EdB

JuOa Inb Cc Cd Ce Df Su

Mag D

1.

zoan

Veterinary Microbiology 155 (2012) 38–43

A

Art

Re

Re

Ac

Ke

Tu

Ca

Dia

Sk

Int

Ba

*

03

do

valuation of the sensitivity and specificity of bovine tuberculosisiagnostic tests in naturally infected cattle herds using aayesian approach

lio Alvarez a, Andres Perez b,c, Javier Bezos d, Sergio Marques e, Anna Grau e, Jose Luis Saez f,lga Mınguez e, Lucıa de Juan d,g,*, Lucas Domınguez d,g

stituto de Investigacion en Recursos Cinegeticos IREC (CSIC-UCLM-JCCM), Ronda de Toledo s/n, 13071 Ciudad Real, Spain

enter for Animal Diseases Modeling and Surveillance, One Shield Avenue, University of California, Davis, CA 95616, USA

ONICET/Facultad de Ciencias Veterinarias UNR, Boulevard Ovidio Lagos y Ruta 33, Casilda, Santa Fe, Argentina

entro VISAVET, Universidad Complutense de Madrid, Avda. Puerta de Hierro S/N, 28040 Madrid, Spain

ireccion General de Produccion Agropecuaria, Servicio de Sanidad Animal, Junta de Castilla y Leon, C/Rigoberto Cortejoso 14, 47014 Valladolid, Spain

bdireccion General de Sanidad de la Produccion Primaria, Direccion General de Recursos Agrıcolas y Ganaderos, Ministerio de Medio Ambiente y Medio Rural y

rino, C/Alfonso XII, 62, 28071 Madrid, Spain

epartamento de Sanidad Animal, Facultad de Veterinaria, Universidad Complutense de Madrid, Avda. Puerta de Hierro S/N, 28040 Madrid, Spain

Introduction

Bovine tuberculosis (BT) is a globally distributedonotic disease of cattle (Michel et al., 2010). Its controld in certain cases eradication has been achieved through

the application of test-and-cull strategies for extendedperiods of time – typically, decades (Collins, 2006; Cousinsand Roberts, 2001). Success of eradication strategies isbased on early detection and removal of infected animalsfrom a herd; thus the use of accurate diagnostic tests is ofcrucial importance.

Diagnostic tests used for detection of infected cattle aremainly based in the detection of the cellular mediatedimmune (CMI) response, which is triggered in the early

R T I C L E I N F O

icle history:

ceived 24 February 2011

ceived in revised form 26 July 2011

cepted 27 July 2011

ywords:

berculosis

ttle

gnosis

in test

erferon-gamma

yesian modeling

A B S T R A C T

Test-and-slaughter strategies have been the basis of bovine tuberculosis (BT) eradication

programs worldwide; however, eradication efforts have not succeeded in certain regions,

and imperfect sensitivity and specificity of applied diagnostic techniques have been

deemed as one of the possible causes for such failure. Evaluation of tuberculosis diagnostic

tools has been impaired by the lack of an adequate gold standard to define positive and

negative individuals. Here, a Bayesian approach was formulated to estimate for the first

time sensitivity (Se) and specificity (Sp) of the tests [single intradermal tuberculin (SIT)

test, and interferon-gamma (IFN-g) assay] currently used in Spain. Field data from the first

implementation of IFN-g assay (used in parallel with SIT test 2–6 months after a first

disclosure SIT test) in infected beef, dairy and bullfighting cattle herds from the region of

Castilla and Leon were used for the analysis. Model results suggested that in the described

situation: (i) Se of SIT test was highly variable (40.1–92.2% for severe interpretation,

median = 66–69%), and its Sp was high (>99%) regardless interpretation criteria; (ii) IFN-gassay showed a high Se (median = 89–90% and 83.5% for 0.05 and 0.1 cut-off points

respectively) and an acceptable Sp (85.7% and 90.3% for 0.05 and 0.1 thresholds) and (iii)

parallel application of both tests maximized the combined Se (95.6% using severe SIT and

0.05 cut-off point in the IFN-g assay). These results support the potential use of the IFN-gassay as an ancillary technique for routine BT diagnosis.

� 2011 Elsevier B.V. All rights reserved.

Corresponding author. Tel.: +34 913943992; fax: +34 913943795.

E-mail address: dejuan@visavet.ucm.es (L. de Juan).

Contents lists available at ScienceDirect

Veterinary Microbiology

jou r nal h o mep ag e: w ww .e ls evier . co m/lo c ate /vetm i c

78-1135/$ – see front matter � 2011 Elsevier B.V. All rights reserved.

i:10.1016/j.vetmic.2011.07.034

swsinth(Eapsinsthvtes

aginEdopliahcne

ewwrEraghtubCachm

sodte

2

2

loarin

J. Alvarez et al. / Veterinary Microbiology 155 (2012) 38–43 39

tages of infection (Pollock et al., 2001); one of the mostidely applied diagnostic technique for this purpose, the

ingle intradermal tuberculin (SIT) test, is based on theoculation of the bovine PPD in the skin of the neck or ine caudal fold (Monaghan et al., 1994). European UnionU) legislation (Council Directive 64/432/EEC as last

mended) establishes conditions for identification ofositive animals but, depending on the epidemiologicalituation, different interpretations can be applied tocrease diagnostic sensitivity (Anon, 2006), although this

ensitivity usually entails a decrease in the specificity ofe test. Accuracy of the skin test is greatly affected by a

ariety of factors, both related with the host and with thest itself; for these reasons accurate estimation of its

ensitivity and specificity in the field has been difficult.In the last 20 years an additional diagnostic tool also

iming at the detection of the CMI immune response, theamma-interferon (IFN-g) detection test, has beencreasingly applied (Wood et al., 1990). In areas of the

U with a high prevalence of BT, its use is recommended toetect the maximum number of infected animals in a herdr region; however, its application in areas in which herdrevalence is low is not recommended due to specificitymitations (Anon, 2006). In an attempt to optimize theccuracy of IFN-g assay different interpretations of the testave been considered (Wood and Jones, 2001), mainlyhanging the threshold for positivity to maximize diag-ostic sensitivity or specificity depending on the aim ofach use.

Latent class analysis is being increasingly applied forstimation of operating characteristics of diagnostic testshen true disease status cannot be determined, i.e., forhich it is not possible to assume the use of perfect

eference tests (Fosgate et al., 2006; Muma et al., 2007;noe et al., 2000). This is the case of BT, as commoneference tests (detection of lesions and/or isolation of thegent) have a limited sensitivity compared to immunolo-ical tests (Vordermeier et al., 2004). Latent class modelsave been used for evaluation of the accuracy ofberculosis diagnosis in meerkats (Drewe et al., 2009),

adgers (Drewe et al., 2010) and cattle (Muller et al., 2009;leggs et al., 2011). However, to the authors’ knowledge,ccuracy of diagnostic techniques currently applied inattle in Spain (SIT test and IFN-g assay applied in parallel)as never been quantified in the field using latent classodels.

The study here presents field estimates of theensitivity and specificity of the different interpretationsf the diagnostic tests currently used in Spain for BTiagnosis on infected herds after the first BT disclosure SITst.

. Materials and methods

.1. Study population

BT diagnostic tests results from 6202 cattle in 42 herdscated in the region of Castilla y Leon, in the west-central

rea of Spain, were available to us. Cattle herds wereandomly selected among those in which BT was detected

types of cattle industry present in the region: beef (n = 11,1141 cattle), dairy (n = 18, 2563 cattle) and bullfighting(n = 13, 2498 cattle). BT-infection was confirmed in the 42herds by either 1, isolation of Mycobacterium bovis/Mycobacterium caprae or 2, detection of macroscopicallesions compatible with BT infection. Tests results wereobtained coincidently with the first implementation of theIFN-g assay on each farm, which always followed theapplication of an intradermal test (performed at least8 weeks before application of the IFN-g assay) in whichpositive reactors were detected (and removed from theherd).

2.2. Diagnostic tests

2.2.1. Single intradermal tuberculin test

SIT test was performed by field veterinarians usingbovine PPD (CZ Veterinaria, Porrino, Espana) according toofficial rules (RD2611/1996, transposition of annex A ofCouncil Directive 64/432/EEC). Cattle were inoculated with0.1 mL of a solution containing 0.1 mg of bovine PPD (2500CTU) on the left side of the neck, and test results weredetermined by measuring the increase of the skin-foldthickness 72 h later. According with the criteria specified inthe Council Directive, results were considered positivewhen an increase of 4 mm or more in the injection site and/or presence of oedema, exudation, necrosis, pain orinflammation of the lymphatic ducts in the region or inthe lymph nodes was observed; inconclusive if clinicallesions were absent and the increase was 2–4 mm; andnegative if the increase was not larger than 2 mm andclinical signs were absent. Two criteria were used fordefinition of the infection status:

– Severe interpretation: all positive and inconclusivereactors were considered infected.

– Standard interpretations: only cattle showing a positivereaction to the SIT test were considered infected.

2.2.2. IFN-g detection assay

Heparinized blood samples were collected from everyanimal before PPDs for skin test were inoculated, anddelivered to the laboratory within 8 h of collection at roomtemperature. Stimulation with avian and bovine PPDs wascarried out as described elsewhere (Wood et al., 1990) andplasma samples were analysed using a sandwich EIA fordetection of bovine IFN-g (BovigamTM Bovine GammaInterferon Test, Prionics, Schlieren, Switzerland). Resultswere interpreted following procedures described else-where (Aranaz et al., 2006). Two cut-off points wereselected to define two interpretation criteria for the test;the ‘‘severe interpretation’’ identified an animal as infectedif the mean optical density (OD) of a sample stimulatedwith bovine PPD minus the mean OD of nil antigen wasgreater than 0.05 and greater than the same value of thesample stimulated with avian PPD (interpretation pre-scribed in the Spanish eradication program); whereas the‘‘standard interpretation’’ considered an animal as infectedif the same value was above 0.1 and greater than the value

btained after stimulation with avian PPD.

the period 2007–2009, including farms belonging to all o

2.3

seIFNcrprm(BavABvapranmDoet19wCa95inthdiPrthanVeth

prexotaswqu

Ta

Pri

ob

D

P

P

Ta

Nu

ea

S

S

S

J. Alvarez et al. / Veterinary Microbiology 155 (2012) 38–4340

. Sensitivity and specificity analysis

A Bayesian approach was used to estimate thensitivity and specificity of both the SIT test and the

-g assay, considering the alternative interpretationiteria of ‘‘standard’’ and ‘‘severe’’ for each test and theevalence of disease in the population. Latent classodels described elsewhere were used in the analysesranscum et al., 2005; Gardner et al., 2000) (codeailable at http://www.epi.ucdavis.edu/diagnostictests/2deptests1popn.html). Prior uncertainty about the truelue of the parameters assessed here was modeled usingobability distributions (prior distributions). Sensitivityd specificity prior values for diagnostic tests wereodeled according to published reports (de la Rua-menech et al., 2006; Gormley et al., 2006; Monaghan

al., 1994; Vordermeier and Ewer, 2006; Wood et al.,91, 1992) and previous experience of experts dealing

ith the Bovine Tuberculosis Eradication Program instilla and Leon (Table 1) using beta distributions. The% credibility intervals (CI) of the prior distributions were

tentionally parameterized wide enough to accommodateem to the expected large variability associated with ofagnostic situations and prevalence expected in the field.ior distribution for the value of disease prevalence aftere first disclosure test (and removal of detected infectedimals) was provided by experts from the Officialterinary Services dealing with eradication of BT withese herds for more than 10 years.Because both SIT and IFN-g tests are based in a common

inciple (detection of cellular immune response) one maypect their results to be conditionally dependent on eachher (Gardner et al., 2000). Thus, a Bayesian model tosess test accuracy for two conditionally dependent testsas adjusted to the data (Branscum et al., 2005) in order toantify the sensitivity and specific of each test and the

intensity of the tests results conditional dependence. Allanalyses were implemented in the WinBugs software(Lunn et al., 2000) and posterior distributions werecomputed after burn-out of the initial 500 iterations.The model was checked for convergence using twodifferent initial chains of values and also for lack ofautocorrelation in the simulations. A sensitivity analysiswas performed assuming uniform prior non-informativedistributions (uniform 0, 1) alternatively for the sensitivityand specificity of each test and the expected prevalence.Overlapping of probability intervals was consideredevidence of model robustness.

3. Results

3.1. Descriptive results

Total number of reactors to the SIT test ranged from 143(2.31%) to 107 (1.73%) using the severe and standardinterpretation respectively (Table 2). Within-herd percen-tage of reactors in the SIT test ranged from 0% to 16%(mean = 2.73%, 95% CI = 1.55–3.91). Reactors to the IFN-gassay were also more numerous (n = 1007; 16.24%) whenthe 0.05 cut-off point (severe interpretation) was appliedcompared to the number (n = 718; 11.58%) when thethreshold was set at 0.1 (standard interpretation).Percentage of positive animals (using 0.05 as the cut-offpoint) varied from 1% to 70% (mean = 19.52%, 95%CI = 13.85–25.18).

3.2. Model results

Estimates of sensitivity and specificity (median and 95%CI) for the tests (SIT and IFN-g) and interpretations(standard, severe) are shown in Table 1. Because allpossible combinations of diagnostic criteria for both tests

ble 1

or (mode and low 95% credibility interval (CI) bound) and posterior estimates (median and 95% CI) for sensitivity, specificity and prevalence of disease (%)

tained for each of the combinations of diagnostic tests and interpretation criteria on 6202 cattle from Castilla and Leon, Spain.

iagnostic interpretation/threshold SIT test estimates IFN assay estimates Prevalence

Sensitivity Specificity Sensitivity Specificity

rior estimates

Severe interpretation 83.9 (95% conf. >60) 95 (95% conf. >75) 92 (95% conf. >80) 90 (95% conf. >80) 10 (95% conf. <20)

Standard interpretation 73.9 (95% conf. >50) 98 (95% conf. >82) 85 (95% conf. >75) 98 (95% conf. >90)

osterior estimates

SIT test: severe

IFN-g assay 0.05 69.4 (40.1–92.2) 99.4 (98.7–99.9) 89.3 (77.5–97.2) 85.7 (84.4–87.6) 2.59 (1.48–4.84)

IFN-g assay 0.1 66.1 (35.3–91.3) 99.3 (98.6–99.8) 83.1 (71.9–91.4) 90.3 (89.1–92.5) 2.52 (1.38–5.17)

SIT test: standard

IFN-g assay 0.05 56.6 (29.2–83.2) 99.7 (99.1–100) 90 (78.9–96.7) 85.7 (84.3–88) 2.58 (1.48–5.04)

IFN-g assay 0.1 53 (27.3–81.5) 99.6 (99–100) 83.5 (73.6–91.6) 90.4 (89.1–92.7) 2.71 (1.38–5.48)

ble 2

mber of reactors to single intradermal tuberculin (SIT) test and interferon-gamma (IFN-g) assay performed on 6202 cattle in Castilla and Leon (Spain) for

ch of the combinations of diagnostic tests and interpretation criteria.

IT test IFN-g assay SIT+/IFN+ SIT+/IFN� SIT�/IFN+ SIT�/IFN�

evere 0.05 113 30 894 5165

0.1 93 50 625 5434

tandard 0.05 87 20 920 5175

0.1 75 32 643 5452

wsdvceds

arbvc(TchthwEbuimpw(�

aarthgoedwin(smdthrthsecccScinsvcs

ePIFinm

J. Alvarez et al. / Veterinary Microbiology 155 (2012) 38–43 41

ere tested (severe and standard SIT test and severe andtandard IFN-g assay), two values for each of the fouriagnostic approaches were computed; both values wereery similar to each in all cases, thus suggestingonsistency of the estimates. The higher variability wasstimated for the sensitivity of the SIT test, with aifference of 3–4 percentiles on the median of both theevere and standard interpretations (Table 1).

Estimates for sensitivity of the SIT test were consider-bly lower than the prior values introduced in the modelegardless the interpretation criteria, with a median valueetween 66.06 and 69.41 (severe interpretation). Con-ersely, estimates for specificity were very high, withomparable values for both severe and standard criteriaable 1). Estimates of parameters of IFN-g assay were

loser to the prior values introduced in the model, with aigh sensitivity (�89% and 83% for a 0.05 and 0.1resholds respectively) and specificity (�85% and 90%ith a 0.05 and 0.1 cut-off point respectively) (Table 1).

ven with the severe criteria specificity was estimated toe below 84% (95% CL). Increase in specificity obtainedsing the standard interpretation (0.1 as cut-off point)

plied a decrease in the sensitivity of approximately 6ercentiles. Estimated overall prevalence in the animalsas low regardless the diagnostic combination applied2.6%, 1.4–5.4 95% CI).Conditional dependence between SIT test and IFN-g

ssay was very low in both infected and non-infectednimals (correlation coefficient of positive and negativeesults ranging, in both cases, between �0.14 and 0.07 for

e different combinations of test interpretations) sug-esting that tests results were independent from eachther. Because of the absence of conditional dependencestimated here, an alternative model, which assumed thatiagnostic tests results were independent from each other,as fitted to the data that were obtained when tests wereterpreted using Spanish official interpretation criteriaevere SIT test, 0.05 threshold for IFN-g assay). Bothodels, assuming conditional dependence and indepen-

ence, led to similar results, as indicated by the values ofe deviation inference criterion (DIC = 25.31 and 25.32,

espectively) and the overlapping credibility intervals ofe estimated parameters (data not shown). These findings

uggest that diagnostic test results were independent fromach other. Assuming lack of correlation between out-omes of both tests, sensitivity and specificity of theombined application in parallel of both diagnostic testsan be computed as: SeðSITÞ þ SeðIFNÞ � SeðSITÞ �eðIFNÞ and SpðSITÞ � SpðSITÞ respectively. Therefore,ombined interpretation of the SIT and IFN tests applied

parallel with cut-off values currently used in Spain (i.e.,evere and 0.05, respectively), and considering the medianalues for individual tests sensitivity and specificityomputed here would be 95.6% and 85.1% for the combinedensitivity and specificity, respectively.

Models converged properly and autocorrelation wasliminated by burning out one every ten iterations.osterior estimates of specificity of the SIT and of theN-g assay were not affected by the use of non-formative priors, which was interpreted as evidence ofodel robustness. However, when non-informative priors

were assumed for prevalence, for sensitivity of either test,or for specificity of the SIT test, the sensitivity of the SITdecreased (median = 14%), whereas the specificity of IFN-gassay and prevalence were higher (median = 99% and 15%,respectively). Although the model was sensitive to theselection of priors, as indicated by the absence of over-lapping in the credibility intervals of the predictions, suchvariation did not affect the conclusions of the analyses.

4. Discussion

A Bayesian approach was used here to estimate the fielddiagnostic sensitivity and specificity of diagnostic techni-ques currently used in the Spanish eradication program ininfected herds after a first disclosure test and in theabsence of a gold standard test. Noteworthy, specificity ofthe BT tests has been assessed here, for the first time, ininfected cattle populations. A one-population approachwas used because all herds were located in the same regionand subjected to the same TB policies (regular skin testingperformed under the supervision of Official VeterinaryServices), and therefore one would expect similar low-level prevalence in infected herds. Most important, theobjective of the study was to estimate the sensitivity andspecificity of the test in the Castilla and Leon, including alltype of herd in the region under study. For that reason, theresults of this analysis will contribute to the evaluation ofthe expectations associated with the application of the TBcontrol program in a region as whole.

Due to the subjective nature of skin tests, a number offactors can affect its specificity and, specially, its sensitiv-ity, what has lead to a very wide range of reported values,estimated using different methods and under differentconditions (reviewed in de la Rua-Domenech et al., 2006).For this reason, to reflect the large uncertainty regardingtrue sensitivity of SIT test, a wide 95% CI was used to modelits uncertainty (Table 1) with both severe and standardinterpretation. Posterior estimates yielded also a wide 95%CI, but also provided evidence pointing to sensitivityvalues considerably lower than those introduced as priorsin the model (median of 66.1–69.4 and 53–56.6 for severeand standard criteria, respectively, Table 1). These resultsare consistent with a field scenario in which SIT test hasbeen performed in an infected herd 2–6 months after aprevious test; most of the SIT reactors have already beenremoved, and an important proportion of the true infectedanimals that will remain in these herds will be in the veryearly stages of infection, when SIT test may have adecreased reliability (Monaghan et al., 1994). In addition,the inclusion in the analysis of a large number ofbullfighting cattle, in which SIT test may be particularlydifficult to perform, may also account in part for thisdecreased sensitivity. Therefore, under the conditionsobserved here, BT-infection may be removed from a herdusing routine SIT testing only after an extended period oftime. Specificity posterior estimates were much closer tothe prior estimates, and confirmed the expected highspecificity of skin testing regardless the interpretationcriteria. Interestingly, even though the standard inter-pretation of the test yielded a higher specificity posteriorestimate (median 99.6–99.7) compared to the severe

crabobsere

prthallaimththFoceteang

thes(malscoinbereththreanexIFNprreatDe0.1im1)apdein0.0wm

prinaltinetdeHoindesuinhiap(AInas

J. Alvarez et al. / Veterinary Microbiology 155 (2012) 38–4342

iteria (median 99.3–99.4), this difference was consider-ly lower than the decrease in diagnostic sensitivityserved if the standard criteria was used instead of thevere interpretation (median 53–56.6 versus 66.1–69.4spectively).

Because the IFN-g assay is a laboratory test, inter-etation of results can be more easily standardized thanose of the SIT. In addition, its quantitative outcomeows the use of different cut-off points depending on the

of the application of the test to, alternatively, increasee test sensitivity or its specificity. However, the result ofe test can be largely affected by the quality of the sample.r those reasons, along with host-related factors, there isrtain uncertainty associated with the accuracy of thest, which has been reflected in the prior estimates for thealysis (Table 1). Posterior estimates of sensitivity of IFN-assay confirmed the usefulness of this test to maximizee number of infected animals detected in infected herds,pecially when the severe cut-off point (0.05) was usededian value 89.3–90%, Table 1). Specificity values wereo among the expected values. These values are inntrast with the low proportion of IFN-g positive animals

the same geographical area from which M. bovis hasen isolated, as low as 30% (Alvarez et al., 2009). Thissult may be explained, as least in part, by the conditionat positive herds were subjected to frequent SIT testsat removed most of the new infected animals. Thus, onlycently infected animals (in which the bacteriologicalalysis has a very low negative predictive value) can bepected to react to this test. Our results indicate that all-g-reactors from an infected herd have a very high

obability of being truly infected, despite the M. bovis

covery rate obtained if microbiological culture istempted, as reported before (TB Advisory Group,partment for Environment, 2009). Choice of 0.05 or

threshold seemed to have a similar (and inverse)pact on sensitivity and specificity of IFN-g assay (Table

. However, under Spanish conditions IFN-g test isplied to maximize the number of infected animalstected, and therefore the aim of its application is tocrease diagnostic sensitivity; in this context the use of5 cut-off point guarantees the maximum sensitivity

hile keeping sustainable specificity levels as recom-ended by the manufacturer.

Both skin test and IFN-g assay are based on the sameinciple, the detection of the cellular immune responseduced by M. bovis (Pollock et al., 2005); therefore,hough its ability to detect different subpopulations of

fected animals has been demonstrated before (Neill al., 1994; Pollock et al., 2005), a certain degree ofpendence between test results could be expected.wever, results of the codependence terms in both

fected and non-infected populations show that thegree of dependence seemed to be very low in animalsbjected to frequent skin testing when the IFN-g assay istroduced for the first time in infected herds, thereforeghlighting the complementarity of this diagnosticproach, as has been previously reported in livestocklvarez et al., 2008; Aranaz et al., 2006; Neill et al., 1994).

this particular setting, proportion of reactors to the IFN-

skin tests, as observed in all herds in this study. Parallelinterpretation of both tests therefore allows a considerableincrease in diagnostic sensitivity while keeping a reason-able specificity.

For some of the parameters, the model was sensitive tothe selection of the priors. However, the conclusions of theanalysis remain unaffected. For example, use of noninformative prior distributions led to even lower estimatesfor sensitivity of the SIT test and higher values for thespecificity of the IFN-g assay and prevalence than thoseestimated when informative prior distributions wereassumed. Such results reinforce the main findings of thestudy, i.e., the high specificity of the IFN-g assay, lowsensitivity of the skin test in the first repetition after thedisclosure test, and usefulness of parallel interpretation.

In summary, our study provided estimations for bothsensitivity and specificity of tuberculosis diagnostic tools(SIT test and first application of IFN-g assay) under fieldconditions. Our data demonstrate that currently usedinterpretation criteria for both SIT test IFN-g assaymaximize diagnostic sensitivity while maintaining anadequate specificity; lack of correlation between theoutcomes of each test reinforces the usefulness of theparallel interpretation. Further studies using data fromfirst disclosure tests would be needed to make an overallassessment of the accuracy of the current diagnosticstrategy used in Spain.

Acknowledgements

This research was funded by European Project 212414‘‘Strategies for the eradication of bovine tuberculosis (TB-STEP)’’. Julio Alvarez is the recipient of a post-doctoralgrant of the University of Castilla la Mancha (UCLM).

References

Alvarez, J., de Juan, L., Bezos, J., Romero, B., Saez, J.L., Reviriego Gordejo, F.J.,Briones, V., Moreno, M.A., Mateos, A., Dominguez, L., Aranaz, A., 2008.Interference of paratuberculosis with the diagnosis of tuberculosis ina goat flock with a natural mixed infection. Vet. Microbiol. 128, 72–80.

Alvarez, J., Marques, S., Saez-Llorente, J.L., de Juan, L., Romero, B., Grau, A.,Mateos, A., Minguez, O., 2009. Evaluacion de las medidas incorpor-adas en el programa de erradicacion de la tuberculosis bovina deCastilla y Leon. Centro VISAVET, Editorial Complutense S.A. Madrid.

Anon, 2006. Working Document on Eradication of Bovine Tuberculosis inthe EU Accepted by the Bovine Tuberculosis Subgroup of the TaskForce on Monitoring Animal Disease Eradication. SANCO/10200/2006.

Aranaz, A., de Juan, L., Bezos, J., Alvarez, J., Romero, B., Lozano, F., Paramio,J.L., Lopez-Sanchez, J., Mateos, A., Dominguez, L., 2006. Assessment ofdiagnostic tools for eradication of bovine tuberculosis in cattle co-infected with Mycobacterium bovis and M. avium subsp. paratubercu-losis. Vet. Res. 37, 593–606.

Branscum, A.J., Gardner, I.A., Johnson, W.O., 2005. Estimation of diagnos-tic-test sensitivity and specificity through Bayesian modeling. Prev.Vet. Med. 68, 145–163.

Cleggs, T.A., Duignan, A., Whelan, C., Gormley, E., Good, M., Clarke, J., Toft,N., More, S.J., 2011. Using latent class analysis to estimate the testcharacteristics of the g-interferon test, the single intradermal com-parative tuberculin test and a multiplex immunoassay under Irishconditions. Vet. Microbiol. 151, 68–76.

Collins, J.D., 2006. Tuberculosis in cattle: strategic planning for the future.Vet. Microbiol. 112, 369–381.

Cousins, D.V., Roberts, J.L., 2001. Australia’s campaign to eradicate bovine

tuberculosis: the battle for freedom and beyond. Tuberculosis(Edinb.) 81, 5–15. say would be usually larger than those reacting in the

d

D

D

E

F

G

G

L

M

M

M

J. Alvarez et al. / Veterinary Microbiology 155 (2012) 38–43 43

e la Rua-Domenech, R., Goodchild, A.T., Vordermeier, H.M., Hewinson,R.G., Christiansen, K.H., Clifton-Hadley, R.S., 2006. Ante mortemdiagnosis of tuberculosis in cattle: a review of the tuberculin tests,gamma-interferon assay and other ancillary diagnostic techniques.Res. Vet. Sci. 81, 190–210.

rewe, J.A., Dean, G.S., Michel, A.L., Lyashchenko, K.P., Greenwald, R.,Pearce, G.P., 2009. Accuracy of three diagnostic tests for determiningMycobacterium bovis infection status in live-sampled wild meerkats(Suricata suricatta). J. Vet. Diagn. Invest. 21, 31–39.

rewe, J.A., Tomlinson, A.J., Walker, N.J., Delahay, R.J., 2010. Diagnosticaccuracy and optimal use of three tests for tuberculosis in livebadgers. PLoS One 5, e11196.

noe, C., Georgiadis, M.P., Johnson, W.O., 2000. Estimation of sensitivityand specificity of diagnostic tests and disease prevalence when thetrue disease state is unknown. Prev. Vet. Med. 45, 61–81.

osgate, G.T., Adesiyun, A.A., Hird, D.W., Hietala, S.K., 2006. Likelihoodratio estimation without a gold standard: a case study evaluating abrucellosis c-ELISA in cattle and water buffalo of Trinidad. Prev. Vet.Med. 75, 189–205.

ardner, I.A., Stryhn, H., Lind, P., Collins, M.T., 2000. Conditional depen-dence between tests affects the diagnosis and surveillance of animaldiseases. Prev. Vet. Med. 45, 107–122.

ormley, E., Doyle, M.B., Fitzsimons, T., McGill, K., Collins, J.D., 2006.Diagnosis of Mycobacterium bovis infection in cattle by use of thegamma-interferon (Bovigam) assay. Vet. Microbiol. 112, 171–179.

unn, D., Thomas, A., Spiegelhalter, D., 2000. WinBUGS – a Bayesianmodelling framework: concepts, structure, and extensibility. Stat.Comput. 10, 325–337.

ichel, A.L., Muller, B., van Helden, P.D., 2010. Mycobacterium bovis at theanimal–human interface: a problem, or not? Vet. Microbiol. 140,371–381.

onaghan, M.L., Doherty, M.L., Collins, J.D., Kazda, J.F., Quinn, P.J., 1994.The tuberculin test. Vet. Microbiol. 40, 111–124.

uller, B., Vounatsou, P., Ngandolo, B.N., guimbaye-Djaibe, C., Schiller, I.,Marg-Haufe, B., Oesch, B., Schelling, E., Zinsstag, J., 2009. Bayesianreceiver operating characteristic estimation of multiple tests fordiagnosis of bovine tuberculosis in Chadian cattle. PLoS One 4, e8215.

Muma, J.B., Toft, N., Oloya, J., Lund, A., Nielsen, K., Samui, K., Skjerve, E.,2007. Evaluation of three serological tests for brucellosis in natu-rally infected cattle using latent class analysis. Vet. Microbiol. 125,187–192.

Neill, S.D., Cassidy, J., Hanna, J., Mackie, D.P., Pollock, J.M., Clements, A.,Walton, E., Bryson, D.G., 1994. Detection of Mycobacterium bovisinfection in skin test-negative cattle with an assay for bovine inter-feron-gamma. Vet. Rec. 135, 134–135.

Pollock, J.M., McNair, J., Welsh, M.D., Girvin, R.M., Kennedy, H.E., Mackie,D.P., Neill, S.D., 2001. Immune responses in bovine tuberculosis.Tuberculosis (Edinb.) 81, 103–107.

Pollock, J.M., Welsh, M.D., McNair, J., 2005. Immune responses in bovinetuberculosis: towards new strategies for the diagnosis and control ofdisease. Vet. Immunol. Immunopathol. 108, 37–43.

TB Advisory Group (Department for Environment, Food and Rural Affairs),2009. Bovine tuberculosis in England: towards eradication. Finalreport of the Bovine TB advisory group.

Vordermeier, M., Ewer, K., 2006. Specificity trial of the BOVIGAMIFN-gamma test in GB cattle. Veterinary Laboratories Agency,pp. 1–29.

Vordermeier, M., Goodchild, A., Clifton-Hadley, R., de la, R.R., 2004. Theinterferon-gamma field trial: background, principles and progress.Vet. Rec. 155, 37–38.

Wood, P.R., Jones, S.L., 2001. BOVIGAM: an in vitro cellular diagnostic testfor bovine tuberculosis. Tuberculosis (Edinb.) 81, 147–155.

Wood, P.R., Corner, L.A., Plackett, P., 1990. Development of a simple, rapidin vitro cellular assay for bovine tuberculosis based on the productionof gamma interferon. Res. Vet. Sci. 49, 46–49.

Wood, P.R., Corner, L.A., Rothel, J.S., Baldock, C., Jones, S.L., Cousins, D.B.,McCormick, B.S., Francis, B.R., Creeper, J., Tweddle, N.E., 1991. Fieldcomparison of the interferon-gamma assay and the intradermaltuberculin test for the diagnosis of bovine tuberculosis. Aust. Vet. J.68, 286–290.

Wood, P.R., Corner, L.A., Rothel, J.S., Ripper, J.L., Fifis, T., McCormick, B.S.,Francis, B., Melville, L., Small, K., de, W.K., 1992. A field evaluation ofserological and cellular diagnostic tests for bovine tuberculosis. Vet.Microbiol. 31, 71–79.