W3C home > Mailing lists > Public > public-hcls-coi@w3.org > April to June 2008

Re: An argument for bridging information models and ontologies at the syntactic level

From: Alan Rector <rector@cs.man.ac.uk>
Date: Tue, 8 Apr 2008 09:39:43 +0100
Cc: public-semweb-lifesci@w3.org, public-hcls-coi@w3.org
Message-Id: <0693452E-2D28-43DD-9ABF-99BAF44F7764@cs.man.ac.uk>
To: Samson Tu <swt@stanford.edu>
Samson, Vipul, All

I saw this by accident and have not been involved in the main  
discussion - so excuse the intron.  However, the issue of the relation  
between ontologies and health records is close to my heart.  There are  
papers about it at both KR-MED 2006 and Medinfo 2007, the KRMed paper  
due to appear in Applied Ontologies RSN.Both papers are available from  
my web site http://www.cs.man.ac.uk/~rector.  An expanded and clearer  
version of the KR-MED paper will be available as soon as the mills of  
the gods grind at AO

Fundamentally, the only interpretation that works is to regard codes  
as being "meta" to the ontology.  I.e. the individuals in the ontology  
are things in the conceptualisation of the world - cases of diabetes,  
people, livers, etc. - individual codes represent classes in the  
ontology.  The entire information structure - HL7 or Archetypes  - in  
fact, is at a meta-level.  It makes sense to talk about a form on a  
patient on which the code or value for body-temperature is missing; it  
does not make sense to talk about a patient without a body  
temperature, even if it is ambient.  It makes no sense to talk about  
the class of hypertensive patients except those that fall into some  
subclass of hypertension, but it makes perfectly good sense to talk  
about the code for hypertension but not its subcodes as being a valid  
filler for, say, a heading.

We are again in the process of doing such representations for both  
OCRe and two commercial collaborations.  One thing I feel confident  
about from this work is that a single level representation of the  
ontology of disorders of patients and the information structures about  
them, including codes, does not work.  We can often get away with  
approximations which ignore the difference for specific applications.  
Because our tools for handling multi-layer representations are poor,  
we sometimes have to, but the problem is fundamental.

it isn't even a question of what formalism one uses.  Medicine involves

*	Pathophysiology - what we know about the patient
*	Clinical care - what we do to the patient based on our assessments  
of the pathophysology of the patient
*	The record of that care and those assessments

Decisions often involve all three levels.  Our actions may be based on  
whether or not a particular piece of information is present in teh  
record , our uncertainty about its value , or is value.

As far as SNOMED-CT goes, to a first approximation, the distributed  
form can be viewed as being "codes" in this sense and should not be  
taken as an "ontology" the codes are individuals representing classes  
of patients.  The "Ontology" is the underlying "stated form" which we  
rarely see. Unfortunately, some of the things people try to do with  
SNOMED ignore this point, and the documentation on the issue is  
confusing at best.

Regards

Alan

On 7 Apr 2008, at 21:45, Samson Tu wrote:

>
> On Apr 3, 2008, at 7:56 PM, Kashyap, Vipul wrote:
>>
>>
>> OK, we disagree on this point. I'd just point out that, if you are  
>> interested in working with HL7 RIM or BRIDG, you have a conceptual  
>> mismatch with them.
>> [VK] I do not view it as a conceptual mismatch as I can get Snomed- 
>> CT the terminology by specifying a transformation on Snomed-CT the  
>> information model.
> Perhaps you can elaborate on your idea of SNOMEDCT the information  
> and what kind of transformations are involved to get SNOMEDCT the  
> terminology.
>
>>
>> If your Acute MI is a subclass of Observation/Problem, then  
>> instances of "Acute MI" class are observations of Acute MI, not  
>> instances of the disease MI. An "observation" does not have  
>> severity, location, and so on. You lose the ability to talk about  
>> properties of the things in the world.  An information model refers  
>> to codes not because of implementation concern, but because  
>> component parts of informational entity are also informational  
>> entities, IMHO.
>> [VK] Would like to separate the issue of incorrect modeling from  
>> the issue of including class analogs of terminological codes into  
>> an information model in general.
>> As far as severity, location, etc are concerned, these could be  
>> implemented as qualifiers to the observations as proposed in the  
>> Clinical Element Model approach by Stan Huff et. al.
>> That said, the issue is not that of accuracy in modeling as I used  
>> Acute MI as an example. was proposing an information architecture  
>> where we create a common framework to model and perform inference  
>> on information models and terminologies.
>>
> Several years ago, I tried to formulate the Clinical Element Model  
> as an ontology without any success. I came to see it as a very  
> flexible data structure for encoding information. If you have better  
> luck formulating it as an ontology, I'd like to know about it.
>
> Thank you.
>
> Samson

-----------------------
Alan Rector
Professor of Medical Informatics
School of Computer Science
University of Manchester
Manchester M13 9PL, UK
TEL +44 (0) 161 275 6149/6188
FAX +44 (0) 161 275 6204
www.cs.man.ac.uk/mig
www.clinical-esciences.org
www.co-ode.org
Received on Tuesday, 8 April 2008 08:40:30 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 April 2008 08:40:30 GMT