How responsible, equitable AI can help us improve global health outcomes

From predicting which patients are most likely to develop heart disease to creating personalised cancer vaccines, artificial intelligence (AI) is adding new capabilities to the health sector with astonishing speed. What’s more, it’s doing so at a time of mounting healthcare challenges, including a severe global shortage of frontline healthcare workers, widening health disparities and putting health systems under financial strain.

The good news is that a targeted analysis of more than 400 healthcare AI use cases shows that the technology exists today to address these and other healthcare ills – especially in low- and middle-income countries, where AI-related innovations could provide an opportunity for healthcare systems to progress at speed. Organizations in developing countries, such as India, Brazil, and Rwanda, are already leading the way in AI in health.

But those benefits can only be realized if leaders address the significant hurdles that stand in the way and form public-private partnership for ensuring responsible, ethical and equitable use, argues the World Economic Forum’s new insight report, Scaling Smart Solutions with AI in Health: Unlocking Impact on High-Potential Use Cases. The report was developed in collaboration with ZS, a global management consulting and technology firm.

There are several ways AI can improve health outcomes. First, AI can improve diagnosis and risk stratification. The best way to bend the healthcare cost curve and allow people to live healthier, longer lives is to treat more people before they become sick. AI holds vast, and largely untapped, promise to diagnose a range of diseases at scale — and earlier than clinicians. It can also suggest early interventions for those whose genetics, environment or behaviours place them at greater risk of falling ill.

Second, it can improve infectious disease intelligence. COVID-19 had a profoundly negative effect on global health in terms of lives lost, declining mental health, strained national health systems and unprecedented healthcare worker burnout. Climate change and human migration threaten to increase the risk of future occurrences of infectious diseases. AI-driven systems exist, however, that can predict outbreaks and map their spread (e.g. by testing wastewater, analysing web traffic and modelling mosquito movement patterns) and deliver customised mitigation suggestions.

Third, AI can increase clinical trial optimisation. Clinical trials are expensive, time-consuming and woefully under-representative of underserved groups and women. AI-powered clinical trials are already helping drug manufacturers select optimal trial sites, recruit and retain participants and create more representative synthetic data. The result will be a faster time to market for new therapies and treatments that work optimally across demographic groups.

In addition, AI-powered chatbots and call centres can answer patient questions and assess whether an evaluation is needed — alleviating the strain on frontline workers and health system resources, which could be particularly valuable in some low- and middle-income countries that have a shortage of frontline healthcare resources. The technology could also automate administrative tasks, which account for up to half of a doctor’s time.

AI tools based on deep learning are already uncovering insights about the mechanisms underlying disease, discovering new therapeutic assets and identifying the patient subgroups most likely to respond to a given treatment. AI also offers the promise of greater transparency into the medical supply chain and better insight into current and future demand for essential products to address shortages, stock-outs and wastage.

We must address four common barriers: insufficient high-quality data, low doctor trust in AI solutions, over-emphasis on flashy pilots at the expense of easily scalable solutions and inadequate technological infrastructure — especially in low- and middle-income countries.

First, governments must strengthen data privacy laws without throttling legitimate use of anonymised patient data to train algorithms. They must also help codify data ownership and security policies to encourage interoperability of data across borders and corporate walls. Meanwhile, providers, payers, technology and life sciences companies must think beyond protected data siloes while improving today’s federated learning solutions to train algorithms collaboratively, without exchanging data.

Second, stakeholders from across healthcare, government and beyond must ensure that algorithms are developed responsibly and transparently and that they work as well as advertised. This means prioritising applications with the highest potential to do good, ensuring data science and design teams are heterogenous and adding AI fluency to medical school curricula. Payers, meanwhile, must continue to encourage AI adoption through greater reimbursement incentives.

Third, governments must incentivise private investment in AI and allocate funds to scale solutions that are already working elsewhere. Partnerships must also be cultivated to ensure AI innovations don’t stay bottled up inside a few countries.

We must come together to ensure AI in healthcare is ethical, responsible and equitable – and results in improved outcomes for all.

This article was first published by the Daily Maverick.