Essays on HIV in Rural KwaZulu-Natal, South Africa: Inference From Measurement to Policy

Abstract

The global response to HIV is in a transition state, moving from the rapid roll out of HIV treatment services to sustained delivery and expansion into new populations. This dissertation analyzes three scenarios in which improved insights into data collection, analysis, and policy are likely to yield improved health outcomes in the context of HIV/AIDS in rural KwaZulu-Natal, South Africa. Firstly, we investigate HIV health system gaps using a longitudinal transition-by-transition cascade of care, a concept which breaks down the pathway from infection to therapeutic response into discrete steps. We demonstrate biases due to transition time in the standard cross-sectional approaches, and recast the cascade using patient flow between stages using survival analysis. We find that the strongest bottleneck in the cascade is becoming linked to care, and that these transitions are likely to be slowing over time. Secondly, we conduct a field experiment to identify whether list randomization can reduce social biases in survey responses related to HIV, using known truths for comparison. List randomization is a survey method intended to mitigate respondent biases by shrouding patient responses in a longer list of questions. We pilot this method using surveillance-derived known status for HIV status. We find list randomization did not accurately reveal known true statuses and behaviors. Finally, we investigate whether the qualification structure of the South African disability grant program, in which qualification for some requires having a low CD4 count, may have negatively impacted HIV health in certain circumstances by disincentivizing antiretroviral therapy adherence. We assess this empirically using a difference-in-difference of recovery rates between disability grant recipients and non-recipients specific to the qualification threshold, finding that manipulation of recovery rates due to the threshold is likely to have occurred.