Populations in Sub-Saharan Africa are highly mobile. Therefore, an individuals’ “social/sexual community” can consist of multiple communities: we define such a social/sexual community as a mega-community (MC). If MCs exist, they would geographically structure a generalized HIV epidemic into loosely connected sub-epidemics. This would have significant implications for designing effective epidemic control strategies. Here we use mobile phone data from Namibia, where HIV prevalence is ~14%, to search for MCs.
Call detail records (CDRs) were collected from 90% of subscribers in Namibia over a 12-month time period; they represent 9 billion communications from 1.19 million unique SIM cards. We analyzed these data at the constituency level, from all constituencies with cell towers: 96 out of 110. Our analyses enabled us to discover countrywide travel patterns, and determine how much time travelers from each constituency spent in other constituencies. We then used a technique from network science (a community-detection algorithm) to determine if MCs exist.
Residents of Namibia spent ~22% (median, Interquartile range: 18-26%) of their time outside their home constituency, over a year. Population-level travel patterns divide Namibia’s population of 2.5 million into eight MCs that vary in size (15,000 to 650,000 individuals) and compactness (figure). Namibia’s generalized epidemic consists of eight connected sub-epidemics: each contained within a MC. We were also able to identify “bridges” that link sub-epidemics: a bridge is a constituency in one MC that is linked, by travel, to another MC. We identified two “types” of bridges (figure): short bridges (linked constituencies are spatially contiguous) and long bridges (linked constituencies are separated by at least one constituency). Notably, the capital of Namibia (Windhoek) is a long bridge and connects to six MCs and sub-epidemics. Oshakati, the capital of one of 14 regions in Namibia, is also a long bridge; it is connected to five MCs and sub-epidemics.
As a result of travel patterns, the population of Namibia is divided into MCs. These MCs are not visible, but they spatially structure Namibia’s generalized epidemic into eight loosely connected sub-epidemics. This suggests that interventions may be the most effective if they are implemented at the level of the MC. Furthermore, it may be very difficult to reduce incidence in Windhoek and Oshakati, as they are connected (by travel) to multiple other sub-epidemics.