A Comparison of Results from Two Sampling Approaches in the South African National HIV Prevalence, Incidence and Behavior Survey, 2012

Abstract

Background: South Africa implements variations of second generation suveilance surveys to monitor human immunodeficiency virus (HIV) epidemic. Objective: This paper compares HIV estimates from two design variations: take all approach and sub-sampling approach to ascertain if any changes in HIV epidemic are due to methodological changes or inherent evolution of the epidemic. Methods: A multi-stage stratified cluster sample of 1000 census enumerator areas was implemented with 15 households systematically sampled within each census enumerator area. In each household, every member was invited to participate (take all approach). To compare to the previous survey designs, a sub-sampling approach of at most four people from each household was implemented by randomly sampling one person from each age group: Results: HIV estimates were comparable with no systematic pattern. Prevalence estimates were slightly higher 12.2% [11.4% - 13.1%] in the take all compared to 11.6% [10.6% - 12.6%] in the sub-sampling approach. Estimates from sub-sampling approach were more variable. The design effects in the take all approach were also slightly higher than those obtained in the sub-sampling. The overall synthetic measure of homogeneity for both methods was ρ = 0.10. Conclusion: In conclusion, as the household size increases the number of people living with HIV in each household increases thus increasing intraclass correlation. Similarity of resulting HIV estimates is re-assuring. However, the take all approach is more preferable than sub-sampling approach as it allows for detailed analyses of HIV data such as estimating discordance between sexual partners and parent-child pair.