The Evolution of the Cuban HIV/AIDS Network

An individual detected as HIV positive in Cuba is asked to provide a list of his/her sexual contacts for the previous 2 years. This allows one to gather detailed information on the spread of the HIV epidemic. Here we study the evolution of the sexual contact graph of detected individuals and also the directed graph of HIV infections. The study covers the Cuban HIV epidemic between the years 1986 and 2004 inclusive and is motivated by an earlier study on the static properties of the network at the end of 2004. We use a variety of advanced graph algorithms to paint a picture of the growth of the epidemic, including an examination of diameters, geodesic distances, community structure and centrality amongst others characteristics. The analysis contrasts the HIV network with other real networks, and graphs generated using the configuration model. We find that the early epidemic starts in the heterosexual population and then grows mainly through MSM (Men having Sex with Men) contact. The epidemic exhibits a giant component which is shown to have degenerate chains of vertices and after 1989, diameters are larger than that expected by the equivalent configuration model graphs. In 1997 there is an significant increase in the detection rate from 73 to 256 detections/year covering mainly MSMs which results in a rapid increase of distances and diameters in the giant component.

💡 Research Summary

This paper presents a comprehensive temporal network analysis of the Cuban HIV/AIDS epidemic spanning the years 1986 to 2004. The authors exploit a unique public‑health surveillance system in Cuba: every individual diagnosed as HIV‑positive is asked to provide a list of all sexual partners from the preceding two years. Consequently, the authors are able to construct two complementary graphs: (i) an undirected “contact” graph Gₜ whose vertices are diagnosed individuals and whose edges represent reported sexual contacts, and (ii) a directed “infection” graph Dₜ that records the inferred direction of transmission (a subset of the contact edges). The dataset comprises 5,389 individuals (4,280 men, 1,109 women) and 4,097 reported contacts; by 2006, 2,122 of these individuals had died.

The study’s primary objectives are threefold: (1) to compare the evolving real‑world network with ensembles of configuration‑model (CM) graphs that preserve the empirical degree distribution at each time slice; (2) to assess how different detection modalities (voluntary testing, contact tracing, blood‑bank screening, etc.) change their relative importance over time; and (3) to determine whether the Cuban epidemic follows the “densification power law” (DPL) and small‑world phenomena documented in other longitudinal network studies.

Methodologically, the authors partition the observation window into 90‑day intervals (with additional snapshots at four‑year marks) and compute a suite of global and local metrics for each snapshot: number of vertices |Vₜ|, number of edges |Eₜ|, average degree, DPL exponent σ (from log|Eₜ| vs. log|Vₜ|), component size distribution, emergence and growth of the giant component, graph diameter and effective diameter (the distance D such that ≥90 % of reachable vertex pairs are within D hops), clustering coefficient, community structure (via the Louvain algorithm), and various centrality measures (betweenness, closeness, PageRank). For the infection graph, they also analyze the size and depth of infection trees. Ten independent CM graphs are generated for each three‑month interval, matching the observed degree sequence, and their statistics are averaged to serve as a null model.

Key findings include:

1. Densification and Degree Growth – The contact graph exhibits a clear DPL with σ ranging between 1.3 and 1.7, indicating that as the epidemic expands, the average degree rises rather than remaining constant. This mirrors observations in citation, internet, and social networks.

2. Giant Component Formation – A giant component first appears around early 1990. Its size continues to increase, eventually encompassing the majority of vertices by 2004. However, unlike many real‑world networks where the giant component becomes increasingly compact, the Cuban network develops long, sparsely connected “degenerate chains” within the giant component after 1989. These chains inflate both the raw diameter and the effective diameter.

3. Comparison with Configuration Model – CM graphs with identical degree sequences display markedly smaller diameters and higher clustering homogeneity. The real network’s diameters are consistently larger than CM expectations after 1989, underscoring the presence of structural constraints (e.g., assortative mixing by sexual orientation) that are not captured by degree‑preserving randomization.

4. Impact of Detection Surge in 1997 – In 1997, the annual detection rate jumps from 73 to 256 cases, driven primarily by intensified contact‑tracing among men who have sex with men (MSM). This influx of newly diagnosed MSM introduces many high‑degree nodes, dramatically increasing the average path length and effective diameter of the giant component. The surge also shifts the community composition: MSM clusters become denser and more central, while heterosexual clusters remain peripheral.

5. Community and Centrality Evolution – Early in the epidemic, communities are roughly divided by gender and heterosexual orientation. Over time, a distinct MSM community emerges with high internal edge density and relatively few bridges to the heterosexual component. Betweenness and closeness centralities initially highlight a few heterosexual men who act as bridges between early clusters; after 1997, MSM nodes dominate the centrality rankings, reflecting their role as new transmission hubs.

6. Infection Tree Dynamics – The directed infection graph reveals shallow, wide trees during the early years (reflecting rapid spread among a few heterosexual individuals). Post‑1990, infection trees become deeper, especially within the MSM subgraph, indicating longer transmission chains facilitated by the degenerate chain structures observed in the undirected contact graph.

7. Detection Method Shifts – The proportion of cases identified via contact tracing rises sharply after 1990, coinciding with the formation of the giant component. Voluntary testing and blood‑bank screening play a diminishing role over time, suggesting that the surveillance system increasingly relies on network‑based case finding.

Overall, the authors demonstrate that the Cuban HIV epidemic cannot be fully explained by simple stochastic growth models that preserve degree distributions. The interplay of detection policies, sexual orientation assortativity, and the emergence of a dense MSM subnetwork creates a uniquely evolving topology characterized by increasing densification, persistent long‑range chains, and a delayed but pronounced small‑world effect.

The paper’s contributions are twofold: (i) it provides a rare longitudinal, high‑resolution view of a sexually transmitted infection network, and (ii) it highlights the importance of incorporating dynamic, attribute‑aware network structures into epidemiological modeling and public‑health intervention design. By showing how a policy‑driven surge in case detection reshapes the underlying contact network, the study offers actionable insights for other settings where contact tracing and targeted testing are central to disease control.