Transcript Document
Interpretation of large-scale stochastic
epidemic models
Iain Barrass
Ian Hall and Steve Leach
Health Protection Agency
14 September 2011
Overview
Stochastic model structure
Source of uncertainty
Ensemble output
Epidemic clustering
Consequences of reporting choice
Interpretation and visualization
Stochastic model structure
Infection
S
I
R
Stochastic transition
or event-driven simulation
Spatial meta-population model
Without interventions, R0~1.6
Pneumonic plague: model
Early
symptomatic
Susceptible
Latent
Removed
Late
symptomatic
Contact tracing
Post-exposure prophylaxis
Isolation
Generic antimicrobial treatment
Specific antimicrobial treatment
Seeding: aerosol release
Variability in
release location (including height)
wind direction
infected individuals within patches
Seeding: disease importation
Decoupled global and UK models – global model acts as a
seed for the UK model.
Variability in importation profile and importee destination.
Pneumonic plague: results
Deaths from “large” release with intervention strategies
Earlier commencement of prophylaxis reduces death count
Clearly interpretable
“Pandemic influenza” spatial spread
Initial seed of 10 cases in resident population of one patch
Solution measures
Final attack size (whole population or typed)
New cases over time
Individuals over time in a state
Duration of “high activity”
Peak of the attack
Consideration of morbidity and mortality (economic cost)
Clustered epidemic curves
50% of epidemics fall within three clusters
Model selection
Model A
Model B
Single wave epidemic
Visualization systems
Summary
High complexity models (or large populations) lead to
event-based simulation with large ensembles
Increasing model structure can increase observation
variability
Consideration of seeding variability and parameter
sensitivity complicates interpretation
Some measures are not very sensitive to model
complexity
Choice of measure may influence model choice through
desire for clarity of interpretation
Highly complex models benefit from specialised
visualization approaches
Acknowledgments
MRA team – in particular Joe Egan and Tim Cairnes
Funding: Department of Health (England), Home Office, EU
FP7 project FLUMODCONT, EPSRC network CompuSteer