See also: IRC log
EN: Alan not on call
Review Demo issues briefly
just 5 wks until workshop
must wittle down to the essentially tasks in the next week
many deliverables alan is expecting next week
questions?
SS: another informal F2F at MIT - April 10
working on modeling of data sets in RDF & OWL
will be checking to make certain things WILL connect as expected
EN: Hope to see lots of code
& tools coming together by then
... any questions re: demo or tasks?
Did put a link on the WIki to solicit more help for Alan?
No outstanding questions
VK: Introduce Stan Huff and Tom Oniki
Showing clinical model work
looking for overlap with OTF - looking for synergies - represent in RDF & OWL
ACPP TF overlap - look for more overlap with CDISC
<vipul> http://esw.w3.org/topic/HCLS/OntologyTaskForce?action=AttachFile&do=get&target=CEM.ppt
SH: Talk from PPT slides
format of speaking on HCLS TCon
(EN: have 45 minutes)
Will go through slides - questions at end
SH: obviously more info than can be covered in 45 minutes
can plan follow-ups if useful
<ericP> slides are http://esw.w3.org/topic/HCLS/OntologyTaskForce?action=AttachFile&do=get&target=CEM.ppt ?
EN: In the audience we have BOTH clinical and basic LS research
recommend going quickly - let non-clinical folks review on their time
SH: Team effort (slide 2)
... Real goals - not just data - trying to enable LIFETIME
e-record (from pre-natal to post-death)
linking to all relevant services
(see slide 3)
SH: Depends on SHARED common info
models coupled to shared terminologies & ontologies
... What is a Detailed Clinical Model? (slide 5 - high-level
overview)
... Slide 5 - simple example (systemic BP)
<ericP> slide 6
SH: re: longitudinal record (slide 6)
what record to support use 100+ years
legal record
data will outlive service, app, message format, AND person
SH: focused on logical structure
of data AND interactions with standard terms &
ontologies
... Real time, patient specific, decision support (slide 7)
All patient specific info - most appropriate course of care for that specific patient
<eneumann> scribe: BillBug
SH: Sharing data (slide 8)
reduce labor within HC enterprise
outside enterprise - many needs (need standard model for it to be computable) - especially to support analysis of this data
use data to improve patient care (outcomes-based - evaluate intervention outcomes - natural disease course)
<ericP> slide 10
<ericP> (i think)
SH: National & Internation sharing (slide 10)
last ref - sharing data (slide 8 & 9)
SH: create NEW kinds of HC marketplace (slide 11)
create apps against standard services using common logical model for patient data
<ericP> slide 12
<ericP> (picture seems applicable)
SH: Order Entry API (slide 12)
same back end - front ends swap out - regardless how complex the app
SH: escape repetative, vertically
stacked, hard-code connection between backend (serializaiton
& data exchange) and specific app
... Shared public standard for DCM & for
APIs/Services
... on slide 18 now
EN: What is COS
SH: Common Order Service
(consider same as Common Object Service)
... Curly brackets problem (slide 18)
<eneumann> Clinical syntaxes: ARDEN, JELLO
SH: Arden Syntax (slide 18)
place shared data within curly brackets
DCM target shared logical model that provide more specific access to detailed data elements
SH: need standard model & repostiory (slide 19)
<eneumann> thinks Stan's desccription of "Shared logical model..." very close to RDF-OWL model
SH: Can't depend on free-text
data (Slide 20)
... Prospective decision support precludes use of free-text
Need shared, precise DCM
SH: Example - SNOMED CT - numbness in specific limb (slide 21)
same codes to represent limb numbness - regardless of which limb - no good for computability
need that in model
SH: Too many ways to say the same thing (Slides 22, 23, & 24)
Need the ability to qualify a measure (e.g. wt) so as to TRULY descriminate TRULY distinct measures
Can end up with totally different representation for same info depending on codes and variable model syntax (slide 24)
SH: RDBMS implications (slide 25)
tuples have different structure
example of using qualification to specify detailed WT type
can't be easily made commensurate across two models (+/-qualification)
end up needing totally different logic
or specific logic JUST to disambiguate
SH: more complex examples (slide 26)
these REALLY encourage a cornucopia of models (if your underlying shared model is NOT sufficient nuanced to capture detail)
<Zakim> ericP, you wanted to note that term mismatch examples all have unequivocal mapping
SH: Need shared terms/ontos AS WELL AS shared, layered DCM (Slide 27)
modelling in layers allows for required variability in application detail
EP: seems previous examples can be unequivovally mapped
SH: though these more complex - and abstract pre-coordinated entities such as syndromes
end up with a very lossy translation
SH: (back to slide 27)
layer model - helps in variety of integration scenarios with differing levels of detail
SH: Slide 31 - How do we use the models?
different use depending on the task - API interprocess comm, Core services, Decision Logic
SH: Bottom line - DCM does NOT dictate physical model
so app sharing and widget sharing can be eminantly re-usable
SH: Data entry validation (Slide 32 & 33)
model helps to insure validation before data reaches backend store
<eneumann> wonders if data storage service (DSS) validation is identical with RDF-S and OWL validation?
known model PRIOR to storing data
need tools that build off that model (validation just one example)
SH: Clinical Element Model (Slide 35)
EIM Subject Areas (high level)
had started with high-level and drilled down to detail
now focus on detail and map it up to the higher-level
<ericP> i envision a panel as an input form for a testor to fill out
SH: Conceptual hierarchichy of Clinical Element Models (slide 36)
from top to bottom - abstract core that gets more detail as you proceed to actual data element
those low-level elements can turn out to be used in many ways in the intermediate layer
EN: LabOpsPanel ?
SH: These are arbitrary collections of data needed for a certain clinical decision process
However, the underlying results may have MANY MORE uses outside of that process node
(e.g., Chem7, Chem20, Serum Sodium)
a STATEMENT - is a single sentence where adverbs& adjectives about single subject.
WHEREAS - a panel is a paragraph with statements about many subjects
EN: 11:50 - time for questions?
Should we jump to later detail slides?
SH: Whatever you like.
EN: Is your work and your consortium - I see XML mentioned
Are folks familiar with RDF?
SH: Familiar with OWL
Alan Rector has been using OWL for same types of models
Very interested to see how they can be used to represent the models
EN: Perusing your models, they
seem to have goals very much commensurate with underlying
SemWeb Tech/RDF goals
... other questions?
EP: Panels recorded in XML?
SH: Our own XML dialect (XSD?)
EP: enables you to extract elements from panels (e.g., person's wt) for other uses?
SH: yes
EP: seems to be exact match to how RDF is designed to let individual STATEMENTS stand on their own in exactly this way - very promising for SemWeb synergy
SH: Statements
both unitary and compound
e.g., BP - may depend on qualification re: posture, etc.
need to deal with individual element inter-dependencies.
<ericP> slide 59: Hard Issues
VK: Is there underlying UML?
SH: Hard issue (second to last slide)
What is actually instantiated in the software object instantiation?
gamut: only core elements represented as objects - all other info are constraints
other end: ALL elements are instantiated as REAL objects (e.g., BP, wt, serium Na+, etc.)
We chose the middle layer as the appropriate layer to model in UML
at the VA - added their own constraint language to UML to model more detail
VK: balance between implementing ALL detail (and including detail in model) for serialization
vs. an Interlingua (where you have to limit detail for it to be relevant)
Of course, too abstract also has minimal utility (HL7 & RIM)
SH: exactly correct.
... Even within HL7 - been work to additionally constrain
models to construct templates that are more immediately useful
as opposed to being too abstract to be useful to a broad
community with a variety of app requirements.
... From purely Modeling standpoint
models should be independent of these detailed application requirements
pick the model detail layer that you require.
EN: How much effort (e.g., code writing) is required to implement
Is it largely automatic
SH: right now, a lot of effort required
There are better ways to add automation to implementation
Should do more education for SW Engineers to communicate what our app requirements are.
is model hierarchy a problem (typical UML & XML Schema problem)
SH: We're very open to SemWeb Tech collaboration
remember: this work is a decade old
also - design decisions made and ran ahead of available tech
in HC IT commercial sector - have immediate needs the
SH: Yes strict model hierachy is a problem
need polymorphism that includes multiple inheritance
both XML & Java get in our way of doing
EN: very useful for us
suggest those interested in HCLS have follow-up discussions
might be useful to have short SemWeb presentation as mapped to the DCM goals
VK: could coordinate that
... Come up with interesting OTF tasks that include OWL
representations of elements from the DCM
<eneumann> ACTION: Vipul to set up on folow-up call with DCM and Stan Huff to describe SW standards and uses [recorded in http://www.w3.org/2007/03/29-hcls-minutes.html#action01]
SH: That would be extremely useful
VK: Aware of Alan Rector work in OWL - will include
SH: We collaborate indirectly with Alan et al (at AMIA meetings, etc.)
<vipul> Gotta go now ...
EN: last question
Work around clinical trials - overlaps with Drug Efficacy HCLS work.
Will see the BP example coming up there too.
We can contribute that the followup
Can help inform how OWL needs to evolve.
<vipul> Will follow up with Stan ...
Leave to VK to setup topics for agenda
VK: Yes - exactly what I had in mind
Examples from CDISC and clinical trials
EN: Has SH et al. worked with CDISC directly
SH: not directly - but very familiar with the work - have worked with them in past.
EN: We can help coordinate work with CDISC
SH: great
EN: Thanks SH et al for
presentation - will be in touch to coordinate
... Next call a week from now
... continue on demo.
This is scribe.perl Revision: 1.128 of Date: 2007/02/23 21:38:13 Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: RRSAgent_Text_Format (score 1.00) Succeeded: s/thniks/thinks/ Found Scribe: BillBug Inferring ScribeNick: BillBug WARNING: No "Topic:" lines found. WARNING: No "Present: ... " found! Possibly Present: BillBug Bill_Bug EN EP Eric_Neumann Olivier_Bodenreider Rachael SH SS Stan_Huff Susie Tanya_Hongsermeier Tonya VK Vipul_Kashyap aaaa aabb aacc aadd aaee aaff chimezie eneumann ericP gamut remember tn_bhat vipul You can indicate people for the Present list like this: <dbooth> Present: dbooth jonathan mary <dbooth> Present+ amy WARNING: No meeting title found! You should specify the meeting title like this: <dbooth> Meeting: Weekly Baking Club Meeting WARNING: No meeting chair found! You should specify the meeting chair like this: <dbooth> Chair: dbooth Got date from IRC log name: 29 Mar 2007 Guessing minutes URL: http://www.w3.org/2007/03/29-hcls-minutes.html People with action items: vipul WARNING: Input appears to use implicit continuation lines. You may need the "-implicitContinuations" option. WARNING: No "Topic: ..." lines found! Resulting HTML may have an empty (invalid) <ol>...</ol>. Explanation: "Topic: ..." lines are used to indicate the start of new discussion topics or agenda items, such as: <dbooth> Topic: Review of Amy's report[End of scribe.perl diagnostic output]