Using mobile phone data to identify social-sexual communities: implications for controlling HIV epidemics

UNAIDS’ Global AIDS update states that efforts to increase HIV prevention are effective when “resources are strategically targeted where they can have the maximum impact”.¹ This is implemented by prioritizing geographic areas where people are at the highest risk of HIV, and is most frequently implemented at the administrative level. It does not consider mobility, although individuals cross administrative boundaries. Mobile phone data can be used to identify circular travel patterns and to reveal hidden spatial structures in populations: to detect meta-communities. A meta-community is a subset of communities that are more tightly linked to each other (due to individuals travelling amongst them) than to other communities; meta-communities are loosely linked (through travel) to other meta-communities. We propose that a meta-community be considered as a single “social-sexual community” (a group of individuals who have the majority of their social/sexual contacts within the group), and that a linked network of social-sexual communities can be used to design geographical targeting strategies for controlling HIV. We discuss this for Namibia.

Namibia has a generalized HIV epidemic: 12·6% of adults live with the virus.² We used mobile phone data from Namibia, previously used to understand malaria dynamics³ and reveal HIV risk networks.⁴ The data are records from 9 billion calls and texts over a 12-month period.

We used methods from network science to analyze the mobility network and identify social-sexual communities.⁵ These methods have been used to find meta-communities in mobility networks in many countries: e.g., Great Britain,⁶ France,⁷ the Ivory Coast,⁷ USA.⁸ Namibia is geographically subdivided into eight social-sexual communities (figure A). They vary in terms of geographic area, the number of constituencies they contain (2 to 33), and size (15,000 to 650,000 individuals) (figure B). They also vary substantially in their degree of urbanicity: e.g., the red and orange communities are predominantly urban (because the majority of travel is between urban constituencies), the grey and green communities are predominantly rural (because the majority of travel is between rural communities).

Figure: Social-sexual communities in Namibia.

(A) Namibia is shown partitioned into eight social-sexual communities. Each constituency is color-coded to show its membership in a specific social-sexual community. Bridge constituencies are denoted with circles; the color within each circle is the color of the linked social-sexual community. For example, the red community containing a purple circle shows the geographic location of the constituency that is a bridge to the purple community. This indicates that many people who live in the purple community spend time in the bridge constituency within the red community. (B) Population size of each social-sexual community by urban-rural status. (C) Network diagram showing all constituencies (circles; color-coded by social-sexual community as in panels A and B). Lines show the links within and across communities.

The social-sexual communities are linked by bridge constituencies; a bridge constituency is defined as having significant mobility ties with a social-sexual community other than the one they belong to. Figure A shows the geographic locations of bridge constituencies and the social-sexual communities they are linked to. Figure A shows two “types” of bridges: short bridges connect spatially contiguous constituencies, long bridges connect non-contiguous constituencies. Notably, the capital of Namibia (Windhoek) is a long bridge constituency connecting six social-sexual communities. The network diagram (figure C) shows the constituencies clustered into the eight social-sexual communities with lines between and across the communities.

Considering a population as a network of linked social-sexual communities provides a new understanding of generalized HIV epidemics. It suggests that they can be viewed as linked networks of sub-epidemics, where each sub-epidemic is associated with a specific socio-sexual community and sub-epidemics are connected to other sub-epidemics in specific geographic locations (at the bridge constituencies). Identifying social-sexual communities may also help identify the most important “type” of transmission occurring: e.g., if the community contains an urban center and rural villages this suggests urban-to-rural/rural-to-urban transmission is important.

We propose social-sexual communities should be targeted, rather than, as currently occurs, geographic areas delimited by arbitrary administrative boundaries. The spatial scale of the targeting strategy should reflect the size of the geographic area each community occupies. The identification of network bridges will show which sub-epidemics are linked; this could be used to determine where it would be most beneficial to implement synchronized strategies. Bridges should be targeted to reduce the probability of “source-sink dynamics” occurring;⁹ these dynamics enable a sub-epidemic in one area to maintain a sub-epidemic in another area where transmission is too low to be self-sustaining. Additionally, targeting bridges could potentially prevent the movement of new strains between two, or more, sub-epidemics.

To date, distribution strategies of HIV prevention modalities in countries with generalized epidemics have considered geography, but exclusively in terms of administrative boundaries. The role of human mobility in shaping connected communities across such boundaries, and the impact of mobility as a driver of HIV risk acquisition, has been ignored. We recommend that to design more effective geographically targeted preventative interventions, large-scale mobility data should be collected, and social-sexual communities identified.