Quantifying HIV transmission flow between high-prevalence hotspots and surrounding communities: a population-based study in Rakai, Uganda

Abstract

Background: International and global organisations advocate targeting interventions to areas of high HIV prevalence (ie, hotspots). To better understand the potential benefits of geo-targeted control, we assessed the extent to which HIV hotspots along Lake Victoria sustain transmission in neighbouring populations in south-central Uganda.

Methods: We did a population-based survey in Rakai, Uganda, using data from the Rakai Community Cohort Study. The study surveyed all individuals aged 15-49 years in four high-prevalence Lake Victoria fishing communities and 36 neighbouring inland communities. Viral RNA was deep sequenced from participants infected with HIV who were antiretroviral therapy-naive during the observation period. Phylogenetic analysis was used to infer partial HIV transmission networks, including direction of transmission. Reconstructed networks were interpreted through data for current residence and migration history. HIV transmission flows within and between high-prevalence and low-prevalence areas were quantified adjusting for incomplete sampling of the population.

Findings: Between Aug 10, 2011, and Jan 30, 2015, data were collected for the Rakai Community Cohort Study. 25 882 individuals participated, including an estimated 75·7% of the lakeside population and 16·2% of the inland population in the Rakai region of Uganda. 5142 participants were HIV-positive (2703 [13·7%] in inland and 2439 [40·1%] in fishing communities). 3878 (75·4%) people who were HIV-positive did not report antiretroviral therapy use, of whom 2652 (68·4%) had virus deep-sequenced at sufficient quality for phylogenetic analysis. 446 transmission networks were reconstructed, including 293 linked pairs with inferred direction of transmission. Adjusting for incomplete sampling, an estimated 5·7% (95% credibility interval 4·4-7·3) of transmissions occurred within lakeside areas, 89·2% (86·0-91·8) within inland areas, 1·3% (0·6-2·6) from lakeside to inland areas, and 3·7% (2·3-5·8) from inland to lakeside areas.

Interpretation: Cross-community HIV transmissions between Lake Victoria hotspots and surrounding inland populations are infrequent and when they occur, virus more commonly flows into rather than out of hotspots. This result suggests that targeted interventions to these hotspots will not alone control the epidemic in inland populations, where most transmissions occur. Thus, geographical targeting of high prevalence areas might not be effective for broader epidemic control depending on underlying epidemic dynamics.

Funding: The Bill & Melinda Gates Foundation, the National Institute of Allergy and Infectious Diseases, the National Institute of Mental Health, the National Institute of Child Health and Development, the Division of Intramural Research of the National Institute for Allergy and Infectious Diseases, the World Bank, the Doris Duke Charitable Foundation, the Johns Hopkins University Center for AIDS Research, and the President's Emergency Plan for AIDS Relief through the Centers for Disease Control and Prevention.

Figure 1

Study design (A) Locations of the RCCS in eastern Africa (left) and the Rakai region of Uganda where the RCCS survey was done (right). The RCCS included an estimated 75·7% of populations in the lakeside area within 3 km of the Lake Victoria shoreline (light brown), and 16·2% of populations in the inland area of the Rakai region (light green). Areas classified as external in this study are shown in light blue. Not shown is one RCCS community northwest outside the map, in which virus sequences were not obtained. (B) The phyloscanner approach for inferring directed HIV transmission networks from deep sequence phylogenies based on ancestral relationships between infecting viruses. With viral deep-sequencing, co-circulating HIV lineages within hosts are represented by many distinct sequence fragments in the data (diamonds, size indicating frequency with which distinct virus was sequenced). In the corresponding phylogenies, sequences from the same individual tend to form subtrees (colours, one for each of the six individuals shown). The ordering of subtrees provides evidence of the direction of transmission. (C) Scale of in-migration into the cohort. For this purpose, RCCS participants were classified as in-migrants if they in-migrated into the cohort in the 2 years before their first visit in the observation period, and otherwise as residents. The panel shows the proportion of in-migrants and residents as well as the size of the population infected with HIV. (D) Key study outcomes including participation, sequencing, and linkage rates. RCCS=Rakai Community Cohort Study.

Figure 2

HIV prevalence and migration in inland and fishing communities (A) Estimates of HIV prevalence in RCCS communities for men (blue) and women (pink) in inland communities (left panel) and fishing communities (right panel). Boxplots indicate central estimates (black bar), IQRs (box), and 95% credibility intervals (whiskers). HIV prevalence was substantially higher in fishing communities for both men and women. (B) Number of RCCS participants in inland and fishing communities by in-migration status. Participants who in-migrated within 2 years before study visit were stratified by the origin of migration, from inland communities (green), from fishing communities (purple), from outside the Rakai area (light blue), and from unknown location (grey). (C) Estimates of HIV prevalence among in-migrants to inland communities to that among in-migrants to fishing communities. HIV prevalence was higher among those individuals migrating to fishing communities than those migrating to inland communities. Sex specific estimates in panels A and C were obtained with Bayesian logistic regression models using the Stan software, version 2.19.

Figure 3

Phylogenetically highly supported transmission flows in the population-based sample, and predicted transmission flows Viral deep-sequence phylogenetics identified 293 source–recipient pairs with strong phylogenetic support for epidemiological linkage and the direction of transmission. Transmission events were geo-located to the communities in which the phylogenetically likely sources and recipients had their households, or to the origin of recent in-migration events. (A) Phylogenetically reconstructed transmission events. 94 phylogenetically reconstructed transmissions events occurred from inland to inland communities, and six occurred from outside the Rakai area to inland communities; seven were observed from fishing to inland communities; 23 occurred from inland to fishing communities; 141 occurred from fishing to fishing communities, and 17 from outside the Rakai area to fishing communities. Not shown are two phylogenetically probable transmission events with unknown source location to inland communities, and three such events to fishing communities. (B) Predicted transmission flow ratio among populations living in inland and lakeside areas of the Rakai region, after adjusting for survey, participation, and sequence sampling bias. The predicted flow ratio of transmissions from inland to lakeside areas compared with the opposite direction was 2·50 (95% CrI 1·02–7·30).

Figure 4

Effect of sex and migration on transmission flows (A) Estimated sources of transmission in inland and fishing communities of the RCCS. (B) Estimated amount of cross-community transmissions between inland and fishing communities originating from residents with partners outside their community and from in-migrants. Estimates in both panels were obtained as described in the appendix (p 11), and adjusted for heterogeneity in participation and sequence sampling. In fishing communities, an estimated 33·6% (95% CrI 26·7–40·7) of transmissions originated from resident women, 43·7% (36·7–51·1) from resident men, 10·5% (6·5–15·8) from in-migrating women, and 11·7% (6·5–15·8) from in-migrating men. In inland communities, an estimated 24·0% (95% CrI 16·7–33·5) of transmissions originated from resident women, 54·2% (44·7–63·6) from resident men, 10·1% (5·1–17·3) from in-migrating women, and 10·9% (5·6–18·4) from in-migrating men. Boxes are 50% CrI and whiskers are 95% CrI. RCCS=Rakai Community Cohort Study. 95% CrI=95% credible interval.