Closed Session of the EuFMD Research Group Kranska Gora, Slovenia 23 th 25 th September 2009 FMD epidemic models: update on recently published/developed models Antonello Di Nardo Institute for Animal Health, Pirbright UK
Why model FMD? Strategic (in peacetime ) Investigate FMD epidemiology and behaviour under local conditions Investigate the effectiveness and efficiency of alternative mitigation strategies and of surveillance and monitoring programmes Investigate the likely cost of outbreaks of different characteristics Provide a resource for training awareness Tactical (during outbreak) Provide information about the likely effectiveness and efficiency of mitigation strategies under existing outbreak conditions Provide information about resource allocation strategies Provide limited scale-dependent forecasting (???)
Model design Deterministic model Values of input parameters are fixed Results do not take account of random variation (i.e. variability) Same result for every iteration Stochastic model Describes processes/events subject to random variation Parameters values are used in form of probability distribution Variability leads to outcome with probability Different results for every iteration (Monte-Carlo sampling)
Model design Compartmental models Consider individual diversity within population Units can exist in mutually exclusive states Modelling the progress of epidemic in compartment/state Transition between states as events occur Susceptible β Infectious ν Recovered β Infection rate ν Rate of recovery
Model design Spatial factors Non-spatial Pseudo-spatial (grids/lattice) Spatially explicit (distance/points/regions)
Baseline epidemic model
Model input (Matthews and Woolhouse, 2005)
Model building Objectives clearly specified Details of data and knowledge of parameters are needed Confirm that type of output designed is generated Test the sensitivity of the model input variables stability Establish if the model behaves like the biological system that it is designated to mirror assessed against data not used in its construction precision specified in advance Assessment of a theory Generality, simplicity, precision, refutedness, testedness (Thrusfield, 2007)
Models used in 2001 Imperial model (Ferguson et al., 2001) Cambridge-Edinburgh model (Keeling et al., 2001) InterSpread model (Morris et al., 2001) Imperial Model Cambridge-Edinburgh Model InterSpread Model Type Deterministic (differential equations) Stochastic (microsimulation) Stochastic (microsimulation) Parameters Few Few Many Spatially explicit Different species Airborne spread Different transmission mechanism Intra-herd transmission dynamics Logistic resources Vaccination strategies
Current FMD model projects AusSpread-DAFF model (Garner and Backett, 2005) North American Animal Disease Spread (NAADS) model (Harvey et al., 2006) InterSpread Plus model (Stevenson et al., 2006) Exodis model (Risk-Solutions, 2006) Automata model (Ward et al., 2007) Multiscale Epidemiological/Economic Simulation (MESA) model (Kostova, 2007) Davis Animal Disease Simulation (DADS) model (Carpenter, 2007)
AusSpread model Stochastic (Monte-Carlo based) State-transition SLIR (micro-simulation) developed from a Marchov-chain Spatially explicit (real farm boundaries or point location data) developed and operates within a GIS environment (MapBasic /MapInfo ) Farm-based and specie-specific (7 default farm types and wildlife) Simulating disease spread Saleyard spread Airborne spread Farm-to-farm spread Modelling mitigation based on AUSVET-PLAN for FMD Quarantine and movement restrictions Stamping-out Surveillance (active/passive) Tracing Pre-emptive slaughter Vaccination Resources
NAADSM model Stochastic (Monte-Carlo based) State-transition (micro-simulation) Spatially explicit Herd-based and specie-specific Simulating disease spread Airborne spread Farm-to-farm spread (direct/indirect contact) Influenced by relative locations and distances Modelling mitigation Quarantine and movement restrictions Stamping-out Surveillance (active/passive) Tracing Pre-emptive slaughter Vaccination Resources (priorities and costs)
NAADSM model 3 experimental versions Cheyenne quarantined units are allowed to be recipients of direct contact Laramie destroyed unites are allowed to be recipients of indirect contact developed to simulate FMD spread among livestock premises by wildlife Riverton units that have progressed to naturally immune state are not included in the destruction programme developed to explore alternative destruction strategies for HPAI
InterSpread Plus model Stochastic (Monte-Carlo based) State-transition (micro-simulation) Spatially explicit (polygonal units or point location data) Farm-based and specie-specific Simulating disease spread Movement spread (distance dependent) Local spread (distance dependent) Airborne spread Modelling mitigation Movement restrictions/standstill Depopulation Surveillance intensity (active/passive) Tracing (backward and forward) Pre-emptive slaughter Vaccination Resources
Exodis model Stochastic (Monte-Carlo based) State-transition (micro-simulation) Spatially explicit (Kernel density function) Farm-based and specie-specific Simulating disease spread Airborne spread Intra-herd spread Farm-to-farm spread Modelling mitigation Quarantine and movement restrictions Stamping-out Surveillance (active/passive) Tracing Pre-emptive slaughter Vaccination Resources
MESA model Stochastic State-transition SLIR (micro-simulation) Spatially explicit Individual-based and specie-specific Simulating disease spread Saleyard spread Airborne spread Farm-to-farm spread (direct/indirect contact) Modelling mitigation based on USDA FMD Response Plan Quarantine and movement restrictions Stamping-out Surveillance (active/passive) Tracing (backward and forward) Minimal time to obtain freedom-from-disease following the last case (42 days)
DADS model Stochastic (Monte-Carlo based) State-transition SLIR (micro-simulation) Spatially explicit Herd-based and specie-specific Simulating disease spread Airborne spread Farm-to-farm spread (direct/indirect contact) Modelling mitigation Quarantine and movement restrictions Stamping-out Surveillance (active/passive) Tracing Pre-emptive slaughter Vaccination Resources
Automata model Stochastic (joint probability threshold) State-transition SLIR (micro-simulation) Spatially explicit (geographical-automata) Herd-based and specie-specific Simulating disease spread Probabilistic inter-herd-distance dependent Product of neighbouring density probability (based on ecological site capacity) Repetitive application of transmission rules (complex spatial behaviour) No modelling mitigation Developed to explore FMD spread within/between domestic and wildlife populations
Models validation current projects Quadrilateral (QUADs) country model comparison project (2005) AusSpread NAADSM InterSpread Plus 1. Comparison of conceptual models (formal review of model description) 2. Comparison of 6 scenarios assessing spread mechanism 3. Comparison of 5 scenarios control measures Statistically significant difference among number of IPs and temporal/spatial spread prediction by the 3 models Difference in the size of outbreak areas EpiLab (2007) Imperial, Cambridge-Edinburgh, InterSpread Plus, DADS DISCONVAC (2009) InterSpread Plus, DADS, NAADSM, others
Models problems and issues Complexity Lack of transparency Impossible to be reproduced Robustness of results difficult to assess Significant number of unknown parameters Still based on expert opinion (< accuracy) Lack of intra-herd/within-herd dynamics modelling Exiguous experimental data Local-scale based Providing regional guidance, but never at tactical level Availability and quality of data Accurate and extensive input data Validation Replication of one outbreak parameterising with data from another Endemic models??? Wildlife??? Multiple FMD serotype incursions???
Models uncertainty and caution All the model are wrong Model is a value-added in Decision Support System Useful for resource planning/allocation Useful for exploring what if scenario Are not good at predicting random behaviour Models only reflect the biases of the modeller and lack of objectivity Be careful of models posing as truth (lack of validation) Keep it simple, keep it flexible, make it transparent
Acknowledges Keith Sumption, Nadia Rumich (EuFMD) David Paton, David Schley (IAH) Michael Ward (University of Sydney) Michael Thrusfield (University of Edinburgh)