Profiling W3C DCAT for use in specific domains - Experience with the HealthDCAT-AP draft standard for the European Health Data Space (EHDS)

The Data Catalog Vocabulary (DCAT) [1], along with DataCite and Schema.org, is one of the popular standards for describing metadata about datasets or data catalogs and is widely used in NFDI [2]. Developed by the W3C to promote findability through harmonized attributes and links between datasets and other relevant resources, DCAT builds on various other vocabularies. Examples include SKOS, ODRL and PROV for organizing knowledge, describing data access and usage restrictions, and tagging data provenance. DCAT itself is already useful for describing medical datasets [3]. However, it is intentionally and technically abstract. Profiling is an instrument for adapting the generic schema to an application schema. Profiling can be used to introduce cardinality restrictions, restrict value ranges, define terminology bindings and also introduce additional attributes. The advantage of this approach is that instances of a derived version of a standard are still valid instances of the superordinate standard without inflating it with rarely used attributes or restricting its application through overly restrictive specifications. One of the most popular profiles is the DCAT Application Profile (DCAT-AP) of the EU, version 3 of which is currently also available in German [4]; other countries such as the USA or UK have similar profiles. However, there are also domain-specific profiles, e.g. for geodata or statistical data. DCAT is to be used in several of the 15 planned Common European Data Spaces [5] to establish a distributed infrastructure of data holders. One of the first data spaces is the European Health Data Space (EHDS). The EHDS is mandated by a regulation of the EU Commission and will enable patients across Europe to access their health data as well as researchers to access pseudonymized health datasets via national access points. Various EU projects are currently preparing the EHDS. In the EHDS2 Pilot project, HealthDCAT-AP was specified as a profile of DCAT-AP [6]. In the TEHDAS project, the specific application is being refined from a technical and content perspective. Public consultations are also being held for this purpose, where stakeholders can provide feedback on the maturity of the draft standard. Both the Medical Informatics Initiative and NFDI4Health have seized this opportunity and also started to map their metadata schemes against HealthDCAT-AP to prepare for the interoperability of their services with the EHDS. For evaluation purposes, three datasets were prototypically modeled with a graphical editor available on the WWW and exported as RDF: a dataset from the field of hospital care in intensive care units, a clinical study with genomic data and a dataset from laboratory medicine. Overall, HealthDCAT-AP proved to be well suited as long as the attributes were kept simple and the desired data was available. Open cardinalities or open ranges are more problematic. For example, many generic stakeholder roles can be mapped, but there is lack of best practices that are desired or sufficiently relevant without being too excessive. The use of Wikidata as a reference terminology was also viewed critically, as the content and governance are untested in contrast to established and widely accepted medical terminologies like LOINC or SNOMED CT. Some attributes that are typically collected in registries for clinical research projects are missing. For advanced topics such as variable writing, sample data or provenance, feedback must be obtained from users as to whether the modeling meets their requirements.