Modeling Presymptomatic Spread in Epidemics via Mean-Field Games

Sukru Yagiz Olmez, Shubham Aggarwal, Jin-Won Kim, Erik Miehling, Tamer Basar, Matthew West, Prashant G. Mehta

Researchers at the University of Illinois at Urbana-Champaign have applied the mathematical framework of Mean-Field Games (MFG) to address a critical public health challenge: understanding how individual "rational" decisions drive epidemic spread, particularly when infections are "invisible" due to a lack of symptoms.

The Human Factor in Epidemiology

Traditional epidemiological models, such as the classic SIR (Susceptible-Infected-Recovered) framework, often treat populations as passive agents reacting predictably to contact rates. However, humans are active decision-makers who adjust their behavior based on perceived risks and social rewards. A significant gap in these models is the feedback loop between individual social behavior and the collective state of the population, especially during presymptomatic spread. If a rational person does not know they are sick, they will continue to be socially active, inadvertently driving the epidemic.

Modeling the "Invisible" Spread

The authors formalize this problem using MFG, a framework designed for systems with a massive number of small agents where each agent's strategy depends on the "mean field"—the average behavior of the entire population.

Model Components

The model defines five health states and a decision-making process based on costs and rewards:

State transition diagram (a) and cost/reward structure (b) for the Mean-Field Game model.
  • States: Agents transition through Susceptible ($s$), Presymptomatic ($a$), Symptomatic ($i$), Recovered ($r$), and Dead ($d$).
  • The Decision: Each agent chooses an activity level ($U$) between 0 (total isolation) and 1 (fully active).
  • The Cost Function: Individuals aim to minimize a "cost" that balances health risks (sickness or death), altruistic costs (the risk of infecting others), and the lost "reward" of social activity.

Information Structures

A core novelty of this research is the comparison of two information scenarios:

  1. Fully Observed: Agents have perfect knowledge of their health status at all times.
  2. Partially Observed: Agents only know they are infected once symptoms appear. Before that, they must maintain a "belief state"—a probability distribution regarding whether they are still susceptible or already presymptomatic.

Key Insights: Rationality and Information

The model reveals several profound insights into epidemic dynamics:

  • The "Vulcan" Paradox: In a world of perfectly rational agents with perfect information, the model shows that $R_0$ (the basic reproduction number) would effectively be zero. Infected people would self-isolate immediately because the cost of infecting others and their own health risk outweighs the reward of social activity. The existence of epidemics suggests that either human behavior is "irrational" or—more likely—agents operate with imperfect information.
  • Rational Presymptomatic Spread: Under imperfect information, a presymptomatic agent shares the same "belief" as a susceptible agent. Consequently, staying active is the optimal choice given their limited data, even though it is exactly what drives the epidemic.
  • The Threshold Effect: The authors identify a critical infection level ($\beta_{crit}$). Below this threshold, rational susceptible agents remain active. Once crossed, the risk-reward balance flips, leading to spontaneous "lockdowns" even without government mandates.

Results and Implications

The paper successfully derives the coupled forward-backward equations—the Hamilton-Jacobi-Bellman (HJB) and Fokker-Planck-Kolmogorov (FPK) equations—for a partially observed MFG. They prove that the optimal strategy is typically a "threshold type": agents remain active until their internal belief of being infected hits a specific percentage, at which point they switch immediately to isolation.

Furthermore, the results suggest that "information is a vaccine." Increasing the speed at which people discover their status (e.g., through faster symptom onset or widespread testing) fundamentally changes the "game," prompting rational agents to isolate sooner and effectively flattening the curve.

Limitations and Future Directions

While robust, the model currently assumes all infected individuals eventually show symptoms, ignoring purely asymptomatic cases who recover without ever knowing they were sick. Additionally, the belief state is currently one-dimensional; adding more states (like recovered) would significantly increase mathematical complexity. Future work may incorporate proactive testing and vaccination into the MFG framework.

Summary updated

Abstract

This paper is concerned with developing mean-field game models for the evolution of epidemics. Specifically, an agent’s decision – to be socially active in the midst of an epidemic – is modeled as a mean-field game with health-related costs and activity-related rewards. By considering the fully and partially observed versions of this problem, the role of information in guiding an agent’s rational decision is highlighted. The main contributions of the paper are to derive the equations for the mean-field game in both fully and partially observed settings of the problem, to present a complete analysis of the fully observed case, and to present some analytical results for the partially observed case.
BibTeX
@inproceedings{olmez2022modeling,
  author = {Sukru Yagiz Olmez and Shubham Aggarwal and Jin-Won Kim and Erik Miehling and Tamer Basar and Matthew West and Prashant G. Mehta},
  title = {{Modeling Presymptomatic Spread in Epidemics via Mean-Field Games}},
  booktitle = {ACC},
  year = {2022},
  doi = {10.23919/acc53348.2022.9867547},
}
Figure