Improving estimates of district HIV prevalence and burden in South Africa using small area estimation techniques

Abstract

Many countries, including South Africa, have implemented population-based household surveys to estimate HIV prevalence and the burden of HIV infection. Most household HIV surveys are designed to provide reliable estimates down to only the first subnational geopolitical level which, in South Africa, is composed of nine provinces. However HIV prevalence estimates are needed down to at least the second subnational level in order to better target the delivery of HIV care, treatment and prevention services. The second subnational level in South Africa is composed of 52 districts. Achieving adequate precision at the second subnational level therefore requires either a substantial increase in survey sample size or use of model-based estimation capable of incorporating other pre-existing data. Our purpose is demonstration of the efficacy of relatively simple small-area estimation of HIV prevalence in the 52 districts of South Africa using data from the South African National HIV Prevalence, Incidence and Behavior Survey, 2012, district-level HIV prevalence estimates obtained from testing of pregnant women who attended antenatal care (ANC) clinics in 2012, and 2012 demographic data. The best-fitting model included only ANC prevalence and dependency ratio as out-of-survey predictors. Our key finding is that ANC prevalence was the superior auxiliary covariate, and provided substantially improved precision in many district-level estimates of HIV prevalence in the general population. Inclusion of a district-level spatial simultaneously autoregressive covariance structure did not result in improved estimation.

Fig 1. Relative standard errors of district-level estimates.

Relative standard errors (RSE) of the 52 South African district-level HIV prevalence estimates obtained from the AIC-best Fay-Herriot area-level model and direct (survey domain) estimation.

Fig 2. HIV prevalence in the districts of South Africa, 2012.

District-level estimates of HIV prevalence and 95% confidence intervals from the 2012 South African National HIV survey, and from Fay-Herriot small-area estimation. Numeric values are provided in S2 Table.

Fig 3. HIV burden in the districts of South Africa, 2012.

District-level estimates of numbers of people living with HIV (PLHIV) and 95% confidence intervals from the 2012 South African National HIV Survey and from Fay-Herriot small-area estimation.

Fig 4. Geographic distribution of HIV prevalence and burden.

Geographic distribution of HIV prevalence (top) and numbers of people living with HIV (PLHIV, bottom) in South Africa, 2012, based upon Fay-Herriot small-area estimation. Numeric values of estimates are provided in S2 Table.