CIMI MTF Minutes 20130404

From CIMI
Jump to: navigation, search

CIMI Modelling Taskforce - Meeting Minutes

Thursday 4th April 2013 @ 20:00-22:00 UTC


Attending

Linda Bird

Stan Huff

Thomas Beale

Mark Shafarman

Harold Solbrig

Patrick Langford

Dave Carlson

Eithne Keelaghan


Agenda

- Weekly news and updates

- Preparation for meeting in Leeds/Rockville

- Terminology bindings and value sets

- Laboratory Panel Specialisations

- Demographics models

Weekly News & Updates

  • Meetings:
    • Leeds/Rockville meeting: Thursday 11th April - Saturday 13th April
    • Post Leeds meeting
  • Tooling Report [Harold]
  • Reference Model [Michael van der Zel]
  • BMM Files [Tom Beale]
  • Lab Results modeling update [Joey Coyle]
  • Other updates?

Detailed Meeting Minutes

Linda: Draft Agenda is to [see above]... and then look at lab panels and specializations and [Demographic Models] ... which I accidently muted for last meeting. Add or subtract from agenda?

Tom: I can stay for 1 hour.

Stan: I couldn't hear. Let me say again. I worked on briefly. The approach is to go through different cases because in different cases there are different clinical models. So quantitative lab reports - what if I put into an arbitrary collection that is a panel? Urine color... need to bind attribute and variable name. Binding discussions to hardest cases... But I want to make sure we understand the binding in simpler cases.

Linda: This relates to LOINC - SNOMED discussion that I wanted to have.

Stan: If defining Hematocrit... "What is the code that corresponds to this binding"? Not a relationship code... No bigger thing because that is just it. I may not be understanding, but...

Linda: Relates to my question - How... seamlessly use LOINC? How semantic binding always done in SNOMED. Whether always used in... Specialize in terms that constraint must be LOINC code... How we handle that is what I want to discuss. So are you OK with this agenda, Stan?

Stan: Yes - I saw it as term-bindings - already in agenda.

Leeds/Rockville meeting

Linda: OK. Start with weekly news and update. Face-to-face next week. I'll be taking time off post-Leeds.

Mark: The past couple of HL7 meetings - CIMI has met after them. Is that planned for the next HL7 meeting?

Linda: Not in May since we will have just met in April.

Stan: Yes - no meeting after the next CIMI...

Mark: OK.

Tooling Report Update - Harold Solbrig

Linda: Harold - summary of tooling report?

Harold: Yes - mulling over... I'd like to rearrange the agenda in face-to-face and give everyone that has content a chance... to get from where is needed... I would like to ask everyone in community to think about what would have to happen to get their content into... and start reviewing.

Linda: Thanks, Harold. Michael is not here. Tom - ADL workbench?

ADL Workbench Update - Thomas Beale

Tom: Last generation - a couple of errors... just program errors... will get these to work next time around... Nest iteration. I have been hand-modifying to what it should be, so I have copies of intermediate working version.

Linda: So not far off?

Tom: I think in next 24 hours he will have his code working straight through.

Linda: Great! Also - adopting things to support the... text data-type.

Tom: Yes - I put stuff in... will bind the simple syntax... we use in openEHR. Should have been in there... will tell the tool which ?code? to bind. Has not been very popular, but...

Linda: But was necessary.

Tom: Was required if you want to use those features... can write... that way.

Linda: Syntax...?

Tom: Some people violently opposed, but I think will be useful. Unfortunately, called the openEHR model. But really... to the AOM. AOM designed to make no assumptions other than... its leaf item... Boolean... The AOM is written in a way... in Health Informatics spent many years coming up with clinical data types. I think we all agree... coded-text on CD and others 15 or 20 known types. The capability - vanilla versions of archetyping... is to add a simplified version of the... ordinal concept and to add... We all talked about those. With these added things, the AOM not only knows about Boolean and strings, but knows about things commonly used. We use a lot in openEHR. Maybe it will turn out another way to do this, but... would be made more useful by having these key... constraint version of a data type. So it is there and works for CIMI.

Linda: Does that involve a new download of the workbench?

Tom: Yes - just testing...

Linda: Great. So should be ready before Leeds?

Tom: Yes.

Linda: Great. Thanks much, Tom. Patrick - update on ADL generation from Mindmaps?

ADL generation from Mindmaps Update - Patrick Langford

Patrick: Yes - Joey is in Chile. We have a Mindmap parser. We ran into a few inconsistencies... so Joey got a hold of you, Linda, and ironed those out. I have been out of country... have not gotten back to Joey. He was going to convert most by hand... Having a bit of a problem getting into ADL workbench and getting it tested. I expect will have more program...

Linda; Yes- how they decompose... Points where I referenced another archetype. We talked about how they are broken up.

Patrick: OK. Just a question of interpreting it. We are parsing....

Linda: Other?

Preparation for meeting in Leeds

Linda: OK - Let's talk about Leeds. On Thursday - modeling taskforce update. Talk about Lab Results. Not sure if we will have something to show... Demographics Model. Reference Model version, After break, Michael van der Zel will talk about Reference model and Gap Analysis. Mike Lincoln will talk about...

Linda (cont'd): Patrick - will you or Joey be in Leeds or Rockville?

Patrick: I will be calling in.

Stan: Joey - calling in as well.

Linda: [continues to go over Leeds agenda]

Linda: Any comments or things you would like discussed during those seminars?

[no response]

Linda: OK. Now Lab Results Specializations or Terminology Binding and Value Sets? I think first... The terminology bindings and value sets...


Terminology bindings and value sets

Linda: You should have seen this already. Example use cases include management of model repositories and quality control of models. Some of the examples were 'active ingredient' and 'basis of strength substance' in the medication example. Both had object meaning of substances... had active ingredient substance and... Don't apply for elements that have a data-type. And we talked about specializations, and we talked about slot filling - the archetype that filled the slot had to subsume the... Hopefully we all remember the general principles.

Linda (cont'd): So - last week we talked about the complete blood count. Good you are on the call, Tom. You might be able to make comments. For CBC examples - both of the 2 cases include CBC with automated count. Wanted to have as separate without differential. And Stan - whether 58410-2 should be represented as... cluster or... and whether appropriate to include as entry...

Stan: Yes. The LOINC... of this data - from HL7 v2, you have ?? segment. Does a lot. Gives you a name for a subset... in that context. ...58410-2... stands for that collection of things that contain leukocytes, erythrocytes, etc... And in each observation segment, would have codes for leukocytes, erythrocytes, and if CBC with manual differential... inside of that would have... Coming from that perspective I was trying to match that world view with whether those exist with same... (?)

Linda: So - last week - discussing whether appropriate to have generalized concept of CBC... automated. Whether... have specialization. Which inherited(?) the CBC by automated count... inherited the panel. So - CBC - have 3 specializations... CBC with no differential, CBC with manual differential, CBC with automated differential. CBC with no differential - changed the cardinality to say there are no subgroups.

Mark: One way to think about this is what was ordered vs. what was done. So - may have CBC ordered, but they decided to do another. So does CIMI want to keep track of what was ordered, or what was ordered and done?

Stan: Both.

Mark: So makes sense to keep track of what or (?)... So want to allow an ad-hoc clustering of what was done or...? May make more sense for clustering... Is stylistic rather than semantic.

Stan: Yes - a couple of ways I've seen. You make a connection between order number and go look at order... or explicitly put a field in the result that says the name of the ordered test. Modeling-wise, it appears to me more (?) to make an association rather than copying.

Mark: Yes. More sense to keep the link as a pointer. Can always follow back.

Linda: Yes. We previously decided to keep a link to associated request. So, in this case, we are doing a grouping of clusters. In CBC with normal differential, we are...

Linda (cont'd): The other aspect of this - where SNOMED code stops and LOINC starts. I'll show on spreadsheet. We agreed - do first 5 as demonstration... simple. Units get mapped to constraint. In terms of these, we talked about whether semantic bindings should SNOMED or LOINC. We have test observable, which could be bound to SNOMED, and then test name with a value from LOINC. Where this is used for heart rate, we use an observable - Perhaps want to allow this to also be a Procedure?

Stan: No - SNOMED clearly distinguishes Hct observed from Hct the (?).

Linda: I like that. May need some CIMI extensions. I like... using the same hierarchy.

Stan: Yes - Karlsson here has been leading observables in SNOMED. True ontologists want to distinguish the procedure... set of steps - from the observable. Should always be an observable, not a procedure.

Linda: We need to look at how this relates to the data elements. Concept model for observables. Concept model for procedures. So, for example, 'point of care test' I thought was more related to the Procedure, than the observable.

Stan: May be a longer conversation. Need correspondence where these concepts can be (?) into another meaning. I think we can include these with the Observation, rather than define them ontologically.

Linda: I agree. Only those things we want to pre-coordinate.

Mark: Point of care would be a point of procedure. Each lab... procedure... Want to document procedures that require some licensed person... are separate from... Every observable implies a procedure to obtain it. But documentation of procedure is not the same as documentation of result.

Linda: Things like reason. Diabetes blood glucose level - whether that is the reason or a pregnancy blood-sugar test. Pre-coordinated... some interface... whether we allow that, or whether we allow an urgent something as a pre-coordinated term or not.

Stan: I have seen the same. In the preferred model these would be separate, but we would allow these to be pre-coordinated in the other isosemantic models. I would argue for preferred model to use separate values, just as modeled here. In both cases, the long-established LOINC codes, LOINC will not make those names.

Linda: Maybe we need to discuss. How much we support mapping from local forms to preferred... Whether support pre-coordinated mapping.

Mark: Like Michael van der Zel... some things are one mapping away. Sometimes... the diagnosis to rule out... Never have all the pre-coordinated reasons. Is more interoperable if more separated out from the thing.

Linda: I think we agree - separated out in preferred model. And need to agree on how much of the pre-coordinated version we want to support.

Linda (cont'd): OK - in test name. If we assume... and then look at test result...

Stan: [can't hear]

Linda: So the specimen for gas and carbon dioxide panel... uses a specific specimen. Should we put as whole blood sample as we did for... So look at SNOMED. Blood specimen... whole blood sample...

Stan: I agree.

Linda: OK. So there are 2 constraints on test observed... and on Test Results... pH of blood. We can leave as an... and inside the name we can set the value of the name as... LOINC code... or we can try to add LOINC code to meaning of cluster... Would be inserting LOINC.

Stan: So ... 2 questions I was asking. Could you show what the bindings look like in partial pressure... in cluster?

Linda: In cluster, is specializing a finding item... So we need it to be a specialization of evaluation finding. So do we leave it as...? Or do we use the LOINC code as the semantic binding? Or do we wait until LOINC added to SNOMED?

Stan: Whether is LOINC or SNOMED... can figure out. But, the cluster... in the IMH way of modeling... different than LOINC code. Is the name of the structure on the right-hand-side. Corresponds 1-to-1 to what code would be... This would not be a LOINC code, but... is representing the structure of collection of elements, as well as the elements themselves. The meaning is described by code and attributes and name. Others may have better words to describe what I am saying.

Linda: You may be saying that a clinical finding is defined using the data element that are contained within the cluster elements. I am not sure whether that will help the iso-semantic problems we have.

Mark: That was a specific test done. And result can be given by set of attributes. ...was a test procedure, and the results of test procedure... Seems the natural way to reference... A SNOMED finding... not just carbon dioxide partial pressure and... but "dangerous". Would assign a diagnostic significance to it... Say you've looked at it and is way outside the ... normal. So I am trying to distinguish test and results and trying to...

Stan: You are talking about more complicated things than I was. Linda - can you click on...? My conceptualization is that the name field is where I would attach the LOINC code... "CO2 partial pressure in blood" and now if I require a modifier, an identifier of a specific template... container result... is the name of the structure, not the LOINC code...

Linda: Quite a different approach. So if the semantic meaning of a cluster is an Information-structure artifact. So becomes hard to do subsumption between parent and child archetypes.

Stan: This example is different thing and has different relationship. Not what we are talking about. This isn't an example of what we are talking about.

Linda: Also we had an alternative way of representing things like 'Indication' - it could be either a Link to another part in the hierarchy, or a just a simple diagnosis type. We had suggested both of them would have the indication-relationship .. Just wondering if there is a relationship here...

Stan: My short answer is I see them as different things. I don't think it applies here.

Mark: The indication "has this test been done and has this observation been made". Could be an ongoing problem... or some physio-logical value. Makes more semantic sense... indication is a Link. That is too big a structure for just a data element... Has a number of different... but different from a single attribute.

Linda: I agree that complete structure required, but...

Stan: If instead of "CO2 Partial Pressure" we were defining a cluster... then I agree. Not just defined by the elements. Defined also by the role that it... But if we want to be... If name represents a LOINC code, some have a method that is part of it... LOINC... allows us to make a...


Stan (cont'd): So our ability to understand the relationship between pre and post coordination is happening in the cluster itself. But in this case... CO2 Partial Pressure in blood... the only relationship between... and CO2 + gas or... This collection of things is included in that bigger cluster - gas and CO2 panel - the relationship... The only semantic is it is contained in... only relationship is by containment. Other than say is contained in... Not like saying is an indication of body location.

Linda: I think of this cluster as all about CO2 Partial Pressure in blood and then contained in this cluster are the defining attributes, such as method; as well as other attributes are non-defining properties. Just as in UML.

Stan: Let me ask a different way. That cluster name is an identifier of the model rather than as a semantic identifier?

Linda: Yes - whereas I have been thinking... even when 2 or more levels... have been thinking semantics from main clinical hierarchies.

Stan: Think about this Use-case. We have 2 different systems doing CO2 partial pressures. They either include LOINC code without method or include LOING code with method. So one where... and another where method... and would be missing isosemantic. And how do we... I would give... In one case, the method would be a defining element and in the other...

Linda: In my approach, I would assume either the pre-coordinated name or the structural representation. And the pre-coordinated definition would be used to define the relationship between these, whether the data was provided one way or the other.

Stan: So - you would have a model where the name could be a name that didn't include method or that does include method... - not have names for things that have one or the other

Linda: Not if coordinated by terminology.

Stan: Want that information in terminology regardless. Use to translate, but if only have one of those... recognizing as distinct. Need to look at element and decide if pre-coordinated. There are systems that are using that. You don't have a name for the model... So, I can have a cluster... pre-coordinated or post-coordinated. But if my system only allows pre-coordination...

Linda: Yes, but if have 5 fields to be used in the pre-coordinated terms, then there is a combinatorial explosion on options.

Stan: If have 5 elements and 1 model then you should be using one of those... then I have to decide.

Linda: That is where normalized terminology expression comes in - flatten out.

Mark: Normalization is not necessarily enough. Have to have indication or reason... structures... have to have those basic structures.

Linda: If pre-coordinated... not defined... then not possible.

Mark: Need iso-semantic basic structures if truly different. Have to have some basic structures before can share data across different systems.

Linda: Depends on where want to push the work. If you mandate a preferred form then you force the sending system to add the data in that form, if you allow it to be sent either way, then the receiving system needs to transform this ... system depends on how much work push to sender or receiver.

Stan: That is why we bother to have preferred form. So sending system sends... from their local model to standard model. If do other way... Can work as well, but is why we want preferred... Work required for sender to do is less.

Linda: Would require sender to be terminology-enabled.

Stan: You could do what you said, but if converting from one form to another, than... data... If going to do completely - 3 models. 1 model - allow pre-coordinated. Another model - never allow (?) to be part of name. 3rd - either way. So have universe of models. You could define in declarative structures... Data with this code in name...

Linda: So in third model would be shared between structure and terminology... Concept model could do the work... So Left Leg in structure and Left Leg in terminology...?

Stan: Yes. In combined model, could have method - optional - filled in or not, and method - pre-coordinated - could have pre-coordinated name and structure

Linda: Need to start with option 1. Would like to think we could evolve to version 3.

Stan: I don't object to the 3rd being a valid model. I just prefer that we represent the subtypes... so I can be explicit when I convert between the models

Linda: So name of model not only represent structural model... but value-set rules...

Mark: Always... some... model... apply the terminology bindings...

Linda: This was useful and interesting conversation, but we could... We were talking about cluster...

Stan: So - show the way you described. So the relationship says Linkage Concept. So this thing contained in next bigger enclosure. You can say "contained in" - is the most you can say. In this one - not very interesting. This is part of the cluster...

Linda: We can define a SNOMED specialization of... or use LOINC code here....

Stan: Yes - a unique identifier. The SNOMED CO2 guy and LOINC CO2 guy - semantically equivalent.

Linda: there is an id... for this...

Stan: I don't have a semantic name for this thing at this level.

Linda: [can't hear]

Stan: No - not the class it comes from. Is the instance... there has to be some identifier of this information model you are showing on this screen. Use the information-model identifier that shows the meaning of the model - Use that for the node.

Linda: We already have the... If going to keep the consistency, will have to leave evaluation finding as it is.

Stan: That archetype id is unique to this model. Not an identifier of parent of this guy.

Linda: Is archetype id of this...

Stan: OK. I don't know where I would use...

Linda: This is also used with heart rate...

Stan: What is the value? A classifier?

Linda: I think so.

Stan: Is arbitrary... a classifier based on structure rather than semantics. This is a lab result because it has a lab result structure.

Linda: So 2 things going on. Structure... So in parent model, will point to finding group... Gets serialized... The archetype-id that gets serialized... defines the structure. But in terms of clinical... observable entity... clinical finding... laboratory... the evaluation finding... Subject of care or... Some higher level of meaning... Look for results of test...

Stan: That helps a bunch. I guess my question is... Why would that be different - whether a heart rate or femur length - whether we called it evaluation binding? Is that the meaning in SNOMED? I am worried about what semantics SNOMED is implying. I worry about that. Use-case specific rather than...

Linda: So general... Have the meaning for results because have specialized to laboratory test... SNOMED. My understanding... and it adds to the concept model.

Stan: I worry that in a bunch of these they represent collections and people will argue with you - it is a lab thing on a patient-finding. I am arguing... more generic... this is where the data is... Arbitrary to say this is lab data.

Linda: But if go to more general result, then this result is a clinical finding.

Mark: This information has to be stored with some kind of recommended form for interoperability and for sharing. So a finding is different from a lab result. Say "here is a lab result". Indication of something... but can't change to one of these other things without an explosion. Going back to my point - is a lab result. Can be used as confirmation of diagnosis. Tell me if I am off track here.

Stan: I think we are talking about something a little different, Mark. But hold that thought. But Linda, I am worried that I can classify things... go back to what we mean by lab data... procedure or analysis.... procedure on patient. Came from a specimen or... I am worried that it implies more than that. Not want to get into argument that transcutaneous (?) is a lab test vs. a finding. People won't use.

Linda: I think that is why you say you prefer... rather than say...

Stan: So, looking at example on screen. Look at second item... object... then object=441742003. If I need to know gas and (?) panel is an "evaluation finding" before I can search... would I think others...?

Linda: No, wouldn't. But searching for Heart Rate finding...

Stan: I am trying to understand what semantics are being implied by evaluation finding.

Linda: Yes... We have 10 minutes left. How best to use time?

Stan: Could we take 1-2 minutes to say what modifier is doing in this?

Linda: Yes. There is no modification to the context here. Not family history, not... so that is just the default.

Stan:...

Linda: The request example is good to look at because modifier does... Anything else to discuss on this model? Put the LOINC code in the name... the tentative decision... Then open for wider debate.

Stan: I apologize for the tediousness of these discussions, but has been useful for me.

Linda: Yes - I agree. But we didn't discuss... Speedy run-through... We start semantic binding for clinical entry...

[Linda describes Mindmap for Clinical Entry]

Linda: The action is a procedure. And the details can be anything that is extended. So a clinical activity can be specialized into requests or procedures - and if look at a request it specializes clinical activity... The request action is defined with the modifier 'requested'... So can request an observation or procedure or medical device... Have kept that general. The modifier of this is 'requested', so If the requested item is an observation, a medical device or procedure, then that observation, medical device or procedure has been requested. We also included a Link to the results.

Linda (cont'd): Any comments on this level? Since this is the first place where we used that modifier...

Stan: Looks good so far.

Linda: Modifier... Want to make sure the meaning of modifier is appropriate.

Stan: We need to talk about the meaning of status. I thought it would be "sent", "awaiting signature"...

Linda: Different types of status.

Stan: We are probably using different... but I meant clinical status. Difference between request status and...

Linda: Should go back to source models and look at status values and see if consistent. Have to do a lot of work here... Whether need to drill down ...

Linda: I do intend to finish presenting this. We can continue this discussion then.

Stan: Again - thanks so much, Linda. I love the work you do.

Linda: Thanks. I'll talk to you at the Leeds meeting.

[end of meeting]