Breaking barriers: harnessing artificial intelligence for a stigma-free, efficient HIV prevention assessment among adults in South Africa

Read the publication

Background

The quality of interactions between healthcare providers (HCPs) and recipients of care (ROC) are important for assessing HIV vulnerability and determining PrEP eligibility. However, these conversations are often limited because of high HCP workload and time constraints. Conversational agents powered by large language models (LLM) offer a promising solution to support such interactions. We explored the potential of an LLM-powered app prototype “Your Choice” to engage people who have recently tested HIV negative in discussions on HIV prevention and PrEP.

Objective

We assessed usability, acceptability, feasibility and appropriateness of the LLM-powered app prototype “Your Choice” in identifying HIV vulnerability and summarizing relevant information to support HCPs in designing personalised HIV prevention plans.

Methods

Using a human-centred design (HCD) approach, we co-developed the “Your Choice” app. Between August 2023 and March 2024, we conducted surveys with ROCs and HCPs following app use. We also analysed app-user interactions for themes and content patterns. Quantitative data were analysed descriptively; qualitative data underwent content analysis.

Results

We enrolled a total of 150 participants (130 ROC + 20 HCPs). Among the ROCs, most were male (77/130; 59.2%) and aged 18–34 years (93/130; 71.5%). ROCs rated the app highly acceptable and appropriate (AIM and IAM scores >4.9/5), with excellent usability scale scores (SUS >92/100, indicating exceptional usability). HCPs (n = 16) experienced in PrEP service delivery reviewed the summaries and the interactions with ROC from the app. In the pre-evaluation survey, only one expressed concern about trusting technology (1/16; 6.3%) and two thirds preferred their own judgement for ROC care (11/16; 68.8%). HCPs rated the app with a high usability scale score (SUS 78/100, indicating good usability). The app fostered open, stigma-free discussions on sex, sexuality, and HIV prevention.

Conclusion:

An LLM-powered conversational agent like “Your Choice” shows promise for private, stigma-free HIV prevention support, including PrEP uptake, supporting the decision-making around PrEP initiation. It can also help providers deliver more targeted care. Future research should address technological trust and clinical integration.