The panelists listen as Gorgens shares takeaways from the session. Screenshot from PMAC session recording.
Bangkok, Thailand–The Asia eHealth Information Network (AeHIN) joined the 2025 Prince Mahidol Award Ceremony (PMAC) plenary session “Harnessing the Power of Data” on January 31, 2025, held at the World Ballroom of Centara Grand & Bangkok Convention Centre.
The panel was a combination of discussions and presentations around three themes: development or creation of public health data, integration and management of public health data, and the use of data, including use cases, challenges, and next steps.
Marelize Gorgens from the World Bank facilitated the session. The panelists were Dr. Boonchai Kijsanayotin, Chair of AeHIN; Peeter Ross, Professor of e-health and head of eMedLab at Tallinn University of Technology (TalTech); Dr. Mary-Anne Hartley, Professor of Artificial Intelligence (AI) at Harvard University and the Swiss Federal Technology Institute of Lausanne (EPFL); and Kara Sewalk, Senior Manager of Innovation and Digital Health Accelerator for HealthMap at Boston Children’s Hospital.
Development and creation of public data
Dr. Kisanayotin explained that paper records can be at the transection and how medical records are mostly in the survey and reporting system. Gradually, health and public health data are generated electronically. Public health data involves health-related data that can be used in several ways, including healthcare, surveillance, and financing.
Following Dr. Kijsanayotin’s emphasis on the available medical, statistical, and financial data that stemmed from various sources from paperwork, Dr. Ross shared the vision of capturing data once and using them several times, which is a challenge. He proceeded to share their experience of learning and discovering the data subdomains and challenges that come with them. Dr. Ross stressed the importance of distinguishing data from the beginning because primary users of data, commonly medical workers, need data quickly for service, experience, and decision-making, while secondary users of data, who could be hospital managers, researchers, insurance, policy-makers, and industry, need good quality of data for business intelligence, disease identifiers, reimbursements, etc. There is tension between these two groups because the medical workers need a user-friendly health information system, and data is time-critical. In contrast, the second group wants aggregated data and has more time to process them.
Gorgens noted the asymmetry in the goals of the two groups and shared that primary care doctors in the United States spend an average of 4 hours a day in an electronic health record (EHR) and another 2.8 hours at night catching up on EHR capture. An emergency doctor, on average, clicks an EHR 40,000 times on one shift. Dr. Ross, in response, raised the opportunities for AI, such as using an AI chatbot to capture data into a standardized format and as a medical scribe or filling in forms and responding to messages.
Dr. Hartley agreed on the opportunities raised by Dr. Ross and reflected on previous questions relating to standardizing, collecting, and managing data, as well as spending time in EHR to the point of health professionals experiencing burnout. As a global researcher, she raised more than data management and standardization; it’s more about data existence, and she talked about how some places do not even have existing data.
Dr. Hartley then shared the technological perspective that, sometimes, paper records get neglected and that data is rarely managed. Even when the data is collected, photocopied, and digitized, without the gold standards, the contents of what the patient actually had could not be determined. She emphasized that the right information at the right time and the right place is the most important thing doctors can give their patients, and she highlighted the importance of respecting information and its source. Dr. Hartley further explained that fragmenting data is fragmenting solutions. Foundation models consolidate information, and it is enough as long as it is representative data, but it will never be the case. Dr. Hartley said foundational models will never be perfect, and there will never be perfect data. Instead, take and use an imperfect model, collect data while using the model, and nudge it to make it adaptive.
Gorgens asked Dr. Kijsanayotin about challenges in data, and he said the challenge in getting quality data is not just its source but the data itself. While big data is available, a large part of it might be ‘trash’ or unusable because they are neither standardized to be integrated nor can a model be created after it. In addition, there is a lack of human resource capacity to do digital health and standards.
Sewalk reflected on the fragmentation of data in the United States, a high-resource country. Boston Children’s Hospital faces challenges in the operability of mother-infant health data; specifically, there’s no easy way to connect the data of infants born in their hospital with the mothers’. The novel data they collect is largely unstructured. They want to structure how they collect data, but with the types of data they collect, it is difficult to do and lacks the contextualization they need.
With regard to fragmented data, Dr. Hartley mentioned federation learning. Instead of just putting data in a centralized setting, there is an option to copy the model to the source of the data and decentralize the learning of models so that people can retain their data ownership, and the problem of interoperability can be partly solved. “Our understanding of sharing data, centralizing it, defragmenting it, and making it interoperable are also changing with the changing technologies,” she concluded.
Integration and management of public health data
Dr. Ross shared that in Estonia, data is generated by clinicians or patients and goes to the EMR or EHR, producing a great dataset of medical data. Hospitals, insurance companies, and population health researchers also collect data through their own data streams. Each data consumer has its own database, so there is fragmented architecture. Estonia hires many people to clean and encode the data, and although this may produce good data, it is resource-consuming.
The solution they are considering is to create more universal data models, collect them, and use them rather than create separate data models and try to merge the data. Data harmonization should be applied in the data capture phase. Then, create virtual registries, produce unlimited digital health services, and have a customer-oriented ecosystem.
Dr. Hartley shared her thoughts on how this could evolve. She said that to represent the diversity that people are demanding and to reflect them in the data, allow the data to categorize itself and to recognize data as a model. Try to specialize the models further and make them nudge to the right environments, representing smaller populations by collecting data systematically.
Gorgens summarized Dr. Hartley’s point of thinking of data in an inductive approach rather than a deductive approach. She added her thoughts about the future, where digital health services would allow patients to access their own health data as a foundational health service, similar to how bank transactions from various merchants appear in one banking application. Gorgens added that there should be a mindset change where people must demand their own health data and their access to it.
When asked about the pitfalls that countries have to anticipate and avoid in terms of access to their health data and what constitutes health data, Dr. Kijsanayotin answered that public health data is all over the place and there is silo due to several applications and systems available. He said foundations are neglected, and it is assumed that data is available. However, there are places that do not even have an established healthcare system. He said there is a close link between providing care, having care, and the data used for it, not just for planning, statistics, and business intelligence. Policy-makers and those with resources use and assume that all data in the hospital are in the same model or structure, but it is not; there are several components.
Dr. Kijsanayotin shared AeHIN’s experience working with countries where the foundation to generate and integrate data has not been laid down. He explained the critical success factors for effective data integration and management through AeHIN’s Mind the GAPS Framework. There is a need for a governance system to convene ministries of health, ICT and finance to work together on seeing the same health data system; an architecture or blueprint that contains the shared vision of centralizing or decentralizing data and sharing data and data ownership; people, both users and developers, need to work together and be knowledgeable in the field; and using standards on terms and data format that will allow data integration. Dr. Kijsanayotin said, “The real world is difficult. People don’t talk. The importance of interoperability is people coming to talk. It is not technical. The technical can be addressed.” In the last 15 years of AeHIN working with countries in the Asia-Pacific, it found that “people interoperability is more important than the data and technical interoperability,” added Dr. Kijsanayotin.
The real world is difficult. People don’t talk. The importance of interoperability is people coming to talk. It is not technical. The technical can be addressed… people interoperability is more important than the data and technical interoperability.”
– Dr. Boonchai Kijsanayotin
Use of data
Sewalk shared that Boston Children’s Hospital uses data not as a separate source but as complementary to traditional data evaluations to supplement interpretations of clinical data and provide information about patients’ health. She then showcased the application they developed, the Biothreats Emergency Analysis and Communications Network (BEACON). It transforms aggregated data and runs it through a language learning model (LLM) to combine data into a report that summarizes what happens on the ground. Subject matter experts review and make sense of what’s happening and provide feedback to the computer about what it gets right and wrong and how to get better.
Dr. Ross shared that in Estonia, in five years of claims data, they found data that indicate diseases applying deep learning algorithms. They concentrated on mental health issues and saw correlations with certain diagnoses in the first and second year and a high probability of a mental issue in the fifth year.
Dr. Kijsanayotin said that Thailand standardized its drug code, and every provider and system used the same language, which made it possible to use it in their reimbursement policy. For continuity of care, they are already using the data, but they are struggling because they cannot integrate it across the healthcare system yet. For the policy or research side, data is already in use but not yet in its full potential.
Dr. Hartley talked about MOOVE. The M stands for massive to validate things in combination of ways, the first O for open to trusted experts, the second O for online for accessibility, and V and E for validation and evaluation respectively. She shared that MOOVE centralizes on people as the ones controlling the model; people can engage AI and co-develop it with their own expertise and not rely on developers, and control it by talking and giving feedback to it.
At the end of the discussions, Gorgens shared the following takeaways:
- There are so many kinds of data that can be used, but there are also foundational issues that need to be addressed
- Agency for people to validate, use, and contextualize data for their purposes. If there is good data, it can be used in multiple ways in multiple settings
- Think and recognize how AI has seeped into the ways health data is used
“We boldly go into this AI-enabled data future for health. We have to really figure out what is that way in which we walk the line between being neither overly careful nor careless about how data are used to benefit health and to benefit health systems,” Gorgens concluded.
Related: