Essentials for Mapping from a Clinical Terminology

Margo Imel, RHIT, MBA, Kathy Giannangelo, RHIA, CCS, and Brian Levy, MD

This paper explains the principles of mapping from a clinical terminology, such as SNOMED Clinical Terms (SNOMED CT®), to other classifications and terminologies. It discusses the benefits and challenges in designing a map and the rationale behind developing a map to support different clinical and administrative purposes in an electronic health record (EHR). It also describes the development of the use case; why a use case needs to be defined for each map; and insight into the application and integration of a map into an EHR.

What Mappings Are Intended to Accomplish

There are a number of maps in existence today that accomplish different goals. Some maps, such as the annual ICD-9-CM¹ map of code revisions, are created to allow facilities and researchers to perform longitudinal studies of diseases and procedures.

In an electronic health record system (EHR), a map from a clinical reference terminology to a classification system is intended to automate the classification process. It is expected that the development of automated coding technologies will greatly impact traditional coding practice by improving productivity and enhancing the accuracy of coded data.

What Mappings May Not Accomplish

While mapping will standardize translation to a certain extent, there will be many challenges to fully automate the coding process due to the nature of mapping a granular terminology to a broad-schemed classification such as ICD-9-CM. In the coming two to three years, significant progress is anticipated in mapping technology. In the short term, human review will be required to verify the results of a map. In the long term, however, mapping will become increasingly automated and rely on robust rules for algorithmic translation. While such advances will significantly increase the efficiency and effectiveness of the coding process, it will not completely eliminate the need for review by skilled professionals, especially for cases involving reimbursement.

Why a Map between Terminologies Should Have a Defined Purpose

Mapping links one classification, nomenclature, or reference terminology scheme to another. At the time of mapping, the map is not specific to a particular patient encounter or event. Each mapping from a source scheme to a target scheme requires an articulated, specific purpose. There may be more than one purpose for mapping the schemes, and this will result in more than one kind of mapping.

Explicitly stating the purpose of the map is important to create shared expectations between the creator of the map and the user of the map regarding what the map is intended to accomplish. Stating the map's purpose is the first step to defining the scope, level of validation, and intended audience for the map. The map creator can then solicit feedback to see how well user requirements will be met by the map, which helps the creator estimate the effort needed. Use cases can be shared well in advance of the actual map product, so the purpose and scope can be refined before work begins.

Once the overall purpose and scope of the map is agreed to be useful, more detail can be added on top of the use case, such as heuristics, guidelines, specific terminology versions, validation approach, quality acceptance criteria, and planned maintenance schedule. In effect, the use case becomes a framework for the additional requirements for the map.

Since a source scheme and the target scheme to which it is mapped may each be serving different standalone purposes, the development of heuristics and guidelines is vital in development of the mapping approach. With consideration to specific target and source conventions, guidelines and purpose are critical in preserving the granularity and flexibility of source and target. The conventions of both the source and target must be respected so as to achieve the desired data outcomes.

What Is a Use Case?

In the software design world, developers write use cases to facilitate the creation of requirements. For example, in the development of an EHR application, developers want to understand how an actual user, such as a clinician would interact with the system. As an example, a use case might describe how a cardiologist signs into the EHR application to order a medication or add patient problems to a list. Use case creation is often tightly coupled to the software development cycle in that the writing, prioritizing, and refining of requirements is linked to the use cases. In fact, entire applications and modeling environments, such as proprietary software packages, and standards, such as UML, are used extensively in creating these use cases. The definition of a use case in this context includes describing a complete course of events that an "actor" or user of the system would experience. Included within use cases are the description of the "actors," priorities, pre- and post-conditions (including input and output), flow of events, user interface issues, and more.

How Does a Use Case Affect Mapping Selection?

In creating a map between different schemes, is it important to understand the purpose of each and how a map from one to another will be used as defined within the use case. Since each scheme has its own purpose, a map becomes a way to link the source and target schemes. The selection of the schema to be mapped is dependent on how the use case is defined and what is required in order to achieve the desire outcomes. Three key principles must be considered in developing a mapping use case:

• Employ data collected for one purpose and use it for another purpose
• Retain the value of data when migrating to newer database formats and schemas
• Avoid entering data multiple times with the risk of driving up cost and errors

As an example, one of the maps that is supported in SNOMED CT® is the mapping to ICD-9-CM. In this map, it is critical to understand each scheme and how each differs in its purposes and design. For purposes of understanding this map, SNOMED CT, a clinical terminology, is the source, and ICD-9-CM, a classification, is the target.

The first is step is to understand the definition of a clinical reference terminology. A reference terminology is defined as "a set of concepts and relationships that provides a common reference point for comparisons and aggregation of data about the entire health care process, recorded by multiple different individuals, systems or institutions."²

A reference terminology is an ontology of concepts and the relationships linking them. A reference terminology allows the concepts to be defined in a formal and computer-processable way. For example, hierarchical relationships can be used to identify which concepts are children of broader concepts. Along with other relationships, such as linking a disease to the body site that it involves, a semantic network is created that is useful for computer representation and processing.

By creating computable definitions, a reference terminology supports reproducible transmission of patient data between information systems. It supports consistent and understandable recording of clinical events, and it is therefore a central feature for the function of computerized patient records. In addition, the computer-readable semantic network inherent in a reference terminology is key to consistent and comprehensive analysis of clinical data.

SNOMED CT is a controlled clinical reference terminology with comprehensive coverage of diseases, clinical findings, etiologies, procedures, living organisms, and outcomes used by clinicians including physicians, dentists, nurses, and allied health professionals in recording and documenting patient data. At its simplest, SNOMED is a coded vocabulary of medical concepts and expressions used in healthcare. It is designed to provide the terminology needed to code the medical record. Controlled means that the content of the terminology is validated with careful quality assurance procedures in place to ensure that the terminology is structurally sound, biomedically accurate, and consistent with current practice. SNOMED CT uses the ANSI Terminology Structure Standard for Healthcare, which is standardized file structure for use in distributing healthcare terminology.

In contrast, a classification is, by definition, designed to group specific types of information into pre-determined categories for specific purposes, such as statistical analysis, payment and fee schedules, and other administrative purposes.

The International Classification of Disease, Ninth Revision (ICD-9) was originally designed to classify patient morbidity and mortality for reporting. Clinical modifications added to ICD-9 provided a way to classify morbidity data for indexing of medical records, medical case reviews, and ambulatory and other medical care programs, as well as for basic health statistics, resulting in the International Classifications of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). ICD-9-CM codes are commonly used for reporting, analysis, and payment of healthcare services. It is a "closed" classification, in which each patient diagnosis is "classified" into one and only one of the broad categories provided by the classification scheme. This type of broad categorization does not support many types of more detailed or clinical analysis, but it does support administrative purposes, such as billing and reporting.

The mapping process employs a standard method in which the terminology contexts, or classification description principles, are interpreted between systems. A mapping considers different purposes, levels of detail, and coding guidelines of source and target schemes. The outcome of the map is influenced by, among other things, the method by which the mapping rules have been defined and applied, the individual mapper, and the resources available. Defining mapping heuristics and applying them consistently is critical for a reliable mapping from one schema to another. Specific examples and guidelines for the mapping rules assure that all persons involved with the mapping process interpret the definitions in the same.

A map is, therefore, created with a specific purpose in mind and must be refined for particular use cases and users in diverse settings.

Several techniques are used to develop the SNOMED CT maps, including mapping files, integration files, and file integration. In addition to providing individual cross-maps, computer programs may use mapping files to translate codes and help automate the process. Full automation that takes into account the coding rules (such as disease during pregnancy) remains an elusive goal that requires knowledge-based software and some kind of human review.

Why Do Multiple Use Cases Occur?

Creating a map between different schemes requires different resources. SNOMED provides mapping resources that allow a cross-walk from SNOMED CT codes to matching codes in other systems. This means that clinical care is recorded in a patient record using SNOMED CT, and that mapping tables can be used to identify the related code(s) in another scheme.

Based upon the philosophy of "code once, use many times," the SNOMED CT mapping resources help to minimize the re-entry of data; thus, maps to other schemes facilitate reuse of data via the map so that data can be sorted, retrieved, and grouped to match different queries without having to resort to manually compiled code lists.

The SNOMED CT cross-mapping schema supports use cases for mappings to more than one target terminology. Each of the following maps or integrations in SNOMED CT has a specific use case defined.

SNOMED CT Maps and Integrations

How the mappings are integrated and used is dependent on the use case and users of this type of data, their needs, and local software implementation. For example, SNOMED CT is the merger of UK-based Clinical Terms Version 3 (CTV3) and US-based SNOMED RT. SNOMED CT data Concept Tables include a "triplet" of identifiers for SNOMED CT, CTV3, and SNOMED RT; thus, no mapping table is required. Likewise, ICD-O-3 reuses the SNOMED RT morphology codes in the M-8 and M-9 range, which are also contained directly in the SNOMED CT data tables. Concepts from the nursing terminologies, by agreement with their respective organizations, are incorporated into SNOMED CT, but a mapping table is required to link the SNOMED CT concept to the corresponding nursing terminology code.

In addition, there can be more than one use case between SNOMED CT and any given terminology. For example, there may be one mapping for all SNOMED CT clinical finding concepts to ICD-9-CM, and another mapping for only active SNOMED CT clinical finding concepts. In one scenario, all concepts may be needed if legacy databases contain codes that have since been inactivated (retired). In another scenario, only active concepts may be needed for mapping current patient data.

How Do Use Cases Affect Mapping Implementation?

The variety of content accessible from within the EHR is now almost as important as the EHR itself. This content includes decision support systems, such as drugs and drug interactions, clinical pathways, disease monographs, protocols, and others, each which may use content based on a coded terminology or classification schemes such as SNOMED CT or ICD-9-CM. The choice of appropriate coded scheme should adhere to similar design principles as the software itself--with the design of use cases and requirements. In particular, the creation of maps among the various schemes should be based upon well-defined use cases. These use cases should define the "actors" or intended users of the maps, the expected scenarios for if and how the users would be exposed to the maps, and the expected inputs and outputs (that is, how the user selects the source of the map and what the expected targets are). The flow of events and user interface issues can be suggested in the use cases, but the EHR vendors will likely embed this content within their existing graphical user interfaces. These use cases for maps should be clearly articulated prior to designing the requirements of the maps themselves.

Once defined, the expected use cases for maps among terminologies should be clearly explained for implementers of the maps. By itself, these maps will simply be stored in databases or terminology servers; thus, the EHR vendors will embed the content of the maps into existing applications and workflow processes. These implementers should incorporate the maps based on the intended use cases. Improper use of these maps could lead to user frustration and unexpected outcomes.

One of the use cases of the SNOMED CT to ICD-9-CM map is to facilitate the selection of an ICD-9-CM code for billing purposes based on the selection of a SNOMED CT concept. In the use case of this map, a SNOMED CT code represents the input to the process with ICD-9-CM codes as the output. The vendor is expected to present one or more suggested ICD-9-CM codes to the end users, possibly including some of the computerized rules built into the maps. If the user is not presented with the ICD-9-CM outputs, then improper billing codes could be sent to payers. Likewise, the computerized rules should be followed and presented to users based on their intended design as described in the use case.

Examples of Maps from SNOMED CT to the ICD Family of Classifications

The following overview provides a high-level summary of some of the use cases that are supported in SNOMED CT. Each demonstrates the different purpose and the importance of maps when SNOMED CT is integrated into an enterprise using an EHR.

SNOMED CT to ICD-9-CM Epidemiological/Statistical Mapping

The mapping from SNOMED CT to ICD-9-CM is designed to support epidemiology and   statistical and administrative reporting needs of organizations employing SNOMED CT as the core terminology for clinical descriptive purposes in an EHR. The file design of SNOMED CT supports automated cross-mapping (in future releases) by providing a technical structure that will support rule-based processing.

Example of SNOMED CT to ICD-9-CM Map

SNOMED CT		ICD-9-CM
35363006   infantile colic
9991008    abdominal colic		789.00 abdominal pain
21522001   abdominal pain

SNOMED CT to ICD-O3

The International Statistical Classification of Diseases for Oncology Third Edition (ICD-O-3) is sponsored by the World Health Organization (WHO) and is used by cancer registries for reporting the topography, morphology, and behavior of neoplasms. The SNOMED CT to ICD-O-3 map has the characteristics of both a map and integration. By agreement between the College of American Pathologists (which owns SNOMED CT) and WHO, the SNOMED CT data tables incorporate morphology codes that are identical to the morphology section of ICD-O. This is considered integration. In contrast, the ICD-O-3 topography codes are not similarly incorporated into SNOMED CT, and thus represent a many- to-one map (where many SNOMED CT topography concepts may be mapped to a single ICD-O-3 topography code).

The link from SNOMED CT to ICD-O-3 is essential because the two most important items of medical information for a cancer patient are the primary site of the tumor (its topography) and morphologic or histologic type of the tumor as diagnosed microscopically by a pathologist (its morphology). Cancer registries receiving SNOMED CT-encoded information from healthcare providers, especially anatomical pathology laboratories, provide for accurate and consistent reporting of cancer cases.

Example of the Integration between SNOMED CT and ICD-O-3 Morphology Codes

SNOMED CT		ICD-O-3
41607009   renal cell carcinoma morphology--M-83123		M83122/3--renal cell carcinoma

Example of a Mapping from SNOMED to ICD-O-3 Topography Codes

SNOMED CT		ICD-O-3
414003   gastric fundus structure-- T-57400		C16.1   gastric fundus

SNOMED CT to Nursing Terminologies

NIC, NOC, NANDA, and PNDS maps demonstrate the best example of   the "convergence" of existing standardized nursing vocabularies. Concepts from nursing classifications "converge" through their relationships with SNOMED CT concepts. The use case was developed to create maps that would enable users of SNOMED CT to interface with standardized nursing vocabularies for point-of-care documentation in electronic patient records.

For example, in the NANDA use case, NANDA Taxonomy II, developed by NANDA International , is a set of nursing diagnoses used to describe clinical judgments about individual, family, or community responses to actual and potential health problems or life processes. An analysis was performed to ensure that all NANDA concepts were included in SNOMED CT. In addition, a notation was made about which SNOMED CT concept related to which NANDA concept. This data was used to create a map between SNOMED CT and NANDA that contains the specific codes of each terminology.

The mapping was accomplished through a collaborative agreement between NANDA and the International and the College of American Pathologists and used the terminology expertise of nursing informaticists in both organizations. The map is distributed only to organizations that license both terminologies.

A similar process was used for other nursing terminologies, with the result that SNOMED CT contains concepts that cover the scope of all and contains code-to-code mappings between terminologies.

Role of a Clinical Terminology and an EHR.

Using a standard terminology within an EHR implementation is important for a number of reasons. The terminology is the linkage between patient data that is recorded and important goals of the EHR software, which may include:

• Access to complete and legible clinical data with links to medical knowledge for real-time clinical decision support
• Interoperability that permits many different sites and providers to send and receive medical data in an understandable and usable manner, thereby speeding care delivery and reducing duplicate testing and duplicate prescribing
• Computer manipulation of standardized data used to find information relevant to individual patients for the purpose of producing automatic reminders or practitioner alerts
• Capability to form queries to look at standards of care for benchmarking, measuring and interpreting effectiveness, and quality improvement.

In order to achieve these ambitious objectives, the clinical terminology captured at the patient care level needs to be integrated or interfaced with other systems or subsystems. Ideally, and over time, this will evolve to be a more seamless and automated process. Today, and in the near future, map files are needed to make these conversions. Many, including those that involve reimbursement, require expert human review.

In the interest of reuse, as mentioned earlier, the same clinical encoding can be input to multiple mappings and conversions for multiple purposes. The advent of the EHR, rather than using paper records, increases the opportunities for automation. As EHR applications become more sophisticated and easier for clinicians to use, entry can be closer to the point of care thereby increasing accuracy and completeness and reducing the cost of transcription and transformation of text-based observations. While this may seem far-fetched, the rise of the telephone in the last century may hold some insight. Telephone calls were connected manually using human operators; at one point, an observer noted that if the rate of phoning increased at the then-present rate, every person in the US would eventually need to be a telephone operator!   Today, electronic switching technology has enabled a huge increase in the number of phone calls that can be processed each day. And, as a result, actually all of us are telephone operators every time that we enter a phone number to make a call. While we take for granted entering a number on a 10-digit pad to "dial" a call, this technology and "user interface" did not always exist. Today, human intervention is rarely required. Similarly, in healthcare technology, we are on the edge of major technology improvements that will let clinicians "make their own phone calls," streamlining operations and freeing up resources that can be focused on patient care.

Endnotes

1. International Classification of Diseases 9 th Revision Clinical Modification© National Center for Health Statistics, Hyattsville, Maryland.
2. 2 Spackman, KA, Campbell, KE, Cote, RA. (1997) "SNOMED RT: a reference terminology for health care," Proceedings of the 1997 AMIA Fall Symposium, October 29, Nashville TN.
3. International Statistical Classification of Diseases for Oncology 3 rd Revision© World Health Organization, Geneva, Switzerland.
4. International Classification of Diseases 10th Revision© World Health Organization, Geneva, Switzerland.
5. Logical Observation Identifiers Names and Codes© Regenstreif Institute, Indianapolis, Indiana.
6. Office of Population, Censuses and Surveys--Classification of Surgical Operations and Procedures 4 th Revision, Crown Copyright National Health Service Information Authority, United Kingdom.
7. Nursing Interventions Classification© University of Iowa, Des Moines, Iowa.
8. Nursing Outcomes Classification© University of Iowa, Des Moines, Iowa.
9. North American Nursing Diagnosis Association © NANDA International, Philadelphia, Pennsylvania.
10. Perioperative Nursing Data Set© Association of Perioperative Registered Nurses, Denver, Colorado.
11. Clinical Terms Version 3, Crown Copyright National Health Service Information Authority, United Kingdom.
12. Systematized Nomenclature of Medicine Reference Terminology © College of American Pathologists, Northfield, Illinois.