Cardiovascular diseases are the leading cause of death globally. Healthcare systems generate heterogenous data (e.g., structured records, clinical text, imaging, physiological signals, genomics) that remain underutilized due to fragmentation, regulatory constraints and limited interoperability.
This paper proposes a federated architecture for cardiac data processing, that preserves local data control, while enabling distributed analytics. Subordinate nodes manage local repositories, Extract, Transform, Load (ETL) pipelines, and virtual datasets enriched with semantic and provenance metadata, supported by local data quality assessment and structural harmonization. A central node coordinates metadata aggregation, query routing, and global model updates, ensuring secure and auditable access. The system extends the OMOP Common Data Model for multimodal data and applies a metadata-driven compute-to-data approach without transferring raw data. This framework is validated on distributed datasets within CVDLINK project. Federated learning implemented via FeatureCloud, supports local training, secure aggregation, and deployment of global predictive models for clinical decision support. By integrating multimodal data and federated analytics, the architecture provides a scalable, secure, and interoperable platform for cardiovascular research and care.