Tim, I get confused with your message, maybe because of langage
subtilities - speak slowly, I'm french ;-)
When dealing with hierarchical structures, there is always a possible
confusion between :
- a structured classification or coding system
- a structured model of medical process (the tree of all the possible
choices in the model)
- a structured document (usually as an instanciation of a part of the above
model)
In Odyssee, our only matter is structured documents since :
- we are working on structured descriptions, not classification (we try to
describe things as they are, not as they can be statistically classified -
we also work with classifications, but never mix both)
- ontologies and classifications are much different paradigms (however, in
France, say 98% of all sofwares use CIM10 (ICD 10) as a list of terms for
description purposes) ; ontologies usually can't have a hierarchical
structure, you need a semantic network.
- structured models of medical processes remains the first delivers of
expert's works, but we immediately ask the expert to take a higher level of
abstraction and build a set of Fils guides out of it - usually the initial
tree can be instanciated using that set of Fils guides, but it is only one
of many - especially when these FG have been mixed with hundreds of others.
As a conclusion, it is always hard to distinguish if we are talking about
the model (expert work) or its instanciation (Mrs Johns examination).
In Odyssee, only instanciations are trees, and there are no hierarchical
structures at all - maybe some hidden one, but I shall kill it as soon as I
find it ;-).
Philippe AMELINE
> We need to be clear about what can and cannot be done with hierarchical
> structures (one to many) and multi-parent structures (many to many).
> Hierarchical structures behave in ways that are relatively easy to
> understand, but I am not so confident with multi-parent structures. This
> applies to any sort of object, whether it be a classification scheme or a
> data model. Performing data extraction and analysis using compositional
> coding and classification schemes (like SNOMED) is more problematic than
> with simple hierarchical schemes (and these have many traps for the
unwary).
>
> Is the answer that every multi-parent scheme must be normalised into
simple
> hierarchies before it can be handled? This is, I think, what is done with
> XML fragments and most software programs. Am I missing something obvious
to
> those with a proper education in information and computer science?
>
> Tim Benson
> Abies e-Health
> 12 St Georges Road, London, NW11 0LR
> 020 8455 8106; 07768 825 012 (mob); 020 8458 9577 (fax)
> [EMAIL PROTECTED]
