Development and Evaluation of a Digital HIV Risk Assessment Tool incorporated within an App-Based Self-Testing Program.

Abstract

Low risk perception is an important barrier to the utilisation of HIV services. In this context, offering an online platform for people to assess their risk of HIV and inform their decision to test can be impactful in increasing testing uptake. Using secondary data from the HIVSmart! quasi-randomised trial, we aimed to identify predictors of HIV, develop a risk staging model for South African township populations, and validate it in combination with the HIVSmart! digital self-testing program. Townships in Cape Town, South Africa. Using Bayesian predictive projection, we identified predictors of HIV and constructed a risk assessment model that we validated in external data. Our analyses included 3095 participants from the HIVSmart! trial. We identified a model of five predictors (being unmarried, HIV testing history, having had sex with a partner living with HIV, dwelling situation, and education) that performed best during external validation (AUC-ROC, 89% CrI: 0.71, 0.68 - 0.72). The sensitivity of our HIV risk staging model was 91.0% (89.1% - 92.7%) and the specificity was 13.2% (8.5% - 19.8%); but increased when combined with a digital HIV self-testing program, the specificity was 91.6% (95.9% - 96.4%) and sensitivity remained similar at 90.9% (89.1% - 92.6%). This is the first validated digital HIV risk assessment tool developed for South African township populations and the first study to evaluate the added value of a risk assessment tool with an App-based HIV self-testing program. Study findings are relevant for application of digital programs to improve utilization of HIV testing services.