Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
Contents lists available at ScienceDirect
Spatial and Spatio-temporal Epidemiology journal homepage: www.elsevier.com/locate/sste
Optimal vaccination strategies against vector-borne diseases Kaare Græsbøll a,⇑, Claes Enøe b, René Bødker b, Lasse Engbo Christiansen a a b
Department of Applied Mathematics and Computer Science, Technical University of Denmark, Denmark National Veterinary Institute, Technical University of Denmark, Denmark
a r t i c l e
i n f o
Article history: Available online 21 July 2014 Keywords: Bluetongue Vaccination Simulation Mathematical modelling
a b s t r a c t Using a process oriented semi-agent based model, we simulated the spread of Bluetongue virus by Culicoides, biting midges, between cattle in Denmark. We evaluated the minimum vaccination cover and minimum cost for eight different preventive vaccination strategies in Denmark. The simulation model replicates both a passive and active flight of midges between cattle distributed on pastures and cattle farms in Denmark. A seasonal abundance of midges and temperature dependence of biological processes were included in the model. The eight vaccination strategies were investigated under four different grazing conditions. Furthermore, scenarios were tested with three different index locations stratified for cattle density. The cheapest way to vaccinate cattle with a medium risk profile (less than 1000 total affected cattle) was to vaccinate cattle on pasture. Regional vaccination displayed better results when index cases were in the vaccinated areas. However, given that the long-range spread of midge borne disease is still poorly quantified, more robust national vaccination schemes seem preferable. Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction Bluetongue has been recognised as a ruminant disease for more than one hundred years (Spreull, 1905), and is present at various levels on all continents except Antarctica (Tabachnick et al., 2004). Bluetongue displays its most severe economic impact when entering non-endemic areas where animal resistance to the disease is little, as was very evident with its invasion of temperate Europe in 2006 (Velthuis et al., 2006), and Australia in 1977 (Ward, 1994). Bluetongue is non-contagious and is transferred by Culicoides vectors (Tabachnick et al., 2004; Wilson et al., 2008). In the case of Denmark, the primary ruminant host of the disease is cattle; as these outnumber sheep ten to one. The dominant Culicoides species in Denmark, which
⇑ Corresponding author. E-mail address: [email protected] (K. Græsbøll). http://dx.doi.org/10.1016/j.sste.2014.07.005 1877-5845/Ó 2014 Elsevier Ltd. All rights reserved.
are known to bite cattle, are the Culicoides obsoletus group and the Culicoides pulicaris group (Lassen et al., 2011). Preventive vaccination strategies can have the objective to prevent or confine the disease to a minimum level; the latter is usually used in a veterinary perspective, where prevention is equivalent to a costly vaccination of all cattle. For the Bluetongue disease (BT) in 2008, Denmark chose initially to vaccinate regionally in order to prevent introduction from nearby countries. However, BT has previously displayed long distance introdof several hundred kilometres such as in Scandinavia, the UK, the Mediterranean and Australia (Sellers and Pedgley, 1977; Braverman et al., 1996; Mellor and Wittmann, 1998; Alba et al., 2000; Ducheyne et al., 2007; Burgin et al., 2012; Eagles et al., 2012); in these cases, regional vaccination close to borders would have had little effect. However, the basic concept of increasing the distance between infectious and susceptible cattle can also be implemented countrywide on a local scale, which is investigated in this paper in comparison
154
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
with more traditional vaccination strategies. Increasing the distance between infectious and susceptible cattle on a local scale enables a vaccination campaign to cover large areas (i.e. country wide vaccination), while using a low number of total doses. In this paper, we aim to rank and quantify the effectiveness of eight vaccination strategies including specially designed strategies to maximise the distance between susceptible animals. Vaccination has been used successfully to control outbreaks in Europe (Zientara et al., 2010). In Italy, a vaccination of 80% of all susceptible animals was able to efficiently halt and prevent transmission (Patta et al., 2004). However, the climate and the vectors differ between southern and northern Europe, and empirical results from Italy can therefore, not be transferred to Denmark. Szmaragd et al. (2010) concluded by simulation that for England, an 80% vaccination of cattle was necessary to halt an epidemic, and that a regional roll-out needed to be done at sufficient speed to prevent Bluetongue from spreading to non-vaccinated areas. In this work, eight preventive vaccination scenarios are investigated with three areas of introduction into Denmark under four different grazing conditions, totalling 96 unique scenarios. For each scenario, the overall national vaccination cover is tested in steps of 10% with 1000 repetitions per step, in total 960,000 simulated introductions of BTV.
Denmark. From the Danish Central Husbandry Register (CHR), information about ownership of cattle could be obtained. Combining this information revealed which farmers could potentially put cattle out to pasture, and where these pasture areas were located. It was not known which farmers put their cattle out to pasture, so in each repetition of the simulation this was sampled randomly; also, the proportion of cattle put out to graze is likely to change in response to an outbreak. Within each grid cell, the transmission of disease is determined by evaluating the probability; given the amount of infectious vectors/hosts and the temperature dependent bite and mortality rates of vectors. Furthermore, the virus development in the vectors and hosts is determined. The spread of the disease is especially dependent upon this, as the period it takes for vectors to become infectious is highly temperature dependent, and vectors most often die before they become infectious. The mortality rate of vectors is also temperature dependent. Furthermore, the vector to host ratio is seasonal with vector season starting at the end of May and ending in late November, displaying four generation peaks. The parameters and full model are presented in Græsbøll et al. (2012).
2.1. Vaccination scenarios 2. Methods The model used for simulating the spread of disease is described in Græsbøll et al. (2012), and a description of Culicoides’ flying parameters for midges are provided in Græsbøll (2013, App. B, SM1). Below a short resume of the model is described, with a detailed description of the vaccination scenarios. The model is a semi-agent based stochastic process model: Individual midges are simulated to fly between cattle. The cattle can be located either on pasture or in stables on the farm’s location. Between flying events, the viraemic stage and the probability of transmission of the virus are evaluated based on temperature, using the Markov Chain Monte Carlo (MCMC) method. The vectors are simulated to fly in two distinct ways: Firstly, an active local flight mode where vectors fly at random on a length scale of hundreds of metres in order to locate cattle. Secondly, a passive mode where vectors can be carried several kilometres by the wind. Over small distances, vectors are assumed to be able to locate cattle with a 100% success rate. Therefore, all of Denmark is divided into a grid with cells of 300 by 300 metres. Within each grid cell, all of the vectors present can locate all of the cattle present. Vectors and hosts located within each grid cell are only discernable by disease status, and therefore, not individually tracked; hence, the term ‘‘semi-agent model’’. The grid also serves further technical purposes in the programming. The overlaying of a grid on pastures in Denmark allows for each grid cell to be assigned a number of cattle, based on the known use of this cell. From the European Union’s arable area subsidies programme, it was possible to obtain information about pasture areas and ownership of these in
Eight preventive vaccination scenarios were selected to represent as different approaches to preventive vaccination as possible, and are defined as follows: P A percentage of cattle on all farms are vaccinated. All farms are vaccinated to the determined in-farm percentage. RH Random Holdings. Farms are selected at random to be vaccinated. All cattle belonging to the selected farms are vaccinated. NN Nearest neighbour vaccination. The two cattle farms closest to each other are identified and by random selection, one of these farms has all of their cattle vaccinated. Then, the two non-vaccinated cattle farms nearest to each other are located and again one is selected at random to be vaccinated. This pattern continues until a certain proportion of farms have been vaccinated. S Vaccination based on farm size. Beginning with the smallest farms in Denmark (in terms of number of cattle), a number of farms are selected to have all of their cattle vaccinated up to the specified vaccination cover. L Vaccination based on farm size. Beginning with the largest farm in Denmark, a number of farms are selected up to the specified vaccination cover. G Vaccination of all cattle put out to pasture for selected farms. Farms are selected at random with cattle on pasture until the specific vaccine coverage, and all cattle from these farms that are on pasture are vaccinated. SJ Regional vaccination of Southern Jutland bordering Germany. Farms in Southern Jutland are selected randomly until a specified coverage was reached.
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
D2B Trench vaccination of all farms within a certain distance from the German border. For an increasing number of steps of 2.3 km from the Danish–German border, all cattle are vaccinated. Vaccination strategies are shown in Fig. 2. Vaccination is assumed to induce 100% immunity. The scenarios were chosen for the following reasons: P and RH are the reference scenarios. Scenario P represents a scenario where a percentage of the cattle population is selected for vaccination, and possibly stratified for age, gender, etc. Importantly, the P strategy can also represent a population vaccinated in the past; i.e., 50% of vaccinated cattle is equivalent to a 100% vaccination scenario that was carried out x years previously; where x is the median lifespan of Danish cattle. Scenario RH is a reference to determine whether scenarios that select individual farms are actually better than random selection. The NN scenario was selected as the most obvious way of establishing distance between susceptible cattle. The L scenario was selected to test a scenario where the fewest possible farms were selected to be vaccinated. When designing the L scenario, the S scenario is easily tested for in the same manner, only in the opposite order, and therefore, this scenario was included. The G scenario tests whether a similar effect to the NN scenario can be achieved by simply vaccinating all cattle on grass rather than taking them into stables. Given that only 35% of the cattle population is on pasture, this scenario is limited to vaccinate at most 35% of the total cattle population. The SJ and D2B scenarios are two different ways of implementing a regional vaccination. SJ with 100% vaccination coverage is similar to the way Denmark initially vaccinated in Jutland in 2008; while D2B is a scenario to test how broad a band to vaccinate from the Danish-German border needs to be for efficient protection. The SJ and D2B scenarios are spatially limited, and can therefore only cover vaccination up to approximately 20% of the Danish cattle population. All vaccination strategies chosen are preventive strategies where the vaccination is completed and immunity is achieved before the disease is introduced. Preventive strategies were selected because both the farmers’ association and agricultural ministries were of the opinion that future outbreaks of BTV in Europe were most likely to start outside Denmark, and vaccination campaigns therefore need to be preventive.
155
(b) 35% of randomly chosen farmers put their cattle out to pasture, the so called ‘pasture’ condition. (c) All farmers with less than 100 cattle put them out to pasture. Furthermore the remaining farmers with more than 100 cattle put 20% of their cattle out to pasture, so that the total number of grazing cattle matches scenario b. This is referred to as the ‘Small on Grass’ condition (SoG). (d) All farmers with less than 100 cattle put them out to pasture. Furthermore, farmers registered as nondairy with more than 100 cattle put 80% of their cattle out to pasture, and farmers registered as dairy with more than 100 cattle put 10% of their cattle out to pasture. In total, the number of grazing cattle matches conditions b and c. This is referred to as the ‘SoG + Milk’ condition. The three different grazing conditions are our best guesstimates of the actual distribution of cattle. The total number of cattle on grass is close within the estimated one third, from a 2010 questionnaire performed on a small sample of Danish cattle consultants by the Knowledge Center for Agriculture, Cattle (www.vfl.dk). The fourth condition of housed cattle was included to compare the effects of an eventual forced housing campaign as a response to an epidemic outbreak, in comparison and/or addition to vaccination. 2.3. Initialisation scenarios Even though Denmark is a small country, the soil quality differs quite substantially from region to region. Therefore, the amount of land used for crops or pasture differs, and hence the density of cattle varies substantially across Denmark (Fig. 1). Most likely because of the low
2.2. Cattle distribution The impact of vaccination scenarios may be affected by the proportion of cattle that are concentrated indoor in stables and put outdoors on surrounding pastures. Given that we do not know exactly how farmers distribute their cattle between stables and pastures now or during a future outbreak, we explored four different grazing conditions: (a) No cattle on pasture, the so called ‘housed’ condition.
Fig. 1. Cattle per km2 in Denmark. Green represents no cattle and light blue is water and non-Danish territories. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
156
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
temperatures and a short vector season, BT has not previously spread throughout the whole of Denmark in one season (Græsbøll et al., 2012), and the final size of an epidemic is therefore correlated with the local parameters at the site of introduction. One parameter that is known to vary considerably over Denmark is the cattle density. To investigate the effects of cattle density, three initialisation scenarios (init) were investigated: D Two farms situated in the cattle dense area south of the Danish–German border were started with five infectious cattle each. M Ten infectious midges were placed in a randomly selected grid cell containing cattle in the middle part of Jutland where cattle density is medium. L Ten infectious midges were placed in a randomly selected grid cell containing cattle on the island of Lolland where cattle density is low. 2.4. Simulations All simulations were started on July 1, using temperatures from 2008. Temperatures were used at an hourly basis interpolated sinusoidally from daily min max
temperatures. The start date was selected due to previous research showing this introduction date to be the one resulting in the largest following outbreak (Græsbøll et al., 2012). Each of the 96 unique scenarios were run 1000 times for 10%, 20%, . . ., 100% vaccination of farms available to the scenario. For the P, RH, S, L and NN, this includes scenarios where all cattle in Denmark are vaccinated (which is 1.6 mil.). For the three pasture conditions, with grass vaccination and G, only up to 35% of the total number of cattle could be vaccinated, as this corresponds to the proportion placed on pasture. For the regional (SJ) and trench (D2B) scenarios, the number of cattle present in the vaccination region represents 20% of the cattle population in Denmark, and therefore, this is the maximum fraction of cattle to be vaccinated in these scenarios. For each of the 96 scenarios, the 10 vaccination cover scenarios provided the 50 and 95 percentile of number of affected cattle. These values were interpolated to estimate the vaccination cover needed to keep an epidemic outbreak less than the specified values of one hundred or one thousand affected cattle. A general linear model (lm() in R 2.15) was used to test which of the settings influence the size of epidemic outbreaks.
Fig. 2. Visualisation of the vaccination scenarios. Green represents pixels with no vaccinated cattle in them, otherwise the brighter the colour the higher the fraction of cattle that are vaccinated within a 5 by 5 km pixel. Shown here is the vaccination cover of 10%, 30%, and 70%, and the pasture distribution scenario. From this it can be seen that farms that are the largest and nearest to each other are in the southern and northern part of Jutland. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
157
Fig. 3. The fraction of all Danish cattle that are to be vaccinated so that the median simulation results are less than a total of 1000/100 (left/right column) affected cattle within a given scenario. The error bars depict the vaccination cover where 95% of simulations reduce the epidemic to the specified size. Rows are the three different initialisation scenarios. There are no data for the ’housed’ condition under the ‘Grass’ vaccination (No black bars over the G scenario) because there are no grazing cattle to vaccinate . Where error bars are missing, there are no achievable levels of vaccination within the scenario that can reduce the epidemic to the desired size; except that all ’housed’ conditions initiated on Lolland (init L) have 95% simulations, showing less than 1000 affected cattle. The vaccination scenarios are listed in Section 2.1. The number of farms to be vaccinated is displayed in Fig. 4.
2.5. Cost comparison of strategies
3. Results
The strategy results for the number of doses and the number of farms to visit to administer these doses for the different uptake scenarios are given. A farm visit and a dose are assigned an index price from zero to one hundred and zero to twenty, so that the total cost of a vaccination strategy can be calculated and compared to the other strategies.
Figs. 3 and 4 show the fraction of all Danish cattle/farms that are to be vaccinated so that the median simulation results are less than a total of 1000/100 affected cattle within a given scenario. The error bars depict the vaccination cover where 95% of simulations reduces the epidemic to the specified size.
158
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
Fig. 4. The fraction of all Danish cattle farms that are to be vaccinated so that the median simulation results are less than a total of 1000/100 (left/right column) affected cattle within a given scenario. The error bars depict the vaccination cover where 95% of simulations reduce the epidemic to the specified size. The rows are the three different initialisation scenarios. Because there are no grazing cattle, there are no data for the ’housed’ condition under the ‘Grass’ vaccination (No black bars over the G scenario). Where error bars are missing, there are no achievable levels of vaccination within the scenario that can reduce the epidemic to the desired size; except that all ’housed’ scenarios initiated on Lolland have 95% simulations, giving less than 1000 affected cattle. The vaccination scenarios are listed in Section 2.1.
Taking the logarithm of the size of epidemic outbreaks plus 1 as response, the general linear model showed significant influences (P < 0.001) from all vaccination strategies, initialisation scenarios, grazing conditions, vaccination levels, and all interactions. There are large differences between scenarios in the vaccination cover needed to achieve an efficient halt on an epidemic outbreak. These differences are both within
vaccination, cattle distribution, and initialisation scenarios. Only the three grazing conditions display similar results across initialisation and vaccination scenarios (although they are significantly different). The regional scenario, SJ, and trench scenario, D2B, are not able to prevent spread of disease to any of the desired levels if the introduction does not take place in the area where vaccination is implemented. Furthermore, these
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
scenarios could not reduce the size of the epidemic with 95% certainty in any pasture scenario (note that there are no error bars on SJ and D2B for the pasture scenarios in Fig. 3). When comparing all cattle distributions the ‘housing’ condition displays the fewest doses needed to protect the Danish cattle to the specified level. Across vaccination scenarios, the Percentage of cattle (P), Small farms (S) and cattle on Grass (G) are the scenarios that need the fewest doses to protect to the specified level (Fig. 3). In order to reduce the number of farm visits (Fig. 4), the most efficient strategy is to vaccinate the largest farms (L scenario). Fig. 5 was created by combining the information in Figs. 3 and 4 by assigning a price for veterinary visits to farms to vaccinate, and a price of the vaccination per animal. From this, the total price of vaccinations could be calculated to reveal the cheapest strategy, which is visualised in Fig. 5. This is by no means a complete economic analysis of the data, but a simple visualisation to demonstrate that the cheapest strategy may depend on many factors. This simplified price estimate of a vaccination campaign was applied to give a first impression of the cheapest
159
vaccination strategy. However, what is apparent from Figs. 5 and S1 is that the cheapest strategy could be any six of the eight suggested, dependent on the scenario and pricing; the scenarios that are never the cheapest are the RH and the NN strategies.
4. Discussion The results presented demonstrate that the different vaccination strategies have markedly different levels of efficiency, depending on the initialisation scenario and grazing conditions; therefore, selecting an optimal strategy is not only a function of the price of doses and herd visits; but also affected by the index case, the grazing conditions, and the tolerable level of affected animals. The two regional vaccination strategies, SJ and D2B, did not perform well when the infection was introduced far from the vaccinated areas, which was to be expected. SJ was unfortunately the initial strategy against Bluetongue in Denmark in 2008, when Denmark was unprepared for a Culicoides borne outbreak. BT rapidly broke through the
Fig. 5. Shows the cheapest scenario when appointing index price to dose and farm visit (Linear function of Figs. 3 and 4). These four plots are example plots, the full 48 plots are given in Supplementary material. All except the NN and RH scenarios can be the cheapest given different conditions.
160
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
region and the strategy had to be abandoned, with a larger scale vaccination campaign to follow. Noticeable among the four cattle distributions was that ‘housing’ of all cattle were able to reduce the size of an epidemic compared to the three pasture scenarios. However, the difference between housing and pasture depends on the ability of vectors to locate cattle. If the distance that midges are able to locate cattle is larger than the average distance between farm locations, then a housing strategy is of no effect compared to having cattle on pasture. Increasing the locating ability for the vectors is equivalent to increasing the grid size of the simulation (Græsbøll et al., 2012). There are no published results on the ability of midges to locate cattle; Sedda et al., 1737 assumed that midges could locate cattle within 200 m, estimated from mosquito data. In this paper, we assume that midges can locate all cattle within a grid square of 300 by 300 metres. There may also be an added effect of housing, given that there have been observations of lower biting rates in stables or lower seroprevalence in areas with fewer cattle (Santman-Berends et al., 2007; Pioz et al., 2012). However, other studies have concluded the opposite effect, and whether the effect of stabling is positive or negative may be influenced by the vector species and possibly the micro climate in the area and in the stables (Barnard, 1997; Meiswinkel et al., 2000; Calvete et al., 2009; Baylis et al., 2010; Ninio et al., 2011). To compare regional versus country-wide vaccination scenarios is difficult, as the long distance spread is not very well quantified. Unsurprisingly, this study finds that regional strategies are good when introduction is in those areas, but very poor when the index herd is outside the vaccinated areas. Different investigations into the long range spread of Bluetongue (Hendrickx et al., 2006; Sedda et al., 1737) have suggested that the number of long distance events is small. All these statistics rest on data based on spread over land. It seems likely that if blown over, sea midges will chose to stay airborne until reaching land, thereby dramatically changing the spread kernel compared to spread over land (Burgin et al., 2012). A further factor that may influence the spread of disease is the time of day that dairy cows are put in to stable (Santman-Berends et al., 2007). The active flying periods of midges are dusk and dawn, with the primary period at dusk (Sanders et al., 2012). In Denmark, the peak of flying in the summertime is therefore quite late in the evening (Kirkeby et al., 2013), which may be after the time that dairy cattle are brought in for the night. From Fig. 3 we see that many of the proposed vaccination strategies can prevent the spread of bluetongue to a tolerable level by vaccinating much fewer cattle than the 80% suggested in other studies (Patta et al., 2004; Szmaragd et al., 2010). However, as also evident when comparing results is that these levels may vary significantly between areas of different cattle densities, so these results are not necessarily transferrable to other parts of Europe. Generally, the P vaccination strategy is one of the strategies that requires the fewest doses for all introductions and cattle distributions . The primary reason for this is probably that with the P strategy, all areas with cattle
are partially vaccinated, which will slow down an epidemic from the start; compared to the other strategies where the chances are relatively high to encounter fully susceptible areas in which the virus can spread quickly. However, the P strategy requires that all farms must be visited, which is costly. Furthermore, should all farms vaccinate there is little reason not to do a 100% vaccination that would prevent any introduction. Other than the P scenario, the small farm (S) and grass (G) scenarios are the scenarios that require the fewest doses to reduce the epidemic to the specified level. These scenarios are thus efficient because they result in the highest number of areas with vaccinated cattle using the fewest doses. A good spatial coverage is the key to halting an epidemic, as this will remove the most possible infectious bites. We had originally hypothesised that the nearest neighbour (NN) scenario would be very effective, especially when housing cattle; however, as this scenario selects small and large farms with equal probability (there is no correlation between size and distance between farms), the effect of this scenario lies somewhere between the S and L scenarios. A source of error in Fig. 5 may be that the calculated price does not take into account that the price of vaccinating stabled and grazing cattle can be substantially different if the preventive vaccination is not started before the grazing season. Furthermore, a complete cost benefit of the system requires multiple variables (Velthuis et al., 2006; Conraths et al., 2012). From the results presented in Fig. 5 and the Supplementary material, two domains of the data can be qualitatively identified. The first one is: if a high certainty of efficiency and low number of affected animals (
Contents lists available at ScienceDirect
Spatial and Spatio-temporal Epidemiology journal homepage: www.elsevier.com/locate/sste
Optimal vaccination strategies against vector-borne diseases Kaare Græsbøll a,⇑, Claes Enøe b, René Bødker b, Lasse Engbo Christiansen a a b
Department of Applied Mathematics and Computer Science, Technical University of Denmark, Denmark National Veterinary Institute, Technical University of Denmark, Denmark
a r t i c l e
i n f o
Article history: Available online 21 July 2014 Keywords: Bluetongue Vaccination Simulation Mathematical modelling
a b s t r a c t Using a process oriented semi-agent based model, we simulated the spread of Bluetongue virus by Culicoides, biting midges, between cattle in Denmark. We evaluated the minimum vaccination cover and minimum cost for eight different preventive vaccination strategies in Denmark. The simulation model replicates both a passive and active flight of midges between cattle distributed on pastures and cattle farms in Denmark. A seasonal abundance of midges and temperature dependence of biological processes were included in the model. The eight vaccination strategies were investigated under four different grazing conditions. Furthermore, scenarios were tested with three different index locations stratified for cattle density. The cheapest way to vaccinate cattle with a medium risk profile (less than 1000 total affected cattle) was to vaccinate cattle on pasture. Regional vaccination displayed better results when index cases were in the vaccinated areas. However, given that the long-range spread of midge borne disease is still poorly quantified, more robust national vaccination schemes seem preferable. Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction Bluetongue has been recognised as a ruminant disease for more than one hundred years (Spreull, 1905), and is present at various levels on all continents except Antarctica (Tabachnick et al., 2004). Bluetongue displays its most severe economic impact when entering non-endemic areas where animal resistance to the disease is little, as was very evident with its invasion of temperate Europe in 2006 (Velthuis et al., 2006), and Australia in 1977 (Ward, 1994). Bluetongue is non-contagious and is transferred by Culicoides vectors (Tabachnick et al., 2004; Wilson et al., 2008). In the case of Denmark, the primary ruminant host of the disease is cattle; as these outnumber sheep ten to one. The dominant Culicoides species in Denmark, which
⇑ Corresponding author. E-mail address: [email protected] (K. Græsbøll). http://dx.doi.org/10.1016/j.sste.2014.07.005 1877-5845/Ó 2014 Elsevier Ltd. All rights reserved.
are known to bite cattle, are the Culicoides obsoletus group and the Culicoides pulicaris group (Lassen et al., 2011). Preventive vaccination strategies can have the objective to prevent or confine the disease to a minimum level; the latter is usually used in a veterinary perspective, where prevention is equivalent to a costly vaccination of all cattle. For the Bluetongue disease (BT) in 2008, Denmark chose initially to vaccinate regionally in order to prevent introduction from nearby countries. However, BT has previously displayed long distance introdof several hundred kilometres such as in Scandinavia, the UK, the Mediterranean and Australia (Sellers and Pedgley, 1977; Braverman et al., 1996; Mellor and Wittmann, 1998; Alba et al., 2000; Ducheyne et al., 2007; Burgin et al., 2012; Eagles et al., 2012); in these cases, regional vaccination close to borders would have had little effect. However, the basic concept of increasing the distance between infectious and susceptible cattle can also be implemented countrywide on a local scale, which is investigated in this paper in comparison
154
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
with more traditional vaccination strategies. Increasing the distance between infectious and susceptible cattle on a local scale enables a vaccination campaign to cover large areas (i.e. country wide vaccination), while using a low number of total doses. In this paper, we aim to rank and quantify the effectiveness of eight vaccination strategies including specially designed strategies to maximise the distance between susceptible animals. Vaccination has been used successfully to control outbreaks in Europe (Zientara et al., 2010). In Italy, a vaccination of 80% of all susceptible animals was able to efficiently halt and prevent transmission (Patta et al., 2004). However, the climate and the vectors differ between southern and northern Europe, and empirical results from Italy can therefore, not be transferred to Denmark. Szmaragd et al. (2010) concluded by simulation that for England, an 80% vaccination of cattle was necessary to halt an epidemic, and that a regional roll-out needed to be done at sufficient speed to prevent Bluetongue from spreading to non-vaccinated areas. In this work, eight preventive vaccination scenarios are investigated with three areas of introduction into Denmark under four different grazing conditions, totalling 96 unique scenarios. For each scenario, the overall national vaccination cover is tested in steps of 10% with 1000 repetitions per step, in total 960,000 simulated introductions of BTV.
Denmark. From the Danish Central Husbandry Register (CHR), information about ownership of cattle could be obtained. Combining this information revealed which farmers could potentially put cattle out to pasture, and where these pasture areas were located. It was not known which farmers put their cattle out to pasture, so in each repetition of the simulation this was sampled randomly; also, the proportion of cattle put out to graze is likely to change in response to an outbreak. Within each grid cell, the transmission of disease is determined by evaluating the probability; given the amount of infectious vectors/hosts and the temperature dependent bite and mortality rates of vectors. Furthermore, the virus development in the vectors and hosts is determined. The spread of the disease is especially dependent upon this, as the period it takes for vectors to become infectious is highly temperature dependent, and vectors most often die before they become infectious. The mortality rate of vectors is also temperature dependent. Furthermore, the vector to host ratio is seasonal with vector season starting at the end of May and ending in late November, displaying four generation peaks. The parameters and full model are presented in Græsbøll et al. (2012).
2.1. Vaccination scenarios 2. Methods The model used for simulating the spread of disease is described in Græsbøll et al. (2012), and a description of Culicoides’ flying parameters for midges are provided in Græsbøll (2013, App. B, SM1). Below a short resume of the model is described, with a detailed description of the vaccination scenarios. The model is a semi-agent based stochastic process model: Individual midges are simulated to fly between cattle. The cattle can be located either on pasture or in stables on the farm’s location. Between flying events, the viraemic stage and the probability of transmission of the virus are evaluated based on temperature, using the Markov Chain Monte Carlo (MCMC) method. The vectors are simulated to fly in two distinct ways: Firstly, an active local flight mode where vectors fly at random on a length scale of hundreds of metres in order to locate cattle. Secondly, a passive mode where vectors can be carried several kilometres by the wind. Over small distances, vectors are assumed to be able to locate cattle with a 100% success rate. Therefore, all of Denmark is divided into a grid with cells of 300 by 300 metres. Within each grid cell, all of the vectors present can locate all of the cattle present. Vectors and hosts located within each grid cell are only discernable by disease status, and therefore, not individually tracked; hence, the term ‘‘semi-agent model’’. The grid also serves further technical purposes in the programming. The overlaying of a grid on pastures in Denmark allows for each grid cell to be assigned a number of cattle, based on the known use of this cell. From the European Union’s arable area subsidies programme, it was possible to obtain information about pasture areas and ownership of these in
Eight preventive vaccination scenarios were selected to represent as different approaches to preventive vaccination as possible, and are defined as follows: P A percentage of cattle on all farms are vaccinated. All farms are vaccinated to the determined in-farm percentage. RH Random Holdings. Farms are selected at random to be vaccinated. All cattle belonging to the selected farms are vaccinated. NN Nearest neighbour vaccination. The two cattle farms closest to each other are identified and by random selection, one of these farms has all of their cattle vaccinated. Then, the two non-vaccinated cattle farms nearest to each other are located and again one is selected at random to be vaccinated. This pattern continues until a certain proportion of farms have been vaccinated. S Vaccination based on farm size. Beginning with the smallest farms in Denmark (in terms of number of cattle), a number of farms are selected to have all of their cattle vaccinated up to the specified vaccination cover. L Vaccination based on farm size. Beginning with the largest farm in Denmark, a number of farms are selected up to the specified vaccination cover. G Vaccination of all cattle put out to pasture for selected farms. Farms are selected at random with cattle on pasture until the specific vaccine coverage, and all cattle from these farms that are on pasture are vaccinated. SJ Regional vaccination of Southern Jutland bordering Germany. Farms in Southern Jutland are selected randomly until a specified coverage was reached.
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
D2B Trench vaccination of all farms within a certain distance from the German border. For an increasing number of steps of 2.3 km from the Danish–German border, all cattle are vaccinated. Vaccination strategies are shown in Fig. 2. Vaccination is assumed to induce 100% immunity. The scenarios were chosen for the following reasons: P and RH are the reference scenarios. Scenario P represents a scenario where a percentage of the cattle population is selected for vaccination, and possibly stratified for age, gender, etc. Importantly, the P strategy can also represent a population vaccinated in the past; i.e., 50% of vaccinated cattle is equivalent to a 100% vaccination scenario that was carried out x years previously; where x is the median lifespan of Danish cattle. Scenario RH is a reference to determine whether scenarios that select individual farms are actually better than random selection. The NN scenario was selected as the most obvious way of establishing distance between susceptible cattle. The L scenario was selected to test a scenario where the fewest possible farms were selected to be vaccinated. When designing the L scenario, the S scenario is easily tested for in the same manner, only in the opposite order, and therefore, this scenario was included. The G scenario tests whether a similar effect to the NN scenario can be achieved by simply vaccinating all cattle on grass rather than taking them into stables. Given that only 35% of the cattle population is on pasture, this scenario is limited to vaccinate at most 35% of the total cattle population. The SJ and D2B scenarios are two different ways of implementing a regional vaccination. SJ with 100% vaccination coverage is similar to the way Denmark initially vaccinated in Jutland in 2008; while D2B is a scenario to test how broad a band to vaccinate from the Danish-German border needs to be for efficient protection. The SJ and D2B scenarios are spatially limited, and can therefore only cover vaccination up to approximately 20% of the Danish cattle population. All vaccination strategies chosen are preventive strategies where the vaccination is completed and immunity is achieved before the disease is introduced. Preventive strategies were selected because both the farmers’ association and agricultural ministries were of the opinion that future outbreaks of BTV in Europe were most likely to start outside Denmark, and vaccination campaigns therefore need to be preventive.
155
(b) 35% of randomly chosen farmers put their cattle out to pasture, the so called ‘pasture’ condition. (c) All farmers with less than 100 cattle put them out to pasture. Furthermore the remaining farmers with more than 100 cattle put 20% of their cattle out to pasture, so that the total number of grazing cattle matches scenario b. This is referred to as the ‘Small on Grass’ condition (SoG). (d) All farmers with less than 100 cattle put them out to pasture. Furthermore, farmers registered as nondairy with more than 100 cattle put 80% of their cattle out to pasture, and farmers registered as dairy with more than 100 cattle put 10% of their cattle out to pasture. In total, the number of grazing cattle matches conditions b and c. This is referred to as the ‘SoG + Milk’ condition. The three different grazing conditions are our best guesstimates of the actual distribution of cattle. The total number of cattle on grass is close within the estimated one third, from a 2010 questionnaire performed on a small sample of Danish cattle consultants by the Knowledge Center for Agriculture, Cattle (www.vfl.dk). The fourth condition of housed cattle was included to compare the effects of an eventual forced housing campaign as a response to an epidemic outbreak, in comparison and/or addition to vaccination. 2.3. Initialisation scenarios Even though Denmark is a small country, the soil quality differs quite substantially from region to region. Therefore, the amount of land used for crops or pasture differs, and hence the density of cattle varies substantially across Denmark (Fig. 1). Most likely because of the low
2.2. Cattle distribution The impact of vaccination scenarios may be affected by the proportion of cattle that are concentrated indoor in stables and put outdoors on surrounding pastures. Given that we do not know exactly how farmers distribute their cattle between stables and pastures now or during a future outbreak, we explored four different grazing conditions: (a) No cattle on pasture, the so called ‘housed’ condition.
Fig. 1. Cattle per km2 in Denmark. Green represents no cattle and light blue is water and non-Danish territories. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
156
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
temperatures and a short vector season, BT has not previously spread throughout the whole of Denmark in one season (Græsbøll et al., 2012), and the final size of an epidemic is therefore correlated with the local parameters at the site of introduction. One parameter that is known to vary considerably over Denmark is the cattle density. To investigate the effects of cattle density, three initialisation scenarios (init) were investigated: D Two farms situated in the cattle dense area south of the Danish–German border were started with five infectious cattle each. M Ten infectious midges were placed in a randomly selected grid cell containing cattle in the middle part of Jutland where cattle density is medium. L Ten infectious midges were placed in a randomly selected grid cell containing cattle on the island of Lolland where cattle density is low. 2.4. Simulations All simulations were started on July 1, using temperatures from 2008. Temperatures were used at an hourly basis interpolated sinusoidally from daily min max
temperatures. The start date was selected due to previous research showing this introduction date to be the one resulting in the largest following outbreak (Græsbøll et al., 2012). Each of the 96 unique scenarios were run 1000 times for 10%, 20%, . . ., 100% vaccination of farms available to the scenario. For the P, RH, S, L and NN, this includes scenarios where all cattle in Denmark are vaccinated (which is 1.6 mil.). For the three pasture conditions, with grass vaccination and G, only up to 35% of the total number of cattle could be vaccinated, as this corresponds to the proportion placed on pasture. For the regional (SJ) and trench (D2B) scenarios, the number of cattle present in the vaccination region represents 20% of the cattle population in Denmark, and therefore, this is the maximum fraction of cattle to be vaccinated in these scenarios. For each of the 96 scenarios, the 10 vaccination cover scenarios provided the 50 and 95 percentile of number of affected cattle. These values were interpolated to estimate the vaccination cover needed to keep an epidemic outbreak less than the specified values of one hundred or one thousand affected cattle. A general linear model (lm() in R 2.15) was used to test which of the settings influence the size of epidemic outbreaks.
Fig. 2. Visualisation of the vaccination scenarios. Green represents pixels with no vaccinated cattle in them, otherwise the brighter the colour the higher the fraction of cattle that are vaccinated within a 5 by 5 km pixel. Shown here is the vaccination cover of 10%, 30%, and 70%, and the pasture distribution scenario. From this it can be seen that farms that are the largest and nearest to each other are in the southern and northern part of Jutland. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
157
Fig. 3. The fraction of all Danish cattle that are to be vaccinated so that the median simulation results are less than a total of 1000/100 (left/right column) affected cattle within a given scenario. The error bars depict the vaccination cover where 95% of simulations reduce the epidemic to the specified size. Rows are the three different initialisation scenarios. There are no data for the ’housed’ condition under the ‘Grass’ vaccination (No black bars over the G scenario) because there are no grazing cattle to vaccinate . Where error bars are missing, there are no achievable levels of vaccination within the scenario that can reduce the epidemic to the desired size; except that all ’housed’ conditions initiated on Lolland (init L) have 95% simulations, showing less than 1000 affected cattle. The vaccination scenarios are listed in Section 2.1. The number of farms to be vaccinated is displayed in Fig. 4.
2.5. Cost comparison of strategies
3. Results
The strategy results for the number of doses and the number of farms to visit to administer these doses for the different uptake scenarios are given. A farm visit and a dose are assigned an index price from zero to one hundred and zero to twenty, so that the total cost of a vaccination strategy can be calculated and compared to the other strategies.
Figs. 3 and 4 show the fraction of all Danish cattle/farms that are to be vaccinated so that the median simulation results are less than a total of 1000/100 affected cattle within a given scenario. The error bars depict the vaccination cover where 95% of simulations reduces the epidemic to the specified size.
158
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
Fig. 4. The fraction of all Danish cattle farms that are to be vaccinated so that the median simulation results are less than a total of 1000/100 (left/right column) affected cattle within a given scenario. The error bars depict the vaccination cover where 95% of simulations reduce the epidemic to the specified size. The rows are the three different initialisation scenarios. Because there are no grazing cattle, there are no data for the ’housed’ condition under the ‘Grass’ vaccination (No black bars over the G scenario). Where error bars are missing, there are no achievable levels of vaccination within the scenario that can reduce the epidemic to the desired size; except that all ’housed’ scenarios initiated on Lolland have 95% simulations, giving less than 1000 affected cattle. The vaccination scenarios are listed in Section 2.1.
Taking the logarithm of the size of epidemic outbreaks plus 1 as response, the general linear model showed significant influences (P < 0.001) from all vaccination strategies, initialisation scenarios, grazing conditions, vaccination levels, and all interactions. There are large differences between scenarios in the vaccination cover needed to achieve an efficient halt on an epidemic outbreak. These differences are both within
vaccination, cattle distribution, and initialisation scenarios. Only the three grazing conditions display similar results across initialisation and vaccination scenarios (although they are significantly different). The regional scenario, SJ, and trench scenario, D2B, are not able to prevent spread of disease to any of the desired levels if the introduction does not take place in the area where vaccination is implemented. Furthermore, these
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
scenarios could not reduce the size of the epidemic with 95% certainty in any pasture scenario (note that there are no error bars on SJ and D2B for the pasture scenarios in Fig. 3). When comparing all cattle distributions the ‘housing’ condition displays the fewest doses needed to protect the Danish cattle to the specified level. Across vaccination scenarios, the Percentage of cattle (P), Small farms (S) and cattle on Grass (G) are the scenarios that need the fewest doses to protect to the specified level (Fig. 3). In order to reduce the number of farm visits (Fig. 4), the most efficient strategy is to vaccinate the largest farms (L scenario). Fig. 5 was created by combining the information in Figs. 3 and 4 by assigning a price for veterinary visits to farms to vaccinate, and a price of the vaccination per animal. From this, the total price of vaccinations could be calculated to reveal the cheapest strategy, which is visualised in Fig. 5. This is by no means a complete economic analysis of the data, but a simple visualisation to demonstrate that the cheapest strategy may depend on many factors. This simplified price estimate of a vaccination campaign was applied to give a first impression of the cheapest
159
vaccination strategy. However, what is apparent from Figs. 5 and S1 is that the cheapest strategy could be any six of the eight suggested, dependent on the scenario and pricing; the scenarios that are never the cheapest are the RH and the NN strategies.
4. Discussion The results presented demonstrate that the different vaccination strategies have markedly different levels of efficiency, depending on the initialisation scenario and grazing conditions; therefore, selecting an optimal strategy is not only a function of the price of doses and herd visits; but also affected by the index case, the grazing conditions, and the tolerable level of affected animals. The two regional vaccination strategies, SJ and D2B, did not perform well when the infection was introduced far from the vaccinated areas, which was to be expected. SJ was unfortunately the initial strategy against Bluetongue in Denmark in 2008, when Denmark was unprepared for a Culicoides borne outbreak. BT rapidly broke through the
Fig. 5. Shows the cheapest scenario when appointing index price to dose and farm visit (Linear function of Figs. 3 and 4). These four plots are example plots, the full 48 plots are given in Supplementary material. All except the NN and RH scenarios can be the cheapest given different conditions.
160
K. Græsbøll et al. / Spatial and Spatio-temporal Epidemiology 11 (2014) 153–162
region and the strategy had to be abandoned, with a larger scale vaccination campaign to follow. Noticeable among the four cattle distributions was that ‘housing’ of all cattle were able to reduce the size of an epidemic compared to the three pasture scenarios. However, the difference between housing and pasture depends on the ability of vectors to locate cattle. If the distance that midges are able to locate cattle is larger than the average distance between farm locations, then a housing strategy is of no effect compared to having cattle on pasture. Increasing the locating ability for the vectors is equivalent to increasing the grid size of the simulation (Græsbøll et al., 2012). There are no published results on the ability of midges to locate cattle; Sedda et al., 1737 assumed that midges could locate cattle within 200 m, estimated from mosquito data. In this paper, we assume that midges can locate all cattle within a grid square of 300 by 300 metres. There may also be an added effect of housing, given that there have been observations of lower biting rates in stables or lower seroprevalence in areas with fewer cattle (Santman-Berends et al., 2007; Pioz et al., 2012). However, other studies have concluded the opposite effect, and whether the effect of stabling is positive or negative may be influenced by the vector species and possibly the micro climate in the area and in the stables (Barnard, 1997; Meiswinkel et al., 2000; Calvete et al., 2009; Baylis et al., 2010; Ninio et al., 2011). To compare regional versus country-wide vaccination scenarios is difficult, as the long distance spread is not very well quantified. Unsurprisingly, this study finds that regional strategies are good when introduction is in those areas, but very poor when the index herd is outside the vaccinated areas. Different investigations into the long range spread of Bluetongue (Hendrickx et al., 2006; Sedda et al., 1737) have suggested that the number of long distance events is small. All these statistics rest on data based on spread over land. It seems likely that if blown over, sea midges will chose to stay airborne until reaching land, thereby dramatically changing the spread kernel compared to spread over land (Burgin et al., 2012). A further factor that may influence the spread of disease is the time of day that dairy cows are put in to stable (Santman-Berends et al., 2007). The active flying periods of midges are dusk and dawn, with the primary period at dusk (Sanders et al., 2012). In Denmark, the peak of flying in the summertime is therefore quite late in the evening (Kirkeby et al., 2013), which may be after the time that dairy cattle are brought in for the night. From Fig. 3 we see that many of the proposed vaccination strategies can prevent the spread of bluetongue to a tolerable level by vaccinating much fewer cattle than the 80% suggested in other studies (Patta et al., 2004; Szmaragd et al., 2010). However, as also evident when comparing results is that these levels may vary significantly between areas of different cattle densities, so these results are not necessarily transferrable to other parts of Europe. Generally, the P vaccination strategy is one of the strategies that requires the fewest doses for all introductions and cattle distributions . The primary reason for this is probably that with the P strategy, all areas with cattle
are partially vaccinated, which will slow down an epidemic from the start; compared to the other strategies where the chances are relatively high to encounter fully susceptible areas in which the virus can spread quickly. However, the P strategy requires that all farms must be visited, which is costly. Furthermore, should all farms vaccinate there is little reason not to do a 100% vaccination that would prevent any introduction. Other than the P scenario, the small farm (S) and grass (G) scenarios are the scenarios that require the fewest doses to reduce the epidemic to the specified level. These scenarios are thus efficient because they result in the highest number of areas with vaccinated cattle using the fewest doses. A good spatial coverage is the key to halting an epidemic, as this will remove the most possible infectious bites. We had originally hypothesised that the nearest neighbour (NN) scenario would be very effective, especially when housing cattle; however, as this scenario selects small and large farms with equal probability (there is no correlation between size and distance between farms), the effect of this scenario lies somewhere between the S and L scenarios. A source of error in Fig. 5 may be that the calculated price does not take into account that the price of vaccinating stabled and grazing cattle can be substantially different if the preventive vaccination is not started before the grazing season. Furthermore, a complete cost benefit of the system requires multiple variables (Velthuis et al., 2006; Conraths et al., 2012). From the results presented in Fig. 5 and the Supplementary material, two domains of the data can be qualitatively identified. The first one is: if a high certainty of efficiency and low number of affected animals (
