> -----Original Message-----
> From: Andrew po-jung Ho
> However, your final caveat regarding vocabulary points out the
> big "BUT" that still remains difficult to resolve. I have
> previously proposed using term-to-term translators rather than a
> "universal language" to solve this problem. What is your view on this?
Vocabulary is one area where the jury is still out. Enrico Coiera has
argued persuasively that "terms are purposive", that "universal
terminological systems are impossible to build" and "while it is deceptively
easy to start to create a terminology, one soon encounters some of the most
subtle and difficult problems at the heart of philosophy, language and
knowledge representation". (see "Guide to Medical Informatics, The Internet
and Telemedicine" (1997) Chapter 12). All true, as I know to my cost! As an
aside, the only two books on health informatics which are genuine classics -
Coiera's and Scott Blois's Information and Medicine (1984)- both deal with
this subject area.
There is NO future in hierarchical coding schemes which are designed to do
more than one job. The original Read Codes were a success because they were
designed for use by GPs in their day-to-day work. Nothing else. Everyone
made a mistake in believing that they could be extended for use in
hospitals. It did not work, costing the British taxpayer some ?20 million
and causing a serious political embarrassment. Coiera provides an
explanation of why.
It may well be that SNOMED CT can provide a universal compositional scheme,
and time will tell. I hope so. LOINC was originally designed to do one job
for laboratory requests and reports. I am doubtful if attempts to extend its
uses will prove satisfactory.
It is rather easy to prepare a scheme of identifiers used to do one clearly
defined job. So one solution is to generate many such schemes. It may well
be possible to take data generated by one scheme and to group it (m to 1) to
meet another requirement. It may also be possible to group the same data in
more than one different way to meet different needs. It goes without saying
on this list that the grouping algorithms should be open and subject to peer
review. This is not simply a term-to-term translator, because the
algorithms may need to take account of other factors. Think about DRGs where
the grouping may include age, sex, co-morbidities and therapeutic procedures
as well as primary diagnosis. ICD classification has some similar features,
but is a bit more subtle.
There are two sorts of grouping schemes. Those were the grouping is done
before data collection (as in a hierarchical coding scheme) and those where
the grouping is done as a subsequent computation. The latter approach is
advocated in Slee, Slee and Schmidt "The Endangered Medical Record" (2000).
It looks promising.
Tim Benson
