Despite this flourishing national and European context, a critical limitation remains. While the BNDMR collects data across all rare diseases, it is primarily designed for care coordination and epidemiological monitoring, thus lacking ophthalmology-specific details such as imaging data or visual function measurements. Conversely, REDgistry aggregates data at the European level but cannot capture the full spectrum of national patient cohorts, potentially overlooking local research priorities and multicenter project feasibility. To address these limitations, comprehensive national rare disease registries should be integrated with disease-specific European infrastructures, creating a continuum from routine clinical data collection to highly specialized ophthalmic research while maintaining international interoperability.

To address this gap, clinicians from the SENSGENE network created FREDD (French Rare Eye Disease Database) [26], a dedicated health data warehouse (HDW) designed to centralize, structure, and enrich clinical data from RED patients across multiple expert centers. FREDD is intentionally designed to follow, as closely as possible, existing and widely adopted principles of rare disease data infrastructures in terms of technical architecture, interoperability standards, and data governance models to ensure alignment with established national and European ecosystems. FREDD aims to provide a nationally coordinated repository for RED data; enable national multicenter research, cohort building, and longitudinal studies; facilitate interoperability with the French BNDMR, ERN-EYE, and the emerging EHDS; and support the development of new diagnostic and therapeutic strategies tailored to RED.

This multiscale articulation—from local clinical data collection to national structuring and European integration—is illustrated in Figure 1.

This manuscript presents the concrete deployment of this national infrastructure, detailing its methodological framework while providing a critical appraisal of both the enabling factors and the operational challenges encountered during its initial phases. To ensure a rigorous and standardized presentation of our system architecture, governance, and deployment outcomes, the manuscript has been formatted as an implementation report and adheres to the iCHECK-DH reporting framework [27].

This article describes the development and implementation of the FREDD database and the assessment of its data collection and integration processes. This did not involve a research project using FREDD data for secondary analysis or the generation of new clinical research findings. The FREDD database has been authorized by the French data protection authority (CNIL) for health data processing under decision DT-2024-008 dated April 6, 2024. Access to FREDD data for secondary research use is subject to a separate governance and data access procedure.

The CNIL HDW reference framework [28] underpins FREDD’s design and operation, guiding both its technical architecture and regulatory implementation. It defines more than 100 organizational, technical, and legal requirements across 12 domains. A data protection impact assessment (DPIA) serves as the core regulatory document demonstrating compliance.

To support multicenter data collection, postcollection quality control, and secure research reuse, FREDD relies on a centralized 3-layer HDW architecture:

First, the collection layer uses a dedicated electronic case report form (eCRF) developed with SKEZIA, a compliant and cost-effective commercial platform developed by SKEZI [29], for standardized clinical data capture and early pseudonymization at participating sites level. Second, the storage and processing layer is hosted within the French National Institute of Health and Medical Research (Institut national de la santé et de la recherche médicale; INSERM) France Cohortes infrastructure [30], where clinical datasets, genetic data, and high-volume imaging batches are securely ingested, harmonized, and curated. Third, the research layer provides project-specific, isolated virtual environments for governance-approved secondary data analysis under strict access permissions.

Data transfers between collection and processing environments currently rely on standardized, semiautomated batch exports via FileSender [31]. Access control follows the principle of least privilege, enforced through a formalized data access matrix. The overall system architecture and end-to-end data life cycle are illustrated in Figure 2. Granular technical specifications are detailed in Multimedia Appendix 1.

While FREDD is designed to ultimately aggregate data for all patients affected by any RED nationwide, the initial pilot phase implements a progressive and controlled deployment strategy. To validate the infrastructure, evaluate the warehouse’s scientific value, and assess center engagement and adoption of the new data-sharing workflows, initial data collection focuses on a selected group of pilot centers and restricts its clinical scope to inherited retinal dystrophies (IRDs). Within this broader IRD population, particular emphasis was placed on retinitis pigmentosa (RP), which represented the largest patient cohort available for early evaluation.

To maximize semantic and technical interoperability, the dataset is built upon national and international standards. As illustrated in Figure 3, the FREDD data model integrates the French BNDMR core framework (via the Set de Données Minimum–Maladies Rares [32]) and overlaps with the European REDgistry schema, which is conceptually derived from the Joint Research Centre common data elements [33,34]. To address the limitations of these generic rare disease frameworks, FREDD extends this core model with ophthalmology-specific variables and imaging data.

To significantly reduce manual data entry, FREDD uses FREDDEX [35], a dedicated parsing and mapping tool that automatically retrieves clinical data already available in each patient’s BaMaRa record upon prospective enrollment into FREDD. Among the 447 variables defined in the FREDD dataset model, 281 are potentially retrievable from BaMaRa, corresponding to a theoretical maximum prefill capacity of 62.9%. This represents the upper interoperability potential of the pipeline, while the effective completion rate remains dependent on source data completeness and quality. Considering duplicated or repeated variable structures within FREDD (eg, main and additional diagnoses), these 281 target variables correspond to 83 unique BaMaRa source variables that can be extracted and mapped to FREDD. Among these source variables, 23 are mandatory BaMaRa fields and are therefore systematically available, providing a guaranteed baseline availability of 28% of BaMaRa-derived variables.

FREDDEX extracts BaMaRa exports, maps the variables to their corresponding eCRF fields, and enforces international domain-specific ontologies and standards, including disease classification using Orphanet (ORPHA) codes [36], phenotypic annotation using Human Phenotype Ontology (HPO) terms [37], and genetic nomenclature following HUGO Gene Nomenclature Committee (HGNC) [38] and Human Genome Variation Society [39] standards for gene and variant identification.

Furthermore, a complementary tool currently under development, REDDEX, leverages these identical interoperability principles to enable automated, structured data transfer from FREDD back toward the European REDgistry infrastructure. REDDEX is currently undergoing production testing and is scheduled to be fully operational by late 2026.

FREDD is an initiative developed and led by the SENSGENE rare disease network, which drove the inception of the project. Administratively and legally, FREDD operates under the responsibility of the INSERM [40], acting as the central data controller.

To demonstrate the feasibility of the architecture and gather early validation data, FREDD is currently deployed across 5 pilot expert centers within the SENSGENE network, which actively feed data into the warehouse:

The FREDD legal framework governs all data life cycle operations through 3 central pillars:

First, governance and data reuse, where a steering committee oversees strategic policies twice yearly, data access for reuse requires a formal review of all research proposals by the scientific and ethics committee, and researchers receive variable-specific completeness reports before extraction to guide data quality strategies.

Second, legal and ethical formalization, where operational standards are defined in a governance charter approved in 2025, complemented by data-sharing agreements with participating centers.

Third, patient rights and transparency, where to ensure full GDPR and CNIL compliance, patients are informed through a dedicated transparency portal and physical brochures in clinical centers, with accessible formats specifically provided for visually impaired patients.

Following the DPIA submission in December 2022, FREDD obtained official CNIL authorization in April 2024. The collection layer became operational and compliant in March 2025, allowing structured data entry to begin while the full France Cohortes infrastructure finalized compliance in November 2025. During this interim phase, collected datasets remained confined to the collection layer; imaging integration was initiated thereafter. The detailed timeline is available in Multimedia Appendix 5.

The launch and development of FREDD were made possible by the RaReTiA project (ANR-21-PMRB-0009), initiated in October 2022. This project was funded through a specific call for proposals launched by the French Ministries of Health and Research under PNMR3 and the France 2030 program, titled Accelerating Research and Innovation on Rare Diseases through Databases.

This dedicated budget successfully covered the entire “build” phase of FREDD, as well as its initial deployment and the beginning of the “run” phase. However, because this institutional funding is time-limited, it does not cover the expenses associated with FREDD beyond the scope and duration of the RaReTiA project.

Over a 1-year data acquisition period spanning from March 2025 to March 2026, FREDD successfully included a total of 1649 patients across the 5 participating pilot centers, demonstrating a steady and robust inclusion dynamic (Figure 4). Image collection was initiated in November 2025, and more than 10,000 ophthalmic images were subsequently integrated into the warehouse during the study period.

Among the 1649 included IRD patients, 763 (46.3%) had a diagnosis of RP (ORPHA:791). According to the February 2026 BNDMR national report [41], 3764 patients with RP have been registered nationwide since the deployment of the registry. Therefore, within its first year of implementation and across only 5 expert centers, FREDD has already enriched data corresponding to 20.3% (763/3764) of the national BNDMR RP cohort. These patients originated from existing BaMaRa records and were further characterized through the systematic collection of disease-specific ophthalmological variables, multimodal imaging data, centralized quality control procedures, and research-oriented governance. This rapid enrichment of a substantial proportion of the national RP cohort highlights the potential of disease-specific data infrastructures to transform routine rare disease records into structured, research-ready resources.

Overall, the FREDD database demonstrates high completeness for mandatory variables (Figure 5A). Figure 5B shows a heterogeneous completeness distribution, with many patients between 30% and 60%, primarily corresponding to FREDDEX-imported records that were prefilled according to the information available in BaMaRa and are progressively enriched through routine curation. Approximately one-quarter (456/1649, 27.7%) of patients have fully completed (100%) the mandatory data.

Data quality analysis (Figure 5A) identified inconsistencies in 7% (110/1649) of patients. These mainly reflect logical inconsistencies related to incomplete records rather than incorrect data entry, such as missing dependent variables.

Slightly less than 40% (605/1649, 36.7%) of patients had at least one associated image (Figure 5A). Image completeness varied across modalities (Figure 5C), reaching nearly 50% (767/1649, 465% and .768/1649, 46.6%) for fundus photography versus approximately 30% (455/1649, 27.6% and 451/1649, 27.3%) .for OCT. This difference reflects both routine clinical practice and operational constraints, as OCT and autofluorescence are acquired less systematically and imaging integration into France Cohortes requires additional workflows outside SKEZIA.

Expanding FREDD from 5 pilot centers to a nationwide infrastructure will introduce organizational and technical challenges. Our implementation experience suggests that clinical data harmonization is unlikely to become the main bottleneck, as the common data model and standardized collection procedures proved transferable across centers despite differing local workflows. Instead, scalability will primarily depend on the storage and processing of ophthalmic imaging, the largest component of the dataset volume. Progressive onboarding should nevertheless allow infrastructure scaling over time.

Beyond structured clinical variables, ophthalmic imaging presents additional challenges for AI applications. Although the current workflow enables secure image integration, image standardization and normalization still require validation through ongoing research projects. Future work will therefore focus on refining these processes while developing scalable, multiexpert annotation frameworks for rare ocular manifestations.

Moreover, as FREDD expands beyond inherited retinal disorders, the data model will require disease-specific extensions. Its modular architecture, however, allows new ophthalmology-specific modules to be integrated while preserving the shared rare disease data core.

Sustaining FREDD beyond its initial funding period presents significant financial and organizational challenges. While maintaining the France Cohortes hosting environment and the SKEZIA platform generates recurring costs, human resources—including project management, data curation, regulatory oversight, quality control, and local data collection—represent the largest long-term expenditure.

As disease-specific HDWs currently receive limited governmental support for routine operations, FREDD is developing a data access and cost-recovery framework aligned with the HDH Starter Kit for HDWs [42]. This framework relies on three principles:

As disease-specific HDWs remain relatively new infrastructures, sustainable funding models are still evolving. FREDD’s long-term viability will therefore depend on iterative refinement based on operational experience, user demand, and evolving national and European funding priorities.

The FREDD initiative demonstrates that strict adherence to the CNIL HDW framework from the very outset does not hinder technological innovation. Instead, this early regulatory integration serves as a foundational catalyst for institutional trust—providing the exact governance guarantees that health care networks now demand before releasing their high-value datasets into collaborative research ecosystems.

While establishing this framework required substantial upfront organizational and financial investments over several years, the rapid inclusion of 1649 patients within the first year of active data collection signals that FREDD has successfully moved past its implementation phase.

At this stage, the critical challenge shifts from infrastructure setup to scientific activation. The true value of the warehouse will now depend on the systematic launch of multicenter research projects, the deployment of AI-driven clinical models, and its deep integration into national and European data ecosystems. By strengthening direct interoperability with cross-border infrastructures such as the REDgistry, FREDD serves as a scalable research engine whose long-term impact will transform collaborative research in the field of rare diseases.