BACKGROUND
To guarantee interoperability, structural and semantic standards must be adhered to. For exchanging medical data between information systems, the structural standard FHIR (Fast Healthcare Interoperability Resource) is recently gaining traction. In terms of semantic interoperability, the reference terminology SNOMED CT as a semantic standard enables a post-coordination in comparison to many other vocabularies. These post-coordinated expressions (PCEs) result in SNOMED CT being an expressive and flexible interlingua, which enables precise coding of medical facts, but at the expense of increased complexity and challenges with storage and processing. In addition, the boundary between the scope of semantic (terminology) and structural (information model) standards blurs, the so-called TermInfo problem.
OBJECTIVE
Although often discussed critically, the TermInfo overlap can also be explored for its beneficial potential by enabling a flexible transformation of parts of the PCEs. In this paper, an alternative solution for the storage of PCEs is presented, i. e., in combination with the FHIR data model. In the end, all components of a PCE should be expressible exclusively by pre-coordinated concepts that are linked to suitable elements of the information model.
METHODS
The approach is based on storing PCEs and/or parts of them conforming to FHIR resources. Using the Web Ontology Language for generating an OWL ClassExpression in combination with an external reasoner and semantic similarity measures, a pre-coordinated SNOMED CT concept that describes the PCE most precisely is determined as a superconcept. In addition, the non-matching attribute relationships between the superconcept and the PCE are determined as the so-called delta. Once SNOMED CT attributes have been mapped to FHIR elements manually, FHIRPath expressions can be determined for the superconcept and the delta, which in turn enable the identified pre-coordinated codes to be stored in FHIR resources.
RESULTS
A web application called PCEtoFHIR was developed to implement this approach. In a validation with 600 randomly selected pre-coordinated concepts, the correctness of the generated OWL ClassExpression could be confirmed. Additionally, two different approaches for calculating semantic similarities were validated, with the approach by Sanches et al. demonstrably yielding more precise results. In a validation of the entire approach, considering 33 already existing PCEs, the correct functionality could also be demonstrated.
CONCLUSIONS
PCEtoFHIR provides services to decompose PCEs for storing them in FHIR resources. When creating structure mappings concerning certain subdomains of SNOMED CT concepts (e.g. allergies) to desired FHIR profiles, the use of SNOMED CT Expression Templates has proven to be very useful. Domain experts can prepare templates with suitable mappings that can be reused in a constrained way by end users more easily.