The IHTSDO seeks to improve the health of humankind by fostering the development and use of suitable clinical terminologies, notably SNOMED CT, in order to support safe accurate and effective exchange of clinical and related health information. The focus is on enabling the implementation of semantically accurate health records that are interoperable.I want to show that coding creates the illusion of semantic interoperability, and that far from improving, it degrades the quality of semantic exchange.
What do codes look like? Heart Failure for example, can be coded as D006333. It can also be coded as 150 or G58 (Read Code v2, widely used in primary care in the NHS). It can be coded 10019280 in the Medra Ontology.
You might ask "Why so many codes for the same thing?", and a plausible answer might be "it doesn't really matter, because the codes all point to the same concept, and this way everyone can use the coding system they are familiar with, and share information when they need to".
And this would be true if heart failure was a single entity. But heart failure is a syndrome. It embodies a collection of concepts - several causes, a range of underlying pathology and a variety of symptoms. It is not a single thing, and the coding schemes reflect this.
Even a relatively simple scheme such as ICD 10 includes a range of possible codes to cover different types and causes of Heart Failure, including I11.0 for heart failure caused by hypertension, 150.0 for congestive heart failure, and 150.1 for left ventricular failure. Do both 150.0 and 150.1 map onto D006333? Or do they map to sub-codes of D06333? Read Codes contain separate codes for acute and chronic congestive heart failure. How do these map to 150.0? Several coding schemes have separate codes for the grades of hypertension as defined by the New York Heart Association scale for grading the severity of Heart Failure. Read Codes 662f - 662i cover these, but they are not represented in all coding schemes (Medra lacks codes for NYHA grades for example). It is recognised that some of the distinctions encapsulated in codes - for example between systolic and diastolic heart failure - are "somewhat arbitrary" (see this ref)
In principle it would be possible to map between each of the codes for heart failure in the major coding schemes. But even if it were done the mapping across schemes would not be 1:1, resulting in a loss of information and the inclusion of code generated uncertainty. As a result, machine based translation would only be reliable if the terms in each coding scheme were kept high level. But that would mean that the high level code for Heart Failure in one scheme would be translated to the high level code for Heart Failure in another scheme. It would be simpler to use the term 'Heart Failure' in this case.
There are two more significant issues with codes. First issue - was the original diagnosis correct? Setting a diagnosis in a code removes the ambiguity and uncertainty that often surrounds diagnoses, which then travels unencumbered across systems.
Second issue - has the diagnosis been coded correctly? Research shows a good deal of variation within and between clinics in the way diagnoses are made and coding is applied. In part this is because of operator error. But also it arises from the often multiple ways of coding a diagnosis. Choice of codes is always a local decision, based on rules which do not travel with the code.
Clinicians are aware of the problems of medical coding, and these problems cover most diagnoses. A receiving doctor will want to review the diagnosis medication and treatment of a newly presented patients. In which case a narrative account will convey enough information for semantic interoperability. Coding adds nothing, and may cause semantic degradation by reducing the amount of information available in the originator document.
Even if all coding were done in a single terminology variation and error would be present. All the studies show variation in the coding for the same condition. Coding therefore introduces new uncertainty on top of the uncertainty already in the underlying narrative record.
There are other justifications for coding. Use in decision support or clinical audit for example. The introduction or error and the loss of information associated with coding mean that coding does not have the value it is assumed to have. In the early days of computing it was useful to code diagnoses because bandwidth and hard disk space were at a premium. But that is no longer the case. The remaining justification is that coding is useful for re-imbursing and other administrative systems. Coding in these circumstances may be valuable, but this is not in itself justification for the continuing development of detailed coding systems. to carry clinical meaning.
Related Post:
Widget by [ Iptek-4u ]
