Clinical Guidelines SIG May 1, 2003
ATTENDEES: Bob Greenes, chair, BWH/Harvard,
Guy Mansfield, IDX,
Rick Shiffman, Yale,
Yasser Alsafadi, Philips,
Jerry Osheroff, Micromedex/Penn,
Guilherme Del Fiol, IHC,
Thomas Beale, Ocean Informatics,
Ken Harvey, Therapeutic Guidelines Ltd,
Ian Purves, SCHN UK,
Evelyn Hovenga, HL7 Australia,
Samson Tu, Stanford,
Jim McClay, UNMC,
Roberto Rocha, IHC/U of Utah,
Thom Kuhn, ACP,
Matt Sailors, UT Houston,
Dongwen Wang, Columbia U,
David Rowed, HL7 Australia,
Peter Haug, IHC/U of Utah,
Jim Campbell, U of Nebraska,
Diparc Kalra, UCL/OpenEMR,
Sam Heard, Ocean/Hl7 Australia,
Stephen Chu, U of Auckland, NZ,
AGENDA:
Q1.
Announcements:
־BOF on Common Services met this a.m., to organize, want to include CDS input (Alsafadi)
־WHO supports “Guidelines International Network”, formerly the AGREE collaboration (Harvard)
־Symposium on guidelines for April 2003, want to have HL7 endorsement, need statement from CDS TC/ CG SIG to BOD requesting this (Tu).
GEM progress report (Shiffman): Schema distributed, use cases based on Sharon Smart’s CPG cases posted. GEM II proposed work discussed. New elements based on COGS conference of 2002. Improved element definitions. Schema instead of just a DTD. Can be put in XML Spy. Improved GEM Cutter tool for multiple versions of Windows, adding version attribute, remove references to ASTM DTDs. Other potential changes.
Implementability Rating Profile, 30 items in 10 dimensions based on review of literature. Now requesting participation in a survey at ycmi.med.yale.edu/irp.
CPGA update (Purves): Schema at Question raised about how to move forward given CPGA and GEM as competing models.
Discussion: Kuhn: Structured Documents TC has agreed this week that CDA should be limited to clinical documents, but that other docs should be included, such as Structured Product Labels that FDA is requiring for meds. Purves: GEM focuses on guideline markup whereas CPGA focuses on building guidelines or taking them apart and rebuilding. Depth and breadth of hierarchy essential. Campbell: There are7 levels of markup from markup to implementation. Purves: Can’t go from GEM to CPGA because all the logic requires many business rules to transform. Kuhn: what are the arguments for this TC/SIG to be interested in GEM? Campbell: Infobutton-like approach of using markup to link to appropriate part of source document. Shiffman: Use cases for GEM don’t all require implementation at point of care: Can be used by guideline development group to improve guideline. Can be used to help develop implementations by popping up at the right time. Alsafadi: How do we pull the different themes together? What are the use cases?
Proposed task: Flesh out a schema, e.g., based on Campbell’s 7 levels plus Sittig’s workflow scheme to determine overall plan for SIG and TC work, workproducts, use cases, requirements, etc.
Q2.
Infobutton revisited (Rocha): Next steps: More input from vendors, esp. CIS. Also start working more actively, con calls, refined proposal. Goal to create specs for the API from clinical system to the resource. Interactions with other committees, e.g, CCOW, Vocabulary, EHR SIG, SD TC; with other SDOs e.g., NISO, Metasearch, OpenURL, Z39.50.
SAGE update (Mansfield): Main deliverables: inoperable guideline model, guideline workbench for authoring, encoding, and maintaining, guideline deployment, and controlled resources. Lifecycle model envisioned. SAGE guideline modeling (Tu): Goals to build on previous work, standards. Have well-defined semantics. Allow CIS integration. Much discussion about workflow aspects, and roles/responsibilities of recipients of guideline recommendations.
Q3.
SAGE update cont’d (Tu). Discussion of controlled resources, particularly use of clinical expression model for defining particular ways of representing concepts via use of constraints.
Workflow process model (Tu): Attempted to map activity graphs to RIM Acts but currently not sure how to proceed. Will come back next time with more examples of activity graphs and definition of requirements to be sure of CG SIG agreement about the approach. May then need to harmonize with RIM by working with MnM.
Task of defining guideline work products, in context of CDS roadmap/foci from Orlando (Campbell, Osheroff). Purves: goal is to make explicit some of the tacit knowledge that went into formation of the CDS TC and its SIGs. Need to cover scope of DSS, process of knowledge engineering, and architecture/infrastructure. Campbell to discuss with Jenders and to plan how to integrate with CDS scope tasks discussed at that meeting.
Q4.
Expression and constraint language discussion cont’d.
Minute of May 1, 2003 Q4 CQ SIG/ BOF on Constraint languages
Dale Nelson and Bob Greenes chaired the meeting
ATTENDEES:
Bob Greenes, co-chair, BWH/Harvard,
Dale Nelson, co-chair,
Geoffry Orberts
Samson Tu
Mark Tucker
Sam Heard
Yasser Alsafadi
Thomas Beale
David Rowed, HL7 Australia,
Dongwen Wang, Columbia U,
Kathleen Connor,
Jennifer Puyenbroek,
Dick Harding,
Pete Routey,
Martin Kernberg,
Diana Perez,
Ghislaine Hernandez,
Graham Grieve,
Shai, Ben-Yehuda,
Heath Frankel,
Mead Walker,
Bob Greenes: Goal of session: to set up framework for understanding different constraint languages and requirements. Today's mission: how to accommodate needs of multiple purposes w/o unduly duplication
Dale: Specific agenda: Try to find common approach to satisfy needs, but urgent need for template project. Need to express template using constraint language within 2 to 3 months. Need straw man to satisfy 80% requirements of template use.
Bob Greenes: DS need expression language for writing decision criteria. Originally wanted to go to ballot, but decided to wait until we have a discussion for common approach.
Languages being considered OCL, GELLO, ASN1, OWL.
Sam: There are needs beyond constraints from EHR and DS perspectives. Knowledge representations requirements.
Graham: Going to publish mapping of HL7 datatype and OCL datatypes semantics.
Martin: people should get somesense what needs these languages satisfy. What are strengths and weaknesses of each, and which has ready-made implementation. OWL captures most of OCL, ASN1 has parallel to Archetypes (allow constraints and aggregations). Do we need to commit to one language? Example, GELLO syntax may be a hindrance to specialist needs.
Dale: Nice to know which language is real. Not clear what is meant by unification of OWL and UML. High cost to be paid if we support more than one languages. HL7 as a whole should look for constraint language.
Martin: OWL is being supported by W3C and can subsume OCL functionalities.
Yassar: There are companies that are making validation and authoring tools.
Thomas Beale: {
Main issue: How to create represent and share knowledge assets. CDA need knowledge assets (templates, archetypes) to give constraints. e.g. CBC statement a message based on Observation RMIM.
Template and archetuypes are pure knowledge assets.
archetypes require the ability to state structure and constraints
template require the ability to state constraints
Archetypes can be converted automatically to:
- RIM/RMIM/CMET e.g. CBC archetype turned into RMIM
Describes a constraint language for archetypes. The language may be used for translation into OCL. Whatever form used for human viewing will represent structure. The constraints to be made can be OCL like.
}
Gunther: What is the difference between this archetype language and HMD? HMD cannot express slot by slot constraints.
Thomas: Believe whole structure can be converted automatically to OCL
OCL not a block-structured language that's easy to read.
Possibility to use OCL as transport mechanism for knowledge models??
Mark Tucker: Structure and expression are separate issues. HMD defines structure.
Graham: OCL is an expression language. It needs to be used in conjunction with a class model. Human readability is not an issue.
Dale: Structure assumed to be HL7 structures.
Person X: Need to have a common language for HL7, CEN, etc. Then have tools to map them to specific technologies.
Graham: Need something for modeling structure, then need constraints. Structures are more convertible (UML to Schema, for example). Predicate Expressions less convertible.
Dale: Need to use HMD as structures. Many people say we can use OCL as the constraint. Want to focus on this. Possible to extend structural and expression languages in the future.
Martin: How about tool support and semantics support?.
Gunther: What needed are structures and structural constraints. Words like semantics and syntax not clear.
Thomas: If we are sharing knowledge assets, need something more general.
Dale: Templates need to constrain currently balloted HL7 artifacts.
Sam: template constraints pointing to artifacts (e.g. cda) is safe to use OCL.
Person Y :Need remember needs of groups like Arden.
Peter: HMDs not visually pleasing or processable. Scheme not friendly to domain experts, but more understandable. Tool to build knowledge important..
Dale: Priority to derive equivalent mapping between HMD and archetype.
Discussion on possibility of domain expert encoding knowledge directly.
Martin: template requirements statement: user friendly artifacts. (general uproar)
Martin: need to have internal computable representation, and mapping to structures that are more user friendly.
Martin shows how to acquire domain knowledge in SRXML Template. Example of acquiring clinical information in XML templates.
Dale: Need to vote on straw-man proposal
Question to be decided: Should OCL be used for as constraint representation to constrain HL7 artifacts for first Template specification ballot,
Scope: Ability to constrain existing balloted RIM-derived artifacts [consistent with expectations of Baltimore Draft and CDA level 2]
Question passed in the affirmative overwhelmingly. One negative vote: OWL is a more powerful language.