🧬 biology

Reconstructing MSM Sexual Networks to Guide PrEP Distribution Strategies for HIV Prevention

Using data from 4,667 MSM participants, this study demonstrates that uncertainty in reconstructing sexual network structures—specifically regarding age and race assortativity—significantly impacts simulated HIV outcomes, revealing that current PrEP distribution is suboptimal compared to targeted strategies based on network centrality.

Original authors: João Brázia, István Z. Kiss, Alexandre P. Francisco, Andreia Sofia Teixeira

Published 2026-03-17

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: João Brázia, István Z. Kiss, Alexandre P. Francisco, Andreia Sofia Teixeira

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine HIV prevention as trying to stop a wildfire from spreading through a vast, complex forest. For a long time, health officials have tried to put out the fire by giving water (PrEP medication) to anyone who looks like they might be standing near a spark (based on individual risky behaviors).

But this new study suggests that to really stop the fire, we need to understand the shape of the forest itself.

Here is the story of the research, broken down into simple concepts:

1. The Problem: We Can't See the Whole Map

The researchers wanted to know the best way to distribute PrEP (a pill that prevents HIV) to Men who have Sex with Men (MSM). The problem is that we don't have a complete map of who sleeps with whom. Asking people for a full list of their partners is hard because it's private and sometimes stigmatized.

So, the team had to build a digital "ghost forest." They took data from 4,667 real people (their age, race, how many partners they had, etc.) and used a computer to build thousands of possible versions of how these people might be connected.

2. The Twist: The Shape of the Forest Matters

The researchers realized that the "shape" of these connections changes everything. They tested different versions of the forest:

The "Age-Bubble" Forest: Where people mostly connect with others their own age.
The "Race-Bubble" Forest: Where people mostly connect with others of the same race.
The "Popular-Cluster" Forest: Where the most popular people (those with many partners) mostly stick together in one tight group.

The Big Discovery:
In the "Age" and "Race" forests, the most popular people act like bridges. They are the only ones connecting different groups. If you don't protect the bridge, the fire jumps from one side of the river to the other.
In the "Popular-Cluster" forest, the popular people are all in one big circle. If the fire starts there, it burns fast inside the circle, but it has a hard time jumping out to the rest of the forest because the popular people are isolated from the "regular" people.

3. The Strategy: Who Gets the Water?

The team simulated giving out PrEP in three different ways to see which stopped the "fire" (HIV) best:

The Random Approach: Throwing water randomly at trees.
The "Risk" Approach: Giving water to people who admit to risky behaviors (the current real-world method).
The "Network" Approach: Using math to find the most important trees (the bridges and hubs) and giving them water first.

The Results:

When the forest is "Age/Race" shaped: The current "Risk" approach was suboptimal. It missed the bridges. The "Network" approach (targeting the bridges) was a huge winner, stopping the fire much faster with less water.
When the forest is "Popular-Cluster" shaped: The current approach actually worked okay because the popular people were already somewhat protected by being in their own group. However, the Network approach was still better.

4. The "Adherence" Catch

There is one major catch: The water has to stick.
The study showed that even the best strategy fails if the people taking the PrEP don't take it consistently.

If people take the pill only 23% of the time (low adherence), it doesn't matter who you give it to; the fire keeps spreading.
If people take it 95% of the time (high adherence), the "Network" strategy becomes a superpower, stopping the fire with half the resources needed by the random approach.

The Takeaway: Why This Matters

Think of the current guidelines as a firefighter looking at a single house and asking, "Did you play with matches?" If yes, they give you a fire extinguisher.

This paper argues that we need a firefighter looking at the whole neighborhood map.

If we know that a specific person is the "bridge" between a young group and an older group, or between two different racial communities, that person is the most critical to protect, even if their personal behavior doesn't look "super risky" on paper.
Because we can't always see the real map, there is uncertainty. But the study proves that guessing the map structure matters. If we assume the wrong structure, we might waste resources or miss the key people who could stop an epidemic.

In short: To stop HIV effectively, we need to move beyond just asking individuals about their risks. We need to use math to understand how communities are connected, find the "bridges" that hold the network together, and make sure those bridges are the first to get the protection they need.

1. Problem Statement

Men who have sex with men (MSM) remain disproportionately affected by HIV, yet optimizing Pre-exposure Prophylaxis (PrEP) distribution is hindered by a lack of granular network data. Current public health guidelines and modeling studies primarily rely on individual-level risk behaviors (e.g., condom use, number of partners) or simplified network topologies. They often fail to account for sociodemographic mixing patterns (assortativity by age, race, and activity level) and the specific structural positions of individuals within sexual networks.

The core challenge is that complete sexual contact networks are rarely observable due to stigma and privacy concerns. Consequently, it is unclear whether current PrEP uptake strategies effectively reach "transmission-critical" individuals (hubs or bridges) or if uncertainty in network reconstruction (specifically regarding demographic assortativity) significantly alters intervention outcomes.

2. Methodology

The authors developed a comprehensive framework combining data-driven network reconstruction, epidemic simulation, and intervention analysis.

A. Data Source and Preprocessing

Dataset: Used data from the ARTnet study (2017–2019), an anonymous cross-sectional survey of 4,904 MSM in the US.
Cohort: After preprocessing (excluding those with zero contacts and imputing missing values), the final cohort consisted of 4,667 individuals.
Attributes Extracted: Age, race, HIV status, number of sexual partnerships (degree), PrEP status, and testing frequency.
Baseline: 13.67% of the cohort reported being on PrEP.

B. Network Reconstruction Framework

Since the dataset lacked actual contact links, the authors constructed synthetic networks that preserved the observed degree sequence while varying demographic mixing patterns:

Baseline Generation: Used the Havel-Hakimi algorithm to generate simple, undirected graphs matching the observed degree sequence ( $V=4,667$ , $E=25,107$ ).
Randomization: Applied Markov Chain Monte Carlo (MCMC) edge-swapping to ensure statistical robustness and randomization, running 10 independent chains with 10 initial configurations (100 total networks).
Structural Optimization: The authors created four distinct network ensembles by maximizing specific structural properties via edge rewiring while preserving the degree distribution:
- Clustering: Maximizing local clustering coefficients.
- Assortativity by Degree: High-degree nodes connecting to other high-degree nodes.
- Assortativity by Age: Nodes connecting to others of similar age.
- Assortativity by Race: Nodes connecting to others of the same race.

C. Epidemiological Simulation

Model: A discrete-time, stochastic Agent-Based Model (ABM) simulating HIV transmission over 50 years.
States: Susceptible ( $S$ ), Infected ( $I$ ), and Treatment ( $T$ , assumed virally suppressed/non-infectious).
Transmission Dynamics: Probability of infection ( $\beta$ ) depended on the number of infected neighbors, sexual role, condom use (30% coverage, 70% effectiveness), and PrEP status.
Intervention: PrEP allocation was simulated under varying adherence levels (0.23, 0.80, 0.95) and coverage (5% to 80%).
Strategies Compared:
1. Random: Uniform distribution.
2. Degree Centrality: Targeting individuals with the most partners.
3. K-Shell Decomposition: Targeting individuals in the network core (nucleus).
4. Empirical: Allocation based on the actual PrEP uptake observed in the ARTnet data.

3. Key Contributions

Novel Reconstruction Framework: First study to systematically incorporate and evaluate demographic assortativity (age and race) alongside degree and clustering in MSM network reconstruction, moving beyond simplified topologies.
Quantification of Structural Uncertainty: Demonstrated that the choice of assortativity dimension (age vs. race vs. degree) significantly alters network topology and, consequently, HIV transmission dynamics and intervention efficacy.
Validation of Targeting Strategies: Provided empirical evidence that network-based targeting (degree/k-shell) outperforms random distribution and current empirical uptake, particularly in networks where hubs bridge demographic groups.
Practical Policy Insight: Highlighted that current eligibility criteria (based solely on individual risk behaviors) may miss critical transmission bridges in demographically assortative networks.

4. Key Results

Network Structural Properties

Assortativity by Degree: Resulted in the highest fragmentation (323 disconnected components), longest average path lengths ( $\langle l \rangle \approx 6.23$ ), and largest diameter. Hubs were isolated within specific communities, creating "transmission reservoirs" but limiting spread to the broader population.
Assortativity by Age & Race: Showed lower fragmentation and modularity. Crucially, hubs were distributed across communities, acting as bridges between demographic groups.
Community Structure: NSBM analysis revealed that age-assortative networks had homogeneous community sizes, while race-assortative networks concentrated ~70% of nodes in a single dominant community.

Epidemic Dynamics

Baseline Prevalence: Assortative-by-degree networks had the lowest equilibrium prevalence (~~63.2%) due to hub isolation. Conversely, age and race assortative networks reached the highest prevalence (~~78–80%) due to efficient bridging by hubs.
Impact of Adherence: At low adherence (23%), all strategies failed to reduce prevalence significantly. At high adherence (95%), targeted strategies showed massive gains.

Intervention Effectiveness

Targeted vs. Random: Targeted strategies (Degree and K-shell) consistently outperformed random allocation.
- In Age/Race Assortative Networks: Targeted strategies achieved equivalent prevalence reductions at 40% coverage, whereas random distribution required 80% coverage.
- In Degree Assortative Networks: Targeted strategies required higher coverage thresholds (~30%) to outperform random allocation because hubs were isolated within clusters.
Degree vs. K-Shell: In age-assortative networks, Degree Centrality outperformed K-Shell decomposition, likely due to lower correlation between the two metrics in these topologies.
Empirical Uptake Analysis:
- In clustered/degree-assortative networks, current PrEP uptake (from data) aligned well with optimal strategies.
- In Age/Race Assortative Networks, current uptake was suboptimal. It failed to target the central bridging individuals. Degree and K-shell targeting substantially outperformed the empirical distribution in these scenarios.

5. Significance and Implications

Policy Optimization: The study argues that PrEP allocation should move beyond individual risk behaviors to include network-informed strategies. Identifying individuals who bridge demographic groups (age/race) is critical for disrupting transmission chains.
Robustness under Uncertainty: Since the "true" network structure is unknown, the study highlights that assuming a specific assortativity pattern can lead to vastly different policy recommendations. It underscores the need for robust strategies that perform well across structural uncertainties.
Data Collection: Future surveillance should collect data on mixing preferences (who partners with whom demographically) rather than just partner counts to refine network reconstruction.
Resource Efficiency: Targeted allocation can achieve substantial epidemic control with fewer resources (lower coverage requirements), making it a cost-effective approach for public health programs.

Conclusion: The paper demonstrates that network topology, specifically demographic assortativity, is a primary determinant of HIV transmission dynamics. Ignoring these structural nuances leads to suboptimal PrEP distribution, while leveraging network centrality can significantly enhance prevention efficiency, particularly in bridging diverse demographic communities.