The push for local data to fuel Vietnam’s medical AI
Vietnamese medical developers are reducing their reliance on datasets drawn from Western populations to build AI tools grounded in local realities. This effort will enable new digital health systems to be accurately tailored to the people they serve.
Local patient data is helping Vietnamese developers build AI predictive models that help deliver personalised care to children with rare diseases. Photo: Beth MacDonald
Something was not right with two-year-old Binh’s* medication. The little boy from Hanoi was battling Kawasaki disease, a rare condition that can dangerously widen the blood vessels around his heart. To keep him safe from complications, doctors prescribed him clopidogrel, a medicine designed to prevent blood clots.
For two months, the toddler was given the drug as prescribed. But when his parents brought him back for follow-up checks, his blood-test results had still not stabilised. The medication, a standard treatment used successfully by millions around the world, appeared to be ineffective for him.
The answer lay in his DNA. A genetic test conducted by GeneStory, a local medical startup, along with the doctor's assessment, revealed that Binh’s unique genotype made it impossible for his body to metabolise clopidogrel properly. His condition finally stabilised when his doctors switched him to an alternative treatment.
For Dr Sy Nam Vo, co-founder and chief scientific officer at GeneStory, the toddler’s experience is a reminder that a one-size-fits-all approach to medicine does not always work.
The startup's AI predictive models were made possible by a landmark project, conducted by Vo and his colleagues at a research institute, to sequence 1,000 Vietnamese genomes. Without this foundational dataset, developing an AI application capable of catching such biological mismatches would have been impossible.
“If a medical AI is trained exclusively on foreign datasets such as Western or Caucasian populations, it might blindly recommend standard treatments that might be ineffective, or actively harmful, to Vietnamese patients,” Vo explains.
To safely treat patients like Binh, Vietnamese developers and hospitals are moving away from Western datasets, seeking ways to responsibly collect and utilise local patient data.
The AI wave in Vietnamese healthcare
Vietnam’s medical AI boom is unfolding within a broader, government-led effort to digitalise healthcare and move towards universal health coverage. The National Digital Transformation Programme, approved in 2020, identified healthcare as one of the priority sectors. A year later, the National AI Strategy set goals for developing and using AI by 2030. In healthcare, this includes improving the sharing of health data and using AI to support diagnosis, treatment decisions, remote monitoring, personalised care, and drug research.
Tech firms in Vietnam are building AI models for health specifically tailored to the Vietnamese population. Photo: GeneStory
Against this backdrop, Dr Pham Duc Phuc, a public health expert at the Institute of Environmental Health and Sustainable Development, describes the current scene as “a hundred flowers blooming”, with tech firms and hospitals racing to create new medical AI applications.
They are focusing on diagnostic imaging, predicting treatment outcomes, community screening, virtual medical assistants, health record analysis, and support for drug and therapy research.
“Medical AI acts as a much-needed lifeline for an overwhelmed workforce,” Tien Zung Nguyen, a professor at the University of Toulouse, France, adds from his experience testing AI-aided disease detection tools developed by his startup Torus AI in Vietnamese hospitals.
“In the context of our current doctor shortage and the massive amount of visits clinicians handle every year, AI is essential. A doctor supported by AI will work much more efficiently and deliver better care.”
Vietnam notched up approximately 15 doctors per 10,000 people in 2025, up from just 9.81 in 2020, the Ministry of Health reported at a conference at the end of 2025. However, the ratio of doctors still falls short of actual demand. Major centres like Hanoi’s Bach Mai Hospital report operating at up to 170% of their original design capacity, placing doctors under constant pressure as they cope with heavy patient loads.
“In the context of our current doctor shortage and the massive amount of visits clinicians handle every year, AI is essential,” Tien Zung Nguyen explains. “A doctor supported by AI will work much more efficiently and deliver better care.”
Hidden risk of foreign data
Building reliable medical AI demands one essential ingredient: local patient data. In Vietnam, however, this information lies in a fragmented, disconnected web.
Vietnam’s healthcare system serves a population of 102.3 million through a vast network extending from central referral hospitals to local primary care. A 2020 World Bank report describes a fragmented digital landscape in which hospital records, health insurance claims, vaccination registries, and communicable disease databases are stored on separate platforms built to different standards. Hospitals commonly operate their own record systems, while commune facilities may use several different programs; records can even be duplicated within the same institution. At one point, Vietnam was dealing with around 1,000 different software systems just for laboratory results.
“These foreign resources could miss the unique genetic makeup of Vietnamese patients, holding back research into personalised care.”
This fragmentation makes it difficult to compile the large, high-quality datasets needed to train AI models. In genetics, a lack of local data forced early researchers to rely heavily on international databases built mostly from people of European descent.
“These foreign resources could miss the unique genetic makeup of Vietnamese patients, holding back research into personalised care,” Vo points out. Bridging the gap, his team had to build a research system from the ground up, figuring out everything from processing raw samples to analysing complex data.
Western datasets do not take into consideration the unique genetic makeup of Vietnamese patients, emphasising the need for collecting local patient data. Photo: GeneStory
When medical AI learns mostly from Western patients, it learns to spot diseases exactly as they look in those specific groups. Drop that same system into a Vietnamese hospital, and it can easily misread deep differences in genetics, local diets, and common illnesses.
Obesity conundrum
Pham cites obesity as a primary example. Consider a rural clinic where a physician examines a patient who appears slim and scores within a healthy range on standard Western body mass index (BMI) charts. These imported metrics, however, are often ill-suited to local demographics. Many Vietnamese patients carry visceral fat around their internal organs, a condition Pham describes as "thin on the outside, fat on the inside". An AI tool trained exclusively on Western datasets would likely misidentify these individuals as healthy, potentially overlooking critical early warning signs of metabolic disease.
This same problem is happening across Southeast Asia. A regional study revealed that neighbouring countries are battling the same messy records and growing fears of AI bias. One Thai policymaker, for instance, noted that several AI tools failed noticeably when applied to local patients, despite working flawlessly on their native Western datasets.
Scramble for datasets
To break this reliance on foreign data, Vietnamese tech teams are scrambling to build homegrown datasets. However, getting that raw data is not easy.
While existing laws permit the use of health data for research, Pham notes that a clear, standardised pathway for accessing hospital archives remains elusive. Medical imaging serves as the most accessible data source for AI developers in Vietnam, as these files follow standardised formats and are easily anonymised.
Conversely, researchers face significant barriers when attempting to access other types of health records. “Hospital directors worry about strict privacy regulations and navigating multiple layers of approval,” Pham notes. “Fearing potential data leaks, they often choose the safest path: locking their files away entirely.”
“I believe AI is the future, and it will be part of every aspect of our lives. Using it is unavoidable. I just think they [the developers of medical AI models] should keep user data protection in mind while developing their products.”
Public trust is another delicate issue. Hanh Nguyen, a 28-year-old office worker who frequents a private clinic in Hanoi, says she would not feel comfortable sharing her medical history unless she was completely sure her detailed information would be kept safe.
"I believe AI is the future, and it will be part of every aspect of our lives. Using it is unavoidable. I just think they [the developers of medical AI models] should keep user data protection in mind while developing their products," she says.
Consequently, a national data drought persists. The fragmented, homegrown datasets currently available remain confined within isolated partnerships. Pham warns that this gap risks fostering new forms of medical bias within Vietnam. If AI models are trained exclusively on data from a few central, urban hospitals, they will reflect only a narrow demographic, failing to capture the diversity of the Vietnamese population across different ages, geographies, and physical characteristics.
A homegrown data foundation
Turning Vietnam’s medical AI dreams into reality will require a major team effort, backed by strong government support, to finally piece together the country's scattered health records.
Partnerships are emerging, with hospitals, universities, and technology groups sharing data for specific projects. Yet, as Tien Zung Nguyen argues, this teamwork must now become systematic. Research teams specialising in the same field, such as cardiology, should pool their data and develop AI tools together rather than building separate systems in isolation.
Pointing to his own team's ongoing work with local doctors, Tien Zung Nguyen notes that their current pool of Vietnamese data remains quite small. “But as we push forward, it will steadily grow,” he adds.
The digital shift in healthcare starts with integrating electronic health books to build a shared national dataset. Photo: GeneStory
Indeed, experts are calling for a united effort to build a shared, national dataset. This foundation should truly reflect the Vietnamese people, weaving together the diverse traits of city, rural, and highland communities, while proportionally representing the country's 54 ethnic groups.
The shift begins on the ground, where hospitals are slowly trading their towering stacks of paper files for electronic health records. The government is driving a massive campaign to digitalise everyday healthcare, rolling out electronic health books and linking patient data directly to a national digital ID app. A patient's medical history will be securely carried on their smartphone, ready to be accessed by any doctor, anywhere.
This digital shift feeds into a much larger ambition. The Ministry of Health is laying the groundwork by building a National Health Database, a massive effort designed to smoothly connect these new digital records across the entire nation.
The ministry launched an intense 90-day nationwide campaign running from July to September. The mission is to meticulously scrub, standardise, and secure 12 major specialised healthcare databases. The initiative also locks down information security to ensure this sensitive data can be safely shared across the national network.
Pham advocates for the creation of a dynamic, coordinated data ecosystem—a unified national network anchored in shared standards, collaborative frameworks, and rigorous ethical guidelines. “Success depends on hospitals fostering mutual trust and policymakers establishing clear, robust privacy regulations,” he explains.
A growing number of local institutions are releasing their cleaned-up datasets to the wider research community. One example is a recent publication in Nature Communications about the 1,000 Vietnamese Genomes Project by Vo and his colleagues. Born from a private sector effort, it provides a much more accurate genetic baseline for the local population.
Vo acknowledges the current sample remains small compared to international projects. Expanding it will require long-term investment, broader participation, and stronger state support.
Vietnamese-owned genome sequencing projects provide a more accurate genetic baseline for the local population. Photo: GeneStory
Besides building a broad data network, Tien Zung Nguyen believes Vietnam must develop its own language models and specialised medical AI to independently control the technologies and information on which its health system increasingly relies.
“Building domestic AI capability must be viewed not just as a technical upgrade, but as a matter of digital sovereignty,” he says.
Above all, it is hoped these shared efforts will ultimately bring comfort and better care to patients themselves. Knowing that new diagnostic tools will be built on data that truly reflects her own people, office worker Hanh Nguyen feels a sense of hope about the future of domestic health technology.
“I’m not afraid to try them,” she says, “as long as they remain under the careful control of doctors.”
*Name changed to protect the patient's identity
