University-based academics from across Asia have been working with Bangladesh policymakers to identify gaps and bottlenecks in using artificial intelligence in maternal healthcare, and to prepare the ground for an upgrade in health services that would in future use advanced technologies such as AI.
The initiative is part of the Association of Pacific Rim Universities (APRU) AI for Social Good project, in collaboration with the United Nations Economic and Social Commission for Asia and the Pacific (ESCAP) in Bangkok.
It involved researchers from the National University of Singapore, the Korea Advanced Institute for Science and Technology, Australian National University and universities in Hawaii in providing research to support the Bangladesh government in developing AI policy and building AI capabilities in pregnancy monitoring.
They worked closely with the Bangladesh Aspire to Innovate (a2i) programme of the government’s ICT Division and Cabinet Division, as well as Bangladesh policymakers and health professionals during the two-year APRU project funded by Google, which released its final report last month. The report focuses specifically on Bangladesh and Thailand.
In the case of Bangladesh, the report concluded that despite technical barriers, an AI system is viable with the development of IT infrastructure to integrate AI and localised AI models using available data from hospitals and clinics.
For example, AI could provide a better understanding of at risk-pregnancies by analysing data collected from different local sources, or through the development of an interactive ‘pregnancy assistant’ or monitoring system.
Arifur Rahman, assistant professor in the department of computer science and engineering at Hawai’i Pacific University, told University World News: “The project was about identifying how artificial intelligence can help in pregnancy monitoring in Bangladesh, then to improve the readiness of Bangladesh to use AI for monitoring.”
Rahman, who is also affiliated to the University of Hawaii, Manoa, said autonomous AI “will be able to predict something is going to happen, and once you can predict, you can take preventive measures to stop it. So, for pregnancy monitoring, there is a prospect in AI. But how much prospect, that still has to be seen”.
While many research groups around the world are working on AI in pregnancy monitoring, he noted that this was very much at the research stage. “It hasn’t been applied yet,” he said.
Rahman added that much ongoing research involves intermediate level AI, which involves input of medical data “and then the [AI] output helps doctors to make a decision – but this is not AI making an autonomous decision”.
Technological challenges to overcome
Rahman’s task within the project was to look at Bangladesh’s current technological challenges and what they have to solve, in order to make some progress in using AI in pregnancy monitoring.
This included what computer hardware is present in hospitals and clinics as well as at village level, how robust the connections are, as well as mobile phone penetration and prospects for digital networking.
While mobile phone penetration in Bangladesh is, surprisingly, almost 100%, he found, computer penetration is lacking and there are other high barriers apart from digital access, such as literacy, social norms and inability to act on information, particularly among the most at-risk groups.
Rahman acknowledged that there was some way to go before autonomous AI systems can be incorporated into maternal care in the country. Nonetheless there were ‘stepping stones’ that can be put in place.
“For predictive AI to be robust could take some time. But instead of just waiting, the country should make some improvements so that when these predictive AI models are ready, they can implement it right away,” he noted.
Bangladesh and SDGs
According to Dr Shabnam Mostari, public health specialist in digital health at a2i, maternal health in Bangladesh has improved greatly in the past two decades. “But the rate of improvement has stalled in recent years and further inputs and efforts are needed to support the health of expecting and new mothers.”
She told University World News that 90% of expectant mothers in the country receive antenatal care, but only 15% receive quality antenatal care and only 30% have access to quality postnatal care by qualified health professionals.
Often complications in maternal and child health arise from delays in getting to quality health care such as doctors or hospitals. This is exacerbated by low levels of doctors per capita – on average just one doctor for 10,000 people, she explained.
“The government’s objective is to reduce maternal mortality and to do that we need to look at a few indicators,” she said. Key among these is to identify pregnant women and reduce the burden on health service providers while also using the findings of the APRU-ESCAP collaboration for a new project for continuous pregnancy monitoring.
“From the beginning of pregnancy, to delivery, [the intention is] to capture the everyday patient data – movement, blood pressure, sleep patterns, temperature, to be stored in the cloud.” Then, at some point in future, “AI will assist the service provider to identify high-risk pregnant women to detect pregnancy complications early,” she said.
“It will help the service provider to focus on high-risk pregnant women.”
Getting ready for AI in healthcare
Continuous pregnancy monitoring is essential in achieving the Sustainable Development Goal 3 of health at every stage of life, according to Professor Olivia Jensen, deputy director and lead scientist (environment and climate) at the National University of Singapore (NUS).
AI will transform treatment and diagnosis as well as the relationship between health professionals and patients. She was the research lead of the NUS team for the project.
Digitalising health systems enhances monitoring and improves data quality. Higher quality data is important for the eventual application of machine learning techniques for AI analysis.
But other foundations need to be laid in the absence of a universal electronic health record system in Bangladesh. The research team assessed options for developing localised AI models using available data from hospitals and clinics.
User-centric design is also important. “Systems and applications should be designed to meet the needs of healthcare providers and expecting mothers, based on a strong foundation of evidence,” she told University World News.
Jensen pointed to the importance of interdisciplinary collaboration, which “enabled us to incorporate perspectives from academic research on risk perception and communication, which were highly regarded by our government partners”.
The project report recommended digitalising standard antenatal care data entry in the public health system, as well implementing a mobile phone-based system for appointment tracking with automated reminders for expecting mothers and community health workers. Jensen also pointed to its use in route planning for health workers in areas of poor transport.
“We gave a lot of emphasis in our work to the community health care workers, people who are not qualified health professionals, but nevertheless play a terribly important role in the frontline of the delivery of health care services, including maternal health care,” she explained.
According to Jensen, an issue that came up during research was the possible use of digital devices by pregnant women themselves. “This has been developed in other countries, and also in Bangladesh, so there are already some apps or online resources like chatbots that provide information to the woman or her family,” she said.
“But we decided that was not going to make a lot of difference to the women and their key outcomes that we were tracking,” she said, pointing to the most vulnerable and marginalised groups.
Anir Chowdhury, policy advisor of the Bangladesh government at a2i, and the government lead for the project, said: “Digital inequality matches exactly maternal health and equality. So, the same people who don’t have access to reliable digital services are the same women who are currently not getting the maternal care they need, and find it most difficult to get to a clinic to receive assistance in the case of a complicated delivery.”
Importance of academic research
Chowdhury explained that the goal for policy-makers was to “collect enough data so that we can identify the high-risk pregnancies and change the visit schedules of health workers to prioritise them. But the question is what is the right data to be collected?”
The research input was valuable in providing comprehensive understanding for policy-makers, and identifying the gaps to narrow, he said. “So hopefully this will lead to a solution that is affordable and scalable – if you collect that data for 100 women in a location, it can scale up to 100,000 or a million, that’s the challenge.”
The second aim, which he noted would take longer, was to train an AI system to help doctors and health workers in their decisions. But this requires a lot of data to be collected “and perhaps a lot of experimentation and permutations and combinations of what the human person does and what the AI does”, said Chowdhury, adding: “It’s a long process but we wanted to get started and some of that has been achieved, but it will take years to get to the point of scalability and accuracy.”
Working with academic researchers was important. “APRU was very careful and deliberate in involving the policy-makers throughout the whole process. In the beginning, the requirements and challenges from the policy-makers went into the design of the research.
“Throughout the whole research process, the guidance and debate between the two parties really helped. There was mutual respect and trust and proof,” Chowdhury said.
“Of course there was a bit of difficulty in the beginning, because the language, the perspectives are different,” he acknowledged. “But we figured out a common language of communication and common vision and this was a good model for us to follow.
“AI is not well understood by policy-makers, there is a great fear about AI, and whether AI is ‘taking over’. And that fear is actually growing. So this collaboration between the academicians who understood much, much better than policy-makers, and could talk about specific examples from other countries that have had successes and failures, really informed the policy-makers,” he said.
APRU’s Chief Strategy Officer Christina Schönleber explained that the research team was a large one with multiple universities involved, brought together by APRU from among more than 60 member universities and based on the expertise needed to frame the research questions.
“For APRU it was about giving governments information that they may already know they need, or they are not aware they need, about the implementation of AI for some of the services that they would deliver, and it’s about the impact,” Schönleber said. “We want to make sure they don’t just use AI for the sake of using AI and because everybody wants to use it.”
Chowdhury also underlined the importance of involving international universities, to learn about successes and failures from other countries “then juxtaposing it to our context”. He added: “That could not have been done by local researchers, they would not have had that exposure from other countries.”
He stressed: “This is not just a research project. It is leading to actionable projects, and interventions that we’ll actually try out.”