Planning Epidemic Interventions with EpiPolicy

Literature Title

Planning Epidemic Interventions with EpiPolicy

Literature Information

• Subject Area: Public Health Policy and Infectious Disease Modeling
• Keywords: Interactive Modeling, Epidemic Simulation, Policy Exploration, Public Health Decision-Making, Monte Carlo Tree Search, Intervention Planning, Model Management System, Deterministic Model, Population Dynamics, Cost Optimization

Research Background and Problem

• Issues and Challenges:

  • In the context of infectious disease outbreaks, public health teams must continuously adjust model parameters, modify intervention strategies, and optimize intervention plans to reduce disease spread while balancing economic costs.
  • Current epidemic simulation tools face significant limitations in conducting "what-if analyses," as models are complex and policy changes are difficult, slowing down policy optimization processes.
  • Modeling new diseases often requires starting from scratch, making it difficult to reuse existing model code. Intervention strategies and model parameters are typically tightly coupled.
  • Decision-making teams use multiple independent tools (e.g., modeling tools, spreadsheets, visualization dashboards) to manage input parameters and simulation results, lacking a unified Model Management System (MMS).
  • Parameter setting and model selection processes are time-consuming and prone to errors, affecting the credibility of policy effectiveness.

• Research Significance:

  • With the continuous emergence of new infectious diseases, formulating efficient and economically feasible public health policies is critical for reducing disease burden and optimizing resource allocation.

• Research Motivation and Related Work:

  • While current mainstream epidemic simulation tools (e.g., STEM, CMS, and GLEaMviz) offer rich modeling functionalities, they perform poorly in key tasks such as multi-policy comparison, intervention plan optimization, and cost analysis.
  • This study proposes EpiPolicy, an integrated collaborative platform for decision-making teams that simplifies the modeling process and supports automated intervention strategy generation and optimization.

Solution

• Core Methods and Tools:

  • Introduced a simulation and policy optimization tool named EpiPolicy, specifically designed for epidemic control.
  • Core Design Principles:
    • High-Level Abstraction: Introduced advanced concepts such as groups (population groups), locales (administrative regions), and facilities to intuitively represent disease transmission and interventions.
    • Separation of Concerns: Clearly separated tasks such as modeling, intervention definition, and scheduling, allowing team members to focus on their respective domain tasks.
    • Minimal but Sufficient API: Provided a Python programming interface, enabling users to define effect() and cost() functions to implement intervention effects and costs.

• Implementation Steps:

  • Step-by-step model definition through a graphical user interface (GUI), including disease transmission models, baseline population characteristics, intervention settings, and schedule creation.
  • Automated exploration and generation of low-cost intervention plans using a Monte Carlo Tree Search (MCTS) optimizer.
  • Post-simulation, results are presented via interactive visualization tools, supporting the comparison of different policies.

• Key Technologies:

  • Deterministic Modeling: Used a multi-patch framework for large-scale population disease transmission modeling.
  • Matrix Storage and Access: Represented scenario elements (e.g., population states, facility configurations, mobility patterns) using multi-dimensional matrices.
  • MCTS Optimization Search: Efficiently identified cost-effective intervention plans in a vast search space.

Research Outcomes

• Specific Achievements:

  • Developed the EpiPolicy tool, which supports local-specific policy and intervention optimization, simplifying the modeling and planning process for policy-making teams.
  • Provided a flexible API interface for customizing complex interventions and cost functions.
  • Enabled rapid simulation of epidemic dynamics and interactive exploration of different policy impacts.

• Advantages:

  • Easier parameter management and construction/comparison of multiple intervention plans compared to existing tools.
  • Facilitated efficient collaboration among teams, reducing dependency on code-level intervention logic.
  • Offered a unified model management system to track all simulations and their results.

• Experimental or Evaluation Results:

  • Successfully validated the tool's applicability in various scenarios, such as hospital capacity planning and vaccine allocation strategies.
  • MCTS identified optimized intervention plans in advance, significantly reducing policy trial-and-error time and manual intervention costs.
  • Expert evaluations indicated that EpiPolicy's separation design principle effectively improved the efficiency and quality of policy discussions.

• Limitations and Future Directions:

  • The current interactive model abstraction in EpiPolicy still requires further simplification for certain domain users (e.g., epidemiologists).
  • Future research should focus on methods to automatically convert agent-based models into deterministic models and enhance tools for parameter integration.
  • Accelerate exploration of more efficient reinforcement learning algorithms to improve intervention plan generation and diversity.