|
Hi,
I think your are not in the right
direction.
My english is not fluent enough to make myself
clear, but we are on a very important point.
You can see medical domain as a cone, since you can
describe patient's state in a very elusive way (sharp end), or very accurate way
(even a microscopic lesion description).
The aim of the ontology is to be able to DESCRIBE
each and every point of that cone, and, as it is a description, it
must have a semantic meaning.
A Classification is made by selecting a plane in
the cone and tracing domains on it ; domains frontiers are inclusion and
exclusion criterions. Each domain is given a code for
indentification.
For EASE OF USE, classifications come with a label
for each domain.
It is a genuine problem, since people use these
labels as description items, thus confusing an artificial domain and a genuine
cone's area.
To be more mater of fact : If you try to build an ontologie out of classification's
labels you will get a big amount of garbage, with no semantic meaning, and
totally pathologies oriented.
I believe in an ontologie with controlled evolution
on the web ; I dont think you can build automatically the core of it (that's to
say the initial kernel).
I can give Odyssee Lexique as a french terminologie
- the semantic network needs lots of work, but it is a starting point ;
maybe Thomas could be more precise on the couple terminologie+semantic network
he uses in GEHR. Lets join energy - and may the force be with you
:-))
Philippe
----- Original Message -----
Sent: Wednesday, May 30, 2001 12:47
AM
Subject: RE: [ontologies] OpenGalen/Grail
and OIO Library, was RE: [ontolo gies] Your opinion
At 06:17 PM 5/29/2001 -0400, John S. Gage wrote:
I'm not sure what "automatically"
means.
Automatically means in this case that you give some
number N of practitioners the coding systems to use. You track their
usage and see what terms are probabilistically associated with which
codes. Where there is absolutely no convergence the terms are probably
meaningless. The thesaurus grows out of tracking
behavior.
But someone must do this tracking. Automation
can help the process, but there is work involved in the tracking and logging
of the terms and their use. This could be very labor
intensive.
It should be noted that that is the
*exact* method by which dictionaries are created. Dictionaries are
*not* prescriptive. They merely record usage. If you think
they're prescriptive, look at the agony of the French Academy and M.
Toubon. The idea is to get a computer to shorten the process by which
the OED was created, for example.
I agree.
Right now, the common
denominator of all terminologies is that health care must be paid for
and two terminologies determine how much will be paid: ICD-9CM and
CPT-4. Every single medical procedure/visit/etc. in the U.S. is
coded with these two codes. That certainly is the place to start
the thesaurus.
The issue is can this be done "legally" for
free, since those codes are proprietary.
ICD-9CM is freely
available on the Duke website. CPT-4 is quite a different
story.
I think not. The next
biggest problem I see is that terms in two different systems aren't really
the same but almost so.
The terms are *complementary* in
ICD-9CM and CPT-4. You need both. Which is a good thing (as
Martha would say). ICD-9CM is for the "supporting diagnosis" and CPT-4
is for what you did to the patient. The "almost so" element is handled
by probability measures: what is the probability that a certain term maps to
another term? If the probability is ~.5 then the term(s) are worthless
and don't have to be mapped.
The ontology space needs to be
mapped out (UMLS?) so that one can tell where the overlaps
are.
Thanks,
Dave
John
David W.
Forslund [EMAIL PROTECTED]
Computer
and Computational
Sciences http://www.acl.lanl.gov/~dwf
Los Alamos National
Laboratory Los
Alamos, NM
87545
505-665-1907 FAX:
505-665-4939
| 