Urgent.News

What's breaking now, across thousands of outlets.

AI

Southern Africa: Smarter Healthcare, Unequal Outcomes? the Risks of AI in Southern Africa

[The Conversation Africa] For generations, the stethoscope has symbolised diagnosis. A doctor uses it to listen to the body, turning faint signals into clinical judgment. Healthcare AI promises a different kind of listening.

In South Africa, mobile clinics employ artificial intelligence (AI) to analyse chest X-rays for signs of tuberculosis (TB). A study screened nearly 10,000 individuals in rural KwaZulu-Natal areas. The Stop TB Partnership reported in December 2024 that an AI-assisted chest X-ray screening programme in South Africa had reached over 450,000 people through 33 mobile clinics nationwide.

Zimbabwe utilizes a similar technology in mining communities to detect silicosis, a lung disease caused by silica dust inhalation. These innovations leverage patterns in patient records, laboratory results, and population data, promising smarter healthcare. However, the question arises: whose health needs do these systems address when data fail to represent everyone equally?

In our research, Shiven Naidoo, a PhD student in bioinformatics at the University of the Witwatersrand, and I investigated whether algorithms employed by AI systems unintentionally perpetuate existing health disparities. We formulated a conceptual framework examining how populations are represented in health data, how algorithms classify and prioritize them, and how these classifications impact access to healthcare.

Our findings indicate that digital health technologies risk exacerbating health inequalities. The issue stems from the limited collection of data from rural and under-resourced communities due to restricted access to healthcare, incomplete medical records, and fragmented digital systems. Disconnected health information systems can result in patients having incomplete medical histories, especially when transitioning between facilities or provinces.

Additionally, most genomic research on how DNA influences health has relied on data from European ancestry. Consequently, tools predicting disease risks may be less accurate for African populations. To ensure technological advancements do not compromise equitable healthcare, incorporating more representative local data, regularly auditing systems for bias, and involving affected communities in their design and evaluation are crucial.

For instance, a screening tool could be evaluated using data from both rural and urban clinics where it will be employed. This testing might reveal that the tool accurately identifies disease in one population but misses more cases in another due to variations in underlying health conditions or disease presentation. Subsequently, the tool can be refined, its screening criteria adjusted, and retested to ensure reliable performance across diverse patient populations.

Drawing on Michel Foucault's concept of biopolitics, we examined how populations are governed through measurement, regulation, and statistical observation. This perspective enabled us to consider not only the accuracy of data but also the consequences of classification. By assigning individuals to risk, vulnerability, and priority categories, healthcare systems can influence who receives interventions and where resources are allocated.

Uneven infrastructure and unequal access to care in southern Africa shape whose experiences are included in datasets. However, representation is merely the starting point. The criteria AI systems use to assess patients and allocate healthcare resources can also lead to unequal outcomes. For example, a 2019 study in the United States found that a healthcare algorithm underestimated the health needs of Black patients because it used healthcare spending as a measure of medical need.

As Black patients historically received less care for comparable health needs, the system assigned them lower risk scores, reducing the likelihood of them being identified for additional care. The concern extends beyond data representation to the assumptions that shape how data are interpreted and utilized. Much of the genomic data informing advanced health technologies originates from European populations, leaving African populations significantly underrepresented.

Polygenic risk scores, which estimate disease risk using genetic information, demonstrate the implications. Scores primarily derived from European genetic data may perform poorly in black African populations. This limits their clinical utility. Furthermore, infrastructure for producing genomic data, such as sequencing capacity, digital infrastructure, and specialist expertise, is predominantly concentrated in major centers.

As a result, data from outbreaks in less-equipped communities may be sequenced late or not at all. This hampers the understanding of important transmission patterns and hinders timely intervention. For instance, during the 2013-2016 Ebola outbreak in West Africa, a delay in viral genome sequencing led to limited real-time tracking of the virus's spread between communities.

Such timely information could have facilitated earlier identification of transmission chains and more effective targeting of surveillance and isolation efforts.

Written by urgent.news from AllAfrica Health's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at allafrica.com →

More in AI

More from Wednesday 18 November →