Mitigate SIR epidemic spreading via contact blocking in temporal networks

Shilun Zhang, Xunyi Zhao, Huijuan WangAppl. Netw. Sci 2022 · OpenReview

TL;DR

This paper proposes novel strategies for mitigating epidemic spread in temporal networks by selectively blocking contacts. They found that targeting contacts between infrequently interacting pairs, especially those occurring early in time, is most effective in reducing epidemic prevalence.

What Problem Does This Paper Address?

Epidemics spread through contact networks, and many real-world networks are dynamic, meaning contacts appear and disappear over time. Traditional methods for controlling epidemics often involve broad measures like lockdowns. This paper tackles the challenge of developing more targeted and efficient strategies to suppress epidemic spread in these dynamic, or "temporal," networks by strategically blocking specific contacts. The goal is to reduce the overall impact of an epidemic, measured by prevalence, peak infection rates, and the time to reach that peak.

Methods

The core of the paper's approach is to define a probabilistic method for removing contacts. The probability of removing a specific contact c(i, j, t) (between nodes i and j at time t) is determined by a function that considers:

  • Link Centrality in the Aggregated Network: They first construct an "aggregated network" where links represent the total number of contacts between any two nodes over the entire observation period. They then calculate various centrality metrics for these aggregated links (e.g., degree product, strength product, betweenness, link weight, eigenvector centrality).
  • Time of Contact: The probability of removal can also be influenced by when the contact occurs, allowing for strategies that prioritize early or late interventions.

They propose 12 different link centrality metrics, each leading to a unique contact removal strategy. A baseline "random removal" strategy is also included for comparison. A key aspect is the iterative re-normalization process used to ensure that a specific fraction (Φ) of contacts are removed on average, regardless of the centrality values.

Datasets

The study evaluates its proposed strategies on six real-world temporal contact networks:

  • HighSchool11 & HighSchool12: Physical contacts between students in a high school.
  • WorkPlace13 & WorkPlace15: Contacts within office buildings.
  • MIT1 & MIT2: Human contact networks among students at MIT, sampled over specific periods.

These datasets vary in size and duration, providing a diverse testbed for the mitigation strategies.

Key Insights and Ideas

  • Targeting "Weak Ties" is Key: The most effective strategies involve removing contacts between node pairs that have fewer overall contacts. This suggests that disrupting "weak ties" is more impactful than targeting heavily connected individuals.
  • Early Intervention Matters: Removing contacts that occur earlier in the epidemic's timeline is significantly more effective at suppressing spread than removing later contacts.
  • Consistency in Removal is Beneficial: Strategies that remove a similar number of contacts across different node pairs tend to perform better. This implies that avoiding overly aggressive removal from a few highly connected pairs is more efficient.
  • Pruned Network Properties: The most effective contact removal strategies result in pruned networks with specific structural properties: a large largest eigenvalue, high modularity, and potentially a small largest connected component. These properties collectively hinder epidemic propagation.
  • SIR vs. SIS Dynamics: The paper highlights that network properties that make a network robust against SIS (Susceptible-Infected-Susceptible) epidemics (like a small largest eigenvalue) don't necessarily translate to robustness against SIR (Susceptible-Infected-Recovered) epidemics, especially in temporal networks.

Key Results

  • The 1/link weight strategy consistently performed best across most datasets and scenarios, significantly reducing average prevalence and peak height. This strategy prioritizes removing contacts between node pairs with the lowest total number of contacts.
  • When removing 10% of contacts, the 1/link weight strategy achieved the lowest average prevalence in most datasets, often outperforming others by a noticeable margin (e.g., reducing prevalence from ~0.05 to ~0.02 in some cases).
  • Time-dependent strategies that prioritized removing early contacts (e.g., f(t) = 4 * 1_{t<T/2} + 1_{t>T/2}) further enhanced the effectiveness of the 1/link weight strategy.
  • The findings remained robust even when introducing uncertainty into the temporal network data (reshuffling contact timestamps or aggregating contacts into larger time bins).

Limitations and Future Work

  • SIR Model Simplification: The study is confined to the SIR model, which is a simplified representation of real-world epidemics. More complex epidemic models and factors (like varying individual susceptibility or behavior) could be explored.
  • Limited Parameter Space: The paper used a limited set of parameters for infection and recovery rates, and a specific fraction of contacts to remove. Exploring a wider range of these parameters could reveal further nuances.
  • Time-Dependent Function f(t): The time-dependent function used to prioritize early or late contacts was a simple form. More complex, policy-maker-friendly time-dependent functions could be investigated.
  • Aggregated Network Assumption: The proposed strategies rely on knowledge of the aggregated network. Predicting or estimating this aggregated network from past observations is a crucial practical challenge that warrants further research.
  • Generalizability: While the findings are promising for temporal networks, direct generalization to all real-world epidemic mitigation scenarios requires further exploration with more complex models and diverse network types.

Summary updated

Abstract

Progress has been made in how to suppress epidemic spreading on temporal networks via blocking all contacts of targeted nodes or node pairs. In this work, we develop contact blocking strategies that remove a fraction of contacts from a temporal (time evolving) human contact network to mitigate the spread of a Susceptible-Infected-Recovered epidemic. We define the probability that a contact c(i,&nbsp;j,&nbsp;t) is removed as a function of a given centrality metric of the corresponding link l(i,&nbsp;j) in the aggregated network and the time t of the contact. The aggregated network captures the number of contacts between each node pair. A set of 12 link centrality metrics have been proposed and each centrality metric leads to a unique contact removal strategy. These strategies together with a baseline strategy (random removal) are evaluated in empirical contact networks via the average prevalence, the peak prevalence and the time to reach the peak prevalence. We find that the epidemic spreading can be mitigated the best when contacts between node pairs that have fewer contacts and early contacts are more likely to be removed. A strategy tends to perform better when the average number contacts removed from each node pair varies less. The aggregated pruned network resulted from the best contact removal strategy tends to have a large largest eigenvalue, a large modularity and probably a small largest connected component size.
BibTeX
@inproceedings{zhang2022mitigate,
  author = {Shilun Zhang and Xunyi Zhao and Huijuan Wang},
  title = {{Mitigate SIR epidemic spreading via contact blocking in temporal networks}},
  booktitle = {Appl. Netw. Sci},
  year = {2022},
  url = {https://openreview.net/forum?id=XHxuOfEGYk},
}
Figure