BILAL MATEEN: African health risks being left behind in AI arms race

The race is on to see how AI can help tackle disease and poor health in Africa and other low- and middle-income regions. Entrepreneurs across the continent are already developing AI tools that could help to close the gap between healthcare access and outcomes for people in high-income countries and those in the rest of the world.

Yet, just as we reach the point where we might begin to see results, the rules and norms that have underpinned global co-operation on innovation, development and technology are being rewritten. Sweeping funding cuts, reductions in foreign aid budgets and a rapidly accelerating AI arms race based on nationalist principles and priorities threaten to scotch potential gains, and instead widen inequality and attendant suffering.

Faced with these headwinds, developing the evidence base for the value AI tools can deliver in healthcare settings, and which AI innovations can be localised and harnessed to reduce clinical workloads and improve outcomes for patients in low- and middle-income countries, is even more urgent.

AI-based approaches to health hold incredible promise but are unevenly deployed, evidenced and understood. AI is making rapid advances in bioscience areas such as drug design and development, but the same is not true in clinical healthcare. While the effectiveness of new technologies in bioscience is continually being validated, the evidence base for clinical applications of AI is meagre at best.

This is particularly true of investments in AI that have the potential to advance equitable health outcomes in low- and middle-income countries. I believe AI at scale could radically reduce the gap in healthcare outcomes between high-income countries and low- and middle-income countries, yet to date just three randomised controlled trials of AI tools have been evaluated in African healthcare settings.

We simply don’t know which AI tools are most effective, or when the development of an AI tool would be the most cost-effective use of resources. This is particularly true when it comes to large language models (LLMs) - a form of generative AI tool trained using huge amounts of data to be able to provide intelligent and creative responses to prompts, like a human. Given that these models pick up information from the environment where they have been primarily used and trained, where and how they are developed has profound implications.

The problem to date is that AI tools in healthcare are being built predominantly for and by people in English-language speaking developed economies — the same people for whom drugs have been developed over the past 100 years. There is no guarantee that these LLMs will be helpful to frontline health workers in African countries (even English-speaking ones), especially when they default to US-based websites (in more than 85% of cases) for reference information.

Furthermore, though maturing, the current standard of evidence for the effectiveness of LLMs in healthcare settings relies too heavily on their ability to answer multiple-choice questions or simulated clinical interactions rather than deliver documented benefits in patient outcomes.

A groundbreaking randomised controlled trial in Kenya, launched by Path, a global nonprofit health organisation, in collaboration with a consortium of Kenyan research and clinical partners, seeks to help plug this evidence gap and demonstrate that African countries can lead the way in innovation in healthcare, rather than be treated as passive beneficiaries of Western-led innovation.

The trial will assess the impact of an LLM-based “co-pilot” designed to support clinicians in primary healthcare settings to make decisions about the care of the 9,000 participants. The AI tool will provide treatment and diagnosis recommendations based on the patient’s health history, clinical visit notes and any laboratory results.

This will be the largest trial of its kind to take place in Africa, in a clinical setting on real patients, in real time, at scale. It will bolster the evidence base to incentivise and derisk greater funding for developing AI tools for healthcare in low- and middle-income countries. It will also help to demonstrate the value that generative AI tools can add: from enabling better quality decision-making, to driving improvements in patient outcomes, and all at a cost-effective price point to ensure we make the most of scarce resources.

The risk is not that investments in AI for health won’t happen, the promise is simply too great to resist. The risk is that without developing the evidence base, we will squander the limited resources we have and undermine efforts to prevent backsliding on the incredible progress made over the past 20 years in tackling some of the world’s deadliest diseases that predominantly affect people in Africa and other developing regions. This trial is an incredibly exciting step towards addressing and mitigating those risks.

• Mateen is chief AI officer at Path and honorary associate professor of machine learning for health at University College London.