A socio-technical assessment of AI-driven clinical data structuring for the EEHRxF: The implementation wall

Abstract

The European Health Data Space (EHDS) establishes the European Electronic Health Record Exchange Format (EEHRxF) as the foundation for cross-border interoperability. While technical specifications are formalising, a significant “Implementation Wall” persists: the challenge of transforming unstructured clinical narratives into structured, interoperable formats. In parallel, a “Technological Push” driven by Large Language Models (LLMs) offers new possibilities for automated clinical structuring. This thesis evaluates the socio-technical feasibility of AI to meet EHDS mandates. Using a dual-methodological approach, it synthesises a systematic review of 103 peer-reviewed AI articles (2020–2025) and an empirical analysis of 357 European stakeholders. Capabilities are evaluated against “First Group” priority categories (Patient Summaries, eP, eD) mandated for March 2029. Findings reveal a stark misalignment between technical potential and organisational capacity. While AI models frequently exceed F1-scores of 0,90 in lab environments, they remain restricted by a “Retrospective Trap”. Notably, over 60% of surveyed stakeholders report zero proficiency in SNOMED CT and HL7 FHIR governance, rendering the assumed “Humanin-the-Loop” architecture a structural misconception. The research concludes that successful 2029 implementation depends on shifting policy focus from algorithmic performance to “Organisational Pull,” specifically through investments in workforce semantic literacy and localised IT infrastructure.