Detailed Clinical Models (DCMs) have been regarded as the basis for retaining computable meaning when data are exchanged between heterogeneous computer systems. To better support clinical cancer data capturing and reporting, there is an emerging need to develop informatics solutions for standardsbased clinical models in cancer study domains. The objective of the study is to develop and evaluate a use case-driven approach that enables a Semantic Webbased cancer study metadata repository based on both ISO11179 metadata standard and Clinical Information Modeling Initiative (CIMI) Reference Model (RM). We used the common data elements (CDEs) defined in The Cancer Genome Atlas (TCGA) data dictionary, and extracted the metadata of the CDEs using the NCI Cancer Data Standards Repository (caDSR) CDE dataset rendered in the Resource Description Framework (RDF). The ITEM/ITEM_GROUP pattern defined in the latest CIMI RM is used to represent reusable model elements (mini-Archetypes). We performed a case study of the domain “clinical pharmaceutical” in the TCGA data dictionary to demonstrate the clinical utility of our approach. We produced a metadata repository with 38 clinical cancer genome study domains, comprising a rich collection of mini-Archetype pattern instances. In summary, our informatics approach leveraging Semantic Web technologies provides an effective way to build a CIMIcompliant metadata repository that would facilitate the detailed clinical modeling to support use cases beyond TCGA in clinical cancer study domains.
Detailed Clinical Models (DCMs) have been regarded as the basis for retaining
computable meaning when data are exchanged between heterogeneous computer systems
(
While the primary focus of these DCM efforts has been on interoperability between
electronic health record (EHR) systems, there are also emerging interests in the use of
DCMs in the context of clinical research and broad secondary use of EHR data. A
typical use case is the Office of the National Coordinator (ONC) Strategic Health IT
Advanced Research Projects Area 4 (SHARPn) (
To better support clinical cancer data capturing and reporting, there is an emerging
need to develop informatics solutions for standards-based clinical models in clinical
cancer study domains. For example, National Cancer Institute (NCI) has implemented
the Cancer Data Standards Repository (caDSR) (
The use case in this study is based on The Cancer Genome Atlas (TCGA) Biospecimen Core Resource (BCR) data dictionary (14). The data dictionary is used to create clinical data collection forms for different clinical cancer genome study domains. TCGA clinical data include vital status at time of report, disease-specific diagnostic information, initial treatment regiments and participant follow-up information (15). The data dictionary groups a preferred set of CDEs per TCGA cancer study domain and renders them as an XML Schema document. All clinical data collected are validated against these schemas, which provides a layer of standards-based data quality control. All the CDEs are recorded in the NCI caDSR repository, the implementation of which is based on the ISO 11179 standard. We envision the definition of a preferred set of CDEs for each clinical cancer study domain is analogous to the DCM modeling effort.
The objective of the study is to develop and evaluate a use case-driven approach that enables a Semantic Web-based metadata repository based on both ISO11179 metadata standard and Clinical Information Modeling Initiative (CIMI) Reference Model (RM). We first used the XML2RDF Transformation technology to transform TCGA data dictionary and caDSR CDE dataset from XML format to RDF-based representation. This transformation allows us to use SPARQL queries to retrieve the caDSR metadata elements that correspond to the CDEs defined in TCGA data dictionary. We then transformed the CIMI Reference Model from UML to a corresponding OWL representation and harmonized it with a subset of ISO 11179 metadata model, from which, we extracted the ITEM/ITEM_GROUP patterns out of the data structures of the CDEs in each TCGA cancer genome study sub-domain, and populated the patterns as the instances of the CIMI Reference Model schema. Finally, we performed a case study in a sub-domain clinical pharmaceutical to demonstrate clinical utility of our proposed approach. 2 2.1
ISO 11179 is an international standard known as the ISO/IEC 11179 Metadata Registry (MDR) standard (12). It consists of six parts. Part 3 of the standard uses a metamodel to describe the information modeling of a metadata registry, which provides a mechanism for understanding the precise structure and components of domainspecific models.
Figure 1 shows a diagram illustrating the high-level data description meta-model in
the ISO 11179 specification. The Data Element is one of foundational concepts in the
specification. ISO 11179 also specifies the relationships and interfaces between data
elements, value sets (i.e., enumerated value domains) and standard terminologies.
Several Semantic Web-based representations of the ISO 11179 Part 3 meta-model
have been created for projects including the XMDR project (16), Semantic MDR in a
European SALUS project (17) and CDISC2RDF in FDA PhUSE Semantic
Technology project (18). In the present study, we utilize a meta-model schema in OWL/RDF
developed in the CDISC2RDF project, which is a subset of ISO 11179 Part 3
metamodel.
The CIMI Reference Model (RM) is an information model from which CIMI’s
clinical models (i.e., archetypes) are derived (
Figure 2 shows the Version v2.0.1 of CIMI Core Model. The classes ITEM, ITEM_GROUP, and ELEMENT form a generic pattern that can be used to represent a wide variety of clinical information. We will refer to this pattern as as the “ITEM/ITEM_GROUP pattern”. ITEM is the abstract parent of both ITEM_GROUP and ELEMENT. ITEM_GROUP represents the grouping variant of ITEM as an ordered list. ELEMENT represents a “leaf” ITEM which carries no further recursion. Figure 3 shows Archetype Definition Language (ADL) (19) definition of a “Body Temperature” archetype, which illustrates how ITEM_GROUP and ELEMENT can be combined when representing a clinical concept.
NCI caDSR is part of the NCI Cancer Common Ontological Representation
Environment (caCORE) infrastructure and uses caCORE resources to support data
standardization in cancer clinical research studies (
NCI caDSR provides the ability to download CDEs in either Excel or XML format (20), which we used to download an XML image of all non-retired production CDEs (i.e., CDEs with Workflow status NOT = “RETIRED”) as of August 7, 2014. Figure 4 shows an XML rendering of the CDE “Pharmacologic Substance Begin Occurrence Month Number”.
The Cancer Genome Atlas (TCGA), a joint venture supported by the NCI and the National Human Genome Research Institute (NHGRI), is a comprehensive and coordinated effort to accelerate the understanding of the molecular basis of cancer through the application of genome analysis technologies, including large-scale genome sequencing. Being a component of TCGA Research Network, the Biospecimen Core Resource (BCR) serves as the centralized tissue processing and clinical data collection center. A BCR data dictionary has been produced using the standard CDEs from NCI caDSR. The CDEs in the data dictionary are publicly available in the XML format. In this project, we will download a snapshot of the data dictionary from the TCGA website (14).
The XML2RDF tool, developed by the Redefer project (21), was used to transform the XML based TCGA data dictionary and caDSR production CDEs into a corresponding RDF representation. We loaded the resulting RDF datasets into a 4store instance (CITE), an open-source RDF triple-store and exposed them via a SPARQL endpoint, allowing us to use the SPARQL query language to preform semantic queries across the datasets. 2.2.2 OWL-based schema for CIMI Reference Model and ISO 11179 We used the latest version of CIMI RM (v2.0.1) in XMI format. We then converted the CIMI RM from XMI to RDF format using the Redefer XML2RDF transformation services (21). We then defined the SPARQL queries to retrieve the UML based elements of the CIMI RM such as classes, attributes and associations. We created a JAVA program that produces an OWL rendering of the CIMI RM using the UML2OWL mappings specified by the Object Management Group (OMG) Ontology Definition meta-model (ODM) standard (22). We finally harmonized and created an OWL-based schema for CIMI RM and ISO11179. 2.2.3 Defining and populating reusable archetype patterns We defined reusable archetype patterns that capture the clinical cancer domains defined in TCGA data dictionary, their associated CDEs and the metadata structures (Object Class, Property, Value Domain, etc.) recorded in the caDSR data repository. We then defined a collection of SPARQL queries to retrieve the metadata elements from both TCGA data dictionary and caDSR CDE dataset. Figure 7 shows a SPARQL query example that retrieves all CDEs of the domain “clinical pharmaceutical” defined in TCGA data dictionary and their metadata recorded in caDSR CDE dataset. We also developed a JAVA program that populates all reusable archetype patterns in TCGA clinical cancer domains into the instance data against OWL-based schema as we created.
We finally performed a case study for the domain Clinical Pharmaceutical to demonstrate clinical utility of our approach. Specifically, we demonstrated how many properties and enumerated value domains are enriched for the domain through the ISO 11179-based data elements recorded in the NCI caDSR. We then evaluate clinical utility of the enriched data elements using a Medication template defined in CDISC Clinical Data Acquisition Standards Harmonization (CDASH) standard (23).
In total, TCGA data dictionary contains 38 clinical cancer domains and 775 CDEs, which covers 21 cancer types. Table 1 shows a list of examples showing the clinical cancer domains and the number of CDEs in each domain. We created an OWL rendering of CIMI RM and harmonized it with the ISO 11179 metadata model schema, in which all classes defined in the CIMI RM are asserted as the subclasses of an ISO 11179 class mms:AdministeredItem. Figure 8 shows a screenshot of Protégé 4 environment illustrating the class hierarchy of OWL-based schema for harmonized CIMI RM with ISO 11179 model. We populated reusable archetype patterns against the OWL-based schema and produced a metadata repository based in RDF format. The repository covers all 38 clinical cancer study domains, comprising 316 distinct object classes, 4719 distinct properties, 1015 non-enumerated value domains and 1795 enumerated value domains (i.e., value sets).
Table 2 shows two pattern examples extracted from the TCGA domain “clinical pharmaceutical”. Pattern 1 captures a number of CDEs asserted in the TCGA data dictionary; Pattern 2 captures equivalent metadata structures (Object Class, Property, Value Domain, etc.) recorded in the caDSR data repository. The 7 CDEs captured in Pattern 1 have their “Object Class” in common that is “Pharmacologic Substance.” The “Pharmacologic Substance” is linked with three “Property” instances: “Begin Occurrence,” “End Occurrence” and “Continue Occurrence.” The properties are associated with 4 Value Domains: “Event Year Number”, “Event Month Number,” “Event Day Number”, and “Yes No Character Indicator”.
As a case study, we looked into the domain Clinical Pharmaceutical that contains 18 CDEs. We retrieved the object classes recorded in caDSR and identified 11 distinct object classes. And then, we retrieved globally in the caSDR CDE datasets for all properties and value domains associated with the 11 object classes. Figure 9 shows a bar graph illustrating the enrichment for the domain Clinical Pharmaceutical by data element, property, value domain and enumerated value domain. The graph indicated that the domain is greatly enriched with properties and value domains associated with those 11 object classes, which forms a pool of data elements that could be used to build detailed clinical models in this domain.
500 450 400 350 300 250 200 150 100 50 0 Data Element
Property
Value Domain 6
124 Enumerated
Value Domain Before enrichment
After enrichment To evaluate clinical utility of our approach, we aligned the data elements between CDASH Medication and TCGA Clinical Pharmaceutical. Table 3 shows the alignment results. Out of 20 CDASH data elements with their data collection questions, 9 of them aligned with the CDEs asserted in TCGA data dictionary whereas 10 of them aligned with those enriched data elements identified from our system. We believe that the results demonstrated the enriched data elements are useful in building a clinical model for the use cases beyond original TCGA data dictionary. Cumulative Agent Total Dose Total Dose Units;Prescribed Dose Units Pharmaceutical Dosage Form Code Number Cycles Route Of Administration Year Of Drug Therapy Start;Month Of Drug Therapy Start;Day Of Drug Therapy Start Agent Administered Begin Time Prior Therapy Treatment Regimen Year Of Drug Therapy End; Month Of Drug Therapy End; Day Of Drug Therapy End Agent Administered End Time Therapy Ongoing The metadata repository system proposed in this study has the following three major implications. The first implication is that the system would enable producing a profile of CIMI-compliant DCM models for TCGA clinical cancer study domains by leveraging the best practice of DCM modeling in CIMI community. Pattern 1 as shown in Table 2 is designed to capture a preferred set of CDEs and metadata for each domain asserted in the TCGA data dictionary. The semantics captured in Pattern 1 should be equivalent to those asserted in TCGA XML Schemas. In other words, Pattern 1 serves as the CIMI-compliant representation of a preferred set of CDEs in a TCGA cancer study domain.
The second implication is that we gained new insights on how ISO 11179 standard could interact with CIMI RM for supporting detailed clinical modeling. The added value would ultimately be the ability to represent ISO 11179 based constructs as constraints on CIMI RM. Pattern 2 is designed to capture equivalent metadata structures (Object Class, Property, Value Domain, etc.) of a CDE informed by ISO 11179. As shown in Table 2, Pattern 2 is represented in a post-coordination manner following certain rules. The approach used in Pattern 2 is similar to the dissection approach that is a common practice used in the terminology space for development of re-usable terminologies. The dissection approach was originally used by the GALEN project (24). In fact, the components in the metadata structure are usually annotated with concept codes from a standard terminology. In NCI caDSR, NCI Thesaurus has been largely used for the annotation purpose. Taking a look at Pattern 2 as shown in Table 2, “Pharmacologic Substance”, an object class, has NCIt code C1909 annotated; “Begin Occurrence”, a property, has NCI codes “C25431:C25275” annotated. In addition, the post-coordination-based approach enabled us to globally retrieve all properties associated with a particular object class. For example, there are globally 40 properties associated with the object class “Pharmacologic Substance” in NCI caDSR, resulting in additional 37 more properties and 5 more associated value domains. Figure 9 also shows such enrichment for the domain Clinical Pharmaceutical. We believe that our approach would produce a rich collection of archetype patterns and constraints (e.g., datatypes, value sets, terminology bindings, etc.) that could be used to facilitate detailed clinical modeling in clinical cancer study domain for use cases beyond TCGA.
The third implication is that we demonstrated the value of using Semantic Web technologies and tools in building such metadata repository. First, we created an OWL rendering of CIMI RM. This allowed us to seamlessly integrate the CIMI RM with an existing OWL-based ISO 11179 model. We envision that CIMI RM and ISO 11179 are two complementary standards that could greatly enhance the DCM modeling and its metadata management. Second, we used XML2RDF Transformation technology to transform XML-based TCGA data dictionary and caDSR CDE dataset into RDFbased format. This allows us to use standard SPARQL query language to define queries to retrieve metadata of a CDE across datasets while this enables a highthroughput approach for globally searching metadata of nearly 50,000 CDEs recorded in the NCI caDSR. Third, we populated reusable archetype patterns against the OWLbased schema using a RDF-based representation. This will allow us to leverage the built-in OWL DL reasoning capability and the RDF validation tools such as Shape Expressions (25) to check the consistency and data quality of CIMI-compliant DCM models. 5
In summary, we developed a use case-driven approach that enables a Semantic Webbased metadata repository in support of authoring DCMs in clinical cancer study domains. All schemas and datasets produced in this study can be accessible publicly at: http://informatics.mayo.edu/caCDE-QA/index.php/Download. Future work will include 1) developing Semantic Web-based RESTful services for the archetype patterns recorded in the metadata repository; 2) building quality assurance mechanism for CIMI-compliant DCMs leveraging OWL DL reasoning and RDF validation tools; 3) creating DCM authoring tools using the metadata repository as the backend; 4) developing tools that enable the transformation of DCM models between RDF/OWL-based format and ADL-based format.
Acknowledgements: The study is supported in part by a NCI U01 Project – caCDEQA (1U01CA180940-01A1). The authors would like to thank Julie Evans and Dr. Rebecca Kush from CDISC, for their kindly support and input. 6 12) ISO 11179 Specification [September 10, 2014]. http://standards.iso.org/ittf/PubliclyAvailableStandards/c050340_ISO_IEC_1117 9-3_2013.zip 13) Warzel DB, Andonaydis C, McCurry B, Chilukuri R, Ishmukhamedov S, Covitz P. Common data element (CDE) management and deployment in clinical trials. AMIA Annual Symposium proceedings / AMIA Symposium AMIA Symposium. 2003:1048. Epub 2004/01/20. PubMed PMID: 14728551; PubMed Central PMCID: PMC1480162. 14) TCGA BCR Data Dictionary 2014 [September 10, 2014]. Available from: https://tcga-data.nci.nih.gov/docs/dictionary/. 15) TCGA Clinical Data 2014 [September 10, 2014]. Available from: https://wiki.nci.nih.gov/display/TCGA/Clinical+data. 16) XMDR Project [September 10, 2014]. http://en.wikipedia.org/wiki/XMDR 17) Semantic MDR Project [September 10, 2014].
https://github.com/srdc/semanticMDR 18) CDISC2RDF Project [September 10, 2014]. https://github.com/phuseorg/rdf.cdisc.org . 19) Body Temperature Archetype in ADL. [September 10, 2014]. https://github.com/opencimi/archetypes/blob/master/miniCIMI/CIMI-COREITEM_GROUP.body_temperature.v1.0.0.adls. 20) caDSR Downloads [September 10, 2014].
https://wiki.nci.nih.gov/display/caDSR/caDSR+Downloads. 21) Redefer Project [September 10, 2014]. http://rhizomik.net/html/redefer/. 22) OMG ODM Specification [September 10, 2014].
http://www.omg.org/spec/ODM/. 23) CDISC CDASH [September 28, 2014]. http://www.cdisc.org/cdash. 24) Rector AL, Rogers JE, Zanstra PE, Van Der Haring E; OpenGALEN.OpenGALEN: open source medical terminology and tools. AMIA Annu Symp Proc. 2003:982. 25) Shape Expressions [September 10, 2014]. http://www.w3.org/2013/ShEx/Primer.