Royal Society Publishing

Endemic cattle diseases: comparative epidemiology and governance

David Carslake, Wyn Grant, Laura E. Green, Jonathan Cave, Justin Greaves, Matt Keeling, John McEldowney, Habtu Weldegebriel, Graham F. Medley

Abstract

Cattle are infected by a community of endemic pathogens with different epidemiological properties that invoke different managerial and governmental responses. We present characteristics of pathogens that influence their ability to persist in the UK, and describe a qualitative framework of factors that influence the political response to a livestock disease. We develop simple transmission models for three pathogens (bovine viral diarrhoea virus, bovine herpesvirus and Mycobacterium avium spp. paratuberculosis) using observed cattle movements, and compare the outcomes to an extensive dataset. The results demonstrate that the epidemiology of the three pathogens is determined by different aspects of within- and between-farm processes, which has economic, legal and political implications for control. We consider how these pathogens, and Mycobacterium bovis (the agent of bovine tuberculosis), may be classified by the process by which they persist and by their political profile. We further consider the dynamic interaction of these classifications with pathogen prevalence and with the action taken by the government.

1. Introduction

In the UK, infectious diseases to which livestock are vulnerable are often classified into two major classes: endemic and exotic [1]. Exotic pathogens, such as rinderpest virus or foot and mouth disease virus (FMDV), are by definition normally absent from the UK, but if an introduction occurs, a potentially fast-spreading, highly damaging epidemic can result. Exotic pathogens are legally notifiable, and border and movement controls are used to prevent their (re)introduction. When an outbreak occurs, strenuous efforts are made to re-eliminate the pathogen, often at huge short-term cost to the industry or to a compensating government [2]. By contrast, endemic pathogens, such as bovine viral diarrhoea virus (BVDV) or Mycobacterium avium spp. paratuberculosis (Map), are present in the UK and typically attract less political debate, media attention, legislation or economic analysis. They usually exist at a stable, if often high prevalence, with low apparent mortality and cause less dramatic outbreaks of disease. Individual farmers may seek to eliminate a particular endemic pathogen, but with a high prevalence of infected herds elsewhere a disease-free status is difficult to maintain. Endemic pathogens are often unnoticed, mitigated or tolerated by farmers. Nonetheless, the presence of endemic pathogens causing a reduction in the performance of infected animals and herds represents a considerable, but often underestimated, drain on farm profitability and reduction in animal welfare [36].

The labelling of a disease as exotic or endemic is essentially a political decision to ‘frame’ or label a disease in a particular way. Diseases (and the pathogens that cause them) are not per se exotic or endemic. For example, FMD is exotic in the UK but was once endemic, and still is in many parts of the world. Once this labelling has occurred, the political debate is framed around it, particularly in terms of how involved government should be in tackling the disease, and the level of resources that are devoted to its research and management. The political distinction between endemic and exotic pathogens is not, however, an immutable rule and can be influenced by particular political agendas. For example, Brucella abortus was recognized as eliminated from Great Britain in 1985 through a government sponsored programme, so that its status has subsequently changed. It was a political decision (motivated by legal requirements and veterinary advice) to use government resources to re-eliminate FMD in 2001 so that it retained its exotic status. The long-term, widespread presence of bovine tuberculosis (bTB) in much of the UK makes it undeniably endemic; yet the resources devoted to its control are more comparable to the response to a newly introduced exotic pathogen. Managing bTB costs taxpayers some £80 million a year, although the true cost is higher because some bTB-related activities are allocated elsewhere [7].

It is evident that there is considerable variation in the political importance attached to a disease. ‘Political importance’ in this context primarily means whether government considers that the disease requires some form of publically funded intervention. The alternative is that the disease is considered as a production disease (a disease that primarily affects productivity and profitability) that is then considered a matter for animal keepers to deal with either individually or through some form of self-generated collective action. ‘Government’ in this context does not refer only to nation-state governments, but also to sub-national governments and international governance arrangements that have relevant jurisdictions such as the European Union, the World Trade Organization and the World Organization for Animal Health (OIE). What is regarded as a production disease might vary between jurisdictions or branches of government and over time, e.g. the Scottish Government conducted a consultation on a BVDV initiative in 2010 (http://www.scotland.gov.uk/Publications/2010/06/29143957/0, accessed 20 July 2010), and has since announced a public intervention (http://www.scotland.gov.uk/News/Releases/2010/09/21145206, accessed 4 October 2010), but in England, disease due to BVDV is being treated as a production disease. Extensive government intervention in bTB may reflect historical commitments and stakeholder pressure as much as policy logic, thus indicating other possible determinants of political importance. Elements of path dependency may enter into the picture because once a disease has acquired a high- or low-profile status it is difficult for a decision to treat it as an appropriate subject for government policy to be reversed because beneficiaries of the policy among stakeholders are created.

In contrast to the political status of a disease, the epidemiological characterization of a pathogen as absent, increasing (epidemic) or relatively stable (endemic) is based primarily on surveillance data. However, the level of surveillance (e.g. whether a disease is notifiable) is largely determined by its political status, so that there is very little surveillance for unregulated endemic diseases. Furthermore, the infection dynamics of pathogens of livestock (a managed host population) are largely determined by socioeconomic processes such as movement of animals between herds, management of herds and decisions to vaccinate.

The endemic pathogens infecting British cattle comprise a diverse, interdependent community of viral, bacterial, protozoan and parasitic species. These infections are not uniformly distributed among British cattle. Farms differ in the scope they provide for the transmission of infection between cattle through, for example, differing management practices and environments. However, herds do not exist in isolation from one another, and cattle movement between herds is a major risk factor for the spread of most pathogens, on which we focus below [810]. Cattle moving on to a farm can bring infection with them; cattle moving off a farm can remove infection; so the prevalence of infection in a herd is a balance between introduction, within-farm transmission, birth, death, recovery and emigration. Farm management varies and this influences the numbers and rates of movement of animals on and off farm. Consequently, the prevalence of any pathogen varies between farms, with the pattern of variation reflecting the nature of the pathogen [11]. As well as being non-uniformly distributed among herds, it is also likely that pathogens are not distributed independently from each other since management practices that increase the risk of introducing one pathogen into a herd may also expose the herd to (or protect it from) others. However, all the studies of transmission dynamics of livestock infections of which we are aware consider pathogens separately. Further, the advice given to farmers seeking to avoid the damaging effects of infection is usually pathogen-specific despite the potential for interdependence among endemic pathogens and the influence of farmer responses on other disease risks.

We have recently discussed the management and regulation of animal disease [1]. However, the characteristics that determine how the social system responds to a particular pathogen and how pathogens respond to social systems have not been considered previously; by social system we mean the combined societal processes (economics, culture, politics, etc.) operating at all levels (farms, governments, etc.). The epidemiology of both diseases and pathogens is both determined by and determines the social system. For example, a farmer may buy a cow (for purely commercial reasons) that happens to be shedding BVDV and his herd subsequently experiences an abortion storm, which changes the management of the herd (e.g. purchase of recently calved cows) that results in the introduction of Map. Here, we seek to address two questions regarding endemic disease prevalence, in order to explore the interlinked dynamics of disease epidemiology and social system change. First, what determines the political response to a livestock disease? Second, what is the relative role of cattle demography, particularly the movement of animals but including birth and death, in determining the distribution of infection between farms? We consider three pathogens: BVDV, bovine herpes virus (BHV, which causes several diseases including infectious bovine rhinotracheitis) and Map (which causes Johne's disease); and we also consider the political responses to Mycobacterium bovis, the aetiological agent of bTB. These four pathogens were chosen because of their differing natural histories, which are expected to produce differing political and epidemiological patterns. Introductions to the properties of BVDV [12], BHV [13], Map [14,15] and bTB [16,17] may be found elsewhere. To address the two questions we present and discuss qualitative and quantitative frameworks, respectively. Finally, we attempt to pull the two strands together and develop a discussion of interactions between natural and social systems that determine and are determined by infectious disease.

2. A political model

(a) Methods

Within the Governance of Livestock Disease (GoLD) project at the University of Warwick, we sought to construct a political model that would provide a framework for understanding the varying political priority given to different endemic cattle diseases. This was seen primarily as an academic modelling exercise but with clear policy implications. It consisted of a list of the various questions that might influence a government's response. Not surprisingly, although developed in-house, the list resembles the criteria used in Defra's Prioritisation project (http://www.defra.gov.uk/foodfarm/farmanimal/diseases/vetsurveillance/documents/dst_summary.pdf accessed 4 October 2010). An initial draft provided by a member of the team from the political science discipline (W.G.) was sent to the other team members covering the disciplines of epidemiology, veterinary science, economics and law. Successive iterations of the model were discussed until a final version emerged (table 1). Some subsequent adjustments were made in the light of comments made by anonymous referees. The questions used in the model were intended to be comprehensive in terms of the factors considered by government. Using the four chosen cattle pathogens as examples, the response to each question was scored, again by discussion within the GoLD team, on a scale of 0 (none), 1 (low), 2 (medium) or 3 (high). In order to consider the effect on the political priority of all questions together, it was necessary to combine the scores in some way. Noting that some questions may be more influential than others, it was decided to give the two questions concerning zoonotic risk a weighting factor of 5 in relation to the others.

View this table:
Table 1.

A framework for assessing the political priority accorded to endemic livestock diseases, with consensus scores for four examples in England. To achieve an overall assessment of an individual disease, each question was scored: 0 (none), 1 (low), 2 (medium), 3 (high); and the scores were aggregated, all questions being weighted equally, except for questions 1a and 1b, which were each given a weighting factor of 5.

The National Archives is the UK government's official archive. Government records that have been selected for permanent preservation are sent to the National Archives when they are 30 years old, but many are transferred earlier. The files are extensive and appear to be a complete sequence [19]. National Archive files relevant to each of the four pathogens were identified and studied (by W.G.), and each pathogen's profile in the National Archives was used as an indication of the attention paid to it by the government.

(b) Results

Consensus scores for each question and pathogen are shown in table 1. Weighted sums of the scores were 53 for bTB, 22 for BVDV, 20 for BHV and 14 for Map. The model predicts, as one would expect and in accordance with its high profile in the National Archives, that bTB will attract by far the greatest political attention of the four pathogens. This is particularly true when the zoonosis weighting is included, but applies even without it. The model also predicts correctly that Map will be considered a lower priority than BVDV. The moderately high score (intermediate between BVDV and Map) given to BHV, however, is inconsistent with its complete lack of profile in either the National Archives or current discussions in the cattle industry. In part this reflects the fact that BHV is easier to diagnose and there are effective vaccines. It can be tackled at farm level without government intervention. The ease with which a disease can be managed can affect government decisions about intervention. In Scotland, the selection of BVDV for a policy initiative was influenced by such considerations: ‘When BVD came in my team were saying we should be looking at Johne's, potential link to Crohn's disease. Tools are all there (for BVDV), don't have a test for (Map) that works in a practical time scale. Not able to detect the animal early enough.’ (Non-attributable interview with policymaker, Edinburgh, 14 July 2010).

3. Epidemiological models

(a) Methods

The primary purpose of the epidemiological modelling was to examine the extent to which the interaction of cattle demography with the natural history of infection explained the observed patterns of occurrence of each pathogen, rather than to make precise parameter estimates or to predict the infection status of particular herds or individuals. The construction of the models is summarized here: full details are available in the electronic supplementary material.

A stochastic simulation model was devised for each of BVDV, BHV and Map with cattle moving between mutually exclusive infection states (figure 1). Most parameters of the model were determined a priori from the literature, but one transmission parameter in each model was adjusted to optimize the fit between the modelled and observed seroprevalence among adults (greater than or equal to 2 years old). Transmission coefficients were estimated separately for dairy and beef herds.

Figure 1.

Epidemiological models for (a) BVDV, (b) BHV and (c) Map. In the BVDV model, animals were susceptible (S), infectious (I), recovered (R) or persistently infected (P). In the BHV model, cattle were susceptible (S), primarily infectious (I1), latent (L) or secondarily infectious (I2). In the Map model, they were susceptible (S), low-shedding (L), high-shedding (H) or clinically infected (C). The +/− superscripts indicate the assumed serological status of each class, subject to test sensitivity and specificity. The subscript ‘ad’ indicates the number of adult (greater than or equal to 2 years old) animals of the given infection class. Parameter values are given in the electronic supplementary material, table S1. Where two infection processes are shown for the same transition, they were combined by calculating the probability of either occurring.

Serological data were available for 60 beef and 53 dairy herds in the southwest of England, each sampled up to four times at approximately annual intervals [10,11,20]. Only animals over 2 years old were sampled. Serum from each sampled animal was tested by ELISA (details in the electronic supplementary material). Samples testing seronegative for BVDV were also tested for the presence of BVDV antigen, and antibody-negative, antigen-positive animals were assumed to be persistently infected with BVDV. Demographic data (births, deaths and transfers between herds) for all cattle on study farms within the simulation period (5 years, from 5 December 2002) were supplied on 25 March 2008 by Defra's Rapid Analysis and Detection of Animal-related Risks (RADAR) project. The modelled pathogen transmission process (figure 1) was then applied to the observed demographic history of the study herds during the simulation period, i.e. cattle movement, birth and death were not simulated, but implemented as they had actually occurred. After each 5 year iteration, the age-specific distribution of infection states among the cattle of each herd was used to populate that herd for the next iteration. Convergence was apparent after 20 such iterations, after which outcomes were recorded for further 20 iterations. These recorded outcomes were the infection status of simulated cattle, on the dates on which they were sampled in reality. Cattle under 2 years of age, not sampled in the observed data, were also sampled in the simulation on the same date as the rest of their herd. Finally, a tally was kept of individual or herd-level infection events in the simulation, recording the class of infection responsible.

Simulated serology results were used to calculate age class- and visit-specific seroprevalences, which were compared with the values from the observed data. Correlations between observed and simulated data (the latter averaged over all iterations) at an individual sample level, at a between-visit level and at a within-visit level, were also calculated. Finally, the possible effects of BHV reactivation upon purchase of a latently infected animal [10,13,21] were investigated.

(b) Transmission dynamics

Observed adult seroprevalences for BHV and Map fell within the range obtained by varying the transmission coefficient (electronic supplementary material, figure S1), allowing the straightforward estimation for each pathogen and herd type (electronic supplementary material, table S1). Simulated adult seroprevalences for BVDV levelled off at just below the observed values with increasing transmission coefficient, which was therefore chosen by eye at the point at which the curve became very shallow.

The match between observed and simulated age profiles for each pathogen (electronic supplementary material, figure S2) was good, with BVDV seroprevalence a little lower than expected at all ages, as expected, given that the overall adult seroprevalence was not reached. In both dairy and beef herds, the simulation overestimated Map seroprevalence in younger age classes and underestimated it in older ones. This is probably because the sensitivity of serological tests for Map increases with age, even within an infection class [22]. The simulation gave a good match to the few BVDV persistently infected animals observed (electronic supplementary material, figure S4) despite not having been optimized for this. The simulations suggest that persistently infected animals under 2 years of age (not tested for in the observed data) vastly outnumber those over 2 years of age.

Table 2 lists the transmission events taking place during the simulation, according to the class of infectious animal to which they were attributed. The results from these simulations suggest that within herds, animals persistently and acutely infected with BVDV (classes P and I, respectively) are responsible for approximately equal numbers of new infections. In utero infections (resulting in the birth of a persistently infected calf) made up less than 3 per cent of transmission in both herd types, and were usually owing to seroconversion of the dam during pregnancy (approx. 90% of cases) rather than to persistently infected dams giving birth. Animals infected with BHV for the first time accounted for about three times as many within-herd transmissions as reactivated latent animals, and the majority (greater than 84% in both herd types) of within-herd Map transmission was due to high-shedding, subclinically infected animals (class H). These animals are abundant (like class L, but not class C; electronic supplementary material, figure S3) and highly infectious (like class C, but not class L; electronic supplementary material, table S1).

View this table:
Table 2.

Within- and between-herd infections attributed to each infectious class. Totals are for all herds over 20 iterations of the 5 year simulation period. The birth of a calf persistently infected with BVDV was attributed to class P if it was due to the calf's dam being persistently infected and to class I if it was due to the dam seroconverting during pregnancy. A herd with no active infection (i.e. all animals susceptible or recovered immune) was considered newly infected when an infectious (or potentially infectious, in the case of animals latently infected with BHV) animal was introduced to it. Within- and between-herd transmissions thus represent counts of cattle infected, and herds infected, respectively. The transmission ratio is the ratio between within-herd transmission and between-herd transmission for a class of infectious animal (n.a., not applicable).

The ratio between within-herd and between-herd transmission events was higher in beef herds than in dairy herds. Among dairy herds, BVDV was transmitted between herds far more often than BHV or Map, both in absolute terms and relative to within-herd transmission. Among beef herds, absolute rates of BVDV transmission between herds were higher than those for BHV or Map, but the difference in within-herd transmission was even greater, such that transmission ratios were lower for BHV and Map than for BVDV. Most between-herd transmission for BHV or Map involved latently infected or low-shedding animals, respectively. These are not very infectious, and must be considered less likely to give rise to secondary cases than an animal that is introduced while infectious with BVDV. Persistently and acutely infected animals were equally likely to bring BVDV to a previously uninfected herd, although a new herd infection owing to an acutely infected animal is more prone to rapid fade-out.

(c) Distributional patterns

Figure 2 shows the herd-level distribution of seroprevalence, and table 3 shows the correlations between simulated and observed herd-level seroprevalence. The three pathogens showed very different distributions of seroprevalence among visits, and the degree to which simulations matched this distribution also varied. For both herd types, observed BVDV data showed a large number of herds with a seroprevalence close to 100 per cent, but a wide range of seroprevalences in other herds, and very few at very low seroprevalence (figure 2a,b). This peak at near total prevalence was successfully reproduced by the simulations, but simulations in dairy herds also predicted a second peak of herds with very low (but not zero) prevalence, which was not seen in reality. Effectively, the simulations predicted a near-dichotomy between high- and low-BVDV dairy herds. Among beef and dairy herds, small herds of fewer than 10 animals almost all actually fell into the highest band of seroprevalence. This surprising result was reproduced by the simulation. The correlation between the observed and (average) simulated correlation for each visit differed hugely between herd types (table 3). There was no correlation in dairy herds, but a substantial correlation in beef herds, meaning that the beef herds with high BVDV seroprevalence in the simulation tended also to have high seroprevalence in reality. Moderate correlation at the within-herd level probably indicates that, in the simulation and in reality, older animals are more likely to be seropositive for BVDV (electronic supplementary material, figure S2).

View this table:
Table 3.

Correlations between the simulated and observed serological status of samples and herds. Overall correlation is the crude correlation at the individual sample level. Between-visit correlations were calculated for seroprevalence at the level of the herd visit, weighted by herd size. Within-visit correlations are calculated at the individual sample level, after recalculating each sampled animal's status relative to the seroprevalence of their herd. Values of β for each pathogen were optimized separately for each herd type. Simulated serological status was averaged over all simulations for each sample and visit.

Figure 2.

Frequency distributions of seroprevalence among visits. The observed (circles) and simulated (bars) distribution of the seroprevalence of (a,b) BVDV, (c,d) BHV and (e,f) Map among visits to (a,c,e) dairy and (b,d,f) beef herds. Filled circles and the entire bars represent data from all herds, whereas the lower part of the bars and the unfilled circles, where distinguishable, represent data only from herds of 10 or more animals. Visits with exactly zero prevalence were counted separately (left-most bars and circles).

Observed distributions of BHV seroprevalence among visits (figure 2c,d) were bimodal, with one peak at around 80 per cent (albeit rather indistinct among beef herds) and another at very low (dairy herds) or zero (beef herds) seroprevalence. In contrast to BVDV, and in accordance with what we might expect from their relative isolation, small herds tended to be free of BHV. The simulations, however, did not reproduce this bimodality, resulting in a very poor fit among dairy herds, but a moderately good one among beef herds (where the bimodality was far less pronounced). As for BVDV, observed and simulated seroprevalences for each visit were strongly correlated (table 3) in beef herds, but not in dairy herds, and correlation at the within-herd level is probably a result of increasing seroprevalence with age (electronic supplementary material, figure S2). Among dairy herds, the introduction of BHV re-activation on purchase with probability ε (and an associated reduction in the background rate of reactivation, α) improved the match between histograms of observed and simulated visit-level seroprevalence by adding a second peak at low (but not zero) seroprevalence. However, the bimodality remained much less pronounced than in the observed data, even for high ε values of 0.5 or 0.75, and the improved distribution shape was not accompanied by any correlation between observed and simulated seroprevalences (data not shown). Among beef herds, the introduction of re-activation on purchase had only a minimal effect, leading to a slight bimodality that did not improve the overall fit to the observed data.

For Map, which had a much lower overall seroprevalence than the other two pathogens, most herds had very low prevalence (or zero, among beef herds), with declining numbers reaching up to about 20 per cent. This was almost perfectly reproduced by the simulation among beef herds (figure 2f) where small herds all had low seroprevalence. Among dairy herds (figure 2e), the match was also good, but the observed variability in Map seroprevalence was slightly underestimated by the simulation. In contrast to the results for BVDV and BHV, there were no substantial correlations at any level between observed and simulated Map status (table 3). Map seroprevalence does not increase with age as do BVDV and BHV (electronic supplementary material, figure S2). The lack of correlation at the between-visit level indicates that although the simulation reproduced the visit-level distribution, this was not achieved by correctly reproducing the seroprevalence of individual herds.

4. Discussion

We discuss the political and epidemiological models separately before attempting to combine them.

(a) Political model

The political diseases model is seen as a heuristic device and a first step in assessing the political importance of a disease. In this analysis it is applied to England, but it could be applied to other jurisdictions and deployed comparatively. It could also be applied to other endemic pathogens of livestock, or to the same pathogen at different points of time. We restricted ourselves here to a very simple system of weightings, which recognizes the salience of zoonotic factors. However, more elaborate differential weightings (including interactions between questions, such that the weight accorded to one depended on the response to another) could be used to refine predictions, or to explore different understandings of a disease. For example, policymakers or any stakeholder could be asked to assign scores and weightings. It therefore can be regarded as a part of a policy toolkit [23] contributing to the development of decision support. The effect of possible future scenarios can be explored by adjusting scores accordingly. For example, there has been some suggestion, as yet unproven, that Map in milk might be a causative agent of Crohn's disease in humans [24]. If the perception of this link increased in the future (perhaps, but not exclusively, as a result of scientific evidence), it might increase the scores for questions 1a and 1b to 2 and 1, respectively, raising the total score from 13 to 28, suggesting political interest considerably in excess of that presently accorded to BVDV. Map thus has the potential to become considerably more politically significant in the future in England, as seen currently elsewhere, e.g. USA, Australia. In this particular sense, the model does have a predictive capability.

(b) How should government classify livestock diseases?

How livestock diseases are classified by government determines the extent to which public finances are used to research and control them, so there are real consequences for disease prevalence and management. We have proposed a framework to capture the most important factors in the current political classification of disease. However, it does not relate directly to the way in which the farming system interacts with the disease. As a consequence, it does not indicate the degree to which government should get involved in disease control in order to improve farm productivity, animal health and animal welfare directly. It should be noted that this is largely a decision about the role of government in agricultural production, and different bodies will take different views. Thus, although the UK constitutes a single epidemiological unit, devolved (governmental) arrangements have seen the emergence of contrasting approaches to disease management reflecting different political priorities. The Scottish Government considers that Scotland enjoys an international reputation for quality meat production and that this requires a more proactive stance towards endemic as well as exotic diseases. Similarly, the Welsh Assembly Government policy ‘has been to improve the ‘branding’ of animals reared in Wales, so that Welsh beef, like Welsh lamb, is synonymous with good quality.’ [25, p. 18]. In relation to the implementation of animal health strategy in Scotland, the Scottish Government comments of endemic diseases in contrast to exotic diseases such as foot and mouth, ‘there are diseases found in GB which have much lower profiles but nevertheless require our efforts to control or eradicate them.’ (http://www.scotland.gov.uk/Publications/2003/12/18658/30622, accessed 17 June 2010).

From an economic perspective, the degree to which livestock disease reduction is beneficial to farmers largely depends on the epidemiological drivers of disease, and especially the role of animal movement. We have shown that movement of cattle between farms explains much of the epidemiology of BVDV and BHV in beef herds. Cattle movement is, therefore, important for these pathogens in this farm type, but is also likely to be important for disease risk for other pathogens in other farm types, so that a reduction in prevalence of infection for these diseases could be beneficial to a large number of farmers. On the other hand, movement does not appear to be a good predictor in dairy herds of the pattern of infection with Map and BHV, whose presence in particular herds seems owing to some unidentified non-demographic herd properties. Reduction in disease level is thus beneficial (and the pathogen a problem) only to those farmers with high-prevalence herds, rather than to the industry as a whole. The extent to which government commits to funding a disease control programme is not necessarily determined by a utilitarian consideration of the greatest level of disease reduction for the greatest number of farmers. It is more likely to be determined by a number of other considerations including cost–benefit analysis, the degree to which the industry commits itself to funding such a programme, the economic manifesto of the government in power and the need to resolve distributional issues that might arise from cost and responsibility sharing debates within the industry [26]. One also has to take account of the attachment or opposition of stakeholder groups to particular programmes or forms of control.

(c) The epidemiological model

The simulations include three processes: the natural history of infection, chance and cattle demography. Other processes, such as genetics and differences in management and husbandry (including, for example, housing) are not included. The same infection model applies to all herds of the same type (dairy or beef), so differences in seroprevalence among simulated herds can arise only though chance, or through their demography. The simulation successfully recreates the distribution of seroprevalence at a herd level (figure 2), as well as the seroprevalence of individual herds (table 3), for BVDV and BHV in beef herds. This strongly suggests that demography (together with stochastic effects) accounts for much of the difference among beef herds in BVDV and BHV seroprevalences. Solis-Calderon et al. [27] reported that BVDV status in Mexican beef cattle was related to an interaction between herd size and purchased/homebred origin (two demographic variables that are accounted for indirectly in our model) but not to several non-demographic variables that they recorded. In the case of Map in beef herds (and to some extent, dairy herds), the observed and simulated herd-level distributions of seroprevalence match, but there is no correlation between each herd's observed and simulated seroprevalences, suggesting that differences in seroprevalence between herds are largely owing to stochasticity, including factors not included in the simulation. The results for BVDV in dairy herds were similar, except that the simulation predicted a peak at very low seroprevalence that was not observed. The purchase of small numbers of animals having recovered from BVDV (class R) might account for this simulated result, while the assortative matching of source and destination farms (i.e. low-BVDV farms preferentially purchasing from other low-BVDV farms) might explain the absence of this effect in reality. As for Map in beef herds, the lack of correlation between simulated and observed data, and the similar heterogeneity (barring the result discussed above) in seroprevalence among herds suggests that stochasticity might be important in determining a dairy herd's BVDV seroprevalence. In the case of BHV and (to a lesser extent) Map in dairy herds, the simulation underestimates the heterogeneity among herds, suggesting that non-demographic differences between herds are of prime importance in determining their seroprevalence. For example, Raaperi et al. [28] reported that the presence of on-farm veterinarians and inseminators, as well as infection with BVDV, increased the risk of BHV in Estonian dairy herds; Scott et al. [29] identified low aridity and soil pH as risk factors for Map in Canadian dairy herds; and Çetinkaya et al. [30] reported that Channel Island breed and the presence of farmed deer were risk factors for Map in English dairy herds.

It is curious that adding reactivation of BHV upon purchase of animals into dairy herds should improve the fit of the visit-level distribution of seroprevalence, without improving the correlation between the observed and simulated seroprevalence at each visit. Purchase of animals is a demographic process modelled in our simulation, but differences between herds in the number of animals purchased do not seem to have been responsible for differences in BHV seroprevalence. It may be that the peak of herds at very low BHV prevalence is caused by another, unmodelled factor, which reactivation on purchase mimics. The concentration of reactivation among the relatively small number of purchased animals, rather than the whole latent population, would allow greater variation in seroprevalence among herds without necessarily identifying those herds that make up the low-prevalence peak in reality.

Where herds differ in seroprevalence owing to factors other than chance, and those factors are within the control of the farmer, changes in management could be used to control the infection. For all three pathogens, our results suggest that the factors influencing seroprevalence in a dairy herd might be different from the factors influencing seroprevalence in a beef herd. While this is not surprising, it does indicate that action to control a pathogen at a national level, which must involve all herd types because of the risk of cross-infection, will require different control measures in different herd types: optimal control of endemic pathogens requires an integrated approach across pathogens and herd types. This result also emphasizes the importance of considering herd type when analysing infection data for risk factors.

Medley et al. [11] likened the dynamics of pathogens among cattle herds to the metapopulation concept in ecology [31]. Individual herds might become clear of infection, but the pathogen persists in the national herd because susceptible herds get re-infected by other herds. Stable metapopulations, however, exist on a continuum between those in which extinctions and recolonizations are rare, and within-patch (within-herd) dynamics (transmission) dominate, and those in which extinction and recolonization are frequent events. By calculating the ratio between within- and between-herd transmission events, we attempted to place each of the three pathogens, in each herd type, on this continuum between those which rely on constant reintroduction for persistence, and those which are relatively stable within herds. The low transmission ratios for BVDV and BHV in beef herds (table 2) suggest that there is considerable transmission between herds, relative to transmission within herds. In the equilibrium situation, which we model, the rate of transmission to previously uninfected herds must be balanced by the rate at which a pathogen is lost from herds. The short infectious period and lack of latency in BVDV means that the virus might relatively easily die out in a herd, particularly if no persistently infected animals are present. Note that such a herd would have residual seroprevalence from the recovered animals, even after active infection was lost. The apparent importance of between-herd transmission is consistent with the suggestion (figure 2 and table 3) that demographic factors play an important role in determining a beef herd's seroprevalence of BVDV and BHV. At the other end of the scale, BHV and Map in dairy herds had the highest transmission ratios, suggesting that within-herd transmission is more important than between-herd transmission. Both these pathogens are capable of a long latent period, which would promote long-term persistence within a herd. This is also consistent with the idea that non-demographic, herd-specific factors are at work (figure 2 and table 3). It is not clear, however, why BHV latency might prevent the loss of infection in dairy herds, but not beef herds. Map in beef herds had a very low transmission ratio, suggesting frequent re-infection of previously uninfected herds. The fact that Map infects very young animals (i.e. before they are purchased) and usually shows no clinical signs until later in life would make it well adapted to move between herds. However, the majority of these between-herd transmissions involved the transfer of animals in the low-shedding class. These animals can easily live their entire life without moving into the more infectious classes, and so it is likely that many of these apparent between-herd transmissions do not lead to any secondary infections. By a similar argument, the intermediate transmission ratio for BVDV among dairy herds might underestimate the importance of between-herd transmission, because BVDV-infected animals, particularly persistently infected ones, are very infectious, and thus likely to generate secondary cases. The lack of correlation between observed and simulated visit-level seroprevalence (table 3) suggested that demographic factors were not particularly important in determining a dairy herd's BVDV seroprevalence, while the heterogeneity generated by the simulation (figure 2a) suggested that non-demographic factors also had limited influence. A dairy herd's seroprevalence of BVDV might be driven largely by the stochastic generation of persistently infected animals within the herd, given an initial herd infection.

The use of real demographic data in the epidemiological model eliminates the approximations and potential biases that would be inherent in reducing a complex system of births, deaths and transfers to a few network parameters and population average demographic rates. However, it has two major disadvantages. First, where demographic data are only known for a short period of time (5 years in our case), the longer simulations necessary to achieve asymptotic behaviour can only be performed by repeating the same demographic time series. Since a farmer's replacement strategy might change from one 5 year period to the next, this approach is likely to overestimate the long-term variability between herds in their demographic properties. The longer the demographic time series available, the less this is a problem. Second, the demographic properties of individuals and herds are inflexible and cannot vary in response to the distribution of the pathogen. Infection with any of the three endemic pathogens which we model has only a minimal effect on culling or transfer rates (D. J. Carslake 2009, unpublished data; [32,33]) and for two rare classes of animal where the effects are more profound (those persistently infected with BVDV or clinically infected with Map), we modelled removal of such animals from the infectious population by transferring them to the most abundant non-infectious group. Such an approach would not be suitable for pathogens that frequently cause the death (including culling) of the animal, such as bTB or most exotic pathogens. It should also be noted that our models were parameterized with data from the southwest of England; differences in farming practices and intensity mean that our inferences should be applied elsewhere only with caution.

The modelling of Map was difficult because of the uncertain test sensitivity, and the likelihood that sensitivity changes with time since exposure (or its proxy, age), the development of clinical signs and the shedding of bacteria (electronic supplementary material, figure S3; [22,34,35]). The poor sensitivity of the test is a severe hindrance to the control of Map [36,37] and a better test would aid both understanding and control of this pathogen. Failing that, a better understanding of the linked development of test sensitivity, infectiousness and clinical signs would improve our ability to model Map.

(d) Combining political and epidemiological dimensions

Since the epidemiology of endemic disease is determined by both natural and social processes (the demography of cattle being the social aspect in this paper), a classification that provides a combination of social and epidemiological dimensions is desirable. In table 4, we propose such a classification system. The two states on the right are those for which there is epidemiologically (and therefore economically) sufficient connexion between herds that it can be argued that disease reduction is a public good. For these diseases, control (restriction) of movement of infection between farms is likely to be an important and effective method of intervention. Diseases in the upper half have a high political profile, for all the reasons previously discussed. Diseases falling in the upper right quadrant have the greatest call on government resources for control—they have a high profile and the epidemiological/economic arguments for creation of a public good. The classification we propose distinguishes different endemic pathogens, but the vast majority of exotic pathogens would fall into the upper right-hand quadrant. This proposed schema thus includes the dichotomous exotic/endemic distinction.

View this table:
Table 4.

Schematic epidemiological and political classification of cattle diseases.

Diseases falling in the lower left quadrant are in the opposite position. However, most interesting are those diseases falling in the upper left and lower right: i.e. those for which there are contradictory indicators for government involvement. Diseases in the upper left quadrant have a relatively high political profile, but they are essentially individual farm-based problems. Historically, when agriculture itself had a higher political salience and there was an emphasis on maximizing production, these diseases might have attracted public funding. However, as they are farm-based problems, public control essentially represents a subsidy to a minority of farmers (if the disease causes production losses), mitigating the impetus for individual farmers to resolve the problem. As government attempts to reduce the resources it puts into livestock disease control, it will increasingly try and persuade farmers, and farmers' organizations, to take responsibility for these diseases, but there will be resistance. For diseases in the lower right quadrant, there is good argument and indication for government intervention—those farmers who are experiencing the disease are doing so because it is in the national herd, and there is little, individually that they can do about it. In this case, farmers and farmers' organizations will be arguing for public resources to help control disease, but there will be resistance from government because the disease has a relatively low political profile. A possible solution is private governance arrangements created within the industry or limited public funding to provide a stimulus to action which is what is envisaged in the Scottish BVDV case (http://www.scotland.gov.uk/News/Releases/2010/09/21145206, accessed 4 October 2010). Our results suggested that the factors influencing a dairy herd's seroprevalence of a pathogen were often different from those influencing seroprevalence of the same pathogen in beef herds, resulting in the placement of each herd type in a different quadrant of table 4. Whether government action (or inaction) on a pathogen can be tailored to different sectors of the cattle industry will depend on the degree of epidemiological connexion between the sectors.

This proposed classification is more interesting when considered as a dynamic system. The political profile of a disease is influenced by its prevalence. Consequently, if an intervention programme is successful, then the political profile will fall. Additionally, the role of movement will also be determined by prevalence. Our simulations suggested that demographic processes, including movement, were not an important influence on the prevalence of Map. However, if the prevalence of Map was much higher, then this result might have been different. Thus, ceteris paribus, increasing prevalence will move a disease towards the upper right quadrant, resulting in greater chance of concerted control, which might reduce prevalence, which will tend to push the disease back towards the lower left quadrant.

A similar process may have occurred with bTB. There was a national control programme during the twentieth century, inspired by zoonotic infections, which resulted in elimination of disease from all but a few confined geographical areas by 1975 [38]. Bovine TB has an environmental component (including badgers; [39]). This non-demographic, farm-specific factor would tend to move bTB to the left-hand quadrants of our scheme, while the involvement of a wild animal with a positive public profile allowed it to retain its high political profile (i.e. position in the upper quadrants) despite reduced prevalence in the 1970s. In the subsequent epidemic, cattle movement, a demographic factor moving bTB from being a farm-specific to an industry-wide problem, has played a key role in dissemination of infection throughout the UK [9]. Consequently, we suggest that bTB has moved from the upper right to the upper left quadrant and is now back in the upper right quadrant as a result of the current disseminated epidemic. Different diseases—depending on their characteristics, the way in which they are framed politically [19] and their economic and environmental impact—may trace different policy paths over time, demonstrating the utility of an integrated model.

Cattle pathogens endemic to the UK are able, by definition, to persist in their socioeconomic and epidemiological environment. We have shown that the mechanisms for this persistence differ between pathogens, and for the same pathogen, between sectors of the cattle industry. Elimination or control of an endemic pathogen requires a change in its environment (as broadly defined here), and we have proposed a scheme by which a pathogen's epidemiology, and the political will to intervene in its control, may be classified. Similarly, their persistence in the political environment is influenced by a variety of factors, not least their zoonotic status, but also whether there is a wildlife reservoir that involves an animal with a positive public profile that only applies to bTB of the diseases examined here. We hope that this will provide a useful framework for future research into endemic livestock disease, and promote governance that is appropriate to each pathogen, within an integrated approach to endemic disease control.

Acknowledgements

We thank the Rural Economy and Land Use (RELU) programme (RES-229-25-0016) for funding. RELU is funded jointly by the Economic and Social Research Council, the Biotechnology and Biological Sciences Research Council and the Natural Environment Research Council, with additional funding from the Department for Environment, Food and Rural Affairs and the Scottish Government. Defra (SE3026) and BBSRC (BBS/B/04854) funded the epidemiological data collection and analysis. We also acknowledge the contributions made by Steve Moore, Sam Mason, Dr Kerry Woodbine and others to the collection and processing of the serological data.

References

View Abstract