Re: Follow up: Clinical Observations Interoperability Telcon @ Tue Oct 30

On Nov 3, 2007, at 10:47 AM, Kashyap, Vipul wrote:

> We definitely have differences in language. Let's try to synchronize.
>> A domain model is something that corresponds to something in the
>> world. So when we use a term from a domain model ontology such as
>> "liver", we mean livers.
> [VK] Actually, this is closer to what I would describe an  
> information model,
> just that it would also contain "epistemological" constructs in  
> addition to
> ontological ones.

Sounds like we need another table for mapping different ways of  
describing such things :)
I think it's helpful considering the stuff that has "epistemological"  
constructs as a different sort of thing.
I tend to worry that having "information" in the name will confuse  
people, since it sounds like it is about information.

Do you have some examples of the epistemological constructs, btw?

>> An information model describes a data structure. Here we would talk
>> about things like records with fields and the values that fields can
>> have.
> [VK] This is where we differ. I would call the above a data model.
> So Detailed Clinical Models are Information Models
> and SDTM models are more of a data model.
> We will need to see if this is indeed an issue but I guess one  
> would have to
> abstract an information model out of a data model (assuming you buy  
> into the
> terminology above)

We differ only in nomenclature afaict. I call this the record level.  
Regarding SDTM, explain what "more" means. Does that mean it mixes  
levels?  I note that SDTM has places where one uses NCI Thesaurus,  
which is, for a large part, a domain model.

> [VK] Observations belongs in the Information model. Contains both
> epistemological and ontological constructs.

I'm not sure on what basis you make this statement. Either it follows  
from your definition of information model, or there is something  
significant about keeping the observations mixed in with the  
statements of fact. If the latter, I'd be curious as what the  
motivations are and any experience that suggests ways in which you  
win if they are together, or lose if they are apart. Barry's  
criticism of these systems from an ontology point view is that they  
land up being full of errors that arise specifically from confusing  
the levels.

Here's the sort of thing he finds (I found this in browsing starting  
at CDISC in about 10 minutes):

In SDTM there is Outcome of Event, to which a definition is given:
"A condition or event that is attributed to the adverse event and is  
the result or conclusion of the adverse event. (NCI)"

This sounds like domain model.

For filling in, reference is made to "CDISC SDTM Adverse Event  
Outcome Terminology", which I gather is in NCIt, and taken from CTCAE  
( There we find

> An AE is any unfavorable and unintended sign (including an abnormal  
> laboratory finding), symptom, or disease temporally associated with  
> the use of a medical treatment or procedure that may or may not be  
> considered related to the medical treatment or procedure. An AE is  
> a term that is a unique representation of a specific event used for  
> medical documentation and scientific analyses. Each AE term is
> mapped to a MedDRA term and code.

I read this as mixing the different levels. At least the parts  
starting  "An AE is any ...." and "An AE is a term" seem to be not  
talking about the same thing, the former a mixture of observation  
(finding) and domain(disease), and the latter a string.

I don't know how much this sort of mixing matters in the work  
clinical informaticists do. From my perspective it would seem to be a  
good idea to be more careful about not mixing in the way that is done.

> A specific implementation of an Observation model in terms of data  
> structures, e.g., relational tables, java classes, etc. are is a  
> data model.

Like anything we do in computer science, things bottom out in bits.  
The tricky part is which bits, and what the path to bottom is :)

>> The practice of linking a field or a field value to a term from a
>> domain model, as SDTM does and as Tom described, is an attempt to
>> link the two. To what extent that is effective, I don't know. I
>> suspect that it goes some, but not all the way, towards addressing
>> Kersten's concerns.
> [VK] In general this is the approach followed by Detailed Clinical  
> Models
> as well and seems to work for now

It would be great to have some documented idea of what "works" means.  
What problems are trying to be addressed with this? Which ones seem  
to be solved? Which ones seem to persist?

> [VK] Would propose that Information Model = Domain Level +  
> Statement Level
>         Data Model = Record Level

If these are definitions, then there is no discussion.

If a choice, then I'd certainly be interested in chatting about the  
pros and cons with other interested parties, at F2F, or via email


Received on Sunday, 4 November 2007 00:22:47 UTC