Shared-Component Model With Application To Mapping Gender Specific Pattern In Hiv Testing And Condom Use In Namibia

Abstract

The main objective of the study was to examine gender-specific and shared spatial variation in HIV testing and condom use in Namibia for targeted health promotion interventions.

The study used data from the Namibia Demographic and Health Survey (NDHS) carried out between 2006 and 2007. Bayesian hierarchical spatial mapping techniques were applied to generate specific and shared spatial patterns in HIV testing and condom use. Particularly, a number of Bayesian Structured Additive Regression (STAR) models were fitted and followed by joint spatial models through the shared component latent variables approach. Firstly, we modelled HIV testing and condom use in males and females with fixed effects such as educational level, frequency of reading newspapers and magazines, frequency of listening the radio, frequency of watching television, wealth index, times away from home, smoking, alcohol consumption, employment status, age at first sexual intercourse, type of residence and marital status whereas spatial references to the communities were modelled as structured and unstructured spatial effects. Secondly, diffuse priors were assumed for the fixed effects, while conditional autoregressive priors were assigned to the structured spatial effects and exchangeable priors for the unstructured random effects. Simulation techniques through Markov Chain Monte Carlo were applied for model estimation. Common and divergent patterns of HIV testing and condom use emerged. Common areas among men and women on HIV testing and condom use were observed in Khomas, Erongo, Oshikoto, and Oshana, while divergent patterns were estimated in Caprivi, Kavango and Karas regions. Urban influence was also captured in the model. Exposure to media was one of the covariates that were found to have a positive effect on the use of condoms and HIV testing.

The study underscore the usefulness of Bayesian hierarchical mapping model in highlighting areas lagging behind in the uptake of HIV testing and condom use with emphasis on

ii

differences between men and women in the same area. We found that compared with gender-specific modeling approach, the shared component model offered useful additional information when modeling HIV testing and condom use in men and women. The information will be valuable for guiding public health actions that are targeted at the overall reduction of risk-sexual behaviours through HIV testing and the use of condoms.