Source-linked AI summary
Exploiting temporal network structures of human interaction to effectively immunize populations
Sungmin Lee, Luis E. C. Rocha, Fredrik Liljeros, Petter Holme
TL;DR
Reducing the number of vaccinations needed for herd immunity requires identifying people likely to become infected and spread disease. The paper proposes immunization protocols based on temporal contact patterns; Weight performs best for email data, while Recent performs best for the other three datasets.
Problem
The paper asks how contact patterns can identify influential individuals to vaccinate and lower the threshold for herd immunity.
Source: Introduction · p. 1Method
The paper proposes practical immunization protocols that use temporal and topological contact information, including recent contacts.
Source: Introduction · p. 2, Discussion · p. 7Results
Weight outperforms Recent and NV on the email dataset, while Recent is the most efficient scheme on the other three datasets.
Source: Discussion · p. 7Takeaways & Limitations
Protocol choice depends on contact structure: Recent suits networks with relationship turnover, while Weight suits strongly overlapping ties with broadly distributed contact rates.
Source: Discussion · p. 8The authors do not claim the email dataset represents disease-spreading contacts, noting it may better reflect information or opinion spreading.
Source: The protocols · p. 3
Abstract
from arXiv · showhide
If we can lower the number of people needed to vaccinate for a community to be immune against contagious diseases, we can save resources and life. A key to reach such a lower threshold of immunization is to find and vaccinate people who, through their behavior, are more likely to become infected and effective to spread the disease than the average. Fortunately, the very behavior that makes these people important to vaccinate can help us finding them. People you have met recently are more likely to be socially active and thus central in the contact pattern, and important to vaccinate. We propose two immunization schemes exploiting temporal contact patterns. Both of these rely only on obtainable, local information and could implemented in practice. We show that these schemes outperform benchmark protocols in four real data sets under various epidemic scenarios. The data sets are dynamic, which enables us to make more realistic evaluations than other studies - we use information only about the past to perform the vaccination and the future to simulate disease outbreaks. We also use models to elucidate the mechanisms behind how the temporal structures make our immunization protocols efficient.
Introduction
The introduction motivates targeting people whose contact behavior makes them influential in disease spread and proposes temporal contact patterns to identify vaccination candidates. It presents contact-based and time-sensitive vaccination approaches as ways to lower the herd-immunity threshold.
Introduction
- Vaccinating a contact named by a randomly selected person can target influential spreaders, who may have unusually high network degree.The paper argues that heterogeneous contact patterns shape people’s importance in disease spread and cites the 2003 SARS epidemic as an outbreak thought to involve a few highly influential spreaders.Source: Introduction · p. 1
- A sufficiently large vaccinated fraction f prevents substantial disease spread, making a lower herd-immunity threshold valuable when vaccines are limited, costly, or carry side effects.Vaccination-preventable diseases are estimated to cause 160 million DALYs annually, around 10% of all disease or twice the burden of HIV.Source: Introduction · p. 1
- Two temporal protocols vaccinate either the person met most often since a specified time or the randomly selected person’s most recent contact who could transmit disease.The recent-contact protocol exploits bursty contact patterns, targeting people who are currently active or will be active soon; it requires no global knowledge and is described as practical to implement.Source: Introduction · p. 2
Results
Recent and Weight use temporal contact information to identify people to vaccinate, and their relative performance depends on the contact sequence: Recent generally leads on the sexual-contact and hospital data, while Weight leads on email. Artificial contact models link this difference to nonstationary activity and partner turnover.
The protocols
- Recent vaccinates a random person’s most recent disease-transmitting contact, whereas Weight targets a frequent contact over a longer timescale, sampled in proportion to tie weight.Recent is intended to favor people in a heightened-activity period; Weight seeks generally more active people.Source: The protocols · pp. 2–3
Simulation of vaccination campaigns on empirical contact sequences
- Using the first three quarters of contacts to select vaccinees and the remainder for outbreak simulations separates vaccination information from evaluation contacts.The campaign assumes immediate, completely effective vaccination and a vaccination timescale short relative to epidemics.Source: Simulation of vaccination campaigns on… · p. 4, The protocols · p. 4
Largest outbreak sizes after vaccination
- For most datasets and vaccination fractions, both strategies reduce worst-case outbreak size versus NV; Weight is almost never worse, while Recent loses on email but clearly leads on the other three datasets.Recent’s relative advantage is strongest in the prostitution data. The email dataset is a communication network, not claimed to represent disease transmission.Source: Figure 2 · p. 3, The protocols · p. 3, Largest outbreak sizes after vaccination · p. 5
Average outbreak sizes in dynamic simulations
- In SIS simulations, Weight is best on email and Recent on the other datasets across the tested infectivity-duration space; strategy differences are smallest when infectivity and duration are both low.The curves resemble the worst-case results, though performance differences are larger for SIS on three datasets and smaller for prostitution data, where about three quarters of contacts occur only once. SIR simulations with the same parameters yield similar curves.Source: Average outbreak sizes in dynamic… · pp. 5–6, Largest outbreak sizes after vaccination · p. 5
Model of artificial contact sequences
- In random-network models, Recent outperforms NV in partner turnover, while Weight outperforms NV in varying activity; the temporal structure explains the distinct empirical winners.The email data are comparatively stationary, whereas dating, hospital, and prostitution data are more nonstationary; the authors link this to Weight’s email advantage and Recent’s advantage on the other datasets.Source: Model of artificial contact sequences · pp. 6–7
Discussion
The discussion argues that practical immunization protocols can exploit temporal as well as topological contact structure, and that the most effective protocol depends on the contact dynamics. Recent is favored in three tested datasets, while Weight performs best on the email data.
Discussion
- The practical protocols exploit temporal and topological structure and outperform neighborhood vaccination; four dynamic datasets separate vaccination history from outbreak contacts, unlike prior studies [10].The datasets include two possible epidemic pathways and two online communication networks, with uncertain correspondence to offline contacts.Source: Discussion · p. 7
- Weight outperforms Recent and NV on the e-mail data, while Recent is most efficient on the other three datasets; the authors recommend Recent when contact structure is unknown.The preferred strategy varies with the contact structure, and Recent is better in the more realistic networks.Source: Discussion · pp. 7–8
- Relationship turnover promotes Recent, whereas Weight is most efficient when ties overlap strongly in time and contact rates vary broadly.The authors identify temporal structure as a factor in immunization effectiveness that merits further study, including for animal disease and email or computer worms.Source: Discussion · p. 8
Disease-spreading simulation
- In SIS simulations, contacts transmit infection with probability λ and infected vertices become susceptible after δ; s counts infections by T, while S is the mean reachable count for λ = 1 and δ = ∞.The quantities are averaged over unvaccinated vertices as infection sources and 1,000 independent runs.Source: Discussion · p. 8
Models of contact dynamics
- The contact-dynamics models hold an Erdős–Rényi topology of 1,000 vertices and 2,000 edges fixed while varying edge-contact timing through activity intervals or partner-turnover bursts.Vaccination and disease start at t* = 3C/4; protocols use only contacts before t*, and the source is selected uniformly from the unvaccinated population.Source: Models of contact dynamics · p. 8