Transcript ppt

Coevolution of Epidemics, Social
Networks, and Individual Behavior:
A Case Study
Jiangzhuo Chen
Joint work with Achla Marathe, and Madhav Marathe
NDSSL Technical Report 10-042
2010 International Conference on Social Computing,
Behavioral Modeling, and Prediction (SBP10)
March 31st, 2010
Network Dynamics & Simulation Science Laboratory
Our group members (NDSSL)
Work funded in part by NIGMS, NIH MIDAS program, CDC, Center of
Excellence in Medical Informatics, DTRA CNIMS, NSF, NeTs, NECO and OCI
(Peta-apps) program, VT Foundation.
Network Dynamics & Simulation Science Laboratory
Talk Outline
• Motivation for the case study.
• Major contributions.
• Background:
– propagation of infectious disease on social contact
networks;
– intervention measures.
• Simulation results.
Network Dynamics & Simulation Science Laboratory
Co-evolution of Behavior, Epidemics
and Social Networks
Policy questions:
• Is there an optimum AV
allocation strategy between
the market-based availability
and public distribution via
clinics that minimizes the
attack rate and recovers the
cost of AV through markets?
•Does “contextual behavioral
adaptation” by individuals
help in controlling the
epidemic?
Price/
Inventory
Disease
Dynamics/
Prevalence
Transmission
Network
Network Dynamics & Simulation Science Laboratory
Demand/Behavior
Susceptibility
Our Contributions
We have developed a
methodology and an
associated modeling
environment to study
the co-evolution of
individual behavior,
disease dynamics and
social networks.
Network Dynamics & Simulation Science Laboratory
Our Contributions (cont.)
We find that there is an optimal AV allocation strategy
between the market and clinics that minimizes the attack rate
and recovers the cost of AV through markets.
We find that markets are inherently unfair in their allocation
scheme but price regulations and other incentives can control
this.
The market based system combined with behavioral
adaptation reduces the epidemic peak as well as delays it.
Network Dynamics & Simulation Science Laboratory
Disease Spread in a Social Network
• Within-host disease model: SEIR
– State transitions are probabilistic and timed.
• Between-host disease model: transmission
occurs along edges of a social contact
network
– People + Locations => Contacts.
– Transmissions are probabilistic.
Network Dynamics & Simulation Science Laboratory
Interventions
• Pharmaceutical interventions: vaccination or
antiviral changes an individual’s role in the
transmission chain
– Lower susceptibility or infectiousness
• Non-pharmaceutical interventions: social
distancing measures change people activities and
hence the social network
– Generic social distancing, school closure, isolation, etc.
Network Dynamics & Simulation Science Laboratory
Interventions: Multiple Models
• Infection reporting/diagnosing model: infected
people show symptom with a prob.; symptomatic
people are diagnosed with a prob.
• Antiviral distribution models:
– Public sector distribution: targeted to diagnosed people,
free.
– Market distribution: mostly prophylaxis, not free.
• Behavioral models for self-interventions:
– Household isolation: when one member is diagnosed.
– Household AV demand: function of local variables (budget)
as well as global variables (price, disease prevalence).
Network Dynamics & Simulation Science Laboratory
Case Study: Simulation Setup
• New River Valley population: 150K
• Total 15K AV supply
– allocated between clinics and market
– from all-to-clinic to all-to-market.
•
•
•
•
20 index cases are randomly picked.
45% attack rate without any intervention.
The price of the AV kit can vary between $50-$150.
Total household budget for the AV is 1% of the
income.
Network Dynamics & Simulation Science Laboratory
Experiment Results
• Attack rate reaches
minimum at 40% clinic
allocation.
• Application of clinic
allocation is upper
bounded by attack rate.
No need to allocate more
to clinics.
• Extra AV can go to
markets. Revenue covers
cost.
Network Dynamics & Simulation Science Laboratory
Effect of Prevalence, Social Distancing
and Individual Behavior
• Dependence of demand on
disease prevalence helps to
delay the peak of the
epidemic by about 30 days.
• Isolation is able to reduce
the peak infection by 30%.
• A combination of market,
public distribution and
contextual behavioral
adaptation is indeed likely to
be more “effective” in
reaching the masses and
controlling the epidemic.
Network Dynamics & Simulation Science Laboratory
Fairness: Who Buys and Who Doesn’t?
• Only top 30 percentile
income households are able
to buy AV through the
market when demand is
sensitive to prevalence.
• Poorer households do not
buy because early in the
epidemic the prevalence is
low; later in the epidemic
when the prevalence
increases, the price
increases making it
unaffordable for the poorer
people.
Network Dynamics & Simulation Science Laboratory
Fairness: Impact of Demand Sensitivity
• Suppose demand is
independent of
prevalence: each
household has positive
demand with a fixed
probability every day,
only subject to budget
constraint.
• AV price rises gradually.
More people are able to
purchase it.
Network Dynamics & Simulation Science Laboratory
Conclusion
• We have developed a system that can study the coevolution of individual behavior, disease dynamics
and social networks. It can study economic and
social effect of an epidemic.
• There exists an optimal strategy for allocating
antiviral between the market and clinics to
minimize the attack rate and to recover cost of
antiviral through the market.
• The market based system combined with behavioral
adaptation lowers and delays the epidemic peak.
Network Dynamics & Simulation Science Laboratory