Several African countries have henceforth one Demographic and Health Survey (DHS) in general population, with HIV testing and GPS data. Traditional spatial interpolation methods are not adapted to the sample surveys. So different ways are necessary to represent the epidemics on a map.
We worked out two distinct approaches. The first reestimates the prevalence of a point from near points until having a sufficient number of observations. The second is a logistic regression including only spatial variables : region, urban/rural and distance to a city. In the both cases, a kriging interpolation is used to make an isopleths map. Precision indicators, confidence interval and seropositive population density are also generated.
All analysis was made under R software (free and open source project).
The results of two countries (Burkina Faso and Cameroon) were, with the collaboration of the national experts of these countries, compared with empirical knowledge of the epidemics. Lastly, this methodology was applied to a virtual country where DHS data were simulated.
Our first results highlight the spatial heterogeneity and diversity of the epidemics at national and regional levels. They point out some assumptions about HIV diffusion (from urban to rural, from some regions to others...).
They also invite to reconsider the weight of antenatal clinical attendees (ANC) surveillance in national estimation and they partly explain differences observed between DHS and ANC data.
Despite several limits (continuity assumption, non-household population exclusion and impossibility to bring to light very localized differentials), this method optimizes the use of DHS data, provides a more precise vision of the epidemics and constitutes a useful tool for programmatic action and analysis of sentinel surveillance biases.
However, general population-based surveys need larger samples to understand the real spatial structure of the HIV epidemics better.