Effects of community structure on epidemic spread in an adaptive network

When an epidemic spreads in a population, individuals may adaptively change the structure of their social contact network to reduce risk of infection. Here we study the spread of an epidemic on an adaptive network with community structure. We model the effect of two communities with different average degrees. The disease model is susceptible-infected-susceptible (SIS), and adaptation is rewiring of links between susceptibles and infectives. The bifurcation structure is obtained, and a mean field model is developed that accurately predicts the steady state behavior of the system. We show that an epidemic can alter the community structure.

💡 Research Summary

The paper investigates how adaptive changes in social contact patterns, driven by individuals’ attempts to avoid infection, interact with community structure to shape epidemic dynamics. The authors construct a two‑community network in which each community has a distinct average degree—one densely connected, the other sparsely connected. On this substrate they place a classic susceptible‑infected‑susceptible (SIS) disease process, and they endow susceptible nodes with a rewiring rule: when a susceptible is linked to an infected neighbor, it may cut that link with probability w and reconnect to a randomly chosen susceptible within the same community. This rule captures the behavioral response of “avoiding the sick” and makes the network topology co‑evolve with the epidemic.

A mean‑field (pair‑approximation) framework is derived. The state variables are the infection prevalences i_A and i_B in the two communities and the fractions of S–I edges, θ_A and θ_B. The resulting set of nonlinear differential equations incorporates infection (rate β), recovery (rate γ), and rewiring (rate w). Fixed‑point analysis reveals two regimes: a disease‑free equilibrium (i_A = i_B = 0) and an endemic equilibrium (i_A, i_B > 0). Linear stability analysis yields a transcritical bifurcation that defines the epidemic threshold β_c as a function of w, the intra‑community average degrees, and the inter‑community link density p_cross. Crucially, the threshold rises when the inter‑community coupling is weak, indicating that a highly connected community can act as a reservoir only if sufficient bridges to the low‑degree community exist.

The authors validate the analytical predictions with extensive agent‑based simulations on networks of N = 10 000 nodes (⟨k_A⟩ = 12, ⟨k_B⟩ = 4, p_cross = 0.05). The simulated steady‑state infection levels match the mean‑field predictions across a wide range of β and w, confirming the accuracy of the closure approximations. Moreover, the simulations expose a dynamic reshaping of the network: as rewiring proceeds, intra‑community links increase while inter‑community links diminish, effectively raising the modularity of the system. In some parameter regimes, susceptible nodes from the high‑degree community migrate toward the low‑degree community to escape infection, creating new cross‑community edges; in other regimes the two communities become more isolated. Thus the epidemic does not merely spread on a static substrate—it actively remodels that substrate.

From a public‑health perspective, the results suggest that individual avoidance behavior (rewiring) can suppress disease spread but also reinforces community segregation, potentially creating pockets of vulnerability. Policies that only target transmission rates (e.g., vaccination, antiviral treatment) may be insufficient if they ignore the structural feedback loop. Complementary measures that modulate inter‑community contact—such as travel restrictions, targeted testing at community borders, or information campaigns that discourage risky cross‑group interactions—could synergize with behavioral interventions to raise the effective epidemic threshold.

In conclusion, the study provides a rigorous analytical and computational treatment of SIS dynamics on an adaptive, two‑community network. The mean‑field model captures the bifurcation structure and predicts steady‑state outcomes with high fidelity. The work highlights that epidemics can fundamentally alter the underlying social network, and that such structural changes, in turn, feed back into disease dynamics. Future extensions could incorporate more than two communities, heterogeneous rewiring strategies (e.g., preferential attachment to low‑risk nodes), or alternative disease models such as SIR or SEIR, thereby broadening the applicability of the framework to real‑world scenarios where both network topology and human behavior evolve together.