Filling A Technology Need
Michael Stearns, M.D., CPC
Although the use of electronic health records is increasing, there has been a tendency to store captured data in an unstructured format. To take full advantage of the benefits of health information technology, clinical data needs to be stored as organized concepts using the constructs of controlled medical vocabularies (e.g., SNOMED CT). There are several issues that arise when attempts are made to capture and store the natural language of free-flowing medical documentation in this type of highly-structured format. This article will review these challenges and introduce the Medicapaedia, a terminology solution developed by e-MDs to overcome barriers to the implementation of controlled terminologies in electronic health records.
There is general agreement that the expanded use of electronic health records (EHRs) is essential to improvements in healthcare, research and cost savings. Although EHRs are gradually gaining momentum in the U.S., their true potential has been limited by the type of data they are gathering. The vast majority of information collected by these systems is in an unstructured format (e.g., free text). This limits the ability of systems to provide improvements in the quality of care, supplement research and improve cost effectiveness; all frequently cited justifications for investing in electronic health record systems. In order to be fully effective, EHRs need to move toward storing data in structured formats such as SNOMED CT. Once this is accomplished, systems will be better able to provide sophisticated alerts and reminders to clinicians in real-time, automatically track orders, perform sophisticated data queries that will benefit research and the quality of care, improve workflow efficiency and reduce cost.
Information stored as structured data in a standardized format (e.g., SNOMED CT, LOINC, others) can also be used by other software platforms that understand the “same language,” as the concepts are associated with codes recognized by both systems. This is referred to as semantic interoperability and is required for true portability of clinical information between systems as patient care moves from one care provider setting to another.
Years ago it was realized that well-entrenched terminologies designed for epidemiology or billing, such as ICD-9-CM and CPT (sometimes referred to as administrative terminologies), lack the structure and level of granularity needed to accurately represent concepts captured during clinical documentation. This has led to the development of highly-structured “reference” terminologies such as SNOMED CT. This terminology consists of hundreds of thousands of “atomic” concepts and synonyms supported by extensive and formal concept interrelationships. It is based on unique concepts that are defined by relationships to other concepts through hierarchical and other types of specified relationships. For example, pneumonia caused by the bacteria streptococcus pneumoniae would receive its own concept code. In addition to its hierarchical relationshipsto lung disorders and infectious disorders, it would be further defined by having site equal lung and etiology equal streptococcus.
Both lung and streptococcus would have interrelationships to other concepts, creating a rich matrix of relationships between concepts that provide the concept with a form of meaning. Description logics applications can take advantage of concept relationships to infer additional relationships and to recognize when two concepts are equivalent.
Although SNOMED CT offers several inherent advantages, electronic health record developers and clinicians have struggled with attempts to capture information in this highly-structured format during clinical documentation. The strict canonical formalisms in SNOMED CT, for example, are necessary to preserve its structure and consistency, but limit its ability to represent the subtleties of expression found in clinical documentation. SNOMED is composed of atomic concepts meant to be used as building blocks to create more complex phrases. It was not designed for use at the point-of-care but rather as a reference for system developers to compose phrases using several SNOMED CT concepts joined together (also referred to as a “code-phrase”).
For example, a commonly used phrase such as, “The patient reports mild fatigue secondary to sleep deprivation, which he feels is caused by stress at work,” could be broken down into multiple atomic reference terminology concepts. In this hypothetical example, the user would be required to select a series of reference terminology concepts including “mild,” “fatigue,” “secondary to,” “sleep deprivation,” “stress” and “work-related.” While the exact meaning and significance of the original sentence is clearly evident to a reader, capturing this accurately as a series of reference terminology codes requires a level of effort that exceeds reasonable expectations. It also lends itself to errors, which could adversely affect the reliability of recommendations provided by decision support algorithms.
Manual coding of documents at some point after the clinical encounter is costly and prevents the system from having the information needed to support medical decision-making during patient care. This is particularly true of automated warnings associated with prescribing medications, an area stakeholders have identified as a leading reason for promoting the use of electronic health records. For example, during an encounter a patient may provide historic information that may indicate kidney dysfunction. Unless this information is provided in real-time, the system will not be able to use it to screen medications for potential contraindications. As reference terminologies are not suitable for thispurpose, less structured terminologies that more accurately match clinician documentation have been developed. These “interface” terminologies consist of natural-sounding and frequently used clinical phrases. For example, an interface terminology may already have a phrase available for “mild fatigue secondary to sleep deprivation” and another for “work-related stress.”
These would be presented to the user as items in pick-lists to increase the efficiency of documentation. Prior to being used in clinical documentation, the text strings would be mapped to one or more reference terminology concepts. This process is referred to as pre-coordination and allows users to quickly and accurately capture clinical information that can be represented in structured format, such as SNOMED CT. Users will still need to combine concepts (also know as post-coordination) from the interface terminology to represent information. However, the level of effort will be greatly reduced. A usable interface terminology will need to strike the correct balance between levels of pre- and post-coordination.
The Medicapaedia is an interface terminology project that will be based on an expanded understanding of terminology usage and challenges, and the data will be hosted by e-MDs Inc. Domain experts will register as editors and be provided with Web-based terminology editing tools. Users will have the ability to add concepts, synonyms, parent-child relationships, allowable (i.e., “sanctioned”) modifiers and map concepts to one or more concepts in other vocabularies (e.g., SNOMED CT, ICD-9-CM). Editors will have the ability to download the full terminology or a subset based on specified criteria. There will be no associated fees for use of the Medicapaedia terminology so long as no charges of any kind which are related to the use of it are levied on end-users of that software.
The Medicapaedia will be based on published and peer-reviewed criteria for developing an interface terminology. For example, Rosenbloom, et al., described the basic desiderata of an interface terminology in their comprehensive review article. This included a statement of scope and purpose, complete domain coverage, concepts that do not change with time but yet evolve as information changes, context-free concept codes, concepts with formal definitions, support for multiple levels of concept detail, methods that allow for the recognition of equivalency between concepts, presence of assertional knowledge and the presence of compositional balance.
We anticipate that this open approach will result in significant contributions from domain experts who will form natural-sounding phrases suitable for use in clinical documents. Users, assisted by terminology domain experts, will map clinical expressions to the appropriate concepts in other terminologies.
Since many of these phrases will map to two or more reference terminology codes, in all likelihood, there will be a terminology explosion. As complete domain coverage is a desiderata for an interface terminology, the work will also focus on creating context-specific (e.g., specialty specific) subsets. For example, a pediatric neurologist has greatly reduced requirements for adult gynecology concepts. This allows this user to navigate through a subset of terms that are specific to the context of their practice. EHRs that use templated documentation can take this a step further. For example, the terminology requirements of a template that is addressing somnambulism (sleep walking) are even more restricted than the domain of pediatric neurology. This context-specific focus will greatly reduce the number of terms through which the user will need to navigate during data entry, but still allow for natural-sounding documentation. The information will also be automatically stored as reference terminology concepts without incurring additional effort from the user.
In the near future, the Medicapaedia tools will allow users to employ assertional knowledge when developing concepts. Assertional knowledge refers to additional information associated with a concept that helps distinguish it from other closely related concepts. For example, the sudden onset of a headache that is maximal at onset might be associated with a subarachnoid hemorrhage, while a typical migraine has a gradual onset that slowly progresses in intensity. The severity at onset and progression patterns are examples of assertional characteristics that allow closely related concepts to be differentiated from each other.
The Medicapaedia project will commence in mid-2007 and will be an ongoing open-ended collaboration designed to meet this terminology need. A large repository of current interface concepts currently in use by over 4,000 clinicians using e-MDs Solution Series EHR will be provided as “seed” content to initiate the process. In the interests of the advancement of the use of electronic tools to improve healthcare, all associated costs will be covered by e-MDs Inc. Editorial policy will be set by terminology and EHR domain experts.
Once the Medicapaedia concepts are instantiated within clinical documents, ongoing testing will evaluate the ease of use of documentation and the accuracy of reference terminology represented in captured information. Another exercise will be to examine the semantic integrity of information captured by the interface terminology, represented locally by a reference terminology and then forwarded to another application that is using a different EHR.
In conclusion, the Medicapaedia will be an open source interface terminology built by individuals with clinical, EHR and terminology domain experience. The terminology will undergo rigorous testing to evaluate the integrity of collected and shared healthcare data. It is our hope that the Medicapaedia will serve as a timely catalyst that increases the use of coded terminology in health records. This is a key component of local and national efforts to improve the quality of care through the use of electronic tools. A well-constructed interface terminology will also enhance the ability of systems to share information across provider and care settings.
Michael Stearns, M.D., CPC, Chief Medical Officer, e-MDs Inc., is a board certified neurologist and certified professional coder. He has 15 years of experience in clinical and academic medicine followed by 10 years of experience in clinical informatics and coding. Michael has presented at several national meetings on medical terminology, electronic health records and coding. He was involved with several projects involving computers in medicine at the National Institute of Health and was a key contributor to the development of SNOMED CT. He is also a voting member of the Health Information Technology Standards Panel.