Moving towards a better understanding of airborne transmission of FMD

Transcription

1 Appendix 36 Moving towards a better understanding of airborne transmission of FMD John Gloster 1*, Isabel Esteves 2 and Soren Alexandersen 3 1 Met Office, UK, based at Pirbright Laboratory, Institute for Animal Health, Ash Rd, Woking, Surrey, GU24 0NF, UK 2 Pirbright Laboratory, Institute for Animal Health, Ash Rd, Woking, Surrey, GU24 0NF, UK 3 Danish Institute for Food and Veterinary Research, Department of Virology, Lindholm, DK-4771 Kalvehave, Denmark Abstract In the event of an outbreak of FMD it is essential for those responsible for controlling and eradicating the disease to quickly assess how the initial animals became infected and its potential for further spread. Once this has been determined the appropriate control measures can be introduced. It has been established that airborne transmission of FMD virus is one of the mechanisms by which disease is transmitted and consequently it is important to develop a capability by which this can be rapidly assessed. This often involves the integration of epidemiological data collected in the field, laboratory investigations and the use of a transport and dispersion model. Before an accurate prediction of airborne spread can be made it is of vital importance to understand each part of the disease chain together with its errors and uncertainties. How the model then represents the data is also key to obtaining a good estimate of disease spread. This paper identifies the key components in the models and identifies current errors and uncertainties. Introduction Laboratory and epidemiological studies have identified airborne transmission of FMD virus as one of the mechanisms by which disease is transmitted (reviewed by Alexandersen et al. 2003). A number of field studies have indicated that airborne virus was the most likely mechanism for disease transmission over many tens of km over land and several occasions involving a hundred or more km over the sea (Hugh-Jones and Wright 1970; Sellers and Foreman 1973; Donaldson et al. 1982; Sørensen et al. 2000; Gloster et al. 2003; Mikkelsen et al. 2003). In principle the airborne disease cycle is easy to understand (emission, transport and infection). However, in practice the inter-relation is far more complex and requires the use of a computer model to assess the potential for disease spread. For example, virus emission rates depend critically on the species involved, the stage of disease and the meteorological conditions. The meteorological conditions may vary significantly throughout one day, let alone over a typical period of virus emission (often a number of days). Definition of areas of risk is often made more difficult in areas of complex topography. The quality of an assessment of airborne transmission will only be as good as the input data and the assumptions and formulation of the model. Inaccurate model output may be misleading or wrong and any operational decisions taken on the strength of it flawed. Consequently it is important to have a detailed understanding of the quality and representivity model input data and how the model handles this information. Materials and Methods There are a number of atmospheric transport and dispersion models in use at centres around the world and potentially there are numerous combinations of inputs, outputs and degree of integration into decision support models (Sørensen 1998; Sanson, Morris and Stern 1999; Sørensen et al. 2000; Sørensen et al. 2001; Morris et al. 2002; Gloster et al. 2003). However, in general they all conform to a similar generic structure. This structure is used as the basis for assessing our current understanding and ability to predict airborne transmission of FMD. Results - Disease transmission and model representation Atmospheric transport and dispersion models are supplied with two major inputs; epidemiological to determine source terms and virus characteristics and meteorological data for virus transport. The output is typically presented in terms of virus concentration as a function of distance from the source. If a Graphical Interface System (GIS) is available other relevant data e.g. livestock distribution can also be presented simultaneously. 227

2 Epidemiological data Epidemiological data is obtained from two sources; firstly, from the infected premises and secondly from the laboratory. The local State Veterinarian is responsible, amongst other duties, for establishing the precise status of disease on the premises and the development of disease from initial introduction. This involves a close examination of all of livestock which may have been moved off the premises and an accurate dating of lesions on infected animals. The local State Veterinarian may be assisted in these crucial examinations by an Epidemiological Team, including field epidemiologists and FMD experts from such organisations as the Institute for Animal Health (IAH). In the UK infected material is passed from the infected premises to the IAH laboratory for confirmation of the presence of disease together with its type/strain. Provided that an accurate pattern of disease on the farm can be established and that the aerosol characteristics can be quantified (Alexandersen and Donaldson 2002; Alexandersen et al. 2003; Alexandersen et al. 2002) an estimate can then be made on the amount of virus released into the atmosphere. Any errors introduced at this stage can have a significant impact on the output from a prediction model. The accuracy of the data and predictions can be improved by deployment of an experienced Epidemiological Team and be further enhanced by extended laboratory testing of serum samples for virus and for antibodies to substantiate the introduction, spread and time course of the infection on the infected premises (Alexandersen et al. 2003b). Accurate lesion dating is often very difficult to perform, especially under field conditions. If the oldest lesion is missed or if the history of disease on the premises is not accurately assessed then the model output is likely to be in error, especially under rapidly changing meteorological conditions. Establishing virus characteristics in the laboratory is hard to characterise and involves making measurements at the limits of detection capability. Daily virus totals are typically calculated from short term measurements. How representative are these of a full 24 hour period? Detailed definition of source terms for example area and height of virus release, particle size distributions etc. Meteorological data Meteorological data is available from observing stations throughout the UK and abroad. There is a trend for the reduction of traditional observing stations and an increase in the introduction of a limited number of automated observing stations. For any given outbreak the location of an observing site may be many tens of km away and even then possibly in different terrain. The representivity of any data must be taken into consideration before it can be used in a transport and dispersion model. The presence of features such as mountains, hills, valleys and proximity of urban development and the coast can seriously influence the representivity of single site observations. Whilst some transport and dispersion models require basic weather observations as input others, such as the longer distance models require data derived from Numerical Weather Prediction models (NWP). These take observed data and then calculate the conditions on a 3-D grid with points being representative of an area. Clearly the greater the distance between grid points the less accurate it is likely to be for a specific location, especially in areas of complex topography. Typically grid points may be tens of km apart, although with increased computing power this is being gradually reduced. It is possible that within the next ten years the resolution may be reduced to around 1 5km. Some transport and dispersion models have a pre-processing stage which calculates a 3-D flow field in the light of topography. The wind speed and direction, and possibly some of the other inter-related atmospheric parameters, is then modified to allow for the effects of hills and valleys. Some models include the provision of using a number of schemes to reflect different flows under stable, neutral and unstable atmospheric conditions. Whilst these can be useful, few if any take into account locally generated drainage currents which may occur during periods of stable conditions (the highest risk of spread of virus in high concentrations). For example in a valley with steep sides the topographical routines may indicate that air is likely to flow up the valley, rather than flow over the hill to the side. However, local winds may in reality introduce an overall down valley flow, thus reversing the direction of the travel of the plume and consequent area of risk to livestock. In reality the situation will depend upon factors such as the orientation of the valley, its latitude and time of year. At this time there are no models which confidently predict these flows in all atmospheric conditions. It will not be until 228

3 model scales are reduced to around a few km, an improved understanding of local flows is established and improved physics is introduced into the meteorological models that this will be overcome. Representivity of either the observed or model estimated data. This is especially important when the atmosphere is stable and the winds are light. The influence of topography, coasts, cities and local meteorological conditions. Meteorological forecast errors if the model is used to predict the future airborne emission pattern from infected premises. Transport and dispersion models There is large variety and considerable complexity in the operation of transport and dispersion models. However for simplicity we can break down the description into six aspects (the distance/range they work over, how they represent source emissions, how they represent turbulence, the surface properties which are treated, the loss processes which are calculated and finally their output). Traditionally the models can be sub-divided into two types (long and short range); however there are an increasing number of models capable of representing a range of scales. The longer range models operate up to thousands of km and have often been developed for other purposes such as to provide guidance for an emergency response to nuclear accidents. These have been adapted to provide guidance on the long distance transport of FMD virus. They require gridded 3D met data, derived from NWP models and operate on either a lagrangian or eulerian reference frame. The former follows individual trajectories as they progress, the later work on a box principle and transport particles from one model grid box to another. Short range models are valid for distances up to about 20km, although their accuracy at its outer limits will depend on the prevailing atmospheric conditions. These models are typically built around a Gaussian plume dispersion equation. They take a meteorological observation and assume that this remains constant for the length of the calculation, often one hour. In strong wind conditions 10msec -1 the outer limits are reached in just over half an hour. Under light wind conditions of 1 or 2msec -1 the particles will have only travelled 3.6 to 7.2km in one hour. In some instances meteorological data is only available at three hourly or greater intervals. Output from models using this information is likely to be considerably less accurate than those using hourly data. Models differ on how they handle the virus source. Some treat the release as an instantaneous release, others as a continuous release over a defined period (typically 24 hours). In sophisticated models there is a capability to release particles as a function of time of day. The spatial release of particles is another variable with particles released from a point, along a line, over an area or into a volume. The height of release can also be input into some models. Some models do not calculate dispersion when calm winds are recorded. More sophisticated models may spread the particles in all directions for a limited distance. This is an important feature as it is under these conditions that high aerosol concentrations are maintained and as a consequence the risk of infection to animals near the source is particularly high. However, dispersion in such conditions is much less predictable and significant errors are likely. Atmospheric transport and dispersion models represent the turbulent characteristics of the atmosphere. In general these can be divided into two groups; those which only take into consideration the conditions in the atmospheric boundary layer and those which take into account the overall conditions in the atmosphere. Some models derive these conditions internally, whilst others accept them from pre-processor packages including NWP. Additional surface properties can also be included in the models and these can influence the representation of turbulence and airflow. For example the surface type and roughness can be included. The airflow over a city can increase the turbulence and as a result reduce the down-wind concentration of particles. Some models can include topography at this stage, but care has to be taken to ensure that this is not included more than once in the model. FMD particles once airborne are subject to loss, either physical or biological. Larger particles (10µm or greater) will, in the absence of turbulence, be deposited on to the ground by gravity within minutes. Smaller particle may remain airborne for many hours. Some models take this into account either by 229

4 assuming a standard or known size distribution. The formation of particles into water droplets or their capture by existing droplets increase their chances of removal from the atmosphere by wet deposition, although this capture process is likely to take some hours to reduce concentrations significantly. Biologically the virus may become inactivated if the RH of the air falls below 60% or the ph of the virus-containing aerosol particles water vapour becomes acidic or alkaline. The output from the models can also vary considerably. Point or gridded data may be provided, with the size of the grid being selected by the modeller. For long distance models this may be in the order of 10km boxes and for shorter range be typically of the order of 1km or even less. The values output may be expressed as a maximum concentration for a period as short as one hour up to the entire emission period. Similarly mean concentrations over a range of times can be output. Dry and wet deposition maps and concentration fluctuations can be produced. In the future it may be possible to produce outputs of probability of infection for specific locations and numbers and type of animal. Whilst this is possible in theory there are too many uncertainties in the overall process to make this available at this time. A more realistic target would be to estimate animal virus challenge time (the time livestock at a given location are exposed to virus above a threshold concentration). Accurate reflection of conditions experienced in the field e.g. virus release, its behaviour once airborne, its interaction with its surrounds. Correct choice of model. Model treatment of virus released during periods of calm winds. Understanding the link between airborne concentration of virus and infection. Production of appropriate output which mimics how animals become infected (time average conc. or instantaneous dosages). Discussion From the above it is clear that there are many variables in the disease cycle and this is compounded by the addition of others introduced at the modelling stage. Model output can look extremely convincing, especially if it is presented in GIS format. However decisions based on this alone may be severely flawed. For example if the model indicates a region, let us say in a sector 30 0 to 90 0 and out to a range of 5km where a pre-determined threshold virus concentration has been calculated it would be easy to identify those premises which were at risk and as a result introduce the appropriate control measures. However if the virus emission was underestimated by a factor of ten (possible) then the area at risk, assuming that the rest of the modelling process was without error (very unrealistic) could extended out to tens of km. The resultant containment strategy may be entirely different. In view of the above it is essential to minimise or at least to quantify the individual errors through a detailed research and education campaign combined with validation against past FMD outbreaks. This is the overall rationale for a three year programme of work sponsored by Defra. In addition a thorough practical knowledge must be developed of the internal workings and sensitivities of the model. It is highly recommended that these activities should be conducted in the absence of an FMD outbreak. The resultant improved prediction models should be tested against a wide range of FMD outbreaks. It has been shown in the literature (see earlier) that individual models perform well in given particular conditions. However it is unwise, at this time, to assume that they will perform equally as well under all conditions. Conclusions Airborne transmission of FMD is complex. FMD airborne prediction models can currently provide useful advice, but their input, internal formulation and output must be handled with considerable care if accurate advice concerning disease spread is required. There is considerable room for improvement before a definitive FMD transmission model is available. This may take 10 or more years to achieve. Recommendations Work should be continued on determining each of the areas of uncertainty identified in this report and their inclusion in decision making tools. When model output is supplied to those responsible for disease control and eradication it should be accompanied by a verbal or written briefing concerning the confidence of the output. 230

5 Meteorologists should include local flows in their transport and dispersion models at the earliest time possible. It is recognised that this requirement may be some years away. Further interdisciplinary collaboration between veterinarians, virologists, meteorologists, epidemiologists is required to fully resolve the contribution played in any outbreak of airborne transmission of FMD. Acknowledgements The authors readily acknowledge the part played by a range of contributors. These include DEFRA for funding the work and colleagues at the Institute for Animal Health and the Met Office, UK. References Alexandersen, S. and Donaldson, A.I Further studies to quantify the dose of natural aerosols of foot-and-mouth disease virus for pigs. Epidem. Infect.; 128: Alexandersen, S., Zhang, Z., Reid, S.M., Hutchings, G.H. & Donaldson, A.I Quantities of infectious virus and viral RNA recovered from sheep and cattle experimentally infected with foot-andmouth disease virus O UK J. Gen. Virol. 83: Alexandersen, S., Zhang, Z., Donaldson, A.I. & Garland A.J.M The pathogenesis and diagnosis of Foot-and-Mouth disease. J. Comp. Path., 129: Alexandersen, S., Kitching, R.P., Mansley, L.M. & Donaldson, A.I. 2003b. Clinical and laboratory investigations of five outbreaks of foot-and-mouth disease during the 2001 epidemic in the United Kingdom. Vet. Rec. 152: Donaldson, A.I., Gloster, J., Harvey, L.D. & Deans, D. H Use of prediction models to forecast and analyse airborne spread during the foot-and-mouth disease outbreaks in Brittany, Jersey and the Isle of Wight in Vet. Rec., 110: Gloster, J., Champion, H.J., SØrensen, J.H., Mikkelsen, T., Ryall, D.B., Astrup, P., Alexandersen, S. & Donaldson, A.I Airborne transmission of foot-and-mouth disease virus from Burnside Farm, Heddon-on-the-Wall, Northumberland during the 2001 UK epidemic. Vet. Rec., 152: Hugh-Jones, M.E. & Wright, P.B Studies on the foot-and-mouth disease epidemic, the relation of weather to spread of disease. J. Hyg. (Camb.), 68: Mikkelsen, T., Alexandersen, S., Astrup, P.,Donaldson, A.I., Dunkerley, F.N., Gloster, J., Champion, H.J., SØrensen, J. & Thykier-Nielsen, S Investigation of airborne foot-andmouth disease virus transmission during low-wind conditions in the early phase of the UK Epidem. Atmos. Chem. and Phys Discus., 3: Morris, R.S., Sanson, R.S., Stern, M.W., Stevenson, M. & Wilesmith, J.W Decisionsupport tools for foot-and-mouth disease control. Rev. sci. tech. Off. Int. Epiz. 21(3): Sanson, R.L., Morris, R.S. & Stern, M.W EpiMAN-FMD: a decision support system for managing epidemics of vesicular disease. Review Scientifique et Technique de l'o.i.e. 18(3): Sellers, R.F. & Foreman, A.J The Hampshire outbreak of Foot-and-Mouth disease (1967). J. Hyg. (Camb.), 7: Sørensen, J.H Sensitivity of the DERMA Long-Range Dispersion Model to Meteorological Input and Diffusion Parameters. Atmos. Environ., 32: Sørensen, J.H., Mackay, D.K.J., Jensen, C.Ø. & Donaldson, A. I An integrated model to predict the atmospheric spread of foot-and-mouth disease virus. Epidemiol. Infect., 124: Sørensen, J.H., Jensen, C.Ø., Mikkelsen, T., Mackay D.K.J. & Donaldson A.I Modelling the atmospheric spread of foot-and-mouth disease virus for emergency preparedness. Phys. Chem. Earth 26: