Comparison of Private vs. Public Interventions for
Controlling Influenza Epidemics
Achla Marathe
Virginia Bioinformatics Institute and
Dept. of Agricultural and Applied Economics
Joint work with Chris Barrett, Jiangzhuo Chen, Stephen
Eubank, Bryan Lewis, Yifei Ma and Madhav Marathe
Acknowledgment
This work has been funded in part by the
following grants: NIH-MIDAS, NIH-R01,
DoD CNIMS, NSF-ICES and NSF-NetSe.
Introduction
• Goal:
– Design effective intervention strategies to
control the spread of Influenza.
• Challenges:
– Lack of compliance for public health directives.
– Lack of accurate knowledge about the global
prevalence and the severity of the disease.
Introduction
• This research considers two sets of interventions
strategies, private and public.
• Evaluates the performance of each intervention strategy
under a variety of scenarios through agent based
simulations.
• Uses a synthetic social network of a large urban city as the
area of study.
• Offers guidance to public health policy makers.
Standard Evaluation Measures
• Effectiveness of intervention:
– Reduce attack rate/peak
– Delay outbreak/peak
• Cost
– Number of antivirals or vaccines consumed. These are often
available in limited supply
– Other costs: e.g. administration of a mass vaccination campaign
(not considered here)
Private Strategy
• Individuals observe the health state of distance-1 (or immediate)
contacts in the social network.
• After a threshold number of contacts become sick, individual
intervenes with an antiviral or a vaccine.
A
Distance 1 neighbors of A
A
Infected neighbors of A
Public Strategy
• Block intervention: take action on all people
residing in a census block group if an outbreak is
observed in the block group
• School intervention: take action on all students
in a school if an outbreak is observed in the school
Private vs. Public Intervention Strategies
Public
Private
• Public health officials use
global incidence data
• Low accuracy on prevalence
• Interventions are imposed
top-down on individuals.
• Compliance is low
• Delay in implementation
• Individuals observe the
health state of local contacts.
• High accuracy on prevalence
• Self motivated to intervene
when encounter sickness.
• Compliance is high
• No delay
Experimental Settings
• Disease propagation through social contact network
on a synthetic population
– Miami network: 2 million people, 100 million people-people
contacts
• Assume unlimited supply of antiviral and vaccine
– One course of antiviral is effective immediately for 10 days:
reduce incoming transmissibility by 80% and outgoing by
87%
– Vaccine is effective after 2 weeks but remains effective for
the season. Vaccine efficacy is 100%.
• Simulation tool used: Indemics
• Indemics is an interactive epidemic simulation and
modeling environment that was developed in our
group.
Within Host Disease Model
•
•
•
•
Individuals move
through disease
states
Incubation period: mean 1.9 days
Infectious period: mean 4.1 days
Symptomatic rate: 0.67
Asymptomatic are 50% less likely
to transmit the disease.
Experiment: A Factorial Design
• 3 different intervention strategies: D1, Block, School
• 2 flu models: 20% (moderate) and 40% (catastrophic)
attack rate
• Diagnosis rate: 2 values 1 and 0.3
• 2 threshold values for taking actions: .01 and .05
– Fraction of direct contacts found to be sick: D1 intervention
– Fraction of block group (school) subpopulation found to be sick:
block (school) intervention
• 2 compliance rates: 1 and 0.5.
• 2 pharmaceutical actions: Antiviral and Vaccination (VAX)
• Delay in implementing interventions: 2 values for Block and
School, 1 day and 5 days; no delay for D1
• 2 x 2 x 2 x 2 x 2 x ( 2 + 2 + 1) = 160 cells
• 25 replicates per cell (4000 simulation runs!)
Experimental Results
Attack Rate: Moderate Flu with Various Interventions
Intervention Coverage: Moderate Flu with Various
Interventions
Attack Rate: Catastrophic Flu with Various
Interventions
Intervention Coverage: Catastrophic Flu with Various
Interventions
Experiment Results: Effectiveness of Actions
• Antiviral is very effective under D1; almost no effect
under two public strategies
• No efficacy delay; protect people from sick contacts
immediately
• Efficacy expires after 10 days; hard to avoid transmissions from
farther-away nodes in the neighborhood
• If only antiviral is available, should motivate people to take
antiviral by themselves
• Vaccine performs best under Block, worst under
School
• Two weeks efficacy delay; sick contacts become less relevant
• Form larger “ring” around “hot-spots”
• Large consumption under Block; little consumed under school
(school students <25% of whole population)
• If sufficient vaccines are available, should apply Block
intervention strategy
Experiment Results
• Compliance: limited impact on attack rate; almost linearly
determine drug consumption
– Higher compliance  more consumption
– Double consumption !  twice reduction in attack rate
• Implementation delay: little difference between 1 day or 5
days
• Nothing is useful under low diagnosis + high threshold
– Campaign to raise concern on epidemic and early action
– Increase diagnosis accuracy and enhance public health surveillance
Antiviral or Vaccine
• D1 intervention is effective with antiviral; Block
intervention is effective with vaccine
• School intervention consumes little: may be most
cost-effective when drugs are available in limited
quantity
Closer look at an
interesting setting…
(catastrophic flu, high diagnosis rate, low
threshold, only vaccines available)
Comparative Performance under Vaccination
Summary
• An interesting comparison study
– Individual behavioral vs. public health level interventions
– Use simulations to guide policy
• Unique capability to run such complex, realistic
studies
– No other tool can apply interventions based on social network
based relationships because it requires
• Detailed social network
• Network relationship based dynamic intervention capability
• An efficient simulation environment
Summary
• Vaccine intervention: Block strategy performs better than
D1. Given the 2 week delay in vaccine efficacy, block
strategy is able to form a larger ring around hot-spots. The
immediate contacts become less relevant. However a lot
more vaccines are needed.
• If the transmissibility is high and vaccines are available in
abundant supply, the Block strategy is likely to be the best
choice.
• Antiviral Intervention: If antivirals are available in limited
supply, it may be best to distribute them to people over the
counter to make them easily accessible.
Thanks!
Indemics: Interactive Simulation
• Indemics: Interactive Epidemic Simulation and Modeling Environment
• Data Models:
– Relational Data about individuals (P)
– Social Contact Network (N)
– Transmission Network/Dendrogram (D)
• Queries on a single data type
– (P) Find all school-ages in area <x>
– (N) Find all neighbors of person <a>
– (D) Find all infected persons at day <t>
• Queries across multiple data types
– Count number of infected persons in zip code 24060 (Blacksburg, VA)
– Find all infectious students on day 20 in Blacksburg high school and their
family members
Dynamic Queries and Interventions
• Users interact with the system using well-defined languages
– Indemics commands: count infected persons : group = seniors,
infected day = between 20 and 22
– SQL statements: select * from social_network SN and infections INF
where SN.pid_a = INF.transmitee_pid and time = 20 (find all
neighbors of all infections at day 20)
– Libraries of queries can be pre-defined by expert users
• Indemics Interventions
– apply interventions: type = antiviral, duration = 10, group = school
age, infected_day = between 24 and 30
– apply interventions: type=work closure, duration = 5, group =
adults, infected day = between 20 and 21; type = school closure,
duration = 5, group = school age