Transmission network structure and newly diagnosed HIV infections: a molecular epidemiology study

BackgroundImproved understanding of network structure—one of the driving forces of HIV transmission—could facilitate the design of interventions. We aimed to assess the structure of HIV transmission networks and their relationship to newly diagnosed cases in Hong Kong, where sexual transmission between men prevails.MethodsWe used HIV gag-pol sequences collected over a 20-year period from all major HIV services in Hong Kong to construct transmission networks. We identified network clusters and decomposed them by modularity if one contained more than half of the connected nodes. We analysed network size, average degree, and epidemiological metrics, and examined sequences of newly diagnosed cases in a cross-sectional study to assess growth of identified clusters.FindingsA total of 2352 archived HIV sequences collected in 1994–2013 were retrieved, alongside HIV sequences from 81 cases diagnosed between Aug 4, 2016, and Jan 17, 2017. The transmission network with a density of 1·58% contained 1801 (74%) nodes that were linked to at least one other node. Network degree followed a power law distribution (correlation 0·988). There were 72 dyads and 64 clusters with at least three nodes. The largest connected component contained more than half of the connected nodes (1282 [71%]) and thus was decomposed into 20 clusters. Of 83 clusters included in the analysis, 10 were linked to 37 (46%) new cases. The sizes of incident clusters were significantly larger (p<0·001, median 15 [IQR 11-81] vs 4 [IQR 3-8]) with a higher mean degree (p<0·001, 10·86 [IQR 4·35–22·64] vs 2·80 [IQR 2·00–4·64]). Clusters with newly diagnosed patients had higher transmission speeds (p<0·001, median 0·97 [IQR 0·72–4·77] persons per year [ppy] vs 0·24 [IQR 0·17–0·44] ppy). Clusters with new cases in 2016 had a higher 3-year cluster incidence in 2013 (p<0·001, 0·33 [IQR 0·33–4·17]) and wider age difference between nodes (40·50 years [IQR 24·75–50·25]) than those without new cases in 2016 (3-year cluster incidence 0·00 [IQR 0·00–0·33]; age difference 17·00 years [IQR 10·00–28·5]).InterpretationHong Kong's HIV transmission was a preferential attachment process as inferred from the power law degree distribution elicited. The HIV epidemic has grown from existing clusters with large size, degree, and transmission speed. Disassortative age mixing could have predisposed younger individuals to transmission risk from older patients.FundingResearch Grants Council of Hong Kong (14103315), AIDS Trust Fund (MSS 229 R, 243 R).