Five Years of the SPHN RDF Journey: FAIR Enough?

1. Introduction

In 2020, the Swiss Personalized Health Network (SPHN) adopted Semantic Web technologies, particularly the Resource Description Framework (RDF) as a common standard format for representing and exchanging health-related data [1]. This stakeholder-endorsed choice fosters semantic interoperability across Switzerland’s decentralized healthcare and research landscape. Over the past five years, the SPHN Semantic Interoperability Framework has evolved into a semantically rich schema, the SPHN RDF Schema [1], encompassing 209 concepts spanning diverse domains such as patient demographics, diagnoses, procedures, laboratory tests, and imaging metadata. The framework harmonizes data provisioning and reuse for research while aligning with the FAIR (Findable, Accessible, Interoperable, and Reusable) principles.

To evaluate and iteratively improve this framework, SPHN has supported a range of projects, including four National Data Streams (NDS) and eleven Demonstrators (DEM) projects, covering diverse research use cases. These projects extended the SPHN RDF Schema to address domain-specific needs, driving its iterative evolution. This study marks a five-year milestone, evaluating both progress and persisting challenges. Specifically, it addresses: i. How has the SPHN framework evolved over five years? and, ii. How widely is it adopted, and which concepts are used?

To answer these questions, we analyzed the temporal evolution of the SPHN RDF Schema, assessed its use across SPHN projects and identified areas for improvement to further enhance the semantic interoperability and data reuse in Switzerland.

2. Methods

This study is based on a longitudinal analysis of the SPHN RDF Schema releases from 2021 until 2025. We examined the evolution of structural and conceptual changes across versions, focusing on quantifying the number and types of concepts (i.e., classes) introduced, tracking modifications in the organization of concepts, and assessing their alignment with external terminologies.

In parallel, we reviewed project-specific metadata available through the SPHN Metadata Catalog [2], a FAIR Data Point that provides an overview of health-related datasets available in Switzerland, currently focused on SPHN projects. This platform hosts quantitative and qualitative metadata, describing RDF data collected primarily from Swiss hospitals. Projects submit metadata using provided scripts to extract relevant information. While qualitative metadata was assessable, quantitative metadata could not be validated against raw data due to data access restrictions. However, we inferred insights from estimated cohort sizes and concepts defined during project planning, which were consistent with the delivered metadata. We analyzed the coverage of key semantic domains using a combination of SPARQL queries and R scripts to extract and aggregate relevant statistics.

3. Results

3.1. SPHN RDF Schema Evolution over the Years

The first release of the SPHN RDF Schema (in June 2021) included 64 concepts, primarily focused on core clinical domains such as encounters, medical devices, and allergies. In parallel, standard terminologies were provided in RDF via the SPHN DCC Terminology Service [3] to facilitate code integration and subsequent data analysis.

Typically, one major release occurs annually in Q1, occasionally followed by a minor mid-year update. Over time, the schema evolved to cover additional domains, expanding to 209 concepts by version 2025.2 (see Figure 1): laboratory tests (2022), genomics (2023), provenance, microbiology, assessments (2024), genomic variants and imaging (2025). With each release, relevant terminologies were included or updated to support the growing diversity and granularity of SPHN data, bringing the total to 18 included terminologies in release 2025.2.

3.2. Projects’ Use of the SPHN RDF Schema

Nine SPHN projects received RDF data from Swiss hospitals conforming to version 2024.1 of the SPHN RDF schema. Data were generated using the SPHN Connector [4], a tool that enables schema-driven transformation and validation. Across six projects with available metadata, we observed that between 26% and 50% of the SPHN concepts were reused (see Figure 2). In total, 113 of the 168 concepts defined in version 2024.1 were used, showing the diversity of data needs across projects. Additionally, 71 project-specific concepts were introduced across these projects, demonstrating the schema’s flexibility to accommodate diverse project-specific requirements. General concepts, such as demographics, diagnoses, and laboratory values, are typically structured and coded in clinical data platforms and consistently used across projects. In contrast, more specific content (e.g., genomic variants or oncology-related diagnoses or assessments) is often only documented in free-text form, requiring project-specific extraction and coding efforts.

4. Discussion

SPHN primarily tackles national requirements while strategically leveraging international initiatives. Adopting RDF and aligning with established terminologies helps in making Swiss health data compatible with broader frameworks (e.g., Fast Healthcare Interoperability Resources, Observational Medical Outcomes Partnership [5]). However, semantic mappings between models are often necessary, introducing potential risks of information loss due to differences in scope and granularity. The challenge lies in balancing the need for local customization with global harmonization. The SPHN schema effectively handles this duality, being used by SPHN projects and attracting international use cases [6].

At the start of SPHN, hospitals had little to no SNOMED CT or LOINC coding in their clinical data platforms. Today, in contrast, at least 6,500 distinct SNOMED CT and 4,000 LOINC codes are in use. Nevertheless, local data heterogeneity in coding hinders full interoperability, highlighting that health data requires not only well-defined semantic models but also harmonized data and coding practices across institutions, as well as supporting tools to ensure quality and usability. Feedback by projects to data providers and adaptation of data collection at the source would be beneficial.

Semantic precision versus usability is a continuous hurdle. For instance, until 2023, body height was modeled as having a performer, which is semantically inaccurate. Body height is a property defined by a value and a unit, while its measurement is a process that may involve a performer. In 2024, the SPHN RDF Schema underwent a major restructuring for better consistency and clarity. It shifted towards a process-oriented modeling approach, aligning with other initiatives [7], to clarify the distinction between entities (e.g., Result, Code) and processes (e.g., Measurement, Medical Procedure, Assessment). While this restructuring introduced more concepts and initially raised concerns among implementers; it ultimately led to predictable concept patterns and schema structural stability. This stability empowers data providers to align data with the schema, guides data users in designing new concepts, and enables automated data transformation tools. As a result, the framework now offers improved coherence, extensibility, and reduced ambiguity for future developments.

5. Conclusion

The SPHN Semantic Interoperability Framework sets a national standard for health data representation, enabling FAIR datasets for personalized health research. The schema’s breadth and maturity provide a solid foundation for data integration and cross-border collaboration. Several projects have successfully received data at scale, demonstrating the practical value of the approach. Future priorities include enhancing data quality at the source, strengthening adoption of RDF-based workflows into research environments, and developing tools for query building and data visualization.