Re: [HCLS-ACPP] Clinical Domain Knowledge using HL7 RIM OWL Ontology - Review presentation and discussion on June 28, 2:00pm

I've found those acronyms (including what we refer to, here as the 
Computerized Patient Record) do more to cause confusion than anything 
else.  Perhaps there is some historical context that I'm not aware of, but 
here in Cleveland Clinic's Cardiothoracic Surgery Research department 
we've been developing a domain-specific model to accommodate our 36+ 
year-old relational database (tracking over 105,000 patient actively) for 
cardiothoracic procedures, demographic data, angiogram measurements, 
laboratory tests, and clinical diagnoses.

Tbe model is expressed in an RDF vocabulary (a hybrid combination of SKOS, 
OWL, RDFS, XSD) that facilitates automated dual-representation of instance 
data as XML and RDF  using XSLT .  This has allowed us to generate user 
interface dialects that understand XML (XForms primarily) automatically 
for data entry  a separate, conversation thread altogether :)

Automation and the ability to accommodate rapid growth into unanticipated 
domains of medicine have been the primary research goals.  Since we didn't 
find any good precedent  to follow (i.e., an *accessible* OWL or RDFS 
upper model of a patient record that could also express domain-specific 
concepts with precision), we used the vocabulary to express a model domain 
sufficient to support our research vehicle.

In the end it was cheaper than taking the proverbial top-down approach. 
Our model has been fashioned by domain experts with experience in the 
primary day-to-day requirements of the data, so for concepts within the 
domain of cardiothoracic procedures and diagnoses it is well modeled.  For 
other domain-agnostic concepts (such as temporal relationships, logical 
composition, demographic terms, etc..) there may have been other 
alternatives but links can be easily expressed with description logics 
(DL) mechanisms or brute force first order logic (FOL) implication rules.

It seems to me like the 'Electronic' and 'Computer' terms in those 
acronyms describe the general advantage of storing health care information 
as variables and values instead of hardcopy and / or narrative rather than 
a modeling convention of some any kind.

When you consider the expressiveness of DL (OWL) and FOL (N3 and RDF), the 
specific set of problems your data is meant to address becomes a more 
prominent issue  especially if you have a good set of modeling best 
practices to follow consistently (one of the major missing components in 
SW literature).

Personally, I think when well-conceived and well-documented modeling 
patterns guide bottom-up vocabulary building you get more bang for your 
buck with SW technologies - so to speak.  This can be as simple as an 
overview of the core components / mechanisms of DL with accessible examples.  I've seen 
too many OWL ontologies that aren't as expressive as they can be about 
their classifications and are essentially flat taxonomies.

At some point, you pay the price for not taking advantage of the 
expressiveness of DL upfront: either while attempting to draw equivalence 
(owl:sameAs, owl:equivalentClass, etc..) links with other vocabularies or 
when composing 'raw' implication rules.

>From my brief perusal, the HL7 V3 RIM model seems precise and concise.  We 
could probably get good mileage out of modeling a subset of it in OWL to 
accommodate fleshing out the scenario for using inference with stroke 
management decision support.

With regard to rules, I have a specific interest in investigating the use 
of forward-chaining inference to derive variables for unanticipated 
clinical study criteria, explicit temporal reasoning (see article below on 
using N3 for temporal reasoning), and in using backward-chaining for 
augmenting experimental hypothesis with logical proofs - an area I believe 
AGFA's Euler Path mechanism is well equipped for.


[1] Practical Temporal Reasoning with Notation 3: 
http://copia.ogbuji.net/blog/2006-04-24/practical_temporal_reasoning_with_N3
[2] AGFA's Euler Proof Mechanism: http://www.agfa.com/w3c/euler/
[3] Desciption Logics  Basics, Applications, and More: 
http://www.cs.man.ac.uk/~horrocks/Slides/ecai-handout.pdf


On Tue, 27 Jun 2006, Eric Neumann wrote:

>
> Helen, Chimezie,
>
> That's great news! If I am understanding you, I see some possibilities for 
> taking one of the EPR scenarios and adding some SW structure into them to 
> augment the scenario with querying and inferencing for decision support...
>
> btw, can you help me (and others) understand how EPR (electronic patient 
> record) is related to EMR (electronic medical record)  and EHR (electronic 
> health record) ? Maybe we should set up a glossary on our HCLS site...
>
> On a related note, I'd like to mention that an author (George Cole) for one 
> of the EDC (electronic data capture) proposals for CDISC, HL7 was asking 
> about SW. They are proposing to use XForms (http://www.w3.org/TR/xforms/), 
> and it seemed to me this would be an interesting area to connect up with the 
> RDFA initiative (think XForms with RDFA embedded in them!). We hope to put 
> this use-case up on our wiki...
>
> Eric
>
>


Chimezie Ogbuji
Lead Systems Analyst
Thoracic and Cardiovascular Surgery
Cleveland Clinic Foundation
9500 Euclid Avenue/ W26
Cleveland, Ohio 44195
Office: (216)444-8593
ogbujic@ccf.org

Received on Wednesday, 28 June 2006 12:10:09 UTC