BioWar: Scalable Agent-Based Model of Bioattacks

1
BioWar Scalable Agent-Based Model of Bioattacks
Project Investigators Kathleen M. Carley
(Project Director and PI) CMU, ISRI, CASOS
Douglas Fridsma University of Pittsburgh,
BMSI Elizabeth Casman CMU, EPP
Students Alex Yahja CMU, ISRI, CASOS, Tiffany
Tummino CMU, EPP Post Doc Li-Chiou Chen CMU,
ISRI, CASOS Programming Staff Boris Kaminsky --
CMU, Demian Nave -- PSC, Neal Altman -- CMU
Potential Usage
Estimating the Impact of Detection
Description Tool for evaluation of response
policies, data efficacy, attack severity, and
detection tools relating to weaponized biological
attacks against the background of
naturally-occurring diseases
Process of aligning BioWar and population-based
Incubation-Prodromal-Fulminant (IPF) model
Generate possible attacks to layer on existing
data or complete simulation of all data Examine
effectiveness/costliness of response
policies Pre-evaluate possible data sets for
detection Pre-evaluate whether more detailed data
collection might be useful Training for intel
officers and health workers about what an attack
might look like
Anthrax attack 19,000 exposed (pop.
170,000) Three (3) kilograms of anthrax,
released in an explosion with an efficiency of
0.05, giving a 150 gram effective mass Time of
 attack 4 PM, Place Stadium in Hampton, VA
Wind speed is low, 0.63 meters/second The
dosage is inversely proportional to the wind
speed The number of people saved is calculated as
0.55 (total_fatalities - fatalities_up_to_that_d
ay) It reflects that the intervention brings the
death_rate down from 0.85 to 0.45. 0.85 is the
untreated death rate published in various papers
including Meselson's Sverdlovsk paper 0.45 is
the quick treatment death rate corresponds to the
US mail case Estimate saved of those likely to
die with general diagnosis -- given detection on
that day by number of days since attack.
BioWar conceptualization city scale
multi-agent network model of weaponized attacks
Validation Tuning Done
Docking Comparison against another model
Generic Pattern Showing pattern for each generated data stream matches observed patterns
Characteristic Matching Showing for each generated output data stream that it has correct seasonal or daily pattern
Relative Timing of Peaks Showing time between peaks for different data streams matches observed difference
Empirical Pattern Showing pattern for each generated data stream matches empirical pattern best for input streams
Within Bounds Showing for each generated output data stream that the mean of simulated stream falls within min/max of that stream for real data
First moments Showing for each generated output data stream that mean is not statistically different than real data yearly, monthly or daily
Conclusion Detection cannot occur soon enough to
prevent a significant numbers of deaths
BioWar vs. IPF, based on time to death for
Sverdlovsk data

• Approach
• Multi-Agent Network Model
• - Cognitively realistic
• Socially realistic embedded in social,
  knowledge,
• task networks
• - Spatio-temporally realistic
• - Organizational network
• - Communication technologies
• Hybrid of many models disease, spatial, network,
  cost, agent, social, geographical, media, etc.

Simulator Comparison
BioWar EpiSims Measured Response CATS NARAC
Sim. Size City City Mult. City Area Area
Geography US real Stylized Stylized World World
Population US Census Stylized Stylized US Census ?
Sim. Type Agent Agent Agent Exposure Exposure
GUI No Yes Yes Yes Yes
Scaleable 0-100 0-100 0-100 N/A N/A
Climate Yes No No Yes Yes
Transport Net Stylized Yes Stylized GIS GIS
Location Net Yes No No GIS GIS
Social Network Yes Household No No No
Agent Learning Yes No Dormant No No
Threats Bio/Ch Bio Bio Bio/Ch/Nuc Bio/Ch/Nuc
Release Air/Ground Direct Direct Air/Ground ?
BioWar vs. Incubation-Prodromal-Fulminant (IPF)
Model
BioWar vs. IPF, based on time to death for US
Mail data
Advantages BioWar the disease progression model
of anthrax can be reused for different cities and
with different scales the disease duration of
anthrax changes with policy responses calculates
secondary data streams such as OTC purchases and
population responses IPF quick disease model
prototyping Disadvantages BioWar BioWar takes
more time and computer resources IPF simulates
only disease progression cannot calculate the
infected rate based on the given demographics of
a population, geography, weather, the released
mass of an anthrax attack
Disease Model Symptom Viral Load Indiv. SIR CHAS ?
Attack Disease 4 1 1 3 ?
Environ. Dis. 58 None None None ?
Simult. Dis. Yes No No No ?
Infection SN rand. Viral Load Direct Direct None(?)
Treatment Yes No No No ?
Response Alert Panic Gov. Advice Advice
Exposure Maps No ? ? Yes Yes
Case Data Yes ? ? No No
Insur. Claims Yes ? ? No No
OTC Drug Pur. Yes ? ? No No
Inf. Over Time Yes ? ? No(?) No(?)