What do we mean by semantic interoperability? By semantic interoperability we understand the ability of a healthcare system to share information and have that information properly interpreted by the receiving system in the same sense as intended by the transmitting system.
Why is semantic interoperability back on the agenda? The recent focus on Machine Learning (ML), Artificial Intelligence (AI) and the European Health Data Space (EHDS) is bringing new impetus to this strategic element of interoperability. It is absolutely vital to ensure that electronic health systems ‘speak’ to each other with absolute clarity.
On 25 April 2022, EHTEL organised an ‘Imagining 2029’ Innovation track webinar, focusing on the views of three different actors active in the field of semantic interoperability: European authorities, a competence centre and a European-funded project that makes use of semantic tools, InteropEHRate.
The three presentations were followed by a discussion, moderated by Luc Nicolas, EHTEL, in which the roles and experiences of different national and private entities in developing semantic interoperability were further explored.
👀Watch the video recordings and download the presentations below to explore the different views:
Konstantin Hyppönen, EU Commission - DG Health and Food Safety (SANTE)
A view from the European Commission: The MyHealth@EU infrastructure – formerly known as the eHealth Digital Service Infrastructure – connects national healthcare providers in ten European Member States. Its common goals revolve around the provision of continuity of care, useful tools and further services, and data protection and security.
MyHealth@EU has moved forward substantially since 2018 – especially in 2020/2021 – on its initiatives on both ePrescriptions and patient summaries.
Almost six million Europeans are able to benefit already from MyHealth@EU’s services. Many more steps, covering around 20 nations, need to be undertaken in 2022-2025.
In terms of semantics, exchange of documentation is now coming to the fore and the use of terminologies is becoming key. Guidelines for these activities have been prepared. They are now either under revision or updating. Others are in the pipeline.
Ten possible next steps are on the agenda. All these developments are likely to be based on solid and reliable semantic interoperability.
Last but not least, the European Health Data Space text – (later) released on 3 May 2022 – offers plenty of further opportunities for both the primary and the secondary use of data, including the use of AI and health data for health policy-making. These advances have implications for both health authorities and for citizens/patients.
What has been done at European level (Myhealth@EU) to allow data to travel across countries? What is required from end-users (and possibly local electronic health record industry members) to validate a patient summary? What are the main developments expected?
A terminology server in the Netherlands
Chantal Schiltmeijer, Nictiz
A view from a national competence centre: Different European Member States are handling the infrastructure underpinning MyHealth@EU in different ways.
Nictiz, the independent Dutch competence centre for electronic exchange of health and care information, described the approach in the Netherlands.
Overall, the focus is on governance and (semantic) standards (e.g., for hospitals, nursing homes, and general practices). In this framework, SNOMED, the international standard plays an important role.
There have been ongoing challenges since the Netherlands started to work on this field back in 2008. Three levels of Dutch legislation influence the field related to terminologies. Keeping up with releases can be especially difficult.
The Netherlands has started on the implementation of a national terminology server. It uses the Australian CSIRO’s Ontoserver. The server is HL7® FHIR®-compliant.
Overall, the Netherlands is optimistic about the advantages brought by the direction it has taken on a central terminology server and a more distributed terminology service ecosystem. It “strives to create an ecosystem of servers which ‘fills itself’ with standards.”
Doing what with whom? For which purposes? For what aspects is public investment absolutely needed? What kind of governance needs to be put in place? What is the role of a national terminology server?
Gabor Bella, IEHR project, University of Trento
A view from a European-funded project, InteropEHRate: The InteropEHRate project has looked at what are the motivations for and problems with health data interoperability, and what are therefore the reasons for any delays in overall implementation in Europe. Instead, the project offers some innovative ideas, that have emerged from its work, on how Europe could make faster progress on the interoperability of health data.
The healthcare sector has not been the earliest adopter of data interoperability. AI – perhaps the latest hype – has chiefly been applied to image processing.
Today’s three main motivations for introducing health data interoperability lie in the fields of care, research, and “non-medical” fields (like cost reimbursement, accountancy, and statistics). Yet – problematically – advances are uneven, and data is diverse and mixed.
Among the solutions, reasonable progress has been made. Some tools, however, are not yet used in care settings.
Why are people ‘not yet there’ with health data interoperability? Four basic reasons were suggested.
InteropEHRate has focused on the solution of deep semantic integration to resolve this set of challenges. Deep semantic integration means that every single data value relevant to the task on hand is understood, made explicit, and is reported on in a formal, language-independent manner.
The five steps used in the “human-centred, yet automated, methodology” were described in detail. Ultimately, the deep semantic system learns from human curation and gradually improves its own performance.
This methodological innovation can reconcile the contradictory needs for precision, traceability, and automation of data. Human data managers are still needed, and the system works by also using AI. All these details are written up in an InteropEHRate White Paper launched in May 2022.
What are the motivations for focusing on the use of health data? What are the practical challenges? What are the solutions, including those created by InteropEHRate? What do health data managers need to do? How does it work with AI algorithms?
Moderated by Luc Nicolas, EHTEL
Three panellists stepped forward.
- Samuel Danhardt (Agency eHealth – Luxembourg)
- Vincent Keunen (Andaman7 – Belgium)
- Benny Van Bruwaene (BT Computing – Belgium)
Further input was also offered by two of the speakers.
The panellists expressed their views on what national and European authorities can do on semantic interoperability. In support of data transformation, they offered opinions on how far human input is needed; what support developers need; and assistance to be offered by AI.
00:34 - How far is Luxembourg planning to go to support semantic interoperability?
Samuel Danhardt (Agency eHealth, Luxembourg) recounted the history of health interoperability in Luxembourg, and its current work on patient summaries and ePrescription. The agency has used medical and care experts to offer it advice on which terminologies to use for which use cases. He focused on Luxembourg’s start with a usable first data set; added value for laboratories; similar descriptions and packaging for medications; a national data lake; and HL7® FHIR®.
04:51 - What to expect from European authorities in terms of supporting the development of semantic interoperability further?
Samuel Danhardt’s plea to the European Commission was, “Any coordinated initiative is welcome. But does it mean it is easily [to do] and can be done quickly?” He applauded the availability of European projects, which can be used to standardise/harmonise approaches.
06:30 - How far is human input still needed at the current stage of evolution of the semantic technical tools?
For Benny Van Bruwaene (BT Computing, Belgium) work should not progress too slowly. In his opinion, always needed is “the human at the first stage to validate things”. He also made detailed points about the current need for learning systems; SNOMED as an Esperanto for medicine; work on what he called pre-coordination and post coordination SNOMED coding; and work on semantics done throughout the whole world e.g., with China and India. He used the example of semantic-related work on acute myocardial infarction and the importance of ICD coding.
10:55 - How should developers expect public/ third party authorities to support them in their work?
A company like Andaman7 draws on standards like HL7® FHIR®, LOINC, and SNOMED to code its personal health record data. This is easy to do in the USA, where Vincent Keunen’s small Belgian company’s product is compatible with 85% of FHIR® data in all US hospitals. Therefore, Europe requires another approach from the one it uses today. The aim should be to impose the use of the most common three health data standards on data users. One could also discuss possibilities for action with those European Member States (e.g., Luxembourg) where activities could be transferred across borders.
14:38 - Under which conditions can AI support the data transformation process?
In the Netherlands, Chantal Schiltmeijer reported that Nictiz talks about AI internally, wants to promote it more proactively, and has AI on its agenda even if it has not yet reached definitive conclusions about what it could or so do. Gabor Bella of InteropEHRate was more sceptical about the use of AI, given the differences between primary and secondary use of data. (With its 35% rate of precision, it was his opinion that AI is still out of the question for use in medical care.) For Benny Van Bruwaene, if a 90% accuracy rate for the WHO’s international classification of diseases (ICD) can be achieved, it should be possible to go further with the use of AI for care.
18:10 - Conclusions
This is a global challenge that covers all aspects of the (healthcare) value chain. A global eco-system and a European terminology server are needed. Certain problems still need solutions i.e., intellectual property rights, and investment in human resources. As an observation, “no one size fits all.” Particularly in relation to AI, citizens and patients may play a role in use cases that are related to the secondary use of health data.
For further coverage of many of the issues discussed in this webinar, see the InteropEHRate third White Paper on semantic interoperability.