This statement is provided by David Berglund, MD, MPH, Medical Officer at the National Center for Health Statistics (NCHS). NCHS is responsible for maintenance of the ICD-9-CM, jointly with the Centers for Medicare and Medicaid Services. NCHS specifically maintains the diagnosis portion of ICD-9-CM, while CMS maintains the procedure portion.
This statement will specifically address questions about PMRI terminology to which testifiers were requested to respond, following this same structure to address the related issues.
Related to grouping subsets of PMRI terminologies, a number of diverse areas are being considered. First, the frame of reference given is based on an original diagram which shows different areas connecting in the middle to “convergence,” which here has SNOMED CT. One issue is what is meant by convergence, and how it differs from clinically specific codes. At face value, convergence could imply moving to combine multiple systems into one, or use one system for multiple purposes. Whether it is better to move toward such a system or rather have multiple systems for different domains would be an issue. At this time, it would not be feasible to use just one system. No one terminology (even SNOMED CT) has all the content that these separate domains have (e.g., in the drug codes, in message specific codes, or in the International Classification of Functioning, Disability, and Health). On the other hand, convergence may refer to the capability to use a reference terminology as a basis for creating formal definitions for code sets, with support for mapping to other systems. For the current purposes, it seems that SNOMED CT could be considered with the clinically specific group, rather than separated into its own group.
The message specific codes and the other codes might be out of scope based on the definition of terminologies. Code sets are not the same as data sets, or specified sets of data elements. Also, message formats are different from codes. These are still an important part of enabling exchange of PMRI. However, these differ, and perhaps should be considered separately, and based on different criteria. One question is the extent to which these message specific codes were covered in previous NCVHS recommendations on message format standards.
A number of new groups or new terminologies could be considered for addition to the current list. Some of these are listed here.
It could be appropriate to consider data sets separately, as noted previously. These are not properly terminologies. This could make it more straightforward to apply different criteria for these, which could be more appropriate.
It would be possible to consider medical literature classification, and the Medical Subject Headings (MeSH). However, it is not clear whether this should be considered in scope for PMRI representation.
A number of terminologies for representing PMRI could be added. The Medical Subject Headings (MeSH) are a terminology for medical literature classification. While this might be of interest in some settings (e.g., for decision support systems), it is not clear whether it would be in scope for PMRI representation. It is one of the systems in the UMLS Metathesaurus.
For mapping between different systems that represent PMRI, the UMLS Metathesaurus has been used, but primarily in research settings. It might meet part of the needs for convergence, with regard to mapping. However, it is based on mapping between specific systems, and is not intended to be a terminology itself.
One system which has been conspicuously absent from consideration is the Generalised Architecture for Languages Encyclopaedias & Nomenclatures (GALEN). It could also be included, either for clinically specific codes, or for convergence. One of the advantages of GALEN is the sophisticated ability to combine terms, using GRAIL (GALEN Representation And Integration Language). GALEN was released in November 1999 (release 1) under an open license allowing free use, through the not-for-profit organization OpenGALEN. This is a Dutch foundation formed by the Universities of Manchester and Nijmegen. One of its functions is to promote the development of GALEN technology, in part “by stimulating the continuing maintenance of the ?open- source GALEN technology,’” according to the OpenGALEN.org web site. GALEN does not have the depth of coverage of certain other clinically specific systems, so it is not clear how this will work for maintenance. The most recent update was release five, in October 2001.
A number of data sets are listed, which technically are not code sets. These typically specify data elements; some may have associated data elements that might constitute code sets. However, it might be appropriate to consider data sets here also. In addition, messaging formats are present, and these also are not code sets. DEEDS is not properly a code set, but rather a data set.
Also, HL7 and X12N are not code sets but specifications for messaging formats. These are important, but should be distinguished from codes or terminologies. These specify structure and format to be used in a messaging, and are relevant in this setting.
The Nursing Management Minimum Data Set (NMMDS), the Patient Care Data Set (PCDS), and the PeriOperative Nursing Data Set (PNDS) are actually data sets rather than code sets. Again, these are important but should be considered separately from terminologies.
Standards for diagnoses and procedure codes have already been specified, as well as standards for messaging. Since these are so widely used, it was appropriate that these be done first. Also, it will be important for these standards to remain current, for example by expeditiously specifying the move to ICD-10-CM from ICD-9-CM.
There is a need for clinically specific codes, as well as drug codes (with retraction of NDCs), and nursing codes. Clinically specific codes would be highest priority of the groups which do not have standards specified. For some domains, there may be advantages to continuing to use systems which are in place (such as use of NDCs for some purposes, but not for all drug related purposes). It may also be of value to specify certain data sets for some uses, but those should be considered separately from terminologies.
Clinically specific codes are needed to represent more detailed information. There is a need to support electronic exchange of detailed clinical information. This is needed for administrative purposes such as claims attachments, and also for clinical use.
Drug codes are needed, and the relatively fast changes to this domain are a potential issue. It may not be feasible to have a single standard for this; use of different systems (as is now the case) can be a reasonable approach. Nursing codes are needed for more detailed tracking of patient care.
None should have too low priority.
Related to the preliminary criteria for selection of HIPAA PMRI terminologies, different criteria are needed for different groups or subsets of terminologies. Selection criteria for messaging formats, and also for data sets, should be different from the selection criteria for terminologies in general. Furthermore, classification systems which are intended to aggregate should have different criteria than clinically specific code systems.
Many of the selection criteria listed are based on Cimino’s desiderata. However, it should be noted that those desiderata are intended to apply to controlled terminologies with multiple purposes. One prime consideration is that these are intended to be clinically specific. Part of the argument is that such a system can then replace other disparate systems, although that is not the current plan. In any case, certain of these criteria would appear to be best applied only to the clinically specific group of codes, or to convergence, if this is not combined with clinically specific codes.
Clinically specific terminologies should support atomic and compositional concepts, but the compositional approach is less important for other types of terminologies. Concept orientation, with the basic unit of a terminology being atomic concepts with only one meaning per concept, is appropriate for vocabularies but not for classifications which are intended to aggregate. Formal definitions and explicit relationships are also part of a clinically specific terminology, and not so applicable for other terminologies. Note that formal definitions are generally still dependent at some level on use of medical terms.
Related to representational forms (and nonsemantic identifiers), a caveat is that at this time, humans may still have to look at the codes to use them. This is despite the contrary idea being expressed in the literature, with acknowledgment that these are not intended for human viewing.
There is the potential to add criteria, such as recognition of redundancy between codes or representations. For clinically specific codes, representing context is important. Some others which might be less important here include synonymy, and multiple granularities. For clinically specific codes, polyhierarchy or multiple classification could be included. Another criteria which could be considered for clinically specific terminologies is the idea of no residual categories; however, this would not be appropriate for general application. The residual categories are needed for practical use in classifications, and enable differentiating between other specified disorders and unspecified disorders in a straightforward way.
The criteria specifying relationship with existing and proposed HIPAA standards, both working with message formats, and mapping with administrative standards, is important for consistency. These are needed for subsequent standards to interoperate with the current standards.
The criteria should not encourage PMRI terminology developers to be ANSI accredited. However, it is important to maintain open meetings for standards development activities, and maintain timely and public notification of standards development meetings. On the other hand, it is not necessary to use the ANSI consensus process for voting / balloting on accepting new versions of each standard. This would not allow timely maintenance, as consensus can be difficult to reach, and conflicting interested parties can push for different changes, and ultimately stall the update process. Such a process could still be susceptible to influence by monied special interests. There would be issues of who should be allowed to vote, whether individual or organization, and what stake they had in the process. The technical nature of these standards limits the usefulness and benefit of voting to decide the standards. The established processes for updates to established standards should not be modified or subverted.
One organization that has become an ANSI accredited standard developer is the College of American Pathologists (CAP), developer of SNOMED CT. The SNOMED CT structure was submitted as a proposed ANSI standard. Even if the SNOMED CT structure is declared an ANSI standard, it does not appear that it would be beneficial to require that all the terminologies under consideration be distributed using this structure.
Licensing for proprietary terminologies should ideally be reasonable and nondiscriminatory. This may require that costs of licensing also be transparent.
Another question to consider is whether a single government mandated standard is needed for each of the specific areas being addressed. For some areas such as drugs, it may be reasonable to use one standard for some purposes, and another for different purposes. This would appear to be the issue behind the rescinding of one standard, the National Drug Code. It would be preferable to recognize such issues prior to mandating a standard, rather than needing to rescind it later.