CIMI MTF Minutes 20131010

Jump to: navigation, search

CIMI Modeling Taskforce - Meeting Minutes

Thursday 10 Oct 2013 @ 20:00-22:00 UTC


  • Linda Bird
  • Stan Huff
  • Joey Coyle
  • Gerard Freriks
  • Patrick Langford
  • Galen Mulrooney
  • Sarah Ryan
  • Eithne Keelaghan

Draft Agenda

  1. Plans for joint meeting with SemanticHealthNet
  2. Any questions or objections to slides presented on last conference call?
  3. Review terminology binding examples - See attached file "binding examples smh 131009.docx"

Detailed Meeting Minutes

Stan: Will talk about joint meeting with SemanticHealthNet. And I want questions from last week on the presentation... that Linda made on models... helps to see... rather than just in Mindmaps. Other things to talk about?

[no response]

SemanticHealthNet - Joint Meeting with CIMI

Stan: So - Gerard and others suggested... SemanticHealthNet folks - joint meeting with CIMI in Februaru or March. I sent out... for all on Modeling TaskForce and Executive Committee. I sent out a Doodle poll... These are people we should invite in my perspective. Have not established an agenda for that meeting... Joint work we are doing, and any conceptual issues. May have a meeting... 1 or 2 days to meet with SemanticHealthNet and then 1-2 days just with CIMI for modeling work. So all should have received Doodle poll and we should know soon when meeting will be held. Any questions?

Linda: So - main purpose is to look for synergies and share ideas?

Stan: Yes - overlapping interests. SemanticHealthNet has bigger scope - integrate across Europe. But also - around specific diseases. So - bigger scope since more than just modeling, but more focused since just disease. But same issues - style, terminology... So - yes - that is right. Looking for synergy... Things to cooperate on, but also share intellectually.

Stan: OK - any other questions?

[No response]

Stan's slides from Cambridge Meeting

Stan: So ...the slides I presented at the CIMI meeting showing how we are trying to get synergy from relationships in SNOMED and... in information model. They were posted with minutes from meeting. I don't know if we had time for questions on this.

Linda: Should I bring them up, Stan?

Stan: Yes - or I have them too.

slide #1 - SNOMED CT Binding Strategy for CIMI Models

slide #2 - CIMI Use of SNOMED CT

Stan: Any thoughts or questions? Different idea of what we are trying to do? I don't want to go off and do something different from what people thought we should do. Pertinent to binding discussion - coordination between SNOMED and the information model.

[No response]

Stan: OK - is it all perfect?

Galen: I've been out of the loop, so no comment.

Stan: OK - take a look and if questions then we can discuss.

Galen: Thank you.

Stan: OK - so document.

slide #4 - Terminology Binding Summary Examples

Stan: We have had formal notes on these.

[Stan reads "SNOMED is primary reference terminology... LOINC is..."]

Stan: Clarifying statements that say - besides the usual bindings, we want semantic bindings...

Stan (cont'd): Any questions?

[No Response]

Stan (cont'd): When we talk about binding - some may come back and say "I agreed, but I did not know what was meant"... so... Any questions about this?

Answer: Nope.

Proposal for Binding

Stan: OK - scroll to proposal.

see slide #4

Stan (cont'd): Thought behind this proposal is that... separated binding into two categories. Constraint binding and value set binding. Second part is relationship binding or real semantic binding... relationships to concepts in SNOMED.

Stan: I've divided the world that way because constraint on value set binding and other use cases that require the semantic bindings. My perception is... IMH can contribute to value set binding...... work on relationships will be harder. So I am thinking... to get models out... in making CIMI - approved models... an officially preliminary... after constraints and value sets, but to be complete - need semantic. Goal is to get models out quickly. Need to get done before model is totally official, but... could get out for FHIR... not sure if they have... So we could work in parallel... work on constraints before semantic, but with commitment to semantic.

Linda: I agree with getting models out quickly. I prefer not having the word observable because... and I assume constraint binding is...(?)

Stan: Yes - if we scroll down to example, your examples... Binding I am interested in... leukocytes... It's those bindings listed in constraint part... and I am fine with taking out observable - that is fine... can do it now.

Linda: OK - thanks.

Stan: Other questions?

Gerard: I agree.

Stan: OK. So I will move on. The clarification - I learned when talk with Linda - the relationship bindings don't show up in CIMI Reference model... the language and (?) to do the (?) are in AOM... constraint model. And the things in Reference model are things that could be set... And things in AOM... could be set... Could be used... Those bindings are static with the model and don't have to be sent. I did not understand that but I want Linda to comment...

Linda: Yes - in some systems, they send model together with data instances, but in this case, no need to if assume know the model.

Gerard: Correct.

Stan: OK. In this file - start with example we have in Mindmap exercise... data... and pull out binding statements... and for... the pattern that would be done. Same set of capabilities to do binding, but binding of (?)... is different from binding of a collection.

Stan(cont'd): So panel collections or bigger things that have a deep hierarchy... like problem lists... So... in the examples I have done, I have copied from Linda's examples. Some I understand, but some I am trying to understand.

Stan(cont'd): One discussion in Cambridge with Dave Markwell - he said examples looked good, but strategy might involve more relationships than in SNOMED. He said they went down that path years ago, but stopped because... So he did not say "don't make more relationships", but he said he and Daniel Karlsson and... might want to be involved... So, extend in a reasonable way those relationships that are there. So keep in mind... would our style lead to new relationships.

Galen: I understand there is a SNOMED observable project going on. How would that affect CIMI?

Stan: Not discussed that exactly and others have asked. One perspective - want to use this observable codes - want to bind to those... and leave the fractal... in the terminology. The other perspective... A certain degree of pre-coordination is involved. So laboratory... pre-coordinated... the timing related to that... the specimen... and other. But what we recognize... in our information model... might do differently... only use... that are independent of... and use the specimen as well. So the ontological model... SNOMED... should be a guide to us... But the real value is in agreeing on the pre or post-coordination.

Linda: Daniel Karlsson is chair of observable group and he made us aware of it, and when Rahil and he and I were doing work on Mindmaps, we had draft-observables - used as a guide.


Galen: You mentioned the document that Daniel Karlsson put together...

Linda: Draft Concept model for the observables - probably on IHTSDO website. If you can email Daniel Karlsson...

Galen: OK

Stan: So - Like to look at bindings underneath this... I have questions about things in yellow. The leukocyte... is a quantitative thing... the idea that this is something I can say about the patient... stands alone... happens in a point in time... someone measured it... class of things... will be all kinds of quantitative measurements, but also cardiac outputs and femur lengths... Seeing it as that class of independent...

Gerard: Why is independent?

Stan: Because there are other independent parts that are in this.

Gerard: I would call this a clinical statement [the independent quantitative statement...] So I am no longer puzzled by independent.

Stan: OK.

Linda: Mindmaps - each of these models based on modeling concept... So every independent quantitative statement has a consistent relationship. So in Mindmaps... inherits from... which is based on patterns and rules for more generic observations... Guidelines for which a class of things can remain consistent.

Stan: Right - there is an inheritance hierarchy behind this. And you would hope we would understand from parent class, and would have been stated there.

Linda: Yes.

Stan: So if we step into... In in the leukocytes in blood by manual count... The terminology binding of this whole cluster... and in that you have the relationship and there is the object... and is a modifier. Linda, you use 2 << arrows for...

Linda: The (ExCG?)... Grammar I was using...

Stan: OK. So within this class of quantitative lab bindings, the relationship would always be resulting in whether Hematocrit or Hemoglobin or... or would those vary?

Linda: No - they would always be the same.

Stan: OK - so you could put that - take away the question and put definitive statement.

[Linda makes changes]

Stan: OK - the question - are leukocytes... Questions I have - is that the text or concept or... Linda?

Linda: I wonder if this is out-of-date... Discussed since that time... focus on... as... Observable object?

Stan: Yes. File I used was from Groningen, and that would be out-of-date. So what would be up-to-date?

Linda: I am trying to get... Are leukocytes in blood by manual count... the leukocytes themselves?

Stan: Would archetype be quantitative lab result? And this would be a template?

Linda: Template is...

[I missed some]

Linda: I am wondering what this came from...

Stan: From...

Linda: But if only on archetype on its own, would not have...

Linda: I need to go back and look at... revised some work done since Groningen on these...

Stan: We can look at what is below, too. Binding relationships...

Linda: Relationship to parent...

Stan: With the modifier... Just see context value... was sort of a place-holder... In this model - the question is... is this thing a general class of things... or... what would be the situation there - choose specific context...

Linda: You choose a tricky example, Stan, because according to Daniel... Observables don't have context... Only procedures and (?) that result from observable have context. So we need to discuss. Have to assume we are reforming to procedure or... Because the context.

[I missed a little]

Stan: The status of the procedure... whether pending or performed... In the modeling itself we had some fields.

Galen: So is that where you would put preliminary... in final?

Stan: I don't think so. I put priority, not status, but we have both priority and status. And I guess that was my question - how does this relate to those explicit... of status and priority... at level of cluster... of result value and name?

Linda: Yes - we need the... In Groningen, Gerard suggested we make sure content modifier is consistent with... However, we need to consider - how define the meaning of a slot not filled in. If have discharge summary with section, such as high priority results - are we going to assume the high priority of the meaning of those... that fill those, or have the context of the meaning of the slot be... independent?

Stan: Yes - I would like to work through. In this quantitative statement... not throwing modifier away, but question... modifier does not seem to serve a role in this class of items.

Linda: We could favor... priority and status, but can't throw away the use case of defining the usage of a slot.

Stan: But I need a concrete example. Do you have in mind one real situation that you would fill in modifier?

Linda: Discharge summary with pending lab results.

Stan: That is the binding of the name of the collection, though.

Linda: Yes - that's true.

Gerard: When you query... template for a screen...

Stan: (?)

Gerard: When you use archetypes for querying... as a result of a query... When you talk about modifiers... When we get into negation I understand, but this is not the case. This is an element with numeric quantity. And when defining priority... would do with other elements, but don't see need to... Check box... done or not done...

[I missed some]

Stan: Would that be using a modifier... Is that a way this would be used?

Linda: It could be. Depends on how much we would want to assume. If always knew subject at each level, then... But... if all records based on some pattern and all... in same place, then could query...

Stan: I think that is our assumption. Defined in CBC... So is assumed that association to subject has already occurred and not element by element in panel.

Linda: May not know specific subject of record. May only know... to patient.

Stan: Yes - I am grouping...

Linda: Yes.

Stan: So for now - we can't see a Use-case for using modifier in this context. If we see a need, then we know where to put it. OK. So - that means at this level... the binding of the cluster to the leukocyte count is... to the observable - right?

[no answer]

Stan: OK - another element that is part of cluster in leukocyte is name...

Linda: This is old, Stan.

Stan: OK - what does new stuff look like?

Linda: Sorry - trying to remember... Discussing what to do with relationship or... And ... leukocyte... or original suggestion was to not define a... but I don't think people liked that.

Stan: I like that. And object binding we are doing with name looks identical to... And I am OK with that. Would it always be true that those are identical?

Linda: I think so.

Stan: Probably worth noting.

[Linda changes on the screen]

Linda: This brings up LOINC vs. SNOMED.

Stan: Yes - my next question... although this might be OK... Constraint said value of name was equal to leukocytes and value of... Concepts we are binding to an object - that is the same meaning... Might be a LOINC code with exact meaning to... Would not do this based on LOINC code that is a synonym for a SNOMED code.

Linda: What are you suggesting?

Stan: Was there intended value? ...separate from object binding?

Linda: The value of the name - what appears in the... whereas intended... then meaning would stay... within SNOMED CT... But this probably needs further discussion.

Stan: It probably does. It is complicated by the new agreement between LOINC and SNOMED. A lot of things that are single code in LOINC will not be a single code in SNOMED... only an expression that, taken together, mean the same things as LOINC code. So if want... tied to SNOMED... would probably be an expression rather than single code.

Linda: Unless we have an expression library with identifiers.

Stan: I think that is what is not allowed. Regenstrief and... wanted... I think it is prohibited by the relationship.

Linda: OK.

Stan: What I would say, for ease of use, easier if I have a single concept I can bind to... and because of observable ontology... and for any LOINC code, there should be an ontological (?). There will be a (?) between the LOINC parts... the components... l all have representation in SNOMED ontology... All are SNOMED concepts as well. Should be able to take LOINC code and use in... So would not have to... since would be known in SNOMED itself.

Linda: So in object of meaning, if had LOINC code, could get...

Stan: Yes.

Linda: Do you know time frame?

Gerard: I think there is a trial version.

Stan: It exists now for a substantial number of things, but I do not know when it will be comprehensive. But I would not wait for this... You would not be able to do the ontological reasoning until it was there, but you would be able to bind... would be waiting for... to do ontological reasoning that is implied by that single concept.

Linda: Sounds reasonable as long as can resolve in near future. But other question... You talked about initial model having constraints...

Stan: Yes. The reason for repeated is to tell someone using the model that this LOINC code could show up in name as opposed to just being in model.

Linda: Yes - is a constraint on the instance data.

Stan: And then we come down to... Let's look at ResultValue [see on screen]

Linda: This has changed... Review... when we decide how to handle the observation group... Not sure if we can cover this next week, Stan?

Stan: Entries in Entries?

Linda: Yes.

Stan: I think I can give my summary, but I am only good for 1 hour next week. We can start, but probably will not finish in one hour. How do you think Result Value will go away based on that discussion?

Linda: In version here, we included clinical finding, but we changed to observable entry... I would have to look it up... Not sure if worth going through...

Stan: We can look below a little... We served my purpose so far... by exposing the questions I had and getting an update on the latest thinking. So if you can correct these - then that would be great.

CIMI SNOMED extensions - constraints on units

Linda: And this is correct - constraints on units.

Stan: The reason I put CIMI SNOMED extension is - what SNOMED has... does not have 10^3 or 10^6 ... as an aggregate... has things per microliter... So 10^3 is not expressible in SNOMED. There are two things... One of the things as unit of measure, we said the units would point to SNOMED concept - would point to UCUM expression - put into convertible environment, but I hadn't realized SNOMED did not have 10^3 or 10^6. Is not an infinite number of these, but none of them are in SNOMED. I did a search and did not see them.

Linda: We... conversion base and conversion factor, so we can convert... If we did this, could define the conversion... We've used a ref set to define concrete... for each of units of measure. So for key, the conversion factor is 1,000. So in this case, could define a conversion factor that...

Stan: At IMH, we have an internal concept code that means 10^3 /ml, and we have a valid UCUM expression for this - a literal... and we put into domains so... the things that are synonyms for this are in the same... set. The question is - do you make concepts for that 10^3, or do it compositional?

Linda: We create separate concepts, but conversion shows relationship.

Priority: Stat, Astat

Stan: OK - so this, what I wrote, is probably accurate. In this example, I stopped. There was priority... status... but I waited to focus on... How was priority bound? And I got confused fast. And I am probably just behind. Do you know what it should look like? You want... the priority... but going down the path of specific relationship... where Mark did not want... There is a LOINC code for priority, but... But there is a SNOMED concept "priorities"... but I don't think it has everything we want, like Stat, Astat... not all the things in... And then are hundreds of things in priorities that not appropriate here, so... Bind to meaningful priority and bind to (?) allowed in this context.

Gerard: The priority of what? State in patient? The blood leukocytes count? No - it's the process. Is appropriate?

Stan: Well - it is appropriate, but maybe not stated right here. Good point - in HL7 - saying there are at least two priorities. Is a statement about - how fast you want phlebotomist to collect blood. And other - processing priority. And then goes to lab... Second - start processing in lab. And third priority - how soon back to patient. Send alert?

Gerard: Are talking about process.

Stan: Meant to be true to life.

[I missed a little here]

Stan: May also exist in order - telling you about process of how done... and should not be in result. But all systems I know of carry this priority in results as well.

Linda: You talk about stat ASAP routine... in SNOMED... I am not sure if any of those synonyms for immediate...

Stan: I assume they are not defined precisely enough to know. In our labs - Stat was abused so much that we created a Critical... so that we could know the difference between hematocrit on patient bleeding out in the ER... That is a value set content question as opposed to binding question. So - looking at what was in SNOMED... had a different feel... had a bunch of things that would not be used in this context. So I think - a minimum would be... a new subset derived from this priority...

Action Item: Need to talk about whether it is accepted de-normalization to copy order into result

Stan (cont'd): So - we may need to break this out into collection priority and... And this may move from here to... may have a link from this to... may be a problem. We may have Use-case where receive result and not know order. In health exchange, we could receive result and not order. So it might be... accepted de-normalization to copy order into result... Need to talk about.

Action Item: To discuss whether we need collection priority vs. process priority vs. reporting priority

Stan (cont'd): Also - whether we need collection priority vs. process priority vs. reporting priority.

Linda: I am trying to work out why... has name and...

Stan: Could be my error.

Patrick put Source up on CIMI browser, and "Node Source", and syntax for all models, all nodes

Patrick: I am glad you removed those 2, Linda - they were worrying me. And I put the... source up on the CIMI browser. A tab called "Node Source" - and you will see syntax for all of the models, all of the nodes.

Linda: Tell us the URL?

Patrick: URL = If you go there, it has complete list of models, and source tab will have these.

Stan: Good to know.

Patrick: So can see the interesting stuff - all laid out in a hierarchy.

Linda: As soon as can resolve the entry in entry - will help us.

Stan: OK - so can discuss for 1 hour next week. And Gerard - will send you... an appointment later...

Stan (cont'd): Yes - the entries in entries is a big problem... Need to understand to get... set up. If you could make other corrections... annotations on those binding examples... I would go on and do other things... these... come into play when we talk about diagnoses and... So can show proper use of modifier and relationship.

Linda: And once we agree, then there is a piece of work for Rahil or another to finish...

Stan: And then we have the information to say... how does this influence the other... in AOM. And would influence... in AML because would need UML representation. We need to decide how it would look... And is a need to understand in... AOM, the syntax that supports...

Gerard - to annotate archeytpes with SemanticHealthNet

Gerard: In next 2 months I will be in process with SemanticHealthNet to annotate my archetypes... the same with NHS... will be interesting.

Stan: Yes - will be interesting to see...

Gerard: It was interesting...

Stan: It is getting clearer to me. Took a long time... I need to see concrete examples to know what we are doing. I appreciate all... The question of binding and entries in entries are fundamental - cannot have ambiguities or it will come back to bite us... Taking a long time, but... need to do it right and all understand the implications of what we are saying.