Introduction: Distributed healthcare research infrastructures face significant challenges when translating routine clinical data into harmonized, research-ready formats using HL7 FHIR standards. State of the Art: Existing FHIR-based pipelines such as the SMART/HL7 FHIR Bulk Data Access API, FHIR-to-OMOP mappings, and analytical services like Pathling demonstrate technical feasibility. However, most assume semantically valid FHIR data, operate within single-institution settings, and lack practical guidance for deployment across heterogeneous, regulated environments. Technical Framework and Deployment: Within the German Medical Informatics Initiative (MII) and the INTERPOLAR project, we developed an open, modular, and participatory toolchain for decentralized FHIR-based data transformation and export across multiple Data Integration Centers (DICs). The toolchain supports FHIR extraction, profile-based transformation, REDCap integration, and OMOP-compatible export. Deployment required adapting to local infrastructures, regulatory boundaries (e.g., de-identified FHIR stores, restricted network access), and clinical domain needs. Configurable modules, proxy support, and site-specific adaptations were essential for integration into operational hospital workflows. Lessons Learned: Key lessons include the necessity of early access to real data, the limitations of synthetic test data, the value of joint workshops for profile interpretation, and the need for adaptable validation tooling. Organizational knowledge gaps, inconsistent FHIR implementations, and performance issues in resource flattening were addressed through co-design and iterative rollout strategies. Validator modules are essential across technical, content, and cross-site consistency levels. Conclusion: Centralized development paired with decentralized, participatory deployment enables scalable, GDPR-compliant infrastructures for embedded clinical research. This approach offers a replicable framework for future multi-site initiatives aiming to leverage real-world data across diverse environments.
Neumann et al. (Wed,) studied this question.