With the support of Xt-EHR.

This session discussed the need to have data that fuels algorithmic tools that are of high quality,  and are meaningfully structured and easily sharable. Achieving impact requires a comprehensive, multi-level strategies – from local hospital workflows up to European semantic frameworks – with human beings and AI working hand-in-hand to progressively refine data ecosystems.

The session brought the two fields of electronic health records (EHRs) and artificial intelligence (AI) closer together. It looked at four themes:

• The merger of primary and secondary uses of data, and bringing them together into an integrated ecosystem which will encourage data fluidity.
• An assurance of better data quality.
• Enablers and blockers that will affect the involvement of the EHRs industry.

These three topics were introduced by EHTEL digital health facilitator, Luc Nicolas, who outlined how they have been covered in detail in a recent EHTEL working paper. He encouraged ongoing contributions to the working paper so as progress the future of algorithm-based tools and AI.

The working paper, entitled “How can EHR system users make the best of algorithm-based tools?”, lay at the heart of this session. If you didn’t yet read it, download it now.

Panellists

Moderator: Michel Silvestri, Swedish eHealth Agency, Sweden

Michel introduced himself and the Swedish eHealth Agency, a government agency which works to digitalise and improving the sharing of data in Sweden. He welcomed and introduced the three panellists who came from the private, public, and research sectors.

Dragan Sahpaski, SORSIX, North Macedonia & Australia

Dragan is the co-founder and Chief Technology Officer of SORSIX, a health technology company and EHTEL member, that has won awards for its approach to the use of the European Yellow Button. Among its strengths, SORSIX has offered more than 250 internships for engineers.

Dragan started from an observation about “garbage in, garbage out”. Alongside other legislation like the European AI Act and the GDPR, in his opinion, it is now a really good time to focus on the EHDS regulation. The legislation provides an opportunity to concentrate on topics like data fluidity, data quality, semantic interoperability, the constructive role that vendors can play, and user-centred design. Dragan’s observations included how no-code tools (like the no-code Pathways Builder that SORSIX uses), with their focus on graphics, can help to generate innovation, and the difficulties that can be posed by being based around legacy IT systems. He ended by proposing that value-based care needs to be aligned with technology and needs to have clear outputs for clinical processes.

Dr. Jonas Flechsig, Fraunhofer Institute & AI4Lungs project, Germany

Jonas is a mathematician and research associate at one of the Fraunhofer’s 70 institutes, working on helping clients whose complex problems can be expressed mathematically. A lot of his work at the Fraunhofer Institute is on decision-making and machine learning, and particularly on the optimisation of life sciences.

Jonas described how clinical decision support can be affected by data quality. Indeed, high quality data is not always available to researchers. He outlined the linkages between health/clinical data, FAIR data, and the 2025 EHDS regulation – having a clear regulatory is vital. He lastly emphasised how medical, data, legal, and EHR experts all need to work together.

Anna Benavent Navarro, Park Taulí University Hospital, Catalunya, Spain

Recognised for her leadership skills, and with 20 years’ experience, Anna is Director of Digital Strategy and Data at a hospital healthcare complex in Sabadell, Catalunya, 20 kilometres distant from Barcelona. She is also a senior advisor at the Fundació TicSalut Social, Catalunya.

Anna focused her comments on the inertia that is generally associated with legacy IT and organisational systems. While it would often be easier to start from scratch on greenfield sites, health systems cannot stop what they are doing today.  Time and timing can be problematic. She considers some of the EHDS timelines, e.g., those associated with the year 2029, to lie too far out in the future and risk the non-adherence of vendors. She also reflected on the barriers and difficulties affecting clinicians (e.g., the demands of co-design). Many clinicians do not have the time available to them to focus on what AI means for them. Yet others – especially younger clinicians – do not necessarily have the expertise to get appropriately involved. In terms of gaps, Anna also referred to the apocryphal phenomenon that “culture eats strategy for breakfast”.

Discussions

Attendees were encouraged to offer their feedback on the EHTEL EHDS Implementers’ Task Force working paper on "How can EHR system users make the best of algorithm-based tools?".

This session’s chief points of discussion were:

• Whether there is a clear distinction between primary and secondary uses of health data.

• What main developments can be expected over the next 2-3 year period.

The panellists agreed that the two types of data use (primary and secondary) cannot generally be separated and there is often a mix of both – indeed, a “data onion” model for efficient and effective data use was offered. Anna emphasised that (health) data spaces work on both uses of data. As a virtuous circle, good data quality means good AI, and good AI models help to have good data quality. Jonas commented on the difficulties faced by researchers in obtaining data for AI training. Dragan drew attention to the advantages of bringing together primary and secondary uses of health data, by treating data as a data onion with at least three layers of data.

The kinds of AI or algorithm-based  use cases that could materialise in common use over the next 2-3 year time-period included:

• Breast (cancer) screening system reminders, and the opportunity to optimise their send-out.
• AI-powered and centralised platforms
• Data structuring and validation for:

Last but not least, in response to a question posed by a member of the audience, the panellists reflected that they would like to see:

• Data to be treated non-deterministically.
• Any data produced to be identified as being augmented or generated by AI.
• Depending on the data use, the training of AI to focus on “standard patients” rather than focusing on outliers.

In conclusion

As a result of this session, everyone is invited to read the Working Paper and share their views on it. 

It is hoped that feedback both from this session and Symposium Session on Structured High-Quality Data with AI will help to:

• Influence the EHDS: Ensure real-world healthcare and innovation needs are reflected in practice.
• Bring practice to policy: Help make the EHDS regulation more realistic, sustainable, and workable.
• Strengthen EHTEL’s collective voice: Contribute to a united position that amplifies EHTEL's influence in European digital health policy.