Abstract
The OHDSI Common Data Model (CDM) is a deep information model, in which its vocabulary component plays a critical role in enabling consistent coding and query of clinical data. The objective of the study is to create methods and tools to expose the OHDSI vocabularies and mappings as the vocabulary mapping services using two HL7 FHIR core terminology resources ConceptMap and ValueSet. We discuss the benefits and challenges in building the FHIR-based terminology services.
Keywords: Reference Standards, Observational Study, Vocabulary, Controlled
Introduction
The Observational Health Data Sciences and Informatics (OHDSI) Common Data Model (CDM) is a deep information model [1] that specifies how to encode and store clinical data at a fine-grained level, ensuring that the same query can be applied consistently to databases around world. The vocabulary component of the CDM plays a critical role through leveraging data integration standards.
HL7 Fast Healthcare Interoperability Resources (FHIR) [2] is emerging as a next generation standards framework for facilitating health care and electronic health records (EHRs) data exchange. The objective of the study is create methods and tools to expose the OHDSI vocabularies and mappings using the FHIR-based terminologies servces.
Methods
We examined both the OHDSI Vocabulary CDM version 5.0.1 [3] and the STU3 Ballot version of the FHIR core terminology resources [2]. We created mappings between the Vocabulary CDM and two FHIR terminology resources with high maturity level – ValueSet and ConceptMap. We used FHIR extension mechanism to capture those fields in CDM (e.g., domain_id, vocabulary_id) that do not have corresponding mappings in FHIR. We installed an OHDSI virtual machine (VM) based on the CDM 5.0.1 version, which contains the full OHDSI Technology Stack and is loaded with the standardized vocabularies. We then created a Java-based transformation tool that invokes the HAPI-FHIR API to transform the OHDSI vocabularies and mappings as the instances of the FHIR ValueSet and ConceptMap, and loaded the instances into a local FHIR server.
Results
The OHDSI VM is loaded with 3,316,702 unique concept ids, in which 1,052,060 concepts are marked as “standard concepts” which have mappings asserted. These concepts are classified by 32 domain ids, 61 vocabulary ids and 210 concept class ids. We have successfully created tools to generate the ValueSet instances that capture the metadata of concepts and the ConceptMap instances that capture the metadata of concept mappings (an example in Figure 1). We loaded 1000 ConceptMap instances and 2000 corresponding ValueSet instances in a FHIR server and examined the search capbility enabled by FHIR.
Figure 1.
A segment of a ConceptMap instance for a mapping between two codes from SNOMED CT and ICD9-CM.
Discussion
The OHDSI Vocabulary CDM and its implementation have been successfully used in supporting clinical observational data integration and systermatic data characterization. It has provided a collection of valuable terminology service requirements that can be generalizable to address the similar needs of broader scientific communities. The FHIR terminology resources and tooling provide a standard mechanism to enable interoperable vocabulary mapping services for the OHDSI vocabularies and mappings.
Acknowledgments
This study is supported in part by NIH grants U01 HG009450, U01 CA180940, and R01 GM105688.
References
- 1.Hripcsak G, et al. Characterizing treatment pathways at scale using the OHDSI network. Proc Natl Acad Sci U S A. 2016 Jul 5;113(27):7329–36. doi: 10.1073/pnas.1510502113. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.HL7 FHIR. http://hl7.org/fhir/2016Sep/index.html. Last visited at November 24, 2016.
- 3.OMOP CDM V5.0.1. http://www.ohdsi.org/web/wiki/doku.php?id=documentation:cdm:single-page. Last visited at November 24, 2016.