Terminology Modeling for an Enterprise Laboratory Orders Catalog

Abstract

Laboratory test orders are used in a variety of clinical information systems at Partners HealthCare. At present, each site at Partners manages its own set of laboratory orders with locally defined codes. Our current plan is to implement an enterprise catalog, where laboratory test orders are mapped to reference terminologies and codes from different sites are mapped to each other. This paper describes the terminology modeling effort that preceded the implementation of the enterprise laboratory orders catalog. In particular, we present our experience in adapting HL7’s “Common Terminology Services 2 – Upper Level Class Model” as a terminology metamodel for guiding the development of fully specified laboratory orders and related services.

Introduction

Partners HealthCare was founded in 1994 by Brigham and Women's Hospital and Massachusetts General Hospital. It is an integrated health care system that also includes community hospitals and health centers, specialty facilities, and other healthrelated entities. Clinical information systems at the various sites, including order entry and electronic medical records, have been maintaining local laboratory test order dictionaries. We are currently working on an enterprise catalog of laboratory orders to reconcile these local dictionaries and provide mappings between local codes, and also mappings to external terminologies. The catalog will become the single ‘source of truth’ for laboratory orders and it will enable translation services between site-specific order codes and to/from external codes. We expect that the catalog will not only improve the clinical data interoperability across the enterprise, but also enable advanced computerized decision support interventions to improve care quality and efficiency.

Terminology modeling and alignment with reference standards were considered critical to the success of the enterprise laboratory orders catalog project. The terminology modeling effort guided the definition and specification of the meaning and structure of the laboratory test orders, leading to consistent definitions, interpretation, and processing. The project goals included investigating whether available reference terminologies could be used to encode the meaning of the laboratory test orders, and whether an extensible terminology meta-model could be used to represent, transform, and reconcile the structure of the local laboratory order dictionaries and the reference terminologies.

In this paper, we will present the terminology modeling process that we have used, including the analysis, selection, and validation of a meta-model. In particular, we will present our experience in adapting the HL7 “Common Terminology Service 2 Service Functional Model – Upper Level Class Model”¹ as a meta-model for guiding the development of consistent laboratory test order concepts and related services. We will also present our findings related to terminology standards applicable to this domain, along with a detailed discussion of relevant challenges and important issues.

Modeling Process

An iterative and incremental process was used in this project, starting with a systematic review and analysis of existing or evolving standards for information and terminology modeling. The analysis generated a set of requirements that were ultimately used as criteria for selecting a terminology metamodel that could be used to create laboratory test order concepts, and subsequently organize these concepts into a comprehensive enterprise catalog. The selected meta-model was later validated using a set of test cases.

Standards Investigation

Information interoperability among diverse healthcare information systems is a key component in leveraging computer technology to improve healthcare quality, efficiency, and reduce costs². In particular, a “Services Oriented Architecture” (SOA)³ enables system interoperability and integration in distributed and heterogeneous environments. Many healthcare organizations, including Partners, have been adopting SOA as the fundamental strategy to achieve clinical systems integration. However, an interoperable SOA has to rely on standards for proper structuring and encoding of the data being exchanged through services. As a result, “standards development organizations” (SDOs), such as “Health Level Seven” (HL7)⁴ and the “Object Management Group” (OMG)⁵, have been making concentrated efforts to define interoperable health care services. An excellent example is the “Common Terminology Services” (CTS) specification⁶.

Many healthcare standards have been developed over the years by SDOs. Table 1 shows some examples of existing or evolving standards created by ISO⁷, W3C⁸, OMG, and HL7. While the domain of interest and the intended scope of each standard may sometimes be different, these efforts provide rules and definitions that specify how to formalize and represent information, carry out a process, or produce a product. Given the focus of this paper, ‘representational standards’ of interest can be generally categorized into there levels: (a) ontology, (b) terminology, and (c) information representation. Standards for information representation (e.g., UML, a modeling language) have been widely studied and adapted by healthcare information applications. We list HL7’s “Reference Information Model” (RIM)⁹ at this level because it provides an abstract representation of the semantic connections that exist between the information expressed in HL7 standards, particularly messages. Only a few clinical applications have applied standards for terminology management. At the level of ontology, most of work is still in active research state.

Table 1.

Relevant Existing or Evolving Standards for Information, Terminology and Ontology Representation^*

	Examples in Each Standard Development Organization
	ISO	W3C	OMG	HL7
Information	-	XML	UML	RIM
Terminology	TC37/11179	Simple Knowledge Organization Systems	Lexicon Query Services (LQS)	Common Terminology Services (CTS)
			CTS 2
Ontology	Common Logic	OWL/RDF	Ontology Definition Model	-

^*ISO: International Organization for Standardization; W3C: The World Wide Web Consortium; OMG: Object Management Group; HL7: Health Level Seven; UML: Unified Modeling Language; RIM: Reference Information Model.

Modeling efforts related to these standards were studied in detail, but taking into account the distinctions between terminology models and information models. A terminology model represents concepts (‘meanings’) at a definitional layer, and it commonly includes constructs such as concepts, attributes, designations, relationships, and the like. An information model represents events and observations at a transactional layer, specifying semantic and structural characteristics of the information. Information models are commonly instantiated with terminology concepts, and can be used to compose, qualify, and contextualize these concepts into complete information statements.

Taking into account this distinction between models, it became clear to us that our initial efforts should be focused on terminology modeling. Given the variety of local or reference terminologies for laboratory concepts, we quickly identified the need for a terminology meta-model, capable of ensuring the necessary representational consistency. In general, a meta-model is a model that explicitly specifies the constructs and rules needed for developing or integrating specific models within a domain of interest. A terminology meta-model also provides a common and consistent abstract framework for developing specifications for terminology service interfaces to query and retrieve terminological content.

Model Identification and Adoption

1. Requirements for Creating an Enterprise Lab Orders Catalog

In order to determine if any of the existing models reviewed could be used or adapted to meet our needs, we examined the details represented by the various local dictionaries. Several use cases, including an editor (e.g., add a laboratory test order) and a set of services (e.g., translate local codes to a common enterprise code), were discussed and vetted with the local sites, along with detailed business and functional requirements. Based on the analysis, the terminology meta-model had to meet a set of key requirements for creating an enterprise laboratory orders catalog, as shown in Figure 1. The model had to represent atomic and composite laboratory test orders (i.e., panels), synonyms, and relationships between orders. In addition, the model had to support concept classifications (e.g., by disease groups), mappings, and complete life cycle management, including proper versioning.

Figure 1.

Modeling Requirements for Creating an Enterprise Laboratory Orders Catalog

2. Requirements for Integrating Various Terminologies

Besides the basic requirements mentioned above, in order to achieve a high level of interoperability, the terminology meta-model should be able to represent the content specified in terminologies within and outside Partners, and it should be able to integrate the constructs and rules specified in each terminology model.

On one hand, the meta-model had to represent, transform, and integrate details specified in the different local laboratory order dictionaries, usually maintained in relational databases. Each laboratory test order often included a unique identifier and a set of attributes such as name, specimen, method, and category, among others.

On the other hand, we also investigated whether one of the existing reference terminologies could be used to encode the laboratory orders and whether the terminology can be integrated using the meta-model. The references terminologies that we studied included LOINC, SNOMED CT, and CPT. While the detailed analysis about their coverage and suitability is out of the scope of this paper, we briefly summarize these terminologies and their underlying model below.

LOINC®¹⁰: a pre-coordinated code system for identifying individual laboratory and clinical observations/results. The LOINC database provides universal identifiers for observations in HL7 messages. Each LOINC name is defined in terms of six major axes: component or analyte, property measured, time aspect, system, scale and method. The initial emphasis of LOINC was on developing codes for laboratory results and clinical observations, and not for laboratory test orders. However, LOINC now includes some single laboratory orders, and a few common laboratory test panels.

SNOMED CT¹¹: a multiaxial, hierarchical, and systematically organized medical terminology covering most areas of clinical information such as problems, findings, procedures, etc. It consists of concepts and their descriptions, hierarchies, relationships and subsets. SNOMED CT is a compositional concept system, which means that concepts can be specialized by combinations with other concepts. In SNOMED CT, “laboratory test” is organized under a subtype of the axis “procedure,” called “laboratory procedure.”

CPT¹²: a medical nomenclature used to describe and report medical, surgical and diagnostic services and procedures for administrative, financial, and analytical purposes. It consists of a list of descriptive terms and identifying codes.

At Partners, the “Clinical Data Repository” (CDR) maps and groups local laboratory order codes to LOINC codes and some clinical information systems retrieve laboratory results from CDR. Partners’ ambulatory electronic medical record maps laboratory orders to CPT (usually many to one mapping) mainly for billing purposes.

3. Important Criteria for Selecting a Suitable Terminology Meta-model

In addition to the specific requirements from the laboratory orders domain, we further developed a set of important generic criteria for selecting a suitable terminology meta-model, as follows:

• Extensibility: Can the terminology meta-model represent different domain of interest? How well concepts from different domains can be represented using constructs and rules from the meta-model? Are there details that cannot be expressed?

• Simplicity: Can the meta-model properly represent terminological content without requiring complex constructs? Are instances of the meta-model easy to understand?

• Preservation of details: Is the terminology metamodel able to preserve the particular details of different terminologies, while ensuring optimal integration and consistency?

4. Adapting the HL7 CTS 2 Upper Level Class Model as a Terminology Meta-model

Based on our analysis of the existing standards and the indentified requirements, we selected the HL7 “Common Terminology Services 2 – Upper Level Class Model (CTS2-ULCM) as the best candidate for our terminology meta-model. We validated this determination using a set of test cases that will be introduced in the model validation section below.

To demonstrate the CTS2-ULCM and later on validate the terminology meta-model, we used “Unified Modeling Language” (UML) version 2.0. The UML models were developed using IBM’s Rational Software Architect (Standard Edition 7.5).

HL7 CTS is an application programming interface (API) specification that describes the basic functionality that will be needed by HL7 version 3.x software implementations to query and retrieve terminological content. CTS defines the minimum set of services required for terminology interoperability within the scope of HL7 messages.

HL7 CTS 2 Service Functional Model (SFM) was developed by the HealthCare Services Specification Project (HSSP), an HL7-OMG joint endeavor. Extending the original CTS standard, CTS 2 SFM defines the functional requirements of a set of services that allow the representation, retrieval, and maintenance of terminology content either locally, or across a federation of terminology service nodes.

HL7 CTS 2 SFM – Upper Level Class Model (ULCM) specifies a concept based terminology model that is capable of representing most types of structured medical terminologies. This model outlines the various components of a terminology, along with the cardinality between these components. Figure 2 shows the CTS2-ULCM, including classes, attributes, and associations between classes. Please refer to the document¹ for more details.

Figure 2.

HL7 CTS2 SFM - Upper Level Class Model (adapted from¹).

Model Validation

In order to validate that CTS2-ULCM was able to meet the identified requirements, we created a set of scenarios and test cases specific for the labotory test orders. These test cases included using CTS2-ULCM to represent concepts, associations, and constraints present in the different order dictionaries, including isolated laboratory test orders (e.g., C-reactive protein), test panels (e.g. electrolytes panel), mappings, and other semantic relationships between order concepts. The test cases also explored lifecycle and versioning requirements.

We concluded that CTS2-ULCM was capable of representing the structures and meanings specified in LOINC, CPT, SNOMED CT, and the various local dictionaries. Components like “code system,” “concept,” “designation,” “value set,” “pick list,” and “versioning” can be efficiently represented using constructs in CTS2-ULCM. Similarly, compositional relationships between a laboratory test panel and its discrete laboratory test orders, hierarchical relationships (e.g., “is_a”) between two concepts, and mapping relationships between concepts were also properly represented. Figure 3 illustrates an example of a mapping relationship.

Figure 3.

UML object diagram illustrating laboratory concepts being mapped to a LOINC concept using a ConceptAssociation. The values of class attributes (e.g., conceptCode) are simplified.

In CTS2-ULCM, a Concept can have zero to many ConceptProperties. ConceptProperties can include focus (analyte), measuredProperty, specimen, method, and so on (see examples in Table 2).

Table 2.

Properties required by laboratory order concepts

Concept Properties	Focus	Specimen	Measured Property	Method
Values*	C-Reactive Protein	Serums, plasma, body fluid, etc.	Mass, arbitrary concentration	high sensitivity

^*We use C-Reactive Protein as an example. The possible values of the four ConceptProperties are listed in the table.

In Table 2, the laboratory concept ‘C-reactive Protein’ (CRP) has focus ‘CRP’. To measure CRP, the specimen can be varied, having values of “serum,” “plasma,” “body fluid,” etc. This phenomenon indicates a relationship that the value of specimen qualifies the value of focus. Similarly, the value of method (e.g., “high sensitivity”) qualifies the value of measuredProperty (e.g., “mass concentration”).

CTS2-ULCM directly supports a compositional structure, where the value of the ConceptProperty is a string (ST), such as “serum.” However, such approach does not promote information reuse because the value of the property is not encoded. Also, semantic relationships between properties are not captured, so relevant semantic details are not represented. Our solution to this problem was to represent the value of a ConceptProperty as a Concept, that is, the “value” of ConceptProperty is also a ‘conceptCode” from the Concept class (see Figure 4). This way, semantic details are properly represented and becomes reusable, plus they are easily retrievable.

Figure 4.

Modeling the value of a ConceptProperty as a Concept, that is, the “value” of ConceptProperty is also a ‘conceptCode” from the Concept class. The values of class attributes are simplified.

Discussion

In summary, we found that a reference terminology meta-model, along with the adoption of the Services Oriented Architecture, is key for improving system interoperability, where corresponding services can be developed to use the meta-model to represent, map, and harmonize the variations in the local dictionaries and the reference terminologies. We further demonstrated the use of the meta-model constructs to represent concepts, associations, and other information present in each order dictionary. Our results showed that the CTS2-ULCM meets our requirements and is capable of ensuring the necessary representational consistency. However, our findings also suggest that future studies should focus on how to use the generic meta-model to represent semantic details (such as semantic relationships) and apply to other clinical domains (e.g., medications).

The ability to search and retrieve terminology resources is a required function for many clinical information systems. Consistent use of terminology resources enables interoperability and facilitates the implementation of advanced functions like computerized decision support. However, similar to Partners, most institutions do not have enterprisewide terminology resources. HL7 CTS 2 defines a standard set of terminology APIs that can be used by existing systems to access terminology resources. Given the diversity of terminology models and data dictionaries currently available, CTS 2 also includes an abstract terminology meta-model that guides the implementation of the APIs. In particular, the CTS2-ULCM serves as a terminology meta-model capable of reconciling disparate terminology models. The CTS2-ULCM meta-model maintains the semantics of each individual terminology, while providing an efficient mechanism for modeling and exchange of terminology content.

In this project, the CTS2-ULCM meta-model was used as a foundation for creating a catalog of laboratory orders, along with the specification of runtime catalog services. Beyond the laboratory orders catalog, we have been considering other domains including orders for medications, nursing, dietary, consults, etc. The methodology and process developed in this project, i.e., the modeling process, analysis of existing standards, modeling languages, and tools, has helped create a set of best practices that we expect to extend to other enterprise terminology content projects.

A challenge that most organizations face is reconciling local dictionaries and mapping them to external terminologies, including reference standards, when the underlying meaning of the concept varies with location and context of use. For example, in the case of laboratory orders for test like “blood cultures,” the definition of the test in terms of the results that will be produced can vary from site to site. A common catalog of laboratory test orders has to be able to represent and map these types of concepts. In addition, associations between orders and results, particularly in the context of laboratory tests, is also challenging. A good terminology metamodel has to be able to represent and maintain these associations, without introducing exessive complexity to the model and to the resulting concept instances.

We also want to acknowledge that a new version of the HL7 CTS 2 SFM was released right before the submission of this paper. The CTS2-ULCM has been modified and updated, and the new model includes more features in terms of version control, sub-setting, and context of use. This new version seems to address some of the critical features explained above and we have started to study it.

Conclusion

In summary, HL7’s CTS2-ULCM is capable of serving as a terminology meta-model for developing an enterprise-wide laboratory orders catalog. Next steps include the systematic instantiation of the proposed terminology meta-model to create the comprehensive catalog of laboratory orders that has been specified. In addition to the catalog content, the meta-model will also guide the development of tools and services, aligning this effort with other terminology efforts within and outside Partners.