Current HIV related challenges: The Potential of AI to Transform
Co-authored by Dr Dino Rech, CEO of Audere, and ?Dr Tigistu Adamu Ashengo, CMO at Jhpiego
IAS 2024: REFLECTING ON HIV AND AI
Current HIV related challenges: The Potential of AI to Transform?
It’s the fortnight after the IAS 2024 conference in Munich where 15000 HIV scientists, advocates, implementers, and government leaders came together once again to take stock of one of the most intense public health battles of our generation: The HIV epidemic and its control and elimination. The delegates had an opportunity to engage on how we address both new and old challenges, and how the latest science is guiding our efforts and directing us to where we can look forward to the impact of new innovations. All in an effort to end HIV as a public health threat by 2030. One new global innovation featured at the conference was the potential of diverse AI functionalities to address existing and emerging challenges facing HIV delivery programs.
The conference theme was “ Putting People First”?
With this theme in mind we will attempt to unpack some of the programming challenges that continue to stand out to us and where we could explore AI as a solution in our toolbox – not as a proverbial silver bullet – but rather as an important augment to a growing list of interventions aimed at genuinely “Putting People First”:?
It’s been obvious for a while that "The single most unused person in health care is the patient"? David Cutler, in MIT Technology Review.
Many advancements were shared at IAS - including exciting new prevention options like Lenacapavir - but we need to ensure strong client demand and access channels for their eventual arrival. Below we unpack some of the AI innovations available to help accelerate demand, ensure quality across decentralised channels, and link these innovations to these programming challenges to highlight the potential for impact.
Not All AI is the Same - but each offers a distinct value
AI is a broad term which encompasses a number of functional areas. Probably the most hyped of all is the emergence of AI language models such as OpenAI’s ChatGPT, Google’s Gemini, Anthrophic’s ClaudeAI and a growing library of large, small, and open language models. These tools have the potential to facilitate bi-directional conversations - mimicking the most knowledgeable, empathetic humans. The ability to provide, monitor, and improve on quality conversations is essential in any healthcare interaction and arguably even more so in HIV.
Another form of AI is computer vision, which uses a camera or other imaging modality as remote “eyes” assisting to quality assure, audit, interpret, and support human diagnostic capabilities. Examples include Computer Assisted Diagnostics (CAD) for X-ray interpretations. Another is the potential of computer vision to support rapid point of care testing and self-testing journeys with linkages to remote care via telehealth.
Another key AI functionality is predictive analytics and big data analyses driven by machine learning, leveraging historical data to identify patterns or predict future outcomes. An example could be individual data related to likelihood of defaulting from HIV treatment. If a model predicts a higher risk of defaulting, this could trigger a pre-emptive intervention - such as an outreach from a dedicated peer mentor.
This is by no means an exhaustive list of AI functionalities or use cases, but is meant to provide a grounding of functionalities leveraged today as we explore opportunities around improvements and impact.
Augmenting HIV Care - how AI can address programmatic challenges
Despite significant efforts to reduce stigma in HIV programming, many youth and key populations still face stigma when accessing care. Personal and institutional biases heavily influence areas of delivery, access points, and HIV-related conversations, creating barriers to effective engagement and support. Sex, STIs, and HIV are deeply personal and sensitive – conversations around these topics within communities are crucial to direct personalised care. However, we know these conversations may not be happening due to fear of judgement or when they are, may be limited in honesty or depth (Mahajan et al., 2008 ; Nyblade, Stangl, Weiss, & Ashburn, 2009 ; Logie & Gadalla, 2009 ; Turan et al., 2017 ).?
In South Africa, a recent study conducted by the University of the Witwatersrand (Indlela HE2RO team) placed an AI counsellor (“Your Choice” ), powered by language models, in the hands of users to provide a stigma-free, private space for conversation in order to extract sexual history and behaviours with the aim of assessing HIV vulnerability counselling on prevention options including PrEP. 130 users engaged with the AI counsellor, which was directed to empathetically gather sensitive information, answer user questions, and summarise? responses for clinicians. The research focused on two key questions: First, how would individuals feel and respond to engaging with a faceless AI on delicate and sensitive matters, and could it reduce stigma to encourage more honest answers? Second, could this approach improve efficiency in busy facilities where time for extended, quality HIV conversations is typically limited? The findings were highly encouraging, with over 90% of clients and health workers finding the AI not only acceptable, usable, and appropriate but also highly desirable (Govathson et al., 2023 ).
The notion that conversational AI tools can partner with humans to help reduce stigma and enhance honest engagement is encouraging and certainly one which merits more investment in scaled research across target populations to understand how we can meet the diverse needs globally.
领英推荐
As mentioned above, differentiated delivery of HIV services has been topical for a number of years. In essence, the notion envisages personalised target services for specific individual needs and context, making accessing and using HIV delivery interventions (both prevention or treatment focused) easier, more accessible, and convenient. This is a good notion in theory but relies on being able to correctly identify and understand the needs and context of the person being targeted and have them reflect and have the insight to understand their own risk profile and circumstances. Current AI tools offer a range of functionalities in HIV care. First, they can use available data to better identify at-risk individuals and target resources and interventions accordingly. Second, in the absence of sufficient data, AI can generate new data and insights through AI-driven conversations. Finally, with quality data, AI can guide decision support and tailor personalised intervention packages and engagement efforts (nudges) to maximise consistent engagement in prevention or treatment modalities.
Phylaxis.ai , a South African healthcare technology company, has developed and validated a machine learning model for HIV risk assessment using data from digital surveys. Their model achieved an AUROC [metric used to assess machine learning model performance] of 82.8% in predicting individuals at risk of contracting HIV, effectively identifying and analysing key risk factors (Majam et al., 2023 ). This work demonstrates the potential of AI to enhance HIV outcomes through targeted interventions and efficient resource allocation. The impact of Phylaxis' work is multifaceted, offering solutions for targeted testing, PrEP prioritisation, and personalised interventions. By accurately identifying high-risk individuals, the model can help allocate limited testing resources more efficiently and guide decisions on PrEP initiation (Fieggen et al., 2022 ). Additionally, the model's ability to break down risk factors at individual and community levels allows for tailored prevention strategies and counselling, aligning with the goal of providing personalised, efficient, and effective HIV prevention and care (Majam et al., 2023).?
These types of tools are already being deployed in diverse contexts, such as supporting HIV case finding in Ukraine (Holt, 2024 ), and are part of the AI HIVE initiative [discussed below]. AI enhances the precision and effectiveness of HIV care by providing personalised, accessible, and convenient interventions tailored to individual risk profiles and circumstances.
Enhanced access to HIV services through decentralised models is a critical strategy in improving the uptake of testing, prevention, and treatment services. By increasing convenience and affordability, these models can significantly boost successful engagement with HIV care. For instance, pharmacy channels and self-testing pathways have shown promise in expanding access and providing confidential, private testing with effective linkages to care.
Pharmacy channels have become essential for delivering HIV services such as testing, prevention, and treatment (PEP, PrEP, and ART). These services are accessible both in physical locations and through innovations like E-Pharmacy and TelePrEP models (Maverick Article, 2023 ; Ortblad et al., 2020 ; Omollo et al., 2023 ). Embedded in communities and trusted by the public, pharmacies offer convenience and privacy, making them ideal access points for HIV care. Furthermore, the ePrEP program , which integrated HIV self-screening with telehealth counselling and home delivery of PrEP, exemplifies how leveraging pharmacy channels can enhance access and ensure continuous engagement and adherence.?
Self-testing and associated care pathways enable individuals to conduct HIV tests in the privacy of their homes, maintaining confidentiality and agency over their health. These pathways provide linkage to care, including telehealth consultations and PrEP access through remote prescriptions and delivery or pick-up options. The ePrEP pilot in Kenya - which partnered with MyDawa, Fred Hutch, Jhpiego and Audere - demonstrated the impact of AI in verifying test results and linking them to individual users, thus building trust and facilitating wider adoption. When considering self-care and testing, AI plays several crucial roles, such as providing advice, supporting testing, offering personalised care, and ensuring linkages to further services. This example showcases the integration of AI in self-testing and care pathways, emphasising the importance of personalised, confidential services that enhance user engagement and adherence (Mendonca et al., 2024 ).
Decentralised models leveraging AI can significantly enhance access to HIV services, ensuring they are more convenient, confidential, and effective. By embedding services in pharmacies and utilising online platforms, these models can overcome traditional barriers to HIV care and support sustained engagement in prevention and treatment efforts.
Sustainability is a critical question in HIV programming, particularly as donor contributions decline and government budgets remain restricted. While there is no single answer to the sustainability question, the solution likely lies in a multipronged approach. Intentional, expedited and technology supported task sharing to the people who need and coordinate their, and their family members, care remain a top priority. PEPFAR programs, for example, are heavily reliant on human resources embedded by implementing partners within communities to reach target populations. Specialised peer navigators form a large and growing part of client communication and information networks. They are effective because they offer empathy and support from individuals who have shared experiences. However, this model faces acute resource challenges. AI offers several avenues to scale successful peer navigator models without requiring additional human resources, thus reducing ongoing delivery costs. AI companions, both for clients and healthcare workers (HCWs) or peer navigators, can provide scalable and affordable information and support tools.?
Audere's projects using computer vision for interpreting malaria rapid diagnostic tests (RDTs) provide valuable insights . These projects have demonstrated how AI can assist HCWs in resource-limited settings, ensuring accurate and timely diagnoses without the need for additional personnel. By utilising AI technology, these projects have shown improved diagnostic accuracy and efficiency, which can be directly applied to HIV care. This concept is referred to as AI-facilitated Supportive Supervision. The "Your Choice" project utilises AI-driven large language models (LLMs) to create empathetic and stigma-free conversational agents. These AI counsellors engage users in private, confidential conversations to assess HIV vulnerability and provide counselling on prevention options, including PrEP. This model not only reduces the stigma associated with HIV testing and counselling but also improves the efficiency of service delivery by providing accurate and empathetic interactions without the need for extensive human resources. Similarly, the AI HIVE (AI for Health Information, Virtual Testing, and Education) project, currently underway in South Africa, aims to extend the capabilities demonstrated in both the "Your Choice" project and Audere's malaria projects by integrating LLM-powered counsellors and computer vision technology into one comprehensive toolkit. These AI-driven tools will support HIV self-testing and counselling, enhancing diagnostic accuracy and providing empathetic interactions.?
AI offers a multipronged approach to enhancing the sustainability and efficiency of HIV care programs. These AI-driven solutions not only address the acute resource challenges faced by traditional peer navigator models but also enhance diagnostic accuracy and reduce stigma. As such, AI can play a crucial role in sustaining and scaling HIV care efforts, ensuring continued progress in the fight against HIV.
Where does this all leave us? What next??
The evolution of AI, particularly Generative AI, is rapidly transforming every sector of modern life. This shift demands careful consideration, extensive research, ethical and inclusive design, and a clear understanding of both its potential benefits and risks.
Ignoring or merely dabbling in AI would be to our detriment and certainly a lost opportunity. We commend optimistic donors for funding innovative AI explorations. However, much of this funding is focused on small-scale pilots. Often the question presented is: how quickly can we see a return on AI investments? We must shift our mindset from questioning if AI will have an impact, to understanding how it will. By embracing AI as an integral part of our HIV delivery strategy and taking a multi-threaded approach, including incorporating the value system of often marginalised communities, we can uncover significant insights and advance toward equitable access, improved sustainability and true epidemic control. Let's not let caution hinder progress; instead, let's boldly invest in learning and exploration as our immediate outcomes to maximise AI's potential in transforming HIV care.
Co-authored by:
?Dr Dino Rech ([email protected] ), CEO of Audere
?Dr Tigistu Adamu Ashengo ([email protected] ) CMO at Jhpiego
#TheNigerianProject
2 个月Great read! One point I don’t see emphasized a lot is the patient’s willingness to accept these tools, given the some times resistance in adopting new technologies. AI would reduce the human interaction, which arguably reduces stigmatization, but the somewhat lack of trust in these AI tools might pose a significant barrier to uptake and scale, if the goal really is to target low resource contexts.
Deputy CEO, Foundation for Professional Development (FPD); Senior Technical Advisor: Climate and Health, Wits RHI
3 个月Very insightful topic. Putting people first is key to successful HIV epidemic control and AI is a critical enabler of this. The authors are correct, the question should not be if we use AI but how we can effectively apply AI to address persistent historical barriers to access to HIV care.
Creating @ NhakaBox | Global Health | Program Management | Human Centred Design | Digital Transformation
3 个月Thank you for this thoughtful reflection on the potential of AI to transform HIV services. There are two caveats I would add to introducing AI moderated service delivery models in LMICs. The first is that we cannot ignore that the quality of AI outputs depends on the data used to train them. The quality of data is important and that is a challenge that needs to be addressed either with AI or before it is widely deployed. Second, the widespread use of mobile phones in many African contexts does not necessarily translate to a readiness to engage in health experiences mediated by digital technologies. Along with leveraging all the strengths of generative AI, it is also important to address the real concerns that people may have about sensitive data being linked directly to them in community settings, how that data is secured and stored, and who has access to their data in the cloud and on their devices. TLDR: AI offers great opportunities for transforming HIV & other care but still has to overcome the challenges of data quality, data protection, and client/patient education.