[COAS-List] Summary of Denver submitters meeting

[Charles Carman <ccarman@medgrid.philips.com>]

		Summary of the
Minutes of the COAS Submitters Meeting at HBOC in Louisville, CO
		7/21- 7/23, 1998

A meeting of the Submitters and Supporters of the response to the CORBAmed
Clinical Observations Access Service (COAS) RFP was held at the HBOC
building in Louisville, Colorado from 21 - 23 July 1998.  The following
people were present:
Submitters representatives
	Tom Culpepper - 3M
	Jon Farmer - Care Data Systems
	Juggy Jugganathan - CareFlow|Net
	Mary Richards - HBOC
	Ron Terrant - HBOC (part-time)
	Chuck Carman - Philips Medical Systems
	Larry Hamel - Philips Medical Systems
	Tim Brinson - Protocol Systems
Supporters
	Vidya Bhat, BHSSF
	Csaba Egyhazy, DoD Health Affairs
	Dave Foard, Sunquest
	Dave Forslund, LANL
	Kent Wreder, BHSSF

The meeting started with introductions and various brainstorming activities
to bring the group to a shared understanding of the scope of COAS.  We
generated lists on the following topics:
· COAS users, user roles, and uses
· Existing and envisioned systems that may become COAS servers
· Existing and envisioned systems that may be COAS clients
· Categories (or kinds) of observations
· Observation data types
· Queries asked of a COAS server
· Architectural patterns for implementing distributed systems
We then went back and analyzed our lists, in particular grouping the
related categories and queries to identify the dimensions of the space of
observation attributes.  The dimensions of observation categories include:
 Quantitative vs qualitative
 Direct vs derived
 Primitive vs aggregated
 Relationship / context
 Domain / department
The dimensions of COAS query parameters include:
 Time [When]: time stamp, time span, time samples, any
 Content [What]: instanceID/example, concept code(s), category/subcategory,
all
 Population [Who]: PID(s), many by sample, many by criteria, all
 Place [Where]: facility / equipment
 Observer [by Whom]
 Encounter / Order [Why]
 Modality [How]
 Association(s)
 State(s)
We then discussed the scope of COAS, resulting in the following statements:
"There is some set of 'Juggy' data connected with an observation (a blob)
that is captured with the observation at the time of creation, e.g. who,
what, when, where, how, by whom,… and may be used to retrieve the
observation. ('Juggy' data is the context information for an observation)"
And (by Kent Wreder),
"The Clinical Observation Access Service (COAS) provides an extensible
framework to provide access to clinical observations. This framework allows
for a uniform way to access contextual information related to the creation
of a clinical observation as well as access to clinical observations of a
simple nature.  COAS, as a framework, allows for more complex clinical
observations to be incrementally added to a computing environment."

We spent some time reviewing a variety of papers related to COAS, e.g.
medical information standards and information modeling papers.  Our
conclusions were that the COSMOS, GEHR, and DICOM Structured Reporting
documents were the most relevant to our information modelling work.  Tom
Culpepper strongly recommended Dr. Stan Huff's work on medical information
modeling, but did not have any of his papers available at the meeting
[Action Item: Tom to find, and send to the COAS-List, a pointer to Stan
Huff's work].  We discussed possible object models for the COAS, starting
with a model that included the notions of Episode, Encounter, Event,
Observation, Measurement, and Quantity.  We ended up simplifying this model
to just include the lowest levels of the previous model: Observation,
Measurement, etc. with many attributes in the Observation object to provide
the context information about where the Observation was in relation to the
higher level objects.  There was not general consensus that this model was
the COAS model. We postponed further discussion until Tom Culpepper had
talked further with 3M about releasing his model, and Tim Brinson and Chuck
Carman met with the European modelling folks at the CORBAmed meetings in
Helsinki the week following this meeting.

Next we discussed the issues related to Federation of COASs.  We all agreed
that COASs should be capable of federating (communicating with other COASs
within an organization and possibly outside of a single organization).
There was much discussion, and no consensus, on issues related to the
question, "Who is responsible for the translation of IDs and ConceptCodes
between ID Domains and Naming Spaces, the COASs or the Client?"  This issue
came up again later in our discussion of Extensibility, and Tom Culpepper
presented some ideas under consideration within an ISO technical committee
to mediate and translate "data" values and structure between two server
systems.

Issues related to Notification were more easily resolved.  Again, we all
agreed that COASs should be capable of supporting the notification of
events when they occur instead of only responding to queries for
observations that occurred in the past.  The discussion quickly dissolved
into a technical discussion between Tim Brinson and Larry Hamel about the
features, capabilities, and design patterns for the notification service,
at which point they took the discussion off-line.  Submitters interested in
the topic will explore the issues that remain first before bringing them
back for discussion by the entire group at a later meeting.

We had a brief discussion of XML, and how COAS might support it.  There was
consensus that COAS should support XML as both a data format, as well as a
format for returning structured data from a COAS.  [Action Item: Juggy was
to explore further the activities regarding XML within the OMG, and report
back to the group.]

Our last major discussion topic was Extensibility, about which we agree
that we want COAS to be extensible. We entered into a long and sometimes
heated discussion of object models, header elements, query parameters and
mechanisms, federation, parameter dimensions, etc.  We appear to have
consensus on the general idea of the proposed query, with at least one
parameter containing a sequence of name-value pairs, but we are not in
agreement on the specific parameter or information model attributes.  We
ran out of time for this discussion, and it will be one of the major topics
to be discussed at the next submitters meeting.

We finished with the discussion of some administrivia.  Tim Brinson and
Chuck Carman were charged with representing the COAS submitters to the
various groups at CORBAmed in Helsinki, and to bring back comments, models,
etc. to the group.  Tim Brinson volunteered to organize the next meeting.
After some discussion, the group (that was present near the end) agreed to
meet in the Portland, Oregon area the three days before the next CORBAmed
meeting, i.e. Fri - Sun, September 11-13, 1998.