Engineering an EHDS-aligned EHR architecture for Romania: integrating openEHR, FHIR, OMOP-CDM, and verifiable clinical RAG

Introduction The electronic health record (EHR) must simultaneously support two paths with divergent requirements: therapeutic personalization, which requires contextualized patient data, and secondary clinical use, which requires standardized data under the pressure of compliance imposed by the European Health Data Space (EHDS). Methods In this study, we adopted the design science research methodology and developed a model, which was evaluated using a descriptive design on a simulation-based clinical scenario. First, we derived five design requirements from the constraints of the Romanian healthcare ecosystem to establish the requirement coverage for the proposed model by systematically comparing realistic alternatives, including the national initiative “My e-Health” (e-SănătateaMea). Results We propose ROMED-EHR, a conceptual reference model that stratifies openEHR (persistence), HL7 Fast Healthcare Interoperability Resources (FHIR) R4 (interoperability), and the Observational Medical Outcomes Partnership Common Data Model (OMOP-CDM) (analysis). The model, which is semantically grounded in Systematized Nomenclature of Medicine - Clinical Terms (SNOMED CT), Logical Observation Identifiers Names and Codes (LOINC), and the Human Phenotype Ontology (HPO), is extended with a clinical support module based on retrieval-augmented generation (RAG) whose outputs are verified against terminologically coded source data. ROMED-EHR is positioned as a complementary solution to the national initiative for the data architecture layer. An explicit mapping contract between levels and a requirements EHDS traceability matrix are proposed. Moreover, a family of metrics for RAG accuracy, a clinically weighted variant, and a separate metric for agreement with clinical guidelines are identified. Of the seven alternatives compared, only ROMED-EHR and the same architecture without the RAG layer covered all five requirements. In a synthetic acute ischemic stroke scenario, the proposed fidelity gate rejected an unsafe thrombolysis recommendation that an assessment based only on the final answer would have accepted. Discussion It should be noted that this article is a conceptual proposal for reference, not an empirically validated system, and therefore, empirical validation is left as a future research direction.