Abstract
This paper reports on the use of possibility theory and agent-based explicit spatiotemporal simulation to compare the effects on each of three real communities given the assumption that a rare disease is carried out of a hypothetical high-containment biological research laboratory sited in that community. The initial event has nonzero possibility, but its probability is not well measurably different from zero. The conditional distributions obtained by making this event an input to the simulation resemble "counterfactual conditionals" that can provide useful information about the relative technical and social desirability of alternative sites even though a conventional risk assessment is not possible.
Introduction: The Challenge of "Black Swan" Events
Traditional risk assessment methodologies typically rely on multiplying the probability of an event by its potential impact. However, for high-containment biological research laboratories, the probability of an accidental release of an exotic pathogen (such as Ebola) is effectively zero. With hundreds of thousands of person-hours logged in U.S. labs without a single community infection, the "signal" of risk is lost within the "noise" of statistical confidence intervals.
To address this, the researchers shifted from probabilistic risk assessment to possibility theory. By assuming a leak occurs (a "counterfactual conditional"), they can ask: If a release happens, which community is most resilient? This approach allows for a comparative analysis of site suitability that traditional mathematics cannot provide.
Methodology: The A-BEST Framework
The study utilized the Agent-Based Explicit Spatial and Temporal (A-BEST) modeling framework. Unlike aggregate models, A-BEST simulates discrete individuals (agents) and their specific interactions across time and space.
Simulation Design
- Three-Phase Daily Cycle: Agents move through Nighttime (home), Daytime (work), and Pastime (service/leisure) phases.
- Two-Scale Network: Interactions are modeled at both the local level (e.g., within a household) and the population level (movement between locations).
- Monte Carlo Approach: 600 simulations (200 per community type) were conducted to account for stochastic variations in movement and transmission.
- Stress-Testing: The model used "worse than worst-case" parameters, including hypothetical sexual transmission of Ebola, to ensure sufficient data for comparative analysis.
Data Integration
The model integrated high-resolution datasets, including 2000 U.S. Census data for demographics, ReferenceUSA and ESRI Business datasets for workplace and household locations, and travel diary surveys to model movement patterns. GIS road networks were used to calculate driving catchments for service centers.
Key Results
The simulation revealed that the primary driver of disease spread was not the total population size, but rather the scale of local healthcare infrastructure. Because the "index case" (the first infected worker) typically seeks medical care, a larger hospital provides more initial contact points with healthcare workers, who then serve as vectors to the broader community.
As shown in the results, the Rural community faced the highest relative risk (3.43 infections per 10,000), as a single infection represents a larger proportion of social links in a smaller network. Conversely, the Urban community exhibited significant resilience; despite having 15 times the population of the other sites, it only saw a marginal increase in absolute cases and the lowest per-capita risk (0.13 per 10,000).
Discussion and Conclusion
The utility of this "possibilistic" framework lies in its ability to generate a risk profile where traditional math fails. By assuming the "impossible" event, researchers can identify structural vulnerabilities—such as the role of hospitals in early transmission—and provide a logical basis for siting critical infrastructure.
While the model uses pessimistic assumptions (e.g., failure of hospital isolation), it serves as a robust tool for comparative resilience analysis. Future work may refine these parameters to include reactive agent behaviors like social distancing, further enhancing the model's predictive value for public health planning.
Summary updated