The PARIS project was carried out as a BMBF‑funded sub‑project within the European ERACoSysMed programme, specifically under the ERA‑Net instrument that promotes system medicine across EU member states. The consortium was coordinated by the Simula Research Laboratory in Norway, with international partners Dr. Kristian Valen‑Sendstad (Norway) and Dr. Maxime Sermesant (France). The University Medical Center Hamburg (UKE) served as the clinical partner responsible for data acquisition and clinical validation. The project ran for 36 months from 1 July 2020 to 30 June 2023, and the funding code was 031L0239. The collaboration was structured so that the UKE team handled the collection and harmonisation of clinical variables, imaging and biomarker data, while the Norwegian and French partners supplied expertise in computational modelling and machine‑learning methods. The partners jointly developed analysis plans, performed retrospective verification in a case‑control study, and subsequently carried out prospective validation in an independent case‑control cohort, thereby ensuring that the project met its predefined objectives without deviation.

Technically, the project built on a rich dataset already available at the University Heart Center Hamburg, which included a 3 % prevalence of atrial fibrillation (AF) in the middle‑aged population and a documented rise over the past five decades. Clinical variables, genome‑wide genetic data, and a broad array of biomarkers were combined with extensive imaging data—echocardiography (including Doppler), cardiac and cerebral magnetic resonance imaging, and computed tomography. The team performed extensive biobanking of blood‑based biomarkers, DNA, and omics data, creating a comprehensive resource for the PARIS analyses. Existing risk scores such as the Framingham AF risk function were found to have only moderate discriminative power, explaining roughly 70 % of the attributable risk, and were based on outdated, population‑level data. To improve upon this, the consortium applied artificial‑intelligence and machine‑learning techniques to integrate the multimodal data and develop more precise predictive models. Computational fluid‑dynamics models of atrial flow were mapped onto the clinical imaging data and refined with patient‑specific clinical variables. The resulting AI‑based models were evaluated for clinical implementation, and their performance was validated against additional clinical datasets. While specific numerical metrics such as area‑under‑the‑curve values were not reported in the summary, the project demonstrated that the new models achieved higher discriminative ability than the traditional scores and provided a more detailed assessment of individual risk for AF and subsequent stroke. The prospective validation confirmed the robustness of the models in an independent cohort, supporting their potential for routine clinical use. Overall, the project advanced the integration of large‑scale multimodal data with machine‑learning to produce a clinically actionable risk index for atrial fibrillation and its complications.