Very little is known at the present time about the role of the place of residence of mobile individuals in the transmission of HIV outside high transmission areas such as mining settlements, transport corridors, or poor urban and periurban communities. The main objective of this study is to bridge a widening knowledge gap that is caused by new mobility dynamics of men and women that live in rural South Africa.

This study makes use of data from one of the most comprehensive demographic surveillance site in Africa that is characterized by high adult HIV prevalence, high levels of poverty and unemployment and frequent residential changes. Its main objective was to determine which places of residence are predictive of HIV acquisition. Between 2004 and 2016, residence changes were recorded for 21,015 individuals over 105,614 person-years. These individuals were HIV negative at baseline. This is one of the largest HIV incidence cohorts in the world in terms of the number of individuals under surveillance, and the number of person-years of surveillance. Over the study duration, there were a total of 3,264 HIV seroconversions. We organized our data in two 48-dimensional contingency tables, one for men and one for women that cross-classify the study participants with respect to the locations of their residencies, their age, and whether they seroconverted. We used state of the art Bayesian methods for structural learning of graphical loglinear models to identify mobility graphs which encode the strongest multivariate predictive relationships supported by the data. Our analysis of the mobility graphs shows that whether men move farther away from their original places of residence is predictive of their likelihood of HIV seroconversion (OR = 2.003, 95\% CI = [1.718,2.332]), but similar residential changes do not seem to predictive of HIV seroconversion in women given their age. The location of the original place of residence is not a strong predictor for HIV acquisition in both men and women given knowledge of age and whether residential moves over longer distances have occurred.

The results of this study which is one of the largest individual-level longitudinal study of mobility patterns and HIV to date, provide evidence that geodemographic segmentation based on the history of residential locations, gender and age can constitute a reliable, objective, cost-effective way to ensure optimal allocation of HIV prevention intervention strategies.