Session supported by XiA.

The EHDS regulation addresses the syntax of health systems transactions, though not yet the use of semantic resources. What are the necessary conditions to implement a meaningful – in particular semantic –  interoperability strategy  in a dedicated organisation? Which infrastructure, tools and competences are needed? Can AI meet the clinician dream (of coding happening in the background) with little or no human intervention?

An EHTEL working paper, entitled “Increasing the availability of high-quality and structured health data, the potential of AI”, lay at the heart of this session. If you didn’t yet read it, download it now.

Introduction

Moderator: Alexander Berler, IHE Catalyst, Greece

Alexander introduced the session by outlining the importance of the EHDS regulation, its implementing acts, and how to better structure health data by using artificial intelligence (AI). The implementation of the regulation – including its implementing acts – is now in the making.

Alexander emphasised that there are already a lot of mature commercial (AI) solutions on the market.

He encouraged the session attendees to reflect on the contents of Working Paper "Increasing the availability of high-quality and structured health data, the potential of AI", which formed the basis for the session’s discussions, to shape the paper’s final recommendations.

The panellists and discussions

Tools

It is important to support the accuracy of health data. Tools like large language models, machine learning, and AI all need to be accurate.

Rachelle Kaye (KSM & K-Tor, Israel) emphasised that, in terms of tools, “we need to walk before we can run.” Start with simple examples of tools like data-gathering from clinicians’ appointments with their patients (and have the clinician monitor the data to ensure that it is accurate). Alexander echoed this sentiment by stating the desire to make the clinician’s work "easier, faster, safer.”

Gabrielle Wong (EMD Digital, Germany) offered some good practice examples, citing the instance of how Merck, a leading science and technology group and an EHTEL member, uses machine learning. She cited the work of a larger cancer institute that has had positive outcomes when applying large language models to a dataset of 8,000+ patients and using it to create a protocol for the treatment of patients 90-days after release from hospital.

Dr. Indrek Koovit (North-Estonia Medical Centre) underlined how health professionals must test and validate tools so as to ensure that “they’re helping us, not stopping us.”

Governance

In terms of governance: How can we manage semantic interoperability? How can we have human beings and AI work together?

Rachelle observed that there has to be human and AI collaboration. A key governance issue is that the use of AI in clinical care needs to be constantly monitored and double-checked, just like checking the results of generative AI.

Gabrielle proposed to have a governance committee whose members would include subject matter experts and who would check that any coding is systematic and accurate. She alerted attendees to the way in which personnel from different specialisms might use exactly the same codes but for different purposes.

For Indrek, medical staff need to find out “what AI can do for you.” On the one hand, double-checking everything will slow down work and staff members need to avoid doing double-work. On the other hand, however, he admitted that he himself would not necessarily trust an AI summary.

Workforce collaboration and training

Semantics offers a strong underpinning of AI. Yet semantic interoperability is not just a technical task, it is a “co-produced outcome”. For this reason, it is vitally important that the workforce be empowered by AI literacy and training on how to use AI. As Alexander emphasised that investment in AI literacy is not a cost. It’s a direct investment in data quality.

For Rachelle, the investment in training is well worthwhile. She suggested that medical schools and nursing schools make AI, data, and digital-related topics all part of the curriculum.

For Indrek, the important aspect is the collaboration between technicians and doctors. “More cooperation equals more understanding and more trust.” 

For Alexander, engineers definitely need to be involved in training as well.

Overall, the importance of not just creating more “more ‘busy-work’” or increasing the clinician burden for medical staff.

Procurement

Among important procurement topics are: How to create marketplace for AI and high-quality data tools? Are there ways to make business procurement requests clearer?

Panellists like to see AI tools drawn into the procurement cycle. They highlighted that attention needs to be paid to obtaining more vendor openness and transparency.

Further suggestions included:

• Creating a certification system for vendors (to enable checking and auditing).
• Exploring two procurement options, i.e., competitive dialogue and pre-commercial procurement.
• Looking at what various areas of legislation and international standards have to offer, like the Medical Devices Act, GDPR, ISO, or the 2014/24/EU Directive on Public Procurement.

Polls

Five short polls answered by 30+ attendees showed  overall agreement  with several statements:

✅ Value that emerges when data carries consistent meaning.

✅ The encouragement of Member States to adopt a semantic governance framework.

✅ The integration of AI-assisted structuring into future EU regulatory artefacts.

In conclusion

The session covered several key issues all related to AI: governance; workforce collaboration and training; and the importance of procurement.

Working Paper 2 on “Increasing the availability of high-quality and structured health data, the potential of AI”, lay at the heart of this session. Following all the session’s brainstorming, there is a desire to evolve the content of the working paper, to improve it, and to make its content more widely accepted.

Look out also for more news in the short-term future on the work of EHTEL’s EHDS Implementers’ Task Force not only on this working paper, but also others!