The Clinical And Registry Entries Semantic Model (CARE-SM) is designed to represent healthcare information stored in patient data registries through the use of Semantic Web technologies, with the objective of facilitating reasoning over federated data sources. Evolving from its origins as the Common Data Element Semantic Model (CDE-SM), CARE-SM improves on this prior art by the standardization and homogenization of its core structure, and also by the addition of a contextual metadata layer with temporal and event-based information. Consistency between data elements' representations allows several implementation improvements, including simplified data transformation and improved data discoverability.
The “Big Data” era provides unprecedented opportunities to undertake large-scale analytics over combined clinical and molecular data. To achieve this, however, there is a need for standardized and interoperable healthcare data models such that federated exploration and analysis can be more easily achieved. There is an increasing number of sensitive registered patient data sources that are intended to be used for research purposes, but the lack of interoperability between data repositories thwarts this goal, causing researchers to invest valuable time finding, preparing, filtering, and combining datasets. [1,
The FAIR Data Principles[
Alarcón-Moreno); https://github.com/markwilkinson (M. D. Wilkinson)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
domain experts in achieving “FAIRness” by providing a pre-defined, generic data model that is
inherently FAIR, and leverages the more complex features of the Semantic Web such as upper
ontologies - specifically, the SemanticScience Integrated Ontology (SIO) [
CARE-SM is an expanded and enriched representation of a prior model, which was primarily
drafted to be capable of representing the Common Data Elements (CDE) for Rare Disease
Registration [
SIO has a well-defined set of design patterns [
Compared to the CDE-SM, there was a need to expand the core set of entity-relationships that were captured. Of particular relevance were the following additions to the core model: The process (e.g. a clinical procedure) is now related to several additional entities beyond the process output, including inputs, agents, routes, protocols and targets (see Figure 1). The output in the CDE model is now enhanced with a unit of measurement. Various kinds of observations will use diferent combinations of these new elements depending on the data element being modeled.
CARE-SM is built upon the Open Biological and Biomedical Ontology (OBO) Foundry [
Figure 2 provides an example of the application of the core semantic model to a specific type of data - in this case, a tumor resection surgery of a patient: A person that has the role of a patient, denoted by a patient identifier, is participating in a tumor resection process. Several entities are associated with this process, such as the intervention protocol and anatomic structure that targets the surgery (defined as lung tissue in this example). Furthermore, the administration of a drug during the intervention (denoted by a drug identifier), followed by its route of administration. Intervention comments can be also added to the clinical process to enrich the contextual information in a human-readable way.
One of the most consequential changes to the overall model when comparing CARE-SM to its predecessor is the introduction of a metadata layer that imparts context on each data element. Semantically, the contextual metadata layer groups every instance, class or property used to describe each data element. As shown in Figure 3, temporal information, in the form of time points or time intervals, is an example of the use of this layer, allowing the definition of a timeline of patient clinical encounters. The use of an encounter identifier can be added to the model to further relate several of these data elements under the same clinical episode or event, for example, a treatment regimine.
As is typical for RDF data, context is modeled using RDF-Quads [12] - that is, a fourth URI element accompanies every RDF triple. This context URI can then be used as the subject for additional triples in order to, for example, add temporal or administrative information about that data element, or to group sets of triples into other higher-level structures.
Although CARE-SM only specifies a generic data model, we have generated a set of tools and
guidelines to assist with the implementation of this model over patient data. The European
Joint Project on Rare Diseases (EJP-RD) [13] has implemented an automated workflow
for transforming tabular data into an RDF representation, which has been adapted to the
requirements of the CARE-SM models in a variety of ways since its initial use with the
CDE-SM. The CDE-SM workflow consumed data-element-specific CSV tables, where the CSV
columns were referenced in data-element-specific templates structured using the YARRRML
specification [ 14]. These YARRRML templates were transformed into the RML mapping
language [
All of these components have been linked into a larger data transformation and publication workflow called FAIR-in-a-Box (FiaB) [ 16], which utilizes a custom daemon to sequentially execute each transformation step within the confines of a docker network, minimizing the exposure of any individual component to the internet, and finally loads the CARE-SM data into a GraphDB-based Triplestore. Thus, the users of CARE-SM within FiaB need only generate a CSV file in order to become FAIR data publishers.
CARE-SM, and its predecessor, have been used in mapping activities against other standardized data models in the clinical data semantic community. Early initiatives had the objective of schema integration and harmonization between CDE-SM and both RDF and non-RDF-based schemas. One of these initiatives, in collaboration with the Critical Path Institute, led to the creation of common SPARQL [17] queries that could map both CDE-SM and Critical Path Institute’s semantic schema by leveraging a “Rosetta Stone” of shared Biolink schema concepts [18]. Other initiatives are currently under development, such as the creation of Extract, Transform, Load (ETL) workflow from CARE-SM to Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) v5.3 [19], which we hope will prove capable of transforming any encounter-based data representation represented in CARE-SM, to an equivalent Observational Health Data Sciences and Informatics (OHDSI) [20] representation.
Other initiatives focus on the creation of federated query tools. A Beacon-v2-compatible API [21] has been created for data discoverability and federation, parsing JSON-based Beacon requests into SPARQL queries executed over Triplestores. The generic adaptation of CARE-SM to the Beacon API was possible due to having a common, predictable semantic data pattern for every Beacon data filter.
Compared with its predecessor, CARE-SM simplifies many aspects of FAIR Data publishing and reuse in the clinical space. Having a single CSV template means the data provider does not have to create multiple export routines for each data element, reducing the time required to generate the data extraction layer. Moreover, this allowed the consolidation of the numerous CDE-SM YARRRML templates into a single template, enabling easier maintenance and evolution. The data model consistency achieved by reusing a single design pattern simplifies query, where the primary diference between data elements are the ontological classes that define the various sub-elements of a data type. Thus through minor adjustments to an overall SPARQL query template, any of the CARE-SM data elements can be explored in the same way. This harmonization assists the creation of toolkits and APIs around the model, for example, the implementation of the Beacon API capable of transforming non-semantic JSON calls into a set of templated SPARQL queries.
CARE-SM allows grouping, through the “context” node of RDF-Quads, of arbitrary data elements, producing linkages between multiple data models, for example, the multiple data elements that arise from a single patient encounter with the healthcare system. Few resources seem to be taking advantage (in the rare disease space) of this RDF-Quad technology, despite it being a well-documented and oficial W3 standard for RDF representation for about a decade. https://www.nature.com/articles/nbt1346. doi:10.1038/nbt1346, number: 11 Publisher: Nature Publishing Group. [10] OBO Foundry, ????. URL: https://obofoundry.org/. [11] Orphanet Rare Disease Ontology - Summary | NCBO BioPortal, ????. URL: https://bioportal.
bioontology.org/ontologies/ORDO. [12] RDF 1.2 N-Quads, ????. URL: https://www.w3.org/TR/rdf12-n-quads/. [13] EJP RD – European Joint Programme on Rare Diseases, ????. URL: https://www.
ejprarediseases.org/. [14] YARRRML, ????. URL: https://rml.io/yarrrml/spec/. [15] CARE Semantic Model Implementation, 2023. URL: https://github.com/CARE-SM/
CARE-SM-Implementation, original-date: 2023-10-09T15:57:30Z. [16] FiaB: FAIR-in-a-box, 2022. URL: https://github.com/ejp-rd-vp/FiaB, original-date: 2022-1212T10:34:05Z. [17] SPARQL 1.1 Overview, ???? URL: https://www.w3.org/TR/sparql11-overview/. [18] P. Alarcon, I. Braun, E. Hartley, D. Olson, N. Benis, R. Cornet, M. Wilkinson, R. L. Walls, Leveraging Biolink as a “Rosetta Stone” Between C-Path and EJP-RD Semantic Models Provides Emergent Interoperability, Journal of the Society for Clinical Data Management 3 (2023). URL: https://www.jscdm.org/article/id/130/. doi:10.47912/jscdm.130, number: 1 Publisher: Society for Clinical Data Management. [19] OMOP CDM v5.3, ????. URL: https://ohdsi.github.io/CommonDataModel/cdm53.html. [20] OHDSI – Observational Health Data Sciences and Informatics, ????. URL: https://www.
ohdsi.org/. [21] J. Rambla, M. Baudis, R. Ariosa, T. Beck, L. A. Fromont, A. Navarro, R. Paloots, M. Rueda, G. Saunders, B. Singh, J. D. Spalding, J. Törnroos, C. Vasallo, C. D. Veal, A. J. Brookes, Beacon v2 and Beacon networks: A “lingua franca” for federated data discovery in biomedical genomics, and beyond, Human Mutation 43 (2022) 791–799. URL: https: //onlinelibrary.wiley.com/doi/abs/10.1002/humu.24369. doi:10.1002/humu.24369, _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/humu.24369.