🗺️🦠 POI Agent-Based Models (ABMs): Precision Epidemiology in Dynamic Environments

Agent-Based Models (ABMs) parameterized by Points of Interest (POIs) represent a sophisticated class of stochastic epidemiological models that move beyond traditional homogeneous mixing assumptions. By integrating dynamic location data and individual behavioral characteristics, these models provide high-resolution insights into disease spread influenced by spatial and social heterogeneity. As a result, they are invaluable tools for evaluating targeted non-pharmaceutical interventions (NPIs) in complex urban environments.

────────────────────────────────────────────

🧬 1. Compartmental Structure

POI ABMs operate by tracking the state of individual agents (N_total) as they move across distinct spatial nodes, referred to as Points of Interest, within a synthetic environment. While population-level summaries can sometimes be approximated using multi-group deterministic frameworks such as meta-population models, the defining feature of POI ABMs is that each agent is simulated explicitly.

Agents typically progress through a classical SEIR (Susceptible–Exposed–Infectious–Recovered) infection structure.

Flow of Agents (Individuals):

• Susceptible (S): Individuals vulnerable to infection. An agent transitions from S to E upon exposure at a POI, depending on the instantaneous local transmission risk.
• Exposed (E): Infected individuals who are not yet infectious. The transition from E to I is governed by the incubation rate σ.
• Infectious (I): Individuals capable of transmitting the disease. Transition to recovery occurs at rate γ.
• Recovered (R): Individuals who have cleared the infection and are immune or otherwise removed from the transmission process.

A defining feature of POI ABMs is that the transition rate from S to E is heterogeneous and depends on the agent’s current POI and the density of infectious agents present at that location.

────────────────────────────────────────────

📐 2. Mathematical Formulation

Because POI models are inherently stochastic and discrete, they do not rely on a single global system of Ordinary Differential Equations for their primary structure. Transmission is governed at the individual level through the instantaneous force of infection applied to each susceptible agent.

When a susceptible agent a visits a POI k, the force of infection λₐ(t) is defined as:

λₐ(t) = Σₖ βₖ · f(Iₖ, Nₖ)

A common proportional-mixing assumption is:

λₐ(t) ∝ Σₖ βₖ · (Iₖ(t) / Nₖ(t))

State Transitions:

• Transition S → E: governed by λₐ(t), which is location- and time-dependent
• Transition E → I: occurs at rate σ
• Transition I → R: occurs at rate γ

────────────────────────────────────────────

📋 Parameter Definitions

Parameter	Definition	Role in the Model
βₖ	Location-specific transmission coefficient	Captures transmission probability scaled by contact intensity at POI k
Iₖ	Number of infectious agents at POI k	Represents the local infectious reservoir
Nₖ	Total number of agents at POI k	Determines local population size or contact capacity
σ	Latent period rate (1 / T_inc)	Controls progression from exposed to infectious
γ	Recovery or removal rate (1 / T_inf)	Governs duration of infectiousness

────────────────────────────────────────────

📊 3. Parameter Ranges (General Viral Disease)

POI ABMs require highly localized parameterization. Clinical parameters are informed by epidemiological data, while βₖ reflects behavioral and environmental characteristics of individual POIs.

Parameter	Typical Range	Unit	Interpretation
Recovery Rate (γ)	0.07 – 0.14	day⁻¹	Infectious period of approximately 7–14 days
Latency Rate (σ)	0.14 – 0.25	day⁻¹	Incubation period of approximately 4–7 days
Location Factor (βₖ)	0.05 – 2.0	Dimensionless or contact-scaled	Low values for households; high values for dense venues such as schools or transit
Basic Reproduction Number (R₀)	1.5 – 3.5	Dimensionless	Emergent quantity derived implicitly from agent interactions

────────────────────────────────────────────

🚧 4. Applicability and Limitations

Applicability – When to Use POI ABMs:

1. Evaluating targeted non-pharmaceutical interventions such as selective venue closures or capacity limits
2. Modeling superspreading events driven by crowded or episodic locations
3. Integrating real-world mobility and contact data
4. Optimizing targeted testing, vaccination, or surveillance strategies

Key Assumptions and Limitations:

1. Data Intensity: Requires detailed mobility, visitation, and contact data
2. Computational Cost: High computational burden for large populations
3. Contact Definition: Transmission-relevant contacts are often simplified
4. Behavioral Modeling: Decision rules may introduce structural uncertainty

────────────────────────────────────────────

📚 References