Source-linked AI summary
Simulated epidemics in an empirical spatiotemporal network of 50,185 sexual contacts
Luis Enrique Correa Rocha, Fredrik Liljeros, Petter Holme
TL;DR
The paper investigates how the dynamical and spatial structure of contacts between Brazilian escorts and sex-buyers affects STI spread. Using simulations on a large empirical network, it finds that temporal correlations accelerate outbreaks early, while the sampled contact structure cannot support an HIV outbreak and appears to affect STI R0 by about 10%. The authors caution against extrapolating these results to society at large.
Problem
The paper addresses limited evidence on how population-level sexual-network structure affects STI epidemics, since large-scale data rarely map connections and their timing.
Method
The paper simulates SI and compartmental epidemic models on an empirical contact network of Brazilian escorts and sex-buyers, comparing original and randomized structures.
Results
Temporal correlations speed epidemics, the contact set cannot sustain an HIV outbreak, and the commercial-sex contact structure would affect STI R0 by about 10%.
Takeaways & Limitations
Within this sampled setting, other transmission pathways are needed to explain Brazil’s STI epidemics, while travel restrictions can be about as efficient as removing highest-degree vertices.
Takeaways & Limitations
The authors caution that results from this commercial-sex contact dataset should not be extrapolated confidently to society at large.
Abstract
from arXiv · showhide
We study implications of the dynamical and spatial contact structure between Brazilian escorts and sex-buyers for the spreading of sexually transmitted infections (STI). Despite a highly skewed degree distribution diseases spreading in this contact structure have rather well-defined epidemic thresholds. Temporal effects create a broad distribution of outbreak sizes even if the transmission probability is taken to the hypothetical value of 100%. Temporal correlations speed up outbreaks, especially in the early phase, compared to randomized contact structures. The time-ordering and the network topology, on the other hand, slow down the epidemics. Studying compartmental models we show that the contact structure can probably not support the spread of HIV, not even if individuals were sexually active during the acute infection. We investigate hypothetical means of containing an outbreak and find that travel restrictions are about as efficient as removal of the vertices of highest degree. In general, the type of commercial sex we study seems not like a major factor in STI epidemics.
1. INTRODUCTION
The paper asks how topological, temporal, and geographic structure in a large empirical commercial-sex contact network affects STI transmission, while emphasizing that this setting cannot be generalized to society as a whole.
- Motivation: Large-scale sexual-network data are difficult to collect, while surveys mainly record partner counts rather than connections between partners.The paper therefore uses Internet data to map contacts and their timing across many people.
- Research scope: The authors caution that commercial-sex contact patterns cannot be generalized to the whole population, limiting society-wide conclusions.They instead focus primarily on transmission within the sampled contact structure, with only a crude population-wide estimate.
- Temporal structure: Temporal correlations can alter epidemic pathways because disease transmission depends on contact order, which static network representations discard.Burstiness and longer-term population turnover are examples of temporal structure in sexual contacts.
- Spatial structure: Geography creates longer network distances and clusters corresponding to densely populated areas, motivating data that cover broad spatial regions.The dataset spans the escort business of Brazil, which the authors believe it covers rather completely.
- Research scope: The study examines how topological, temporal, and geographic structure affects transmission pathways within this type of commercial sex.It also briefly investigates strategies for limiting epidemic outbreaks.
2. OUR EMPIRICAL SEXUAL NETWORK
The empirical network is reconstructed from an anonymous online forum linking Brazilian escorts and sex-buyers through claimed encounters, with timestamps and activity categories.
- Data construction: Forum threads organized by encounter location and prostitution type are used to connect sex-buyer posts with escorts in a bipartite network.Edges represent claimed sexual contacts between the two participant groups.
- Measurement: Post timestamps are treated as estimates of encounter times, although sex-buyers may report several encounters in the same session.The dataset also records oral sex without condom, kissing, and anal sex as activity categories.
- Dataset: 50,185 contacts were recorded between 6,642 escorts and 10,106 sex-buyers from September 2002 through October 2008.The contacts span twelve Brazilian cities.
- Dataset: The largest connected cluster covers over 97% of individuals despite the network spanning twelve cities.This indicates that most sampled participants belong to one connected component.
3. SIMULATION OF EPIDEMICS WITHIN THE DATA
Simulations on the empirical contact sequence compare epidemic spreading under SI and SIR models and under randomized temporal and topological structures. The results show distinct effects of temporal correlations, network topology, and transmission thresholds.
- Topology and ordering: Network topology slows outbreaks initially, because topologically randomized data produce faster and more pervasive early outbreaks.The original data later reach about 70% infected, below topology-only randomization but above the original temporal comparison.
- Outbreak variability: Outbreak sizes remain highly diverse even at ρ = 1, with a local maximum infecting about 0.75 of the sample population.The paper hypothesizes that temporal constraints restrict possible infection paths and increase outbreak-size diversity.
- Thresholds: Epidemics are practically absent below transmission rate 0.19, with an estimated threshold ρ* = 0.19±0.02.The threshold estimates converge toward this value as the starting time increases.
- HIV implication: The estimated acute-infection threshold δ* = 31±1 days exceeds the estimated duration of acute HIV infection, so the contact pattern is not dense enough to support an HIV outbreak.The conclusion assumes an overestimated transmission probability, making the result conservative within the model.
4. EFFECTS OF TARGETED CAMPAIGNS
The paper evaluates targeted vertex removal and time-limited vaccination as hypothetical outbreak-control strategies. Removing highly connected vertices and restricting travel can substantially reduce outbreaks, while campaign timing and duration trade off against one another.
- Targeted removal: Deleting 2% of the highest-degree vertices decreases outbreak sizes by 90%.The authors note that partner counts may be estimable in online communities, making this intervention technically possible there.
- Travel restrictions: Travelers are less individually effective targets than degree-ranked vertices, but removing them is about as efficient overall because travelers comprise only about 5% of vertices.The comparison concerns deletion strategies in the empirical network.
- Campaign trade-offs: Halving vaccination coverage from 0.8 to 0.4 requires a fourfold increase in campaign duration from 28 to 112 days.A reduction in infected fraction from 1 to 0.6 is equivalent to increasing campaign duration from 56 to 168 days.
5. AUGMENTING WELL-MIXED MODELS
The paper augments well-mixed models by estimating how the observed sexual contacts alter R0, while noting that this community is connected to broader sexual-contact patterns. A crude calculation suggests the recorded contacts contribute little to the R0 correction factor and cannot alone sustain an HIV epidemic, though the estimate is uncertain.
- Results: The observed contact structure alone is not enough to sustain an HIV epidemic, although an overestimated average contact value could make its contribution sizable within the crude model.The latter possibility concerns STI transmission via acute or chronic infections.
- Model augmentation: The analysis asks how much the recorded sexual contacts add to background sexual activity and compares the resulting change in R0.In the simplest models, R0 is the number of secondary infections in a completely uninfected population; R0 = 1 marks the epidemic threshold.
- Model augmentation: An effective R0 correction factor is estimated as Λ = 1 + σ^2/c^2, using the variance and average of contact rates.The paper applies this criterion to populations with distributed contact rates.
- Parameter estimation: The Brazilian data provide σ+ = 16.3 year−1 and c+ = 5.0(1) year−1 for the recorded contacts.These values are used to compare the correction with and without the claimed contacts.
- Results: The recorded sexual contacts contribute about 5% of the R0-correction factor and are therefore rather insignificant in this sketchy calculation.Figure 7 plots the difference d over a range of average contact rates and contact-rate variances around the estimated values.
6. DISCUSSION
The study finds that temporal correlations accelerate early disease spreading, while temporal ordering and network topology slow epidemics and produce subexponential outbreak growth. The analyzed commercial-sex contact structure is insufficient to sustain HIV and appears to have limited influence on STI reproduction, but results should not be broadly extrapolated beyond this network.
- Network and scope: The empirical network is among the largest recorded sexual-contact networks, covering Brazilian escorts and sex-buyers across twelve cities.The data come from an Internet-mediated prostitution community and represent a subset of a larger dynamic network.
- Temporal and topological effects: Temporal correlations speed up epidemics, especially during the early phase of superlinear growth, compared with randomized contact structures.The authors argue that temporal correlations should be considered in disease modeling and could inform vaccination protocols.
- Temporal and topological effects: Network-topological correlations and contact time-ordering slow epidemics compared with random contact structures, yielding subexponential outbreak growth.High densities of short cycles and community structure are identified as likely contributors to slower diffusion.
- HIV implications: The contact structure probably cannot sustain HIV outbreaks, including when acute infection is modeled with unrealistically high transmission probability.For the acute phase to support spread, its duration would need to be at least one month, compared with a median of about two weeks in previous studies.
- Containment: Deleting high-degree vertices and removing the most frequent travellers are both efficient outbreak-control strategies, with travel restrictions about as efficient as high-degree removal.Restricting encounters to selected types still leaves the network connected, so disease could spread through most of it.
- Broader implications: The studied commercial-sex contact set would, from a sketchy estimate, affect STI R0 by about 10%, consistent with studies that reduce the role of prostitution in STI epidemics.The authors caution against extrapolating these findings to society at large or to other forms of commercial and non-commercial sexual contact.