Adrian Midgley wrote:
>I may be showing my confusion and ignorance here, but to what extent do the
>ideas of GEHR archetypes overlap with the ideas of an ontology as typified by
>other projects that label themselves ontologies?
>
only the same confusion we all have! Good question. This is a complex
area. The way we see it is that there are "level 0" ontologies which are
broad descriptions of "facts" or "principles" in medical and clinical
topics. Any such ontology should "generate" or have associated with it a
"terminology". In the case of things like SNOMED, it is trying to be the
same thing, and will probably suffer through its lack of design as
either an ontology or terminology (they just want to get it out, and it
is not being designed with EHRs in mind). Better examples include the
Galen approach, but the funding is not available to populate all its
ontological descriptions.
Anyway, the purpose of these ontologies is to describe non-volatile
facts in the domain, such as one would find in an anatomy, biochemistry,
or other textbook. Facts about clinical process (e.g. obstetrics) or
pathology (e.g. the structure of test results, sampling protocols) may
or may not appear in level 0 ontologies, depending on how globally
"true" they are.
The current situation is that there are just terminologies and EHR &
other systems which try and use them. Hence the problem of
terminologists trying to define everything in terminologies (leading to
combinatorial explosions and pollution with volatile concepts; all
well-documented by Alan Rector in various papers).
This is where archetypes come in. HL7 calls them "templates", and
various other systems use something similar (Philippe Ameline's Odyssee
is one of the nicest partial implementations I have seen so far; there
is also an XML-based archetype system being develped at InterMountain
Health in Utah by Stan Huff's group).
I consider archetypes and all equivalents to be higher level ontologies.
THey do have a point of view, are more volatile, and their scope of
applicability may be narrower, e.g. only locally applicable. What
archetypes do is define structure for "information in use", what this
means formally is that they define _compositional_ structures of
information as it is captured & used, as opposed to _defining_ its
meaning, which is mostly with classificatory structures, which you find
in level 0 ontologies and their terminologies. Archetypes depend for
their existence on well-defined underlying ontologies. So an archetype
about blood pressure just references (via coded terms) the concepts
"systolic blood pressure" and "diastolic blood pressre", which are
defined in an underlying ontology, which also defines the real meaning
of such things, as in the following example from Alan Rector:
(A) Concept Representations
Pressure
Pressure which <isOf-Fluid in-Container>
Pressure which <isOf-Blood in-Vessel> :: sensibly isAt-Phase %%adds a
'slot'
Pressure which <isOf-Blood in-SystemicArtererialTree atPhase-Systolic> ==
'SystemicSystolicBloodPressure'
Pressure which <isOf-Blood in-PulmonaryArterialTree atPhase-Systolic> ==
'PulmonarySystolicBloodPressure'.
...
(B) Terms
'SystemicSystolicBloodPressure' - usual_term - "Systolic blood pressure"
'PulmonarySystolicBloodPressure' - usual term - "Pulmonary systolic pressure"
...
(C) Archetypes
common_blood_pressure_archtype = {'SystemicSystolicBloodPressure',
'SystemicSystolicBloodPressure'}
So from this, Alan suggests that A & B are the job of ontologies /
terminologies, and C is the job of archetypes - as you can see, it is an
association of concepts in a form that they would be recorded in, rather
than a definition of their meaning, which is the job of A, or even a
definition of their terms, which is the job of B.
Underlying ontologies are used / should be used for inferencing, while
archetypes should be used to define and standardise lumps of information
according to knowledge models, so that we can achieve interoperability
at the knowledge level. But note: EHR systems are not passing around
definitinos of BP, they are passing around actual BP measurements, which
is why archetypes correspond to this interoperability requirement, not
underlying ontologies.
Hope this makes some sense...
- thomas beale
