herd-level true Mycobacterium avium subspecies paratuberculosis prevalence on sheep, beef cattle and deer farms in New Zealand using a novel Bayesian model

Preventive Veterinary Medicine 117 (2014) 447–455

Contents lists available at ScienceDirect

Preventive Veterinary Medicine journal homepage: www.elsevier.com/locate/prevetmed

Estimation of flock/herd-level true Mycobacterium avium subspecies paratuberculosis prevalence on sheep, beef cattle and deer farms in New Zealand using a novel Bayesian model Cristobal Verdugo a,b,∗ , Geoff Jones d , Wes Johnson e , Peter Wilson c , Lesley Stringer b,f , Cord Heuer b a

Instituto de Medicina Preventiva Veterinaria, Universidad Austral de Chile, Valdivia, Chile Epicentre, Massey University, Palmerston North, New Zealand c Institute of Veterinary, Animal and Biomedical Sciences, Massey University, Palmerston North, New Zealand d Institute of Fundamental Sciences, Massey University, Palmerston North, New Zealand e Department of Statistics, University of California, Irvine, USA f Veterinary Epidemiology Unit, Department of Agriculture and Rural Development, Belfast, Northern Ireland, United Kingdom b

a r t i c l e

i n f o

Article history: Received 13 May 2013 Received in revised form 3 October 2014 Accepted 4 October 2014 Keywords: Mycobacterium avium subsp. paratuberculosis Pastoral systems Prevalence Bayesian analysis

a b s t r a c t The study aimed to estimate the national- and island-level flock/herd true prevalence (HTP) of Mycobacterium avium subsp. paratuberculosis (MAP) infection in pastoral farmed sheep, beef cattle and deer in New Zealand. A random sample of 238 single- or multi-species farms was selected from a postal surveyed population of 1940 farms. The sample included 162 sheep flocks, 116 beef cattle and 99 deer herds from seven of 16 geographical regions. Twenty animals from each species present on farm were randomly selected for blood and faecal sampling. Pooled faecal culture testing was conducted using a single pool (sheep flocks) or two pools (beef cattle/deer herds) of 20 and 10 samples per pool, respectively. To increase flock/herd-level sensitivity, sera from all 20 animals from culture negative flocks/herds were individually tested by Pourquier® ELISA (sheep and cattle) or ParalisaTM (deer). Results were adjusted for sensitivity and specificity of diagnostic tests using a novel Bayesian latent class model. Outcomes were adjusted by their sampling fractions to obtain HTP estimates at national level. For each species, the posterior probability (POPR) of HTP differences between New Zealand North (NI) and South (SI) Islands was obtained. Across all species, 69% of farms had at least one species test positive. Sheep flocks had the highest HTP estimate (76%, posterior probability interval (PPI) 70–81%), followed by deer (46%, PPI 38–55%) and beef herds (42%, PPI 35–50%). Differences were observed between the two main islands of New Zealand, with higher HTP in sheep and beef cattle flocks/herds in the NI. Sheep flock HTP was 80% in the NI compared with 70% (POPR = 0.96) in the SI, while the HTP for beef cattle was 44% in the NI and 38% in the SI (POPR = 0.80). Conversely, deer HTP was higher in the SI (54%) than the NI (33%, POPR = 0.99). Infection with MAP is endemic at high prevalence in sheep, beef cattle and deer flocks/herds across New Zealand. © 2014 Elsevier B.V. All rights reserved.

∗ Corresponding author at: Instituto de Medicina Preventiva Veterinaria, Universidad Austral de Chile, Valdivia, Chile. Tel.: +56 63 222 1117; fax: +56 63 229 3233. E-mail address: [email protected] (C. Verdugo). http://dx.doi.org/10.1016/j.prevetmed.2014.10.004 0167-5877/© 2014 Elsevier B.V. All rights reserved.

448

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455

1. Introduction

2. Materials and methods

Paratuberculosis (Ptb), caused by Mycobacterium avium subspecies paratuberculosis (MAP), is a chronic granulomatous enteric disease, which occurs worldwide and affects domestic ruminant species including deer, sheep, and cattle (Harris and Barletta, 2001). A large proportion of infected animals will remain subclinically infected during their entire productive life, with no measurable effect on production (Nielsen and Toft, 2008). The isolation of MAP from intestinal tissue of Crohn’s disease (CD) patients, a human chronic inflammatory bowel disease, has suggested a possible link between the two diseases, with MAP-contaminated animal produce a possible source of CD in humans (Mishina et al., 1996; Feller et al., 2007). The apparent association between MAP in animals and humans has raised concerns within the agricultural industry about the public health relevance of MAP, and the potential effect of high MAP prevalence on trade or consumption of milk and meat. The first Ptb case in New Zealand was recorded in 1912 in an imported cow (de Lisle, 2002). The disease was first reported in New Zealand sheep in 1952, and in the 1980s Ptb was confirmed in farmed deer (de Lisle et al., 1993). Despite MAP infection being thought to be endemic in the New Zealand, no population-based estimates of MAP infection prevalence or disease incidence are currently available for sheep, beef or dairy cattle. Conversely, in deer, a national herd-level true prevalence (HTP) of 59% has been estimated, based on tissue culture of randomly selected lymph nodes from abattoir surveillance (Stringer et al., 2013b). Moreover, a significant prevalence difference was reported between New Zealand North (NI) and South (SI) Islands, 44% and 67% respectively. However, no farm-based estimate is available for this species. Infection prevalence estimation is a key element to assess the disease impact, and for the design of control programmes. MAP prevalence has been previously estimated in other countries using Bayesian latent-class modelling (Sergeant and Baldock, 2002; Nielsen et al., 2007; Dhand et al., 2010; Okura et al., 2010). This methodology has the advantage that adjustments can be made for sensitivity and specificity of diagnostic tests, in a flexible probabilistic framework, allowing researchers to model complex testing designs to obtain true prevalence estimates. In New Zealand, domestic ruminants are commonly farmed in multi-species pastoral systems, where sheep, beef cattle and/or deer are often grazed on the same pastures, either concurrently or successively. Therefore, Ptb research in New Zealand addresses all species in an overarching strategy, in order to develop an integrated approach to Ptb control (JDRC, 2011). The objective of this study was to estimate HTP of MAP infection in sheep, beef cattle and deer flocks/herds in New Zealand, at national and island level (North Island (NI) and South Island (SI)). The study also aimed to assess the distribution of the infection in the populations, and to explore/validate prevalence differences between islands, as it has previously been reported for deer herds.

2.1. Selection of farms Farms were selected in two stages. In the first stage (a), from December 2008 to March 2009 a postal survey was mailed out to 7998 farmers. The survey simultaneously gathered information about two different diseases (Ptb and leptospirosis). The contacted farmers were clients of 28 large animal veterinary practices, located in four administrative regions in the NI (Waikato, Wairarapa, Hawkes Bay, Manawatu-Wanganui), and three in the SI (Marlborough, Canterbury, Southland). The survey targeted ‘commercial’ sheep, deer, and/or beef cattle operations (i.e. those with a minimum of 40 deer, 400 sheep, and/or 40 beef cattle). In New Zealand, there are 24,644 farms meeting these criteria with at least one of the three species under study, based on 2009 AgribaseTM data (national population database). The questionnaire gathered retrospective information about animal demographics, reproduction performance, Ptb & leptospirosis incidence, and grazing management information on all ruminant species present on the farm. Those data are reported elsewhere (Verdugo et al., 2010). Correctly completed questionnaires were returned from 1940 (24.3%) commercial single- or mixed-species farms, constituting the sampling frame (source population). In the second stage (b), farms were randomly selected from the source population with the aim of sampling of one group of each ruminant species present (sheep, beef cattle and/or deer).

2.2. Sampling protocol and laboratory testing Sampling was carried out by contracted veterinary practitioners from June 2009 to July 2010. Twenty animals from each species present on farm were randomly selected.1 Paired faeces and serum samples were collected from sheep (ewes, 2-years and older), beef cattle (cows, 2-years and older) and deer (yearlings, i.e. 12–24 months, either sex). A single pool was prepared from sheep faeces (20 samples/pool), and two pools were prepared from beef cattle and deer (10 samples/pool) from each farm. Culture was performed by the Wallaceville Animal Health Laboratory, Upper Hutt, using BACTEC 12B liquid culture medium containing egg yolk and mycobactin, after a decontamination step with cetylpyridinium chloride, as described by Whittington et al. (1999). If a pooled faecal culture (PFC) was positive, the entire flock/herd was classified as being positive. Individual serum samples from culture negative flocks/herds only were tested by ELISA tests: Pourquier® ELISA for sheep and cattle (Institut Pourquier, Montpellier, France), and ParalisaTM for deer (Griffin et al., 2005). In a given species, a flock/herd was defined as positive if a PFC pool or ELISA test (group level interpretation, cut off = 1 + ve animal) was positive.

1 Animals belonging to the target age group were rounded up and yarded. Then, animals were selected using a systematic random sampling procedure as they were released from the yard.

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455

2.3. Statistical analysis 2.3.1. Model The design, which involved first testing pooled samples using a PFC test followed by individual ELISA testing for PFC negative flocks/herds, required the development of a novel statistical model. Specifically, a Bayesian latent class model was developed to estimate HTP for each species at national and island levels. The HTP was defined as the percentage of flocks/herds with at least one MAP-infected animal. The model used (a) flock/herd-level results from the twotest protocol described above, (b) prior information about test performance characteristics and (c) prior information about animal-level true prevalence (TP), derived from independent or peer-reviewed studies, to obtain posterior distributions of HTP. For each flock/herd, denoted as (H), a sample, denoted as (I), of size n animals (n = k × r corresponding to r pools of size k), was collected. Since one infected animal determined an infected flock/herd, a sample of animals that had at least one infected animal (I+ ) could only be drawn from an infected flock/herd (H+ ). On the other hand, a sample containing only non-infected animals (I− ) could be drawn from either a non-infected flock/herd (H− ) or an H+ . Let T1 denote the binary flock/herd level PFC test outcome and T2 the binary flock/herd level ELISA test outcome, where T+ denotes a test-positive flock/herd, and T− a test-negative flock/herd. For each flock/herd, the outcome is one of the following; (T1+ ; T1− ; T2+ ; T1− ; T2− ). When combined across flocks/herds for a given island (i), the flock/herd level testing protocol resulted in the vector of corresponding flock/herd level counts (yi ), that was assumed to be distributed as multinomial, namely: yi ∼multinomial(Xi , nhi ) where Xi is the vector of probabilities corresponding to the three possible test outcomes, and nhi is the number of flocks/herds sampled on island i. The vector of probabilities Xi was modelled under the assumption of conditional independence between PFC and ELISA outcomes, given infection status I of the sampled animals. Thus, the multinomial cell probabilities for Xi are: Pr(T1+ ) = Pr(I − ) × Pr(T1+ |I − ) + Pr(I + ) × Pr(T1+ |I + )

(1)

Pr(T1− T2+ ) = Pr(I − ) × Pr(T1− |I − ) × Pr(T2+ |I − ) + Pr(I + ) × Pr(T1− |I + ) × Pr(T2+ |I + )

(2)

Pr(T1− T2− ) = 1 − Pr(T1+ ) − Pr(T1− T2+ ).

(3)

Thus what is required for computation is Pr(I+ ), the probability that the sample of n contains at least one infected unit, and the conditional probabilities Pr(T1+ |I + ), Pr(T1− |I − ), Pr(T2+ |I + ) and Pr(T2− |I − ), which might be referred to as the sample level sensitivity and specificity of the PFC and ELISA tests, respectively. The rest of the inputs in equations 1–3 are complementary probabilities. The above probabilities are determined by several factors, namely: (a) herd level true prevalence (HTPi ) in a given island (i = 1, 2), (b) the TP fraction in an average

449

infected flock/herd, assumed to be the same in NI and SI flocks/herds, (c) the pool level sensitivity (Pse) and specificity (Psp) of the PFC test, and (d) the animal level sensitivity (Se) and specificity (Sp) of the ELISA test. In order to perform the statistical analysis, it is necessary to be able to express all of the components in equations 1–3, in terms of the basic model parameters HTPi , TP, Pse, Psp, Se and Sp. These expressions are derived in Appendix. For the two-test scenario, joint herd-level sensitivity (Hsej) and specificity (Hspj) were derived using basic probability calculations, as above, and using the facts that Pr(Tt+ |I − , H + ) = Pr(Tt+ |I − ), Pr(Tt+ |I + , H + ) = Pr(Tt+ |I + ), for t = 1, 2, and Pr(T1− |I − , H − ) = Pr(T1− |I − ), resulting in: Hsej = Pr(I − |H + ) × (Pr(T1+ |I − ) + Pr(T1− |I − ) × Pr(T2+ |I − )) + Pr(I + |H + ) × (Pr(T1+ |I + ) + Pr(T1− |I + ) × Pr(T2+ |I + )) Hspj = Pr(T1− |I − ) × Pr(T2− |I − ) Prevalence estimates for each island were weighted by their sampling fractions to obtain weighted averages as national level HTP estimates. Weights are the proportion of flocks/herds present in each island among all flocks/herds present in the national population. The weighted national estimate for a given species is: HTPN ∼(HTP1 × WNI + HTP2 × WSI ) where HTPN is the adjusted national flock/herd level true prevalence for a given species, and WNI and WSI are the respective NI and SI weights rescaled to add to one. Comparison of HTPi distributions between NI and SI was done by computation of posterior probabilities (POPR); testing the hypothesis that NI has higher HTPi than that of SI. A POPR is approximated by the proportion of Monte Carlo (MC) samples where the hypothesis tested was true (Okura et al., 2010). A POPR value close to 1 (or 0) represents the probability that the first stratum has a higher (or lower) HTPi than the second one, meaning that the difference between the two prevalence is positive (or negative). For example, a POPR of 0.95 will indicate that 95% of the model iterations had a greater HTP in the NI than SI. Whereas, a POPR of 0.05 represents that 5% of the posterior distribution had greater HTP in the NI than SI. A POPR value around 0.5 implies that the two prevalences under comparison are similar. A difference was regarded as being unlikely due to chance, or statistically important, when POPR was either smaller than 0.1 or greater than 0.9. 2.3.2. Model priors For each species-specific model six parameters were used: TP, HTPi , Pse, Psp, Se, and Sp. Prior scientific information on these parameters was modelled using independent beta distributions. Available TP and HTPi information for sheep and beef cattle flocks/herds in New Zealand is sparse and estimates from other countries are not appropriate for New Zealand’s farming conditions. To overcome the lack of published information for these two parameters, data from an independent study was used to form the prior distributions. The data set is constituted by individual ELISA test results from 2274 animals, sourced from 67

450

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455

single and mixed species farms, belonging to a large farming corporation. These farms included 63 sheep flocks and 49 beef cattle herds. This data set was not included in the prevalence estimation reported in the present study, contributing exclusively to the construction of model priors. Test results were adjusted using a Bayesian latent class model, proposed by Branscum et al. (2004), to obtain TP and HTPi distributions. In this model, for the jth flock/herd, the number of test positive animals Zj , out of nj sampled, was assumed to be distributed as binomial: Zj ∼binomial(Wj , nj ) where Wj = TPj × Se + (1 − TPj ) × (1 − Sp) and TPj = TP ∗ × Vj Vj ∼bernoulli (HTP) Diffuse independent beta priors were used to model TP* and HTP. For deer herds, prior information was obtained based on a recent study conducted by Stringer et al. (2013b), which estimated TP and HTP in New Zealand’s NI and SI. ELISA and PFC tests characteristics were obtained from peer-reviewed literature. Beta parameters (a, b) were obtained using the free software BetaBuster available at http://www.epi.ucdavis.edu/diagnostictests/. Input parameter values, beta distributions, and references are presented in Table 1.

2.3.3. Sensitivity analysis Stringer et al. (2013a) reported faecal culture Se and Sp in deer, rather than Pse and Psp. Results from that study were used as a proxy to form Pse and Psp prior distributions in this species. Since posterior results may depend heavily on the prior, the effect of selecting different values was assessed by varying their respective values by ±10%. In the case of +10% Psp, new values were truncated to avoid values greater than one. POPR values were computed to assess differences in the HTP estimates between models. 2.3.4. Model computation Three parallel chains with different starting values were used to assess model convergence. The Gelman–Rubin statistic and the Gelman–Rubin-Brooks plot were used for assessing the convergence of the three chains. Models were fitted using the WinBUGS software (Spiegelhalter et al., 1996), which uses a Markov Chain Monte Carlo (MCMC) sampling algorithm to obtain a MC sample from the posterior distribution. Models were run for 25,000 iterations, after discarding the first 5000 MC samples as a burn-in period. Point and interval estimates were obtained as the median of the posterior distribution and the associated 95% posterior probability interval (PPI), respectively. 3. Results A total of 238 farms were sampled, representing 162 sheep flocks, 116 beef cattle herds and 99 deer herds (7540 animals). Fig. 1 shows the spatial distribution of the sampled farms. Four veterinary practices provided services to neighbouring regions, thus some farms located

Table 1 Prior parameters, mode and 95% percentile of Pool sensitivity (Pse) and Pool specificity (Psp) of pooled faecal culture of MAP in liquid media (BACTEC). Sensitivity (se) and specificity (sp) of ELISA and ParalisaTM tests for MAP, and animals (TP) and flock/herd (HTP) level true prevalence for sheep, beef cattle and deer flocks/herds in New Zealand. Mode

5%/95% percentile

a

Sheep Pse Psp ELISA se ELISA sp TP HTP

0.23 0.999 0.349 0.988 0.13 0.78

0.54 0.985 0.430 0.983 0.50 0.45

2.74 226.11 36.78 1862.65 1.73 6.22

6.82 1.23 67.73 23.61 5.87 2.47

Dhand et al. (2010) Dhand et al. (2010) Sergeant et al. (2003) Sergeant et al. (2003) Calculated Calculated

Beef cattle Pse Psp ELISA se ELISA sp TP HTP

0.36 0.995 0.28 0.99 0.06 0.40

0.46 0.99 0.48 0.90 0.44 0.20

25.57 1137.51 5.94 31.08 1.32 5.03

44.68 6.71 13.71 1.15 6.05 7.04

Tavornpanich et al. (2004) Tavornpanich et al. (2004) Okura et al. (2010) Okura et al. (2010) Calculated Calculated

0.77 0.99 0.19 0.94

0.92 0.98 0.30 0.93

5.76 560.73 14.12 1690.58

2.42 6.65 57.01 108.85

Stringer et al. (2013a) Stringer et al. (2013a) Stringer et al. (2013a) Stringer et al. (2013a)

0.29 0.51

0.45 0.66

8.86 14.64

20.25 14.10

Stringer et al. (2013b) Stringer et al. (2013b)

0.44 0.67

0.64 0.85

7.90 7.26

9.78 4.08

Stringer et al. (2013b) Stringer et al. (2013b)

Deer Pse Psp ParalisaTM se ParalisaTM sp TP –NI –SI HTP –NI –SI

NI: New Zealand’s North Island, SI: New Zealand’s South Island.

b

Reference

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455

451

Fig. 2. Distribution of farm type strata for those with ≥400 sheep, ≥40 beef cattle and ≥40 deer, sampled in the present study compared with New Zealand farm statistics (AgribaseTM ). SHP: only sheep, BEE: only beef cattle, DEE: only deer, S&B: sheep & beef cattle, B&D: beef cattle & deer, S&D: sheep & deer, and SBD: sheep, beef cattle & deer.

in the regions of Otago and Taranaki (close to the border) were also sampled. The main regions represented in the study were Manawatu-Wanganui in the NI, and Canterbury and Southland in the SI. The distribution of the source population and AgribaseTM data is presented in Fig. 2 by farm species composition (FSC). For each sampled species, the flocks/herds size distributions and their flock/herd level apparent prevalences (HAP) are presented in Tables 2 and 3, respectively. At the farm level, 69% presented at least one test positive flock/herd, and an overall HAP of 71, 29, and 57% was observed for sheep, beef cattle and deer flocks/herds, respectively.

Fig. 1. Spatial distribution of sampled farms (N = 238).

Table 2 Number, minimum (Min), first quartile (Q1), median, third quartile (Q3), and maximum (Max) of flocks/herds size by Island. Species

Flock/herd size Island

No. flocks/herds

Min

Q1

Median

Q3

Max

p-Valuea

Sheep

NI SI Both

96 66 162

402 495 402

814 789 565

2940 2585 2925

13,270 8198 13,500

30,250 14,435 30,250

<0.01

Beef

NI SI Both

86 30 116

43 44 43

70 87 63

350 262 303

1285 611 1285

2314 948 2314

<0.01

Deer

NI SI Both

41 58 99

40 83 40

114 225 83

338 530 439

958 1950 3100

3640 7000 7000

<0.01

a

F-test for the null hypothesis that flock/herd size population in NI vs. SI have the same variance.

Table 3 Sampled population frequency and MAP flock/herd-level apparent prevalence (HAP) at national level (NAT), North Island (NI) and South Island (SI) strata. Sheep flocks

Beef cattle herds

Deer herds

Strata

N

HAP % (95% CI)

Strata

N

HAP % (95% CI)

Strata

N

HAP % (95% CI)

NAT NI SI

162 96 66

71 (67–75) 76 (72–80) 64 (58–70)

NAT NI SI

116 86 30

29 (25–34) 31 (26–36) 23 (15–31)

NAT NI SI

99 41 58

57 (52–62) 49 (41–57) 62 (56–68)

CI = Confidence interval.

452

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455

Table 4 Posterior median and 95% posterior probability interval (PPI) for the true flock/herd-level MAP prevalence of sheep, beef cattle and deer flocks/herds, at national level (NAT), North Island (NI) and South Island (SI). Flock/herd level true prevalence (%) Sheep flocks

Beef cattle herds

Deer herds

Strata

Posterior median %

95% PPI

Strata

Posterior median %

95% PPI

Strata

Posterior median %

95% PPI

NAT NI SI TP

76 80 70 20

70–81 71–87 62–77 6–53

NAT NI SI TP

42 44 38 13

35–50 34–54 30–47 2–46

NAT NI SI TP

46 33 54 40

38–55 24–44 44–67 24–58

Table 5 Herd level sensitivities and specificities of the PFC test (Hse and Hsp) and PFC + ELISA joint testing (Hsej and Hspj) of sheep, beef cattle and deer flocks/herds. Sheep

Hse Hsej Hsp Hspj

Beef cattle

Posterior median

95% PPI

0.50 0.90 0.99 0.78

0.34 0.63 0.98 0.70

0.61 0.99 1.00 0.85

Deer

Posterior median

95% PPI

0.42 0.69 0.99 0.94

0.14 0.25 0.98 0.84

The HTP estimates and posterior test performance distributions are summarized in Tables 4 and 5, respectively. Sheep flocks had the highest HTPN of the species under study, with 76% (95% PPI 70–81%) of flocks infected. In this species, estimated HTPi was higher in the NI than SI (80 and 70%, respectively), with an associated POPR of 0.96, representing a statistically important difference in prevalence between islands. The use of the ELISA test increased the flock/herd level sensitivity (Hse) from 0.50 (PFC only) to 0.90 (Hsej), while decreasing the flock/herd level specificity (Hsp) from 0.99 to 0.78 (Hspj). In beef cattle herds, a HTPN of 42% (95% PPI 35–50%) was estimated, this being the species with the lowest HTPN . There was a moderate trend for a difference in prevalence between islands (POPR = 0.80), with HTP estimates of 44 and 38% for NI and SI, respectively. Similar to the pattern observed in sheep flocks, ELISA testing of PFC negative herds increased Hse from 0.42 to 0.69 and decreased Hsp from 0.99 to 0.84. The HTPN in deer herds was estimated at of 46% (95% PPI 38–55%). Deer herds in the NI had lower HTPi (33%) than the SI (54%), with a POPR of 0.01 showing a high probability that the prevalences in the two islands are different. ParalisaTM testing in PFC negative herds resulted in a Hse increase from 0.88 to 0.98. However, a statistically important decrease in Hsp, from 0.98 to 0.38, was observed when ParalisaTM testing was used. Finally, the sensitivity analysis did not yield any important differences between the scenarios under comparison, validating the use of Se and Sp estimates as a proxy to inform Pse and Psp priors in the deer model. 4. Discussion Mycobacterium avium subspecies paratuberculosis infection has been diagnosed in all ruminant livestock species in New Zealand. However, there has been no field level study to quantify the distribution of MAP, based on a random selection and sampling of farms and animals. Hence, this study presents the first robust prevalence estimate of MAP infection in New Zealand sheep, beef cattle, and deer

0.64 0.98 1.00 0.99

Posterior median

95% PPI

0.88 0.98 0.98 0.38

0.63 0.91 0.96 0.32

0.99 1.00 0.99 0.44

flocks/herds. Our analysis uses a novel Bayesian method for modelling the combined results of two tests, where one is a pooled test and the other is based on individual animal tests, to obtain HTP estimates. Model outputs were adjusted by their sampling fraction to obtain national level estimates. Results support the contention that MAP is an endemic, well established and widely spread infection in New Zealand, with a high proportion of sheep flocks, deer and beef cattle herds infected (76, 46 and 42%, respectively). Group level interpretation of two testing protocols was used to obtain HTP estimates in this study. A flock/herd was defined as infected, if at least one animal in the flock/herd was infected. At animal level, this target condition potentially includes latent infection, shedding and/or clinically affected animals (Nielsen and Toft, 2008). PFC needs viable MAP organisms to render a positive result. It is thus able to identify a flock/herd where infectious, affected or passive shedding animals are present. Due to the pass-through phenomenon (Sweeney, 1996), a non-infected animal (passive shedder) could render a positive PFC result (false positive). However, previous studies have used the presence of MAP in the farm-environment as an indicator of MAP infection in livestock (Raizman et al., 2004; Lombard et al., 2013). Therefore, PFC was regarded as a highly specific test for the assessment of the flock/herd MAP status. In PFC negative flocks/herds, the ELISA test was applied at individual level to obtain a group level assessment. The presence of at least one sero-positive animal was used as an indicator of a positive flock/herd. The combination of PFC and ELISA test had the advantage of increasing the Hse performance (Table 5). However, the repetition (20 animals per flock/herd) of an imperfectly specific test had the effect of decreasing the Hsp. This effect was observed most strikingly in deer herds, in which Hspj was 0.38, the effect being attributable to a relatively poor test Sp of ParalisaTM (Stringer et al., 2013a). In comparison, there was a lesser impact of the Pourquier® ELISA, in sheep and cattle. Indeed, this effect could explain the different patterns observed

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455

between HAP and HTP. Deer herds presented a higher HAP than HTP (57 and 46%, respectively), suggesting that a proportion of ParalisaTM positive herds were false positive. Conversely, sheep and beef cattle flocks/herds presented a higher HTP than HAP; representing a lack of adjustment for lower Se (mainly in beef cattle) of the testing protocols. The testing design used in this study was not suitable for use of the Rogan–Gladen methodology (Rogan and Gladen, 1978), partly because that method assumes a much simpler testing setting with a single test and that its sensitivity and specificity are completely known, and moreover because in any generalization to the more complex two stage testing protocol described here, it would be even more difficult to assume perfect knowledge of the performance of these diagnostic tests when used in series, under New Zealand conditions. In addition, the Rogan–Gladen estimator can easily give prevalence estimates that are negative or larger than one. To overcome these limitations a Bayesian approach was adopted. The serologic testing of only PFC negative flocks/herds precluded the use of standard models. Therefore, a novel model was developed, assuming conditional independence between culture and serologic tests at sample level (pool of animals). This approach differs from previous models where conditional independence has been assumed at flock/herd level (Branscum et al., 2004). Parameters used for the priors in this study were gathered from different sources. Test characteristics of PFC and ELISA tests were obtained directly from peer-reviewed studies, as were the TP and HTP estimates for deer herds. In the case of sheep and beef cattle, test results from an independent study of 112 flocks/herds were used to obtain prior information for TP and HTP. We regarded these 112 flocks/herds to be typical of New Zealand livestock farming systems, and thus consider that prior distributions are well informed. Outputs from a Bayesian model, based on independent data, have been previously used to form prior distributions for the estimation of MAP prevalence (Lombard et al., 2013). The sampled flocks/herds were randomly selected from a sampling frame of 1940 farms of the immediately preceding postal survey (source population). The source population represented a convenience sample of the general population (national). Therefore, model outputs should be interpreted with caution, due to the chance of differences in MAP distribution between the source and the general populations. However, the 28 veterinary practices enrolled in the study serviced 32.5% of ‘commercial’ farms with at least one of the three species and the surveyed regions are representative of different agroecological zones present in New Zealand. Furthermore, a comparison between the national and source populations (Fig. 2) shows a similar demographic distribution. We therefore considered that the source population was fairly representative of the general population, and therefore adequate to permit inferences at the national level. The response rate to the postal survey was 24.3%, which might suggest that the survey dataset was biased towards farmers experiencing Ptb. If present at all, such bias was expected to be reduced by inclusion of questions relevant to leptospirosis in the questionnaire. Leptospirosis is widespread in deer, beef cattle and sheep in New Zealand (Dorjee et al.,

453

2008; Subharat et al., 2009; Heuer et al., 2010), and has a different epidemiology to that of Ptb. Sampling fractions were applied to adjust the prevalence estimates, so that conclusions could be drawn at the national population level. The authors realize that an ideal study should involve random sampling of all premises that farm sheep, beef cattle, and/or deer, rather than a convenience sampling from the general population of farms. However, the approach used in this study was a pragmatic one, chosen because of the convenience of contacting farms through the veterinary practices, due to the pre-existing bond with farmers, facilitating their enrolment in the study and sampling activities. In particular, this last point was a key element, due to the logistical complexity of sampling flocks/herds in several regions of New Zealand. Further, the contact information held in the national population database is frequently outdated, hampering its direct use for the enrolment of farms into the study. Model outputs indicate a high MAP prevalence in New Zealand, mainly in sheep flocks. High MAP prevalence has also been reported elsewhere (Nielsen and Toft, 2009; Pozzato et al., 2011; Lombard et al., 2013; Stringer et al., 2013b). However, comparisons with such studies are difficult to draw due to differences in the target population, farming systems or sampling methods. Lombard et al. (2013) used a Bayesian latent class model to adjust for test characteristics of faecal-environmental sampling, reporting a high HTP on dairy farms across the United States (91%, 95% PPI 82–99%). Similarly, Pozzato et al. (2011) used Bayesian modelling to estimate the HTP in dairy herds in two regions of Italy, with estimates of 70 and 71% for Lombardy and Veneto, respectively. In a systematic review of MAP prevalence in European countries (Nielsen and Toft, 2009), the authors found the HAP country prevalence ranged from 3 to 68% in cattle. Due to study design differences, lack of information, or errors in test interpretation, those authors were unable to obtain HTP directly from the reviewed studies and could only report an approximate HTP estimate of over 50%. A similar situation was described for deer by Nielsen and Toft (2009), in the Czech Republic. However, due to scarcity of available information those authors did not provide any HTP estimate for this species. In sheep flocks, HAP of 24 and 29% were reported in Switzerland and Spain, respectively. However, HTP estimates are not available (Reviriego et al., 2000; Muehlherr et al., 2003). Sergeant and Baldock (2002) estimated sheep HTP in Australia, based on abattoir surveillance data, with estimates ranging from 2.4 to 4.4% Australia-wide and 6–10% in New South Wales. The comparison of HTP showed differences between NI and SI in all species, although the degree and direction of this difference was species-specific, with POPR values of 0.96, 0.80, and 0.01 for sheep, beef cattle, and deer, respectively. Differences between islands, as noted in this study, have not been previously identified or suspected for sheep or beef cattle. However, in an abattoir-based study, Stringer et al. (2013b) found a similar island difference between deer herds, with HTP estimates of 44 and 67% for NI and SI, respectively. Risk factors explaining the observed differences between NI and SI have not been investigated for any of the three species, and the study of these factors was

454

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455

beyond the scope of this research. Differences in prevalences across islands could be explained by differences in flock/herd size (Table 2), as has been suggested in previous studies (Jakobsen et al., 2000; Wells and Wagner, 2000; Daniels et al., 2002; Hirst et al., 2004; Woodbine et al., 2009). Soil characteristics have also been proposed as a MAP risk factor (Ward and Perez, 2004; Dhand et al., 2009), and potentially could play a role in the differences in prevalences. A recent molecular survey in New Zealand, using VNTR and SSR techniques in tandem, observed significant differences in the dominant MAP strains isolated from the NI and SI (Heuer et al., 2012). Hence, the observed pattern may be explained by the interaction between soil type, flock/herd size, and differences in MAP strains between islands. Such factors were not considered in our study and may therefore deserve future research attention.

5. Conclusion Mycobacterium avium subspecies paratuberculosis infection is widely spread in New Zealand, especially among sheep farms, with an estimated national flock prevalence of 76%. Infection prevalence of deer and beef cattle herds is also relatively high at 46 and 42%, respectively. Prevalence differences were observed between islands for the three species. Sheep and beef cattle HTP estimates were higher in the NI than SI, while deer herds located in the SI had at greater estimated HTP than those in the NI.

Conflict of interest statement The authors declare they have no conflict of interest.

Acknowledgments This study was funded by the Johne’s Disease Research Consortium New Zealand (Grant no. JDRC.5.1.Epidemiology). The authors would like to acknowledge farmers and veterinary practices for their participation, as well as numerous post-graduate student volunteers and casual workers for sample processing and data entry. Staff at the Disease Research Laboratory (DRL) at Otago University and AgResearch Laboratories, Wallaceville, especially Dr. Geoff DeLisle, Gary Yates and Simon Liggett, are acknowledged for diagnostic capability and support. And special thanks go to Paulina Guzman, Saskia Prickaerts, Daniela Tapia, Neville Haack, and Raewynne Pearson for their practical help and invaluable support to this research, and to Dr Rebecca Mitchell for her valuable comments on the manuscript.

Appendix A. We derive here the expressions needed in order to completely specify components in the equations 1–3, in terms of the model parameters HTPi , TP, Pse, Psp, Se and Sp. For ease of exposition we drop the island-specific subscript i for HTP and its dependent terms.

The probability of no infected animals out of n sampled from a flock is Pr(I − ) = (1 − HTP) + HTP × (1 − TP)n using the fact that, if the herd is not infected, the sample must be non infected; then Pr(I+ ) = 1 − Pr(I− ). The sample level specificities of the tests are Pr(T1− |I − ) = Pspr and Pr(T2− |I − ) = Spn . Deriving the sample level sensitivities is more complex. Note that Pr(T1+ |I + ) = Pse if there is only one pool per herd, but this is not always the case (only sheep flocks), and careful conditioning arguments are required. First we derive the herd level specificity (Hsp) and sensitivity (Hse) for the PFC test as Hsp = Pr(T1− |H − ) = Pr(T1− |I − , H − ) = Pr(T1− |I − ) = Pspr and Hse = Pr(T1+ |H + ) = 1 − [1 − (1 − TP)k × (1 − Pse) r

+(1 − TP)k × Psp] ,respectively. Conditioning now on the status of the sample, we can also express Hse as

Pr(T1+ |H + ) = Pr(T1+ |I + , H + ) × Pr(I + |H + ) + Pr(T1+ |I − , H + ) × Pr(I − |H + ) = Pr(T1+ |I + ) × [1 − (1 − TP)n ] + (1 − Pspr ) × (1 − TP)n By equating these two expressions for Hse we can solve for Pr(T1+ |I + ) in terms of the model parameters; then Pr(T1− |I + ) = 1 − Pr(T1+ |I + ). Similar derivations can be made for the ELISA test. Two alternative expressions for herd-level sensitivity of the ELISA, applied to each of n sampled animals, are Pr(T2+ |H + ) = 1 − [(1 − Se) × TP + Sp × (1 − TP)]n and Pr(T2+ |H + ) = Pr(T2+ |I + , H + ) × Pr(I + |H + ) + Pr(T2+ |I − , H + ) × Pr(I − |H + ) = Pr(T2+ |I + ) × [1 − (1 − TP)n ] + (1 − Spn ) × (1 − TP)n These are equated and solved for Pr(T2+ |I + ) in terms of the model parameters; then Pr(T2− |I + ) = 1 − Pr(T2+ |I + ) References Branscum, A.J., Gardner, I.A., Johnson, W.O., 2004. Bayesian modeling of animal- and herd-level prevalences. Prev. Vet. Med. 66, 101–112. Daniels, M.J., Hutchings, M.R., Allcroft, D.J., McKendrick, I.J., Greig, A., 2002. Risk factors for Johne’s disease in Scotland – the results of a survey of farmers. Vet. Rec. 150, 135–139. de Lisle, G.W., 2002. Johne‘s disease in New Zealand: the past, present and a glimpse into the future. N. Z. Vet. J. 50, 53–56. de Lisle, G.W., Yates, G.F., Collins, D.M., 1993. Paratuberculosis in farmed deer: case reports and DNA characterization of isolates of Mycobacterium paratuberculosis. J. Vet. Diagn. Invest. 5, 567–571. Dhand, N.K., Eppleston, J., Whittington, R.J., Toribio, J., 2009. Association of farm soil characteristics with ovine Johne’s disease in Australia. Prev. Vet. Med. 89, 110–120. Dhand, N.K., Johnson, W.O., Toribio, J., 2010. A Bayesian approach to estimate OJD prevalence from pooled fecal samples of variable pool size. J. Agric. Biol. Environ. Stat. 15, 452–473.

C. Verdugo et al. / Preventive Veterinary Medicine 117 (2014) 447–455 Dorjee, S., Heuer, C., Jackson, R., West, D.M., Collins-Emerson, J.M., Midwinter, A.C., Ridler, A.L., 2008. Prevalence of pathogenic Leptospira spp. in sheep in a sheep-only abattoir in New Zealand. N. Z. Vet. J. 56, 164–170. Feller, M., Huwiler, K., Stephan, R., Altpeter, E., Shang, A., Furrer, H., Pfyffer, G.E., Jemmi, T., Baumgartner, A., Egger, M., 2007. Mycobacterium avium subspecies paratuberculosis and Crohn’s disease: a systematic review and meta-analysis. Lancet Infect. Dis. 7, 607–613. Griffin, J.F.T., Spittle, E., Rodgers, C.R., Liggett, S., Cooper, M., Bakker, D., Bannantine, J.P., 2005. Immunoglobulin G1 enzyme-linked immunosorbent assay for diagnosis of Johne’s disease in red deer (Cervus elaphus). Clin. Diagn. Lab. Immunol. 12, 1401–1409. Harris, N.B., Barletta, R.G., 2001. Mycobacterium avium subsp. paratuberculosis in Veterinary Medicine. Clin. Microbiol. Rev. 14, 489–512. Heuer, C., Dreyfus, A., Wilson, P., Benschop, J., Subharat, A., AyaneguiAlcarreca, A., Fang, F., Collins-Emerson, J., Midwinter, A., 2010. Epidemiology and Control of Leptospirosis in New Zealand. Society for Veterinary Epidemiology and preventive Veterinary Medicine, Nantes, France. Heuer, C., Verdugo, C., Price-Carter, M., Collins, D., French, N.P., Wilson, P.R., 2012. Epidemiological inferences from typing Mycobacterium avium subsp. paratuberculosis (MAP) isolates from mixed sheep, deear and cattle farms. In: Proceedings of the 11th International Colloquium on Paratuberculosis, Sydney, Australia, 5–10 February, p. 250. Hirst, H.L., Garry, F.B., Morley, P.S., Salman, M.D., Dinsmore, R.P., Wagner, B.A., McSweeney, K.D., Goodell, G.M., 2004. Seroprevalence of Mycobacterium avium subsp paratuberculosis infection among dairy cows in Colorado and herd-level risk factors for seropositivity. J. Am. Vet. Med. Assoc. 225, 97–101. Jakobsen, M.S., Alban, L., Nielsen, S.S., 2000. A cross-sectional study of paratuberculosis in 1155 Danish dairy cows. Prev. Vet. Med. 46, 15–27. JDRC, 2011. Johne’s Disease Research Consortium. Lombard, J.E., Gardner, I.A., Jafarzadeh, S.R., Fossler, C.P., Harris, B., Capsel, R.T., Wagner, B.A., Johnson, W.O., 2013. Herd-level prevalence of Mycobacterium avium subsp. paratuberculosis infection in United States dairy herds in 2007. Prev. Vet. Med. 108, 234–238. Mishina, D., Katsel, P., Brown, S.T., Gilberts, E., Greenstein, R.J., 1996. On the etiology of Crohn disease. Proc. Natl. Acad. Sci. U. S. A. 93, 9816–9820. Muehlherr, J.E., Zweifel, C., Corti, S., Blanco, J.E., Stephan, R., 2003. Microbiological quality of raw goat’s and ewe’s bulk-tank milk in Switzerland. J. Dairy Sci. 86, 3849–3856. Nielsen, S.S., Toft, N., 2008. Ante mortem diagnosis of paratuberculosis: a review of accuracies of ELISA, interferon-[gamma] assay and faecal culture techniques. Vet. Microbiol. 129, 217–235. Nielsen, S.S., Toft, N., 2009. A review of prevalences of paratuberculosis in farmed animals in Europe. Prev. Vet. Med. 88, 1–14. Nielsen, S.S., Toft, N., Jørgensen, E., Bibby, B.M., 2007. Bayesian mixture models for within-herd prevalence estimates of bovine paratuberculosis based on a continuous ELISA response. Prev. Vet. Med. 81, 290–305. Okura, H., Nielsen, S.S., Toft, N., 2010. Prevalence of Mycobacterium avium subsp. paratuberculosis infection in adult Danish non-dairy cattle sampled at slaughter. Prev. Vet. Med. 94, 185–190. Pozzato, N., Capello, K., Comin, A., Toft, N., Nielsen, S.S., Vicenzoni, G., Arrigoni, N., 2011. Prevalence of paratuberculosis infection in dairy cattle in Northern Italy. Prev. Vet. Med. 102, 83–86.

455

Raizman, E.A., Wells, S.J., Godden, S.M., Bey, R.F., Oakes, M.J., Bentley, D.C., Olsen, K.E., 2004. The distribution of Mycobacterium avium ssp. paratuberculosis in the environment surrounding minnesota dairy farms. J. Dairy Sci. 87, 2959–2966. Reviriego, F.J., Moreno, M.A., Dominguez, L., 2000. Soil type as a putative risk factor of ovine and caprine paratuberculosis seropositivity in Spain. Prev. Vet. Med. 43, 43–51. Rogan, W.J., Gladen, B., 1978. Estimating prevalence from results of a screening-test. Am. J. Epidemiol. 107, 71–76. Sergeant, E.S.G., Baldock, F.C., 2002. The estimated prevalence of Johne’s disease infected sheep flocks in Australia. Aust. Vet. J. 80, 762–768. Sergeant, E.S.G., Marshall, D.J., Eamens, G.J., Kearns, C., Whittington, R.J., 2003. Evaluation of an absorbed ELISA and an agar-gel immunodiffusion test for ovine paratuberculosis in sheep in Australia. Prev. Vet. Med. 61, 235–248. Spiegelhalter, D., Thomas, A., Best, N., Gilks, W., 1996. BUGS: Bayesian Inference Using Gibbs Sampling. Medical Research Council Biostatistics Unit, Cambridge. Stringer, L.A., Jones, G., Jewell, C.P., Noble, A.D., Heuer, C., Wilson, P.R., Johnson, W.O., 2013a. Bayesian estimation of the sensitivity and specificity of individual fecal culture and Paralisa to detect Mycobacterium avium subspecies paratuberculosis infection in young farmed deer. J. Vet. Diagn. Invest. 25, 759–764. Stringer, L.A., Wilson, P.R., Heuer, C., Hunnam, J.C., Verdugo, C., Mackintosh, C.G., 2013b. Prevalence of Mycobacterium avium subsp. paratuberculosis in farmed red deer (Cervus elaphus) with grossly normal mesenteric lymph nodes. N. Z. Vet. J. 61, 147–152. Subharat, S., Wilson, P.R., Heuer, C., Collins-Emerson, J.M., 2009. Leptospirosis: the facts. In: Deer Branch of the NZVA, Christchurch, pp. 85–88. Sweeney, R.W., 1996. Transmission of paratuberculosis. Vet. Clin. North Am. Food Anim. Pract. 12, 305–312. Tavornpanich, S., Gardner, I.A., Anderson, R.J., Shin, S., Whitlock, R.H., Fyock, T., Adaska, J.M., Walker, R.L., Hietala, S.K., 2004. Evaluation of microbial culture of pooled fecal samples for detection of Mycobacterium avium subsp paratuberculosis in large dairy herds. Am. J. Vet. Res. 65, 1061–1070. Verdugo, C., Heuer, C., Prickaerts, S., Wilson, P., 2010. Paratuberculosis prevalence in New Zealand pastoral livestock: a deer industry perspective. In: Association, N.Z.V. (Ed.), Proceedings of the Deer Branch of the New Zealand Veterinary Association, Queenstown. , pp. 35–38. Ward, M.P., Perez, A.M., 2004. Association between soil type and paratuberculosis in cattle herds. Am. J. Vet. Res. 65, 10–14. Wells, S.J., Wagner, B.A., 2000. Herd-level risk factors for infection with Mycobacterium paratuberculosis in US dairies and association between familiarity of the herd manager with the disease or prior diagnosis of the disease in that herd and use of preventive measures. J. Am. Vet. Med. Assoc. 216, 1450–1457. Whittington, R.J., Marsh, I., McAllister, S., Turner, M.J., Marshall, D.J., Fraser, C.A., 1999. Evaluation of modified BACTEC 12B radiometric medium and solid media for culture of Mycobacterium avium subsp. paratuberculosis from sheep. J. Clin. Microbiol. 37, 1077–1083. Woodbine, K.A., Schukken, Y.H., Green, L.E., Ramirez-Villaescusa, A., Mason, S., Moore, S.J., Bilbao, C., Swann, N., Medley, G.F., 2009. Seroprevalence and epidemiological characteristics of Mycobacterium avium subsp. paratuberculosis on 114 cattle farms in south west England. Prev. Vet. Med. 89, 102–109.