Martin Bøgsted, Professor and Head of Centre for Clinical Data Science (CLINDA) at Aalborg University, delivered a presentation on the ARISTOTELES project at a high-profile conference on AI in healthcare held at the Novo Nordisk Foundation headquarters in Copenhagen on 15 September 2025.

The one-day event, organised under the auspices of the Danish EU Presidency and co-hosted by the European Economic and Social Committee (EESC) and Danish Society of Engineers, examined AI infrastructures and their economic and societal implications. The day was divided between a morning roundtable on AI investment and computing power for EESC delegates and Danish stakeholders, and an afternoon public conference focusing on AI and big data advancements in rare diseases, followed by discussions on quantum computing applications.

During the afternoon session focused on "Harnessing AI and Big Data for Advancements in Rare Diseases," Bøgsted presented the ARISTOTELES project's work on dynamic risk prediction in atrial fibrillation. His talk highlighted how the project is leveraging harmonised real-world data from approximately 2 million AF patients and 10 million controls across European databases, including the UK Biobank and Danish national registers.

"Europe, particularly Denmark, has world-class health data," Bøgsted noted in his presentation. "But we have too many small, local multimodal studies. To unlock our full competitive advantage, we need to discuss how we scale data collection across Europe."

He emphasised the importance of the European Health Data Space in addressing key challenges around legal frameworks, infrastructure, computational capacity, and methodology. The ARISTOTELES project, funded by Horizon Europe, exemplifies this approach by using harmonised data through the OMOP common data model on a platform hosted at the Danish National Genome Centre.

Building the Foundation: Danish Partners' Data Harmonisation Work

The data infrastructure underlying Bøgsted's presentation represents a major work by the project's Danish partners. Researchers from Aalborg University, working with the Danish Medicines Agency and the ARISTOTELES team, recently completed the first national-scale mapping of Danish healthcare data to the OMOP Common Data Model, achieving 94.7% overall mapping coverage across core clinical domains. This systematic transformation encompasses 3.8 billion records from 11 national registries covering 9.1 million individuals from 1977 to 2024. The mapping of the Danish registries was carried out primarily by Tomer Sagi, whose work was central to achieving this level of coverage and quality.

Danish healthcare uses country-specific coding systems—ICD-10-DK for diagnoses, SKS for procedures, NPU for laboratory tests, and ATC codes for medications—that aren't directly compatible with international OMOP standards like SNOMED CT, LOINC, and RxNorm. The team employed a hybrid methodology combining clinical expert curation, semi-automated tools (including Usagi), and Large Language Model-assisted classification. The results were impressive: drug exposure mapping achieved 99.93% coverage (1.34 billion records), laboratory measurements reached 99.24% coverage (1.59 billion records), procedures achieved 93.35% coverage, and diagnoses reached 89.3% coverage.

Notably, the team applied GPT-4 to classify 113,163 healthcare organisations, demonstrating AI's potential for large-scale medical terminology mapping whilst also highlighting current limitations compared to expert annotation. This technical work directly enabled the research Bøgsted presented: the Danish registry mapping identified 366,918 atrial fibrillation patients with comprehensive longitudinal data, which were then integrated with European databases for multi-national comparative studies. All mappings and transformation code have been made publicly available through the OHDSI Denmark GitHub repository.