Challenges of AI in Maternal Healthcare: Lessons from Zambia

Author Min’enhle Ncube is an anthropologist and Mozilla Africa Mradi Research Grantee. Ncube is investigating the challenges and dangers posed by biased datasets in the delivery of quality and timely care to pregnant people. Learn more.

Every expectant mother deserves access to the highest quality of care. In our increasingly digital world, the fusion of artificial intelligence (AI) and maternal health holds the promise of achieving this goal. AI has been applied broadly in healthcare in areas such as medical imaging, decision support systems, remote patient monitoring, virtual health assistants, genomics for precision medicine and other examples spanning diagnostics, treatment, patient management and healthcare administration.

My ethnographic enquiry on how AI is taking hold in supporting maternal health in Zambia offers some important lessons on what representative datasets might be and some ethical considerations for deploying equitable AI applications in maternal health. Hypertensive disorders of pregnancy are the leading cause of maternal mortality in Sub-Saharan Africa - preeclampsia and eclampsia contributing to adverse maternal and foetal outcomes. These ethnographic insights are valuable when reflecting on emerging technologies in Zambia using AI to leverage the timely diagnosis of pregnancy disorders such as preeclampsia towards better maternal outcomes.

Zambia's urban and peri-urban inequalities are socioeconomic, infrastructural and linguistic. One observes how places of residence and socioeconomic status determine the accessibility of the Internet and the type of mobile device one has. Electrical and solar electricity is more fragmented or less abundant in high-density compounds compared to low-density residential areas. One's economic status determines smartphone ownership—a phone capable of running apps—and access to mobile data or Wi-Fi to take advantage of AI-driven health apps, such as the DawaMom product. While software developers have initiated the provision of mobile apps in vernacular languages, some users, especially those outside the city, have limited ability to read.

I came across Dawa Health, a medical technology startup in Lusaka, during the first year of my PhD in anthropology. Dawa Health's vision to support maternal care using AI began at BongoHive, a startup accelerator hub in Lusaka that brings together entrepreneurs from various disciplines to brainstorm applications for digital technology. Dawa Health's clinicians undertake home visits to gather information, such as users' primary data on their blood pressure, level of haemoglobin, symptoms of anaemia, signs of hypertension or cardiac disorders and others—the aim of this exercise being to predict the likelihood of disorders such as PE and eclampsia to improve the parity and gravity outcomes from pregnancy and reduce the possibility of miscarriage or stillbirth. Their DawaMom is an app presenting automated personalised information from the health data gathered that supports patients and health practitioners get better clues on potential high-risk pregnancy disorders such as PE.

A recent publication describes preeclampsia (PE) as a multi-organ disease that presents in women after the twentieth week of gestation. Elevated blood pressure (systolic blood pressure equal to or more than 140mmHg, diastolic blood pressure equal to or more than 90mmHg) and proteinuria (excess protein in the urine) are common indicators of PE. In its severity, PE is characterised by higher blood pressure levels and additional symptoms such as visual disturbances, swelling of the hands and face, impaired liver function, abdominal pain, headaches and other clinical signs. In the case of proteinuria, laboratory tests may be conducted to assess organ function, and accompanying blood tests may be performed to check platelet counts, liver and kidney function, and clotting factors. Determining other cases of preeclampsia may involve foetal heart monitoring and ultrasound that assess foetal growth and well-being. Eclampsia is an advanced complication of preeclampsia and a medical emergency characterised by the development of seizures and convulsions in a patient with preeclampsia, posing a danger to the mother and baby, with a successful delivery being the only cure. A health facility-based study in Lusaka estimated the prevalence of PE at 18.9%, and another study from the University Teaching Hospital in Lusaka reported a 12% prevalence. A government assessment showed that out of the 398 maternal deaths per 100,000 live births in Zambia, about 18% are directly attributed to PE or eclampsia, making it the second leading cause of maternal mortality after haemorrhage.

Disruptive technology has yet to become mainstream in diagnosing PE, as the current diagnosis has been the same for decades and involves the conventional systolic and diastolic blood pressure measurement, despite this disorder being widely complex and requiring lots of information for clinicians to make a precise diagnosis. Dawa Health's team of clinicians and software developers believes that standard statistics alone are inadequate in determining a broad disorder such as PE and are collating prenatal data for machine learning and augmented intelligence to assist clinicians in timely diagnosing PE. Their DawaMom app aims to bring this multimodal information to their expectant mothers' phones as feedback on their clinician home visits. The continued expansion of their localised datasets means that their AI-driven DawaMom produces tailored feedback, personalised for each user. However, one sees broader nuances that are otherwise excluded or cannot be considered in this data.

When considering representative datasets for maternal care, I am drawn to multiple narratives that emerge from the approaches to maternal care that are not limited to biomedicine and extend to traditional and religious practices. A few respondents who are DawaMom users testify to adopting conventional means of managing prenatal complications, such as drinking herbal medicines otherwise disqualified on biomedical grounds. These prescriptions potentially produce abnormal readings by DawaMom and a range of disorders, all clustered into one group of the app's raw data. As the app uses a biomedical approach to determining the well-being of its users, there is no room to factor in or quantify other epistemologies of care and pluralism present in communities. This means that while the app holds the potential to provide a better quality of care, it is tailored into one biomedical framework and is delivered to users based on socioeconomic status.

Gathering big healthcare data for AI from individuals presents significant logistical challenges for a small medical technology startup such as Dawa Health. Firstly, accessing comprehensive and reliable healthcare data in a country with limited digital infrastructure and fragmented healthcare systems can pose challenges. The need for standardised electronic health records and interoperable systems complicates data aggregation and integration efforts. Moreover, cultural and linguistic diversity among patients may necessitate tailored data collection approaches to ensure inclusivity and accuracy. Additionally, ensuring data privacy and security in compliance with regulatory frameworks adds another layer of complexity, requiring robust protocols and infrastructure. Furthermore, more resources and expertise may be needed to maintain startups’ capacities to deploy sophisticated data collection and analysis techniques, necessitating creative partnerships and capacity-building initiatives. Overcoming these challenges requires strategic collaborations with local stakeholders, innovative data collection and management approaches, and a steadfast commitment to ethical and responsible data practices.