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

TL;DR

It is demonstrated that integrating demographic mixing patterns into network reconstruction fundamentally alters optimal intervention design, offering a practical framework for improving HIV prevention in MSM populations where complete contact data are unavailable.

Abstract

Men who have sex with men (MSM) remain disproportionately affected by HIV, yet optimizing the distribution of pre-exposure prophylaxis (PrEP) in this population remains a major public health challenge. Current PrEP eligibility guidelines and most modelling studies do not incorporate sociodemographic or network-level factors that shape transmission. We present a novel network reconstruction framework that generates MSM sexual contact networks from individual-level behavioral data, incorporating clustering and demographic assortativity by age, race, and sexual activity. Using data from 4667 MSM participants, we reconstructed networks with varying topological properties and simulated HIV transmission over 50 years. Network structure strongly influenced outcomes: assortative by degree networks showed 18% lower equilibrium prevalence (63% vs 80% in assortative by race networks) due to hub isolation within communities. Targeted PrEP strategies based on degree or k-shell centrality achieved similar reductions with 20 to 40% coverage, matching random allocation at 60 to 80% coverage, particularly in assortative by age and race networks where hubs bridge demographic groups. Empirical PrEP distribution was suboptimal, underperforming by up to 30% compared with network-based strategies. Our findings demonstrate that integrating demographic mixing patterns into network reconstruction fundamentally alters optimal intervention design, offering a practical framework for improving HIV prevention in MSM populations where complete contact data are unavailable.

Figures (15)

• Figure 1: Workflow from data preprocessing to simulations of epidemic spreading on networks and deployment of PrEP under different intervention scenarios.
• Figure 2: Number of nodes per community detected by NSBM. The barplots represent the proportion of nodes in each community for the most clustered (a), and assortative by degree (b), age (c), and race (d) networks. The dashed horizontal line indicates the average node proportion per community.
• Figure 3: Node degree distribution within each community detected by NSBM. Each boxplot presents the median and the interquartile range of the node degrees found within the the most clustered (a), assortative by degree (b), age (c) and race (d) networks. The y-axis is log-scaled for clarity. The outliers detected correspond to nodes whose degree was significantly higher than the median node degree of the respective community (hubs).
• Figure 4: HIV final size reduction computed for several PrEP distribution scenarios. HIV final size reduction was defined as the ratio between the final proportion of PLHIV for a certain implemented strategy $F_s$ and the baseline model $F_0$ computed for varying PrEP efficacy (0.23, 0.80, and 0.95), coverage (0.05, 0.10, 0.20, 0.40, 0.60, 0.80), and targeting strategies (random, highest degree centrality, and $k$-shell). Panels a-c, correspond to the most clustered networks, d-f assortative by degree, g-i assortativity by age and j-l assortativity by race.
• Figure 5: HIV dynamics comparison between intervention scenarios and current PrEP distribution retrieved from the data. Proportion of PLHIV across 50 years in clustered (a), and assortative by degree (b), age (c), and race networks (d), considering a PrEP adherence of 95%. Each colored line corresponds to a different PrEP distribution strategy (empirical PrEP distribution, random, highest degree and k-shell). Prevalence is reported as the median with interquartile range at each time step, based on $N=50$ stochastic simulations.