month 1 (January 2002)–35 (Novem-ber 2004). ‘Herd size’ was centred on the overall mean andadded to the model in a polynomial form using a linear and aquadratic term. The effect of ‘number of herds perveterinarian’ was estimated as linear.

To visualize the shape of the potential within herdeffect of PMWS, the scatterplot of the effect of time ofdiagnosis on the usage of antimicrobials included onlyherds that had received the diagnosis of PMWS during thestudy. The scatterplot and smoothed curve indicated thatthe potential within herd effect of PMWS on the usage ofantimicrobials in general occurred at the same time as thediagnosis was obtained. Typically, the usage of antimicro-bials was somewhat higher within the first months afterthe diagnosis compared to the successive months. Afterabout 1 year, however, the usage was at the same level asbefore outbreak with PMWS. When estimating the effect ofoutbreak of PMWS, initially, the effect was estimated by 3-months dummy variables (1–3 months, 4–6 months, 7–9months, 10–12 months and >12 months after outbreak ofPMWS) relative to the time before outbreak of PMWS.Based on the parameter estimates and significance of thefit of the model including the dummy variables (results notshown), we reduced the number of variables used todescribe the effect by successively reparameterizing themodel. For weaners the final model included an overalleffect of 1–12 months after outbreak of PMWS (and noeffect after 12 months) and an additional effect of 4–12months after outbreak of PMWS. Using this parameteriza-tion, the mean effect during the first 3 months afteroutbreak is the estimate of the dummy variable ‘outbreakto the effect of 1–12 months after outbreak’. The meaneffect on the antimicrobial usage during the periodbetween the 4th and the 12th months after outbreak ofPMWS can be obtained by adding the estimate of effect of4–12 months after outbreak to the effect of 1–12 monthsafter outbreak. In Model VI, a random effect for ‘effect ofPMWS 1–12 months after outbreak’ was evaluated at theveterinarian and herd level.

In the modelling of the antimicrobial usage in finishers,none of the time-dependent dummy variables describingthe effect of outbreak of PMWS were significant. Therefore,the effect of outbreak of PMWS was not included in ModelV for finishers and no Model VI was estimated.

Boxplots were made to compare the usage of anti-microbials in herds diagnosed with PMWS before theyreceived the diagnosis with herds not affected with PMWSduring the study. These boxplots revealed a potential effectof PMWS on the usage of antimicrobials before thesyndrome was diagnosed. This effect – ‘PMWS-herd’ –was estimated in a categorical form at herd-level.

The estimated model described the usage of anti-microbials on a logarithmic scale. Exponentiation ofregression coefficients yielded estimates of relative,multiplicative effects, e.g. of the presence versus absenceof a factor, which could be converted into percentageincreases or decreases by multiplication with 100%followed by subtraction of 100%.

3. Results

3.1. Data

After editing and organising, the data used in theanalysis constituted 56% of the total amount of anti-microbials prescribed to weaners in Denmark and 82% of

Table 1

Structure of the data from 71,585 study-months of antimicrobial usage in weaners and 138,801 study-months of antimicrobial usage in finishers in Danish

pig farms 2002–2004.

Level Weaners Finishers

Number Average number per

unit at next-higher level

Range Number Average number per

unit at next-higher level

Range

Vet 347 – – 522 – –

Herd 3,371 10 1–63 7,434 14 1–159

Study-month 71,585 24 1–34 136,801 18 1–35

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109 103

the total amount of antimicrobials prescribed to finishersin Denmark during the study period. The proportion of sowherds (assumed to have weaners) included in the studyrose from 29% of the total population at the start of thestudy to 42% at the end and these herds contained 38–54%of all weaners. The proportion of finishing herds includedin the study rose from 38% of the total population at thestart of the study to 46% at the end and these herdscontained 66–80% of all finishing pigs. The major cause ofdata loss was the lack of registration of the number of pigsin the herds. The presence of weaners in each herd in eachstudy-month (according to the CHR) was tabulated againstthe usage of antimicrobials in weaners in the same study-month within the same herd. Within 40% of the recordsthere were no weaners registered in the herd even ifantimicrobials has been prescribed to weaners in thatstudy-month in the particular herd. These records wereeliminated from the data set. In 10% of the cross-tabulatedrecords there were no antimicrobial use in weaners even ifweaners were registered in the herd and these recordswere eliminated from the data set, under the assumptionthat within the production of weaners antimicrobials willbe used at some level during a longer time period (it ispossible to rear weaners without the use of antibiotics, butit is undoubtedly exceptional). The data set used in theanalysis of use of antimicrobials in weaners included71,585 records (study-months) in 3371 herds serviced by

Table 2

Descriptive statistics of independent fixed variables. Based on data from 71,585 st

of antimicrobial usage in finishers in Danish pig farms 2002–2004.

Variable Scale of measurement Weaners

# records

Fixed effects

Calendar month Sine–cosine curve –

Study-month Continuous –

Health status Conventional 33,558

SPF-status 38,027

Type of herd Production 66,105

Multiplying 3,462

Organic 2,018

Herd size Continuous –

PMWS-herd Not affected 63,829

Affected 7,756

PMWS-outbreak Not affected months 67,999

Affected months 3,586

Number of herds per veterinarian Continuous –

347 veterinarians (Table 1). Of these herds, 321experienced an outbreak of PMWS during the study. Thedata set used in the analysis of use of antimicrobials infinishers included 136,801 records (study-month) in 7434herds serviced by 522 veterinarians (Table 1). Of theseherds, 409 experienced an outbreak of PMWS during thestudy. Table 2 presents descriptive statistics andunconditional associations (with the outcome variables)for all independent variables used in the analyses.

The median amount of antimicrobials used per pig-dayin a herd each month in the study was for weaners 1.2ADDkg (Q1 = 0.5, Q3 = 2.7) and for finishers 0.5 ADDkg(Q1 = 0.1, Q3 = 1.6). The median duration of prescriptionperiods for weaners was 36 days (Q1 = 29, Q3 = 45), and forfinishers 39 days (Q1 = 28, Q3 = 44).

3.2. Data analysis

The plot of the empirical autocorrelation functionsobtained in the models, assuming no serial correlationbetween repeated measures within herds, indicated thatthe within herd measures were correlated, and that thecorrelation decreased with time in both weaners andfinishers (Fig. 1, upper row). The plot of the empiricalautocorrelation functions obtained from the models basedon the AR(1) structure revealed a small, but significant,dependency between consecutive months (Fig. 1, middle

udy-months of antimicrobial usage in weaners and 136,801 study-months

Finishers

Median Range # records Median Range

– 1–12 – – 1–12

19 1–35 – 18 1–35

– – 87,455 – –

– – 49,346 – –

– – 130,755 – –

– – 4,419 – –

– – 1,627 – –

840 60–26,867 – 827 20–32,343

– – 126,929 – –

– – 9,872 – –

– 132,186 – –

– – 4,615 – –

38 1–141 – 36 1–141

Fig. 1. Empirical autocorrelation corresponding to the normalized residuals of the model ignoring the repeated structure of the data (no correlation) (upper

row), the normalized residuals of the model taking the serial dependency into account using an AR(1) parameter (middle row) and ARMA(1,3) (bottom row).

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109104

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109 105

row). Using the AR(1) structure the serial correlationbetween the antimicrobial usage in consecutive monthswas estimated to �55–60% in all models. Re-estimation ofthe model using the more complex correlation structureARMA did result in a significantly better model (smallervalue of the AIC) (Fig. 1, bottom row), but it had minimaleffect on the fixed effect estimates and was omitted for thesake of simplifying the variance estimates.

Table 3 lists the regression estimates for the sixregression models of the log-transformed antimicrobialusage in weaners. Including the herd-level variablescaused the largest improvement in the fit of the modelduring the modelling. In Model V three PMWS related

Table 3

Regression coefficients with standard errors for multilevel repeated measureme

2004 (71,585 observations).

Variable Model I

(null model)

Model II

Fixed effects

Intercept �0.178 (.061)* �0.184 (.031)*

Study-month 0.008 (.001)*

Season

Sin (2�p� calendar month/12) 0.051 (.009)*

Cos (2�p� calendar month/12) 0.018 (.008)*

Herd size (per 100 pigs)

Linear

Quadratic

Health status of the farm

Conventional

SPF

Production type at the farm

Conventional

Out door

Breading and multiple

Number of herds per veterinarian

PMWS-herd

Not diagnosed

Diagnosed

Outbreak of PMWS

Before outbreak

1–12 months after outbreak

4–12 months after outbreak

Random effects

Level 1 intercept variance (time) 1.375 1.370

Level 2 intercept variance (herd) 0.865 0.866

Level 3 intercept variance (vet.) 0.119 0.121

Level 2 effect of 1–12 months

after outbreak of PMWS variance

Model fit

Deviance 202,426 202,308

Number of parametersa 5 8

DDb 118

Ddfc 3

Prob. x2 d <0.001

AICe 202,439 202,324a Number of parameters estimated in modeln.b Deviance (�2� log Likelihood Deviance) modeln�1�Deviance modeln.c Number of parameters added to modeln compared to modeln�1.d The tail probability of DD in a chi-squared distribution with Ddf.e Akaike’s Information Criteria.* Significance at 0.05 level.

variables were included: ‘PMWS diagnosed at the herd’(herd-level) and ‘effect of 1–12 months after outbreak ofPMWS at the herd’ and ‘effect of 4–12 months afteroutbreak of PMWS at the herd’, respectively (study-month-level), which improved the fit of the model significant.Finally, random effects of ‘effect of 1–12 months afteroutbreak of PMWS at the herd’ to account for the variation inthe effect between veterinarians and herds were added tothe model. The model with random effects at bothveterinarian and herd level did not fit the model signifi-cantly better then a model with random effect only at theherd level (p-value = 0.10). The model with random effect of‘effect of 1–12 months after outbreak of PMWS at the herd’

nt model of log antimicrobial usage in weaners in Danish pig farms 2002–

Model III Model IV Model V Model VI

�0.236 (.037)* �0.215 (.037)* �0.253 (.037)* �0.253 (.037)*

0.010 (.001)* 0.010 (.001)* 0.010 (.001)* 0.010 (.001)*

0.046 (.009)* 0.045 (.009)* 0.045 (.009)* 0.044 (.008)*

0.016 (.008)* 0.015 (.008)* 0.016 (.008)* 0.016 (.008)*

�0.033 (.001)* �0.033 (.001)* �0.034 (.001)* �0.034 (.001)*

0.001 (.000)* 0.001 (.000)* 0.001 (.000)* 0.001 (.000)*

Ref. Ref. Ref. Ref.

0.067 (.035) 0.068 (.035) 0.078 (.035)* 0.078 (.035)*

Ref. Ref. Ref. Ref.

0.140 (.098) 0.142 (.098) 0.145 (.097) 0.144 (.080)

0.073 (.080) 0.076 (.080) 0.082 (.080) 0.083 (.097)

0.002 (.000)* 0.002 (.000)* 0.002 (.000)*

Ref. Ref.

0.319 (.063)* 0.319 (.063)*

Ref. Ref.

0.194 (.049)* 0.195 (.052)*

0.125 (.053)* �0.129 (.053)*

1.343 1.343 1.342 1.340

0.797 0.797 0.790 0.790

0.152 0.140 0.134 0.134

0.100

201,526 201,500 201,450 201,445

13 14 17 18

782 26 50 5

5 1 3 1

<0.001 <0.001 <0.001 0.019

201,552 201,528 201,484 201,481

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109106

(Model VI) had a significantly better fit than the Model V.There was no significant variation between herds in theeffect of ‘PMWS-herd’ (results not shown).

The fit of Model VI did not improve when two-terminteractions between the effect of ‘outbreak of PMWS’ andthe herd level confounding variables were added to themodel.

We found a substantial negative covariance betweenthe two random effects at herd level. However, allowing forcovariance in the model had only minimal effect on thefixed effect estimates and was omitted for the sake ofsimplifying the variance estimates.

After a herd had experienced an outbreak of PMWS, themean effect on the antimicrobial usage during the threefirst months after the outbreak of PMWS was an increase of22% (95% CI: 10–35%), and the mean effect on the anti-microbial usage during the period between the 4th and the12th months after the outbreak of PMWS was an increase of

Table 4

Regression coefficients with standard errors for multilevel repeated measureme

2004 (136,801 observations).

Variables Model I (null model) Model

Fixed effects

Intercept �1.114 (.071)* �1.070

Study-month 0.011 (

Season

Sin (2�p� calendar month/12) 0.018 (

Cos (2�p� calendar month/12) 0.057 (

Herd size

Linear

Cubic

Health status of the farm

Conventional

SPF

Production type at the farm

Conventional

Out door

Breeding and multiple

Number of herds per veterinarian

PMWS-herd

Not diagnosed

Diagnosed

Outbreak of PMWS

Random effects

Level 1 intercept variance (time) 1.506 1.450

Level 2 intercept variance (herd) 1.368 1.374

Level 3 intercept variance (vet.) 0.330 0.330

Level 2 effect of 1–12 months

after outbreak of PMWS variance

Model fit

Deviance 395,902 395,548

Number of parametersa 5 8

DDb 354

Ddfc 3

Prob. x2 d <0.001

AICe 395,915 395,565a Number of parameters estimated in modeln.b Deviance (�2� log Likelihood Deviance) modeln�1�Deviance modeln.c Number of parameters added to modeln compared to modeln�1.d The tail probability of DD in a chi-squared distribution with Ddf.e Akaike’s Information Criteria.* Significance at 0.05 level.

7% (95% CI: 4–19%) compared to before the outbreak. Thevariation between herds in the effect of an outbreak ofPMWS was significant. Predictions from Model VI of theeffect of an outbreak with PMWS showed that during thethree first months after an outbreak of PMWS the minimumchange was a decrease of 21%, the 1st quartile was a 13%increase; median was a 22% increase; the 3rd quartile was a33% increase and maximum was a 93% increase. During the4th to the 12th months after outbreak of PMWS theminimum change was a decrease of 31%, the 1st quartilewas no change; median was a 7% increase; the 3rd quartilewas a 15% increase and maximum was a 69% increase.

Additionally, in herds experiencing an outbreak ofPMWS, the usage of antimicrobials before the outbreakwas 37% (95% CI: 22–56%) higher compared to herds notexperiencing PMWS during the study period.

Table 4 lists the regression estimates for five nestedthree-level regression models of the log-transformed

nt model of log antimicrobial usage in finishers in Danish pig farms 2002–

II Model III Model IV Model V

(.036)* �1.184 (.039)* �1.145 (.038)* �1.153 (.038)*

.001)* 0.013 (.001)* 0.012 (.001)* 0.012 (.001)*

.006)* 0.013 (.006)* 0.012 (.006)* 0.012 (.006)*

.006)* 0.054 (.006)* 0.054 (.006)* 0.054 (.006)*

�0.044 (.001)* �0.044 (.001)* �0.044 (.001)*

0.001 (.000)* 0.001 (.000)* 0.001 (.000)*

Ref. Ref. Ref.

0.111 (.031)* 0.110 (.031)* 0.109 (.031)*

Ref. Ref. Ref.

0.276 (.124) 0.278 (.124) 0.273 (.124)

0.005 (.079) 0.008 (.079) 0.011 (.079)

0.002 (.000)* 0.002 (.000)*

Ref.

0.174 (.065)*

–

1.469 1.468 1.468

1.282 1.283 1.282

0.379 0.355 0.354

394,318 394,269 394,262

13 14 15

1,230 50 7

5 1 2

<0.001 <0.001 0.006

394,345 394,297 394,292

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109 107

antimicrobial usage in finishers. The estimated effects ofthe potential confounding variables were very similar tothose observed in weaners (Models I–IV). Also, in finishingherds, we found that in herds experiencing outbreaks ofPMWS, the usage of antimicrobials before the outbreakwas higher (19%, 95% CI: 5–34%) compared to herds thatdid not experience PMWS during the study period. Therewas no significant overall effect of an outbreak of PMWS onantimicrobial use in finishers.

According to Model V for weaners, the variation ofantimicrobial usage residing at the veterinarian-level wasonly 6%, whereas 35% was at the herd-level. According toModel V for finishers, 11% of the variation resided at theveterinarian-level and 40% at the herd-level.

There was no high correlations between the estimatedparameters of the independent variables, which indicates alow degree of linear correlation between two or moreexplanatory variables in the regression models.

3.2.1. Diagnostics

There were no indications that the amount of anti-biotics used was more variable in herds experiencingPMWS than in herds not experiencing PMWS. Neither wasthe amount of antibiotics used more variable in largerherds than in smaller herds.

The assumption of normality seemed reasonable for theresiduals at veterinarian-level and herd-level althoughthere is some asymmetry in the distributions indicatingshorts ends of the data distribution when compared to anormal distribution. The assumption of normality seemedreasonable for the residuals at the study-month-levelthough the curved point pattern of the residuals from theanalysis of antimicrobial usage in weaners indicates thatthe data distribution is slightly skewed to the right.

The normality of the residuals were somewhatimproved by the Box–Cox transformation, however, theestimated parameters of all fixed and random effects werevery close to those obtained from the log-transformedresponse variable and graphic assessment of the residualsindicated only slight improvement of the normality of theresiduals. Consequently, the details of these models havenot been presented.

4. Discussion

4.1. Data validity and representativeness

The objective of this study was to determine how theincreased occurrence of PMWS in the Danish pig industryinfluenced the use of antimicrobials. The study capitalizedon the extensive animal production and health databaseswhich are maintained for the pig industry in Denmark.When records on number of pigs in a herd were notcomplete, it was not possible to estimate the amount ofantimicrobials used per pig-day in that herd. Due to thatrecord deficiency, particularly for weaners and finishers,the study included approximately half of the amount ofantimicrobials prescribed for weaners and 80% of theamount of antimicrobials prescribed to finishers in thewhole Danish pig production during 2002–2004. There isno indication that the lack of registration of the number of

pigs is related to either the amount of prescribedantimicrobials or outbreak of PMWS, and we assume thatthe registration is missed by random and does not bias theestimated effect of PMWS on the usage of antimicrobials.

The increased proportion of herds included at the endcompared to the beginning of the study period is aconsequence of an increased proportion of herds with thenumber of weaners and finishers registered in the CHR.That a herd undertook registration of the number of pigswas assumed to be independent of the usage of anti-microbials and the occurrence of PMWS, and we assumethat the herds included in the study during the studyperiod were similar to herds that were included at thebeginning of the study period.

The fact that the percentage of the total usage ofantimicrobials included in the study was larger then thepercentage of the total population included in the studyindicates that herds using relatively large amount of anti-microbials were overrepresented in the study compared tothe general population (selection bias). In general, herdsincluded in the analyses were larger than herds excludeddue to missing data. For both weaners and finishers thedistribution of herds in relation to health status and produc-tion type was very similar for included and excluded herds.

4.2. Dependent variable

In this study we used the amount of prescribedantimicrobials to the specified age groups in the herdsas a proxy for amount of antimicrobials used in thespecified age group in the herds, and it was assumed thatthe prescribed antimicrobials were used throughout theperiod until a new prescription was registered for the agegroup. The assumption is supported by the function of theveterinary authorities’ regulation and control system. It ismandatory for the veterinarian to register use and deliveryof drugs to farmed animals and keeping records availablefor inspection by veterinary officials for 3 years. Thereforethe veterinarians act careful about the prescription ofantimicrobials, not allowing large amount of antimicro-bials at the farms.

Another source of potential bias might be error (missingor wrong) in the registration of prescriptions. There is noindication that these types of error are correlated to thestatus of PMWS in the herds. By assuming independencebetween exposure and dependent variable in relation toerror in the registration of prescription, the missing andwrong registrations will not bias the effect of PMWS, butreduce the power to detect a significant effect.

Herd health contracts between the veterinarian andfarmer allowed prescription, in such herds, for up to 35days of treatment of diagnosed disease or disease that wasexpected in the herd.

Prescription of antimicrobials to weaners and finishersare, however, typically done at the monthly visitsaccording to herd health contracts between the herdsand veterinarians. Prescription of antimicrobials to wea-ners and finishers outside these visits are very unusual.

The ADDkg defined in this study was based on Danishcriteria and may differ from other national ADD. Thenumber of ADDkg was considered to be a measure of

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109108

treatment. However, the measure should not be inter-preted as an absolute measure of treatment frequency,because treatment administration in n pigs in the earlieststages of the rearing period would contribute less to thetotal number of ADDkg than a treatment administrated inthe same number of pigs at later stages. The variationbetween veterinarians and herds in ADDkg/pigday, can,except from differences in disease frequency, be caused byvariation in administrated dosages, treatment length andage at treatment.

By using a hierarchical model we decomposed thevariation in usage of antimicrobials into veterinarian, herdand time-levels. Theoretically, the estimated variancecomponent at each level is a result of the effect of unknownfactors related to the respective levels. In our study we usedan estimated value of the usage of antimicrobials as thedependent variable, and error in the data used for theestimation of the dependent variable might have creatednon-biological variation in the dependent variable. Thisvariation will be absorbed into the variance componentswhereby they will be overestimated. For instance, the valueof the dependent variable in two equally sized herds usingthe same amount of antimicrobials will vary if one of theherds has the wrong number of pigs recorded in the CHR,and this created variation will be absorbed into the variancecomponent at herd-level. Furthermore, if a prescription ofantimicrobials to a herd with constant usage of antimicro-bials is missed in the VetStat database, the value of thedependent variable will vary and this created variation willbe absorbed into the variance component at the time-level.The veterinarian is responsible for writing the prescriptions,including correct CHR number, species and age group, andany errors will be transferred to VetStat. Due to variabilitybetween the carefulness about writing the prescriptions,variation (error) will be created in the dependent variable. Inthe analysis, this error will be absorbed into the variancecomponent at the veterinarian-level. Based on our generalknowledge on the quality of the databases, we believe thatthere is about the same proportion of overestimation at allthree levels.

4.3. Result discussion

Bias in evaluating the effect of PMWS on the amount ofantimicrobials used can occur from errors in classificationof herds having PMWS. In our study it can be assumed thatseveral herds having PMWS was misclassified as nothaving PMWS. It can, however, be assumed that thismisclassification of the exposure of PMWS biases the effectof PMWS towards zero.

The study has nevertheless shown that the diseasesyndrome PMWS had a significant influence on theantimicrobial usage in the pig industry during 2002–2004. The high usage of antimicrobials both amongweaners and finishers in the period before the farmshad a diagnosis of PMWS may indicate that these herds hadgeneral disease problems. These herds might also be moreprone to having outbreaks of PMWS – perhaps becausethese herds had a management system with high risk forintroduction and establishment of infectious diseases ingeneral. However, it is also possible that there was a

significant effect of PMWS in herds before they werediagnosed and that the disease syndrome might have beenpresent for long time in some herds. In this case, the totaleffect of PMWS on the antimicrobial usage is estimatedboth in the variable ‘PMWS-herd’ and ‘outbreak of PMWS’.

Only in weaning herds, there was a significant increasein usage of antimicrobials after the outbreak of PMWS. Thesignificant time dependency indicates that the effect ofPMWS changes over time. Immediately following theoutbreak the usage of antimicrobials peaked, after whichthe effect declined. On average, 1 year after the outbreak theusage of antimicrobials is equal to the usage before in theherd. The time-dependency of the effect is probably a resultof a very complex interrelation between the pathogenspresent in the herd, development of immunity in the herdand adjustment of management to prevent the increase ofsecondary bacterial infections as a result of PMWS.Furthermore, the significant variation between herds inthis effect probably reflects a number of different factorsthat influence the animal health and usage of antimicrobialsin pig herds in general such as (i) variation in occurrence ofviral and bacterial infectious diseases that interact withPMWS and (ii) variation in the production systems andintensity. The fact that we did not find a significant between-veterinarian variation in the usage of antimicrobials as aresponse to an outbreak of PMWS contradicts the belief thatpractitioners have a stereotypic way of prescribing anti-microbials as a response to disease problems.

In this study, the effect of PMWS was adjusted for theeffect of season, long-term time effect, herd size and healthstatus. In both weaners and finishers, these variables had asignificant effect on the usage of antimicrobials. The reasonfor including them in this study was to adjust forconfounding and not to model their relationship toantimicrobial usage explicitly. A general discussion ofthese effects is out of the scope of this paper.

To obtain an estimate for the national effect of PMWS inweaners on the usage of antimicrobials, we multiplied theestimated effect of PMWS and the estimated nationalprevalence of herds with weaners that were diagnosedbetween 2001 and 2004 (19%) (Vigre et al., 2006). In thiscalculation we did not take into account the time-dependent decrease in the effect of PMWS that wasestimated in our study. Instead we assumed that the effectof an outbreak of PMWS was time-independent and equalto the estimated effect of PMWS 1–3 months post outbreak(22% increase). The estimate of the national effectindicated that the increase of PMWS infected herds fromalmost zero to 20% caused an increase in use ofantimicrobials by �4–5% in weaners.

The relative magnitude of the variance residing at herd-level indicates that herd-level factors are important interms of antimicrobial usage in the Danish pig populationand suggest that more work still needs to be done toinvestigate why there is so much variation between herdsusing the same veterinarian and how one can reduce ittowards a generally lower usage.

We found that there was a significant serial correlationof 60% in the amount of antimicrobials used per weaningand finishing pig in consecutively months. This suggeststhat the usage of antimicrobials within herds in general are

H. Vigre et al. / Preventive Veterinary Medicine 93 (2010) 98–109 109

homogeneous in relation to time, and this probably reflectsthe slow changes in the manifestations of differentinfectious diseases within pig herds.

4.4. Modelling method

In this study we utilised a large amount of data toestimate the effect of PMWS on the usage of antimicrobialsto pigs. Throughout the study we wanted to balance theneed to find a model which ‘best fit’ the data with the needfor simplicity. To obtain the most valid estimate possiblefor the variable of primary interest, we paid attention topossible confounding effects of other independent vari-ables, whereas these variables were of less interestthemselves. Therefore, random effects were only esti-mated for the effect of PMWS. Inclusion of random effectsfor all independent variables would have complicated themodelling process unnecessarily.

This study makes a contribution to current methodologyresearch in the field of pharmaco-epidemiology in animalproduction accounting for all variations among the three-level structure of the animal production system (veterinar-ian/herd/time) in relation to use of antimicrobials.

References

Anon., 2002. Sales of Antimicrobial Products used as Veterinary Medicines,Growth Promoters and Coccidiostats in the UK in 2000. 27 July 2004.http://www.vmd.gov.uk/general/publications/amrrpt2000v51.htm.

Bækbo, P., 2005. PMWS Case Definition (Herd Level). http://www.pcvd.org/news.php.

Clark, E.G., 1996. Post-weaning multisystemic syndrome: preliminaryepidemiology and clinical findings. In: Proc. Western Can. Assoc.Swine Pract. pp. 22–25.

DANMAP, 2002. Use of Antimicrobial Agents and Occurrence of Antimi-crobial Resistance in Bacteria from Food Animals, Foods and Humans inDenmark. July 2003. http://www.dfvf.dk/Files/Filer/Zoonosecentret/Publikationer/Danmap/Danmap_2002.pdf.

DANMAP, 2003. Use of Antimicrobial Agents and Occurrence of Antimi-crobial Resistance in Bacteria from Food Animals, Foods and Humans inDenmark. July 2004. http://www.dfvf.dk/Files/Filer/Zoonosecentret/Publikationer/Danmap/Danmap_2003.pdf.

DANMAP, 2004. Use of Antimicrobial Agents and Occurrence of Antimi-crobial Resistance in Bacteria from Food Animals, Foods and Humans inDenmark. July 2005. http://www.dfvf.dk/Files/Filer/Zoonosecentret/Publikationer/Danmap/Danmap_2004.pdf.

Jensen, V.F., Jacobsen, E., Bager, 2004. Veterinary antimicrobial-usagestatistics based on standardized measures of dosage. Prev. Vet.Med. 64, 201–215.

Madec, F., Eveno, E., Morvan, P., Hamon, L., Blanchard, P., Cariolet, R.,Amenna, N., Morvan, H., Truong, C., Mahe, D., Albina, E., Jestin, A.,2000. Post-weaning multisystemic wasting syndrome (PMWS) in pigsin France: clinical observations from follow-up studies on affectedfarms. Livest. Prod. Sci. 63, 223–233.

Pinheiro, J.C., Bates, D.M., 2000. Mixed-Effect Models in S and S-PLUS.Springer, USA.

Segales, J., Domingo, M., Chianini, F., Majo, N., Dominguez, J., Darwich, L.,Mateu, E., Jestin, A., 2004. Immunosuppression in postweaning multi-systemic wasting syndrome affected pigs. Vet. Microbiol. 98, 151–158.

Statistics Denmark, 2003. November 21, 2004, http://www.statbank.dk/statbank5a/default.asp?w=1024.

Stege, H., Bager, F., Jacobsen, E., Thougaard, A., 2003. VETSTAT—the Danishsystem for surveillance of the veterinary use of drugs for productionanimals. Prev. Vet. Med. 57, 105–115.

Trujano, M., Iglesias, G., Segales, J., Palacios, J.M., 2001. PCV-2 fromemaciated pigs in Mexico. Vet. Rec. 148, 792.

Vigre, H., Baekbo, P., Jorsal, S.E., Bille-Hansen, V., Hassing, A.-G., Enøe, C.,Bøtner, A., 2005. Spatial and temporal patterns of pig herds diagnosedwith Postweaning Multisystemic Wasting Syndrome (PMWS) duringthe first two years of its occurrence in Denmark. Vet. Microbiol. 110,17–26.

Vigre, H., Bækbo, P., Jorsal, S.E., 2006. Estimation of the national pre-valence of PMWS using two different data sources. In: Proceedings ofthe 19th IPVS Congress, vol. 2, Copenhagen, Denmark.