CIMI MTF Minutes 20141218

CIMI Modeling Taskforce – Meeting Minutes

Thursday 18 December 2014 @ 20:00-22:00 UTC

Attendees

• Stan Huff
• Linda Bird
• Deepak Sharma
• Patrick Langford
• Thomas Beale
• Joey Coyle
• Gerard Freriks
• Jay Lyle
• Harold Solbrig
• Eithne Keelaghan

Action Items

1. * ACTION ITEM: Thomas to think about and work out what this change would look like in AOM -- if we changed AOM to add this item-category (coded_label? See text below)
2. * ACTION ITEM: Harold & Deepak to make changes to remove language as an attribute in CODE_TEXT but support as an attribute of statement and as attribute that can be associated as plain text

Draft Agenda

Proposed Agenda Items

• Reference model changes – Harold Solbrig
  • We agreed to remove structure_type
  • We agreed to include “link” and “participation” only at the level of ITEM_GROUP
• Proposal from Thomas on how we should add “coded_label” to the Reference Model
• Progress on lab models – Patrick Langford
  • The first topic of discussion for this call is about pre and post coordination of specimen type, method, timing, etc.
• Language attribute – add it to statement and remove it from the coded_text data type in the reference model?
• Discussion of archetypes and templates – Thomas Beale, All
• Finalize the essential meta data attributes that need to be present in the models – Sarah Ryan
  • Copyright information
    • Notes out to IHTSDO and LOINC
    • Plan to include as url
  • Import source, etc. for management of lifecycle
  • Acknowledgement of original source
  • Should not include realm specific approval dates, etc. in the data file
  • List of metadata items from the ISO standard
    • Gerard will send the ISO list, and his own list
• How do we indicate which models are part of the same isosemantic family?
• Different coded_text data type for single code versus code expressions?
• Representation of “leaf” models as model files or as a tuple?
• Any other business

Meeting Minutes

[I missed about 5 minutes]

Reference model changes – Deepak Sharma

Thomas: Links onto participations - I don't think

Stan: I think Thomas is right. Let me bring up the brief notes.... What I thought we were trying to do is different from that. Or maybe I don't understand...

Deepak: We agreed to... LINK and PARTICIPATION will... So ITEM_GROUP has LINK and... PARTICIPATION... so LINK w ITEM_GROUP... the PARTICIPATION will lose...

Stan: I think we don’t want LINK in PARTICIPATION.

Deepak: So you wanted to keep the LINK in PARTICIPATION...?

Linda: I think it was 2 separate operations that have been merged... PARTICIPATION meant to come down to ITEM_GROUP... 2 separate operations.

Deepak: So this will come here...

Linda: Can we roll back to the previous one..? So, see the 2 previous associations...?

Deepak: So this?

Linda: Yes. And the target... and if you get rid of structured type... Yes - I think that is right.

Stan: Yes - Thomas?

Tom: Yes.

Deepak: OK... I thought maybe we have to preserve the link between .. and PARTICIPATION... Why he did that... I can change accordingly.

Stan: Thank you.

Stan (cont'd): Thomas - is there value in showing the BMM representation?

Tom: I can put up... tomorrow... so people can look at it in the tool. Right now I don't think it is very interesting.

Deepak: You can take back the screen...

Linda: Other - I would love to see details as the second attribute... should be the second attribute since is adding additional details to what was there.

Deepak: Yes - one detail in.. and one in ...

Linda: Can you move to second attribute? Would you mind? Thank you. So details... second... if that is OK with everyone else. Just change the order if you could.

Tom: You might want to do that offline... an annoying operation.

Stan: Yes - maybe offline. But - under the link it would say meaning and details instead of details and meaning. And the same under... just the order of those attributes would be changed.

Deepak: Does it give us any benefit?

Stan: No, but... not change the functionality... it is just cosmetic.

Linda: Easier for a human.

Stan: Thank you.

Proposal from Thomas on how we should add “coded_label” to the Reference Model

Stan: So - Thomas - do you want to... a different diagram.

Tom: I just did it in the BMM. That translates to the workbench... so if you could make me presenter. Deepak : What does the archetype .. mean.

Tom: When you are in this... properties... it is the list of properties of whatever the type is. On the Right-Hand-Side, is the type of the property. On Left-Hand-Side - the class declared in. So these - ancestors... locatable... and this... item...

Deepak: What is that icon..?

Tom: The ellipse means... is a subtype... I would have to go in and look at the logic... let me check that. Maybe the icon is not doing the right thing... have not thought about that for ages.

• ACTION ITEM:: Tom to check the logic of the icon (ellipse?) that Deepak asked about

Deepak: OK. Thank you.

Tom: On item... category...

Stan: We called it CODED_LABEL before.

Tom: A bad idea... I have made it an enumerated type. I just made up some values. The types of values is what makes sense... qualifier... I made them integer enumerations. Could be string emumerations... or code inside and make it a heavy-weight way of doing it... I would be more inclined... to do it into the reference model... An enumerator type... what the values are really going to be. There is a new property called - whatever we want ... has a value of that enumerator type. So if you go to an actual archetype... such as... observation under a... if you constrain that, you can see...a method... code... so I've constrained the main category of code... something like focal or... the key... I think...

Joey: Our key concepts... we use icon(?) as a key... focal concepts...

Tom: This is category method... is called method... would put that on method.... some other things part of method... normal range... I called it qualifier... that would be the cue. We have to develop a theory of what would be the real set of values here... and where they would really be here... In terms of visualization... we had discussions... not really going to be had... like IMH or some other way... could see these elements instead of just being a green dot... could be an I or M or... just GUI workup...

Stan: So - my intuition would have been to make this a coded-element rather than an enumeration in the model... [Not]? Need to be controlled but... allows us to have synonyms for those things... Allows you to make relationships... so if parent-child relationships... if wanted to ... method is a certain type of qualifier... could make those kind of parent-child relationships... that would have been my thinking. But maybe you could tell me advantages of making it in the... model?

Tom: It depends on what we are trying to do... I think... trying to do with a course-grained... at IMH... not really many types... a sort-of course-grained... We do the same kind of thing with our state... reasoning... course-grained... the idea is to separate out what are the focal set of items... what are the interpretive ones... What isn't needed... like protocol... method... you don't need to know... but you do need to know that the patient just ran the marathon... those type of course-grained... can see.... I don’t think it is useful to replicate a fine-grained... like parent-child relationships... can be done in SNOMED... can be attached by binding... as they are now.. I don't think that is what we are trying to do here.

Joey: I agree with Thomas. This is closer to what we do, but not mean you can't do both. Method and reason... just because we have an internal code to that represent focal concept.... we can still have in our terminology... a code for that... and look in terminology server... and look at... all the qualifiers are this... Having a small list... Thomas can do tooling on it... and we can do things like... we don't want a focal ... and if you... it is harder for him to work on...

Stan: OK. I think I am persuaded. But part of my confusion is... method would not be here... What would be here is modifier... and the fact that you threw method in there... is no different than status or priority...

Joey: Yes - was a mistake.

Linda: Yes - I agree.

Stan: I thought we wanted to... be able to know the openEHR type of thing... we could know that this thing is the protocol.... So could have the openEHR protocol... Were you thinking that too, Thomas? Or only for clinical... compatibility?

Tom: ...Try to make it a single-list again... course-grained... What Joey just said... the runtime... So I was hoping that we could agree on a set of types... So interpretive... what we in openEHR call "State". Patient-state. You have to know some things about the patient. Just taken a glucose challenge... totally depends on ... taking a glass of glucose. The generic concept of that is interpretive...

Stan: Well - I think you would get a lot of disagreement from people about what is interpretive or not... The method... there are known clinical differences between the methods... depends on your use-cases... any qualifier could be interpretive... it tells you more about something... And whether that more is... depends on the use-case rather than something you can derive from first principle.

Tom: If in some case the... changed the meaning of... you could mark the method as...

Stan: Is not determined by the kind of thing it is, but by the use-case that uses it...

Tom: Example?

Stan: There are use-cases where I say - I don't care if used BP cuff or was an in-dwelling catheter for BP. But in some cases this is important. In-dwelling catheter is more dynamic... than what you get from cuff. It depends on the program... they say - for my purpose, that is important... how I want to apply logic...

Tom: Then I would say you have to query for... with that method.

Stan: Another way to say this is I don't know when they DO care about...

Tom: My... is to have a pretty good chance of knowing when something should be displayed and when it shouldn’t be. We have... generating screen forms... they tend to do things like... method... then it just generates a little icon you can click and it expands... what was the position... if it is there... So that is one potential reason to do that.

Stan: That to me is a whole different class of thing. You started off with interpretive - whether it was important or essential... to whether or not I want it by default to display... I agree with that, but that to me is a different... interpretation... access than qualifier.... modifier...

Tom: You couldn’t compete on the focal... without looking at...

Stan: That is the point. There is no agreement about that. No... to say this set of things is interpretive and that is not... no set of rules...

Tom: I would not have seen this as a first interpretive thing... mark things as you want... completely individualized.

Stan: That is the point. Is not a characteristic of the object... whether it is interpretive or not is dependent on what logic I am using... So if you specify it then... not something that is declared as part of the object...

Tom: It is used in... but we don’t need to..

Stan: The Use-case you said... whether we want to display it or not... I can override it.... not whether it is interpretive or not... display it.

Tom: How would you classify glucose test... the fact that the state of the patient is 1-hour post-consumption of the glucose consumption?

Stan: Now that is pre-coordinated in the LOINC code.... The LOINC code says....

Linda: But if not pre-coordinated, it would become a qualifier.

Stan: Yes.

Linda: I would prefer these values be objectively defined, and these.... that can't be defined.... would be... the modifier shifts the hierarchy of the meaning... and presumably something that is not there.... So - not get a category or...

Stan: Yes - those are the ones I would have said we need.... the focus... qualifier....

Linda: Yes... not... value of the data...

Tom: To me, the focal item - the thing it is about - is the value... you could have something like...

Linda: I guess the challenge with the value is... the distinction between the value and the data itself is sometimes not clear.... say, observable of hair color with finding of brown hair, or... different ways of breaking up the...

Stan: The only other things I would say, Thomas, is ... just to match the grammatical thing... I would say focus instead of focal.

Linda: Other thing - other than main category, I would say item-category... is category at the top of items.

Stan: yes.

Tom: My main... was to show the mechanics...

Stan: Make that focus instead of focal...

Linda: Put something else into the mix.... to understand why doing it this way... these items appear in the data instance... lab test.... My question - if every archetype has same focus concept and same qualifiers... why not doing it in ADL(?).... why inference where... do we want to... every instance of the model...?

Stan: Really good question. I would tend to agree with you... want to do that in the model and not in the instances...

Tom: Why do we think this would appear in instances?

Joey: I didn't think this would appear in the instance.

Linda: Aren't you adding to the attribute... ?

Tom: Yes - but not talking about... not a really... have to go through... into CEMs... 13606 CDAs... before the instances.

Linda: But it would appear in the CIMI instances because we are planning to... instances of this archetype... because...

Tom: ...

Stan: The example data.

Linda: Just want all to understand... a conscious decision... normally... you would say once about the archetype and not put into each instance...

Tom: I was unaware we were talking about CIMI instances.

Linda: We talked about them in Amsterdam.

Stan: I think you were correct - not planning that people would implement CIMI model in their production database. But we were planning to... sample instance... that could... and populate... so could test software with the example data....

Joey: So - Thomas - could we do a...terminology-binding... so you could bind to qualifier... to modifier... to value... would be a... terminology that just has those things... so each ID - you could identify what it is... instance?

Tom: Yes, we could also, in theory, put this classifier in the... archetype mini-model...? Could put any... in the archetype... I need to better understand what the requirement is.

Gerard: What we are talking about here is the meaning of things, and I want to reserve the archetype model... to express the meaning of things and not use attribute to change the meaning of things... So I think Stan wants to see in the archetype model... that is one thing... another is... in the pattern that tells me... what the focus point is... the pattern I use will make that clear... so I have no need of... like these.

Tom: We could do it... Could create a special category of bindings that... for any path... that have to have a binding that marks them as some code... focus... modifier and qualifier... Gerard - you have an alternative way of achieving the same thing?

Gerard: Yes - and I have my own internal... that tells me what it is...

Tom: A binding?

Gerard: Yes- the normal way in which we do bindings with the archetype model.

Tom: So - maybe we need to get a better idea of authoring phase... runtime phase...

Stan: Yes - we can say that. At the authoring time, what you want to be able to do is tag these things and say - this thing is a modifier, this thing is a value... and you want each of the items to have those things specified... so don’t want one of them to not have one of those things done.... We use the difference between qualifier and modifier... at the time of query, is important... most of the time... unless you want to... couldn’t... how often a term was used, searching for body location and know whether it was the body location associated with a BP or fracture or Heart Rate... is not meaningful . So at the time of query - you use these to tell people they need to specify a focal concept and... And this is the meaning. ...the modifier will tell you this is the measurement on the patient or on a transplanted organ or on a baby but stored in the mother's record.... that is the real value when you are querying... to allow people to get what they intended from the query.

Tom: OK - would it make sense if I did this in a more term-binding way...?

Stan: Let me ask - does it make a big difference to do it that way vs. putting it into AOM? If we did it as a term-binding, the proper use will depend on a tool where the tool requires a... rather than the semantics... whereas in AOM you could say - this has to be present and it has to have one of those values.... would have to be in the tool... rather than in an explicit part of what we are doing?

Tom: That is 100% correct. So if we feel that this is... a health informatics requirement... and I am inclined to agree... we can... the enumeration... and, like you say, any tool that... implements the...

Stan: That would be my preference... but I want to get Joey and Gerard and Linda and Patrick and... to say what they think.

Deepak: ... tried to model this... is another meaning type... observation is...? Category is an enumeration of something?

Tom: It is an enumeration.

Joey: My vote is for AOM.

Stan: I think Gerard was for AOM as well.

Deepak: Are we agreeing to go from...?

Stan: That change occurred because we took out structured type...a breaking change...

Deepak: OK - we will change the meaning according to...

Tom: I think... we are still at a pretty-experimental way with CIMI really... I was just being... by doing that...

Stan: So - if we are going to have rules, I would like to obey them... [laughter]

Tom: Makes us look like good citizens...

Stan: So - everyone has a preference for AOM. And Harold will have trouble as well...

Harold: I just joined - what kind of trouble are you giving me?!

Stan: Don’t tell Harold. [laughter]

Gerard: ...just come up with a set of codes...

Stan: It is the difference that I just said - if you do it that way... our rule is... you have to specify the categories for every item. If you do it as a terminology-binding, then tool... If put into AOM, then any tool that uses AOM will have to respect that...

Gerard: The archetype... I am using a pattern... all archetypes I make are using the same pattern... most of the... types are there... don't change much... so - I don't care. Either one.

Stan: So - Harold - we are talking about the thing we were talking about... coded-label, last conference call... Allows us to tag things... this is the focus concept... this is that value.... these are qualifiers or modifiers... rather than the focus... and we were going through the ways we could do that. We had 3 ways. One was - we were adding a new attribute to the reference model. The second way - do those bindings... meaning bindings... or change AOM so reflected... And part of the reason this discussion came up was adding a new attribute in the model would let this appear... data -instances... Any cases we knew... didn't want them in instance... just... so a term-binding or a...would accomplish that. And the difference between AOM and terminology-binding.... whereas if done in AOM, then it is explicit... what these things are used for and what is required.

Gerard: There is another... that is... I don't want to have many divergences from AOM... when compared with AOM and the 13606... the renewed one...

Stan: I have the same goal but I would keep it in line by having them do it the way we do it...

Gerard: That is fine but... if try to convince...

Tom: Talking about 13606, Gerard?

Gerard: Yes.

Tom: I think we agreed- openEHR and 13606... between 1.4 and... 2.0... Each of those intermediate versions captures... a bunch of useful things... there is a big (wiki?) page... (?) and Diego had a good look at it... I agree that... really an AOL type of thing... 1.6 or 1.7... It will have to hit 2 before it will do this kind of thing. Like anything - we did not think of it at the beginning. We have to manage it in a reasonable way.

Stan: So - Harold - we were asking - if we changed AOM to add this item-category... how much trouble does it cause you in AML?

Harold: Would be hard to answer without seeing what it looks like in AOM... How difficult is it going to be to transform... I need a better idea of what it looks like.

Tom: ... Have to work out what it looks like in ADL... You can see [on screen]... syntax does not matter... Imagine...

Harold: So, what is the '2' you are pitching there?

Tom: '2' is from... we can have strings and letters and... these kind of classifiers... something like the IMH classifiers... various elements... Probably have to think about... what a reasonable default value is... You might say the default is qualifier... And you have a rule that says you must have a... in the flat... or archetype... a tricky thing to get right...

Harold: There are 2... As far as reference model vs AOM. So far all I am seeing could be implemented either way except... this is model and we don’t' expect it in the instance. So the question - is this the only thing that will be in this category... or will we find other things in a model that we don’t want in an instance as well... what triggered my thought is... underscore in it... do we really need this in the reference model or can we have our own enumeration? The other issue when it comes to AML, this is probably the only thing that is health-care specific in AML.

Stan: I don't think this is health-care specific.

Harold: Generic?

Tom: Yes - I think so too... the concept is pretty deep.

Stan: Yes. So - let me say this. We could continue to talk now... if there are more questions of what we have to accomplish... but Thomas, we'll let you think about it... what it might look like in AOM...

Tom: Yep. I will bug some of you offline.

• ACTION ITEM:** Thomas to think about and work out what this change would look like in AOM -- if we changed AOM to add this item-category (coded_label? See above) **************

Stan: And if he has that, Harold -- a concrete proposal for AOM... this would let you... think about... AOL.

Stan: Thank you, Thomas... want to show our appreciation... would like more than at-a-boy, but...

Joey: Follow-up question. I think this solves what we talked about in Amsterdam... Harold brought up... We can see the actual value. The question is - for all elements... do we want to say for every element... term-binding to SNOMED or LOINC... now we can tell what value is... LOINC for focal concept... but do we still want terminology-binding... and say is...?

Stan: Yes - when doing CDL with GE... the feeling was... yes - I guess what ... made it more symmetrical... every stinking thing had a ... rather than... and never have a code..

Joey: So - in SNOMED extension, we would add these in?

Stan: Would add 1 code that said... data or... one thing in code system...

Tom: What problem is this solving?

Joey: So - does every IP [ID?} have a terminology-binding? ...could not really tell what was the value before you came up with... But question is - do we still have terminology--- to tell what...

Tom: That gets back to...

Joey: Could say this is the value to all LOINC concept... the results value of a LOINC concept.... we could have different flavors of the specific... not have to be...

Gerard: I have a question for Linda... this discussion of category... how does it relate to your...?

Linda: If you had all the SNOMED bindings you would not need this...category. However, there is one area where it was causing us contention with these.... in terms of qualifiers and modifiers and results... You can do this in SNOMED, but I understand... if want an easy way to categorize in models themselves... would be... However, would want the rules around... in the model... semantics bindings that are consistent... would not want binding... temporal context...

Gerard: So should harmonize that?

Linda: Yes - as long as is consistent with each other. Makes sense?

Stan: Yes - it does to me. OK Joey? ... Item or data... have a term-binding as well. Progress on lab models – Patrick Langford

Stan: OK - done with that. Patrick - do you have things to show and tell?

Patrick: One - thing - to clean up SNOMED CT... gave multiple columns when found different types of things... so I cleaned up so only LOINC and SNOMED CT... both lower case... and the other - units put into the correct place. So I corrected those... So now - I am in a bit of a holding pattern.... how going to consolidate the archetypes.

Stan: So, nothing to show now?

Patrick: No - not anything of consequence.

Stan: OK - let's go on.

Language Attribute – Add to statement, remove from the coded_text data type in ref model

Stan: I don't know if want to do this now... and maybe we decided and I forgot... language attributes... statement and take out of coded text... basically saying... at the level of entry... you need to declare your language... not see a Use-case... part of language in one and part in the other... No - it was worse than that... we said in a reference model... What do people think about language attribute.?

Gerard: In 136060 - there was a discussion. If I remember correctly - was a data type... would propose a data-type... in text... but not in coded_text.... and if language is an issue -would be part of the archetype itself.

Stan: And where would it be in the archetype itself?

Gerard: ...It’s a fixed point in the archetype... the language is specified... we have not talked about where in the archetype... at the data-type level... only for...not for coded text... All the attributes... what the language is...

Stan: The reason this came up - as we looked at models, and especially as you think about actual instance data, if language is in the data type, then everywhere you have a code, you are redundantly specifying a language - when 99.95 percent of the time the language will be the same. I think that was the thinking behind this. I think it is a good addition. Language associated with the text blob... dealing with coded stuff... Specify at the... of entry or bigger... container... every code in here is English or French or... Linda - you may have raised this initially, or I may be attributing it to you unjustly.

Linda: Sorry - are we talking about the language... moving it to statement?

Stan: Yes.

Linda: I can see benefit in doing that. We were talking about looking at some European Use-cases... to see if any benefit in keeping it at coded text. ...is the most useful level, in terms of sharing. So I would support moving it to there.

Stan: I think that is my preference, too. The only downside to taking that of coded-text is that now we would have a data-type that is... slightly different from FHIR or openEHR... Language is in the openEHR data-types, right Thomas?

Tom: It is on the entry. It is the equivalent of... and you can override it on a text because you can get these funny... can get in Norwegian... I don't know if anyone has convincingly articulated... with language... bits of text. But the theory is... if you can't know the language, there is.... but the meaning is not the same.

Linda: The other way ... in CIMI we are defining logical data-types... So could take the CIMI... and interpret it as...

Harold: I was just checking - language is optional....

Linda: We were talking about removing that. Cannot think of any time we would want to... if someone wanted to add, then they could...

Harold: I had to look... So you are saying... removing it all the way? Making it not possible is our goal because if it is possible... how does it hurt us?

Stan: If not needed, then people have to program complexity that they do not use.

Harold: OK.

Stan: And that is our argument. We don’t have a single use-case that people can express when needed.

Harold: From a political standpoint - going to hit the checkbox issue, whether use it or not... if you support multi-language... But I guess I am not opposed...

Linda: I think we have had a philosophy of minimalist... and particular use-cases and implementations - can add to that. So, in the spirit of ... I think it is not needed.

Deepak: So does that mean that... what if someone wants to know what language...?

Stan: I think we were planning to do... You could associate language to a text blob, but could not associate language to a coded-text.

Linda: Text-block or statement?

Stan: Also - with a free-text string, could associate language, but not with free-form text.

Linda: Oh.

Tom: ...just put it on plain-text...

Linda: Oh -ok. That is good, too

Tom: Depends on whether you have the rubric on coded text...

Deepak: Moving it from...

Tom: Time (?) is a language-dependent element of coded-text...

Linda: Stan - are you suggesting that language is... ?

Stan: Yes.

Tom: That is how we do it. Also - 136060 does this.

• ACTION ITEM:: Harold & Deepak make changes to remove language as an attribute in CODE_TEXT but support as an attribute of statement and as attribute that can be associated as plain text, and we will look at this next time we meet **********************

Stan: So - that is the proposal. We would go ahead and make the change and look at it next time to make sure we agree.... To remove language as an attribute in coded-text, but support language as an attribute of statement and as an attribute that can be associated as plain text. So - Harold and Deepak?

Harold: Yes - I was just typing Deepak.

Stan: So - I like that. So - interesting questions....

Discussion of archetypes and templates – Thomas Beale, All

Stan: Archetypes and templates. When you get to the level in modeling, where the structural part is not changing - only the constraints are changing, terminology-constraints or magnitude constraints - those kinds of things. What is our style for doing that? And part of this - I need to understand - I would like to understand your best thinking about this, Thomas, and then get other viewpoints. I know we have been thinking about this - how associated in FHIR as well as in CIMI models - so I am interested in what Joey thinks and Harold and others.

Tom: So - my first question. At the moment, we are doing a 1-to-1... code... I think we need to determine whether that will work out... The right level of mapping archetypes... LOINC is... And the second question... a bunch of examples... like... 25 archetypes... that are bacteria... identified in X where X is tissue or body fluid... where Y is some kind of method... And at the moment... done in some sort of... fashion... so generate 25. Do we want to do that or do we want to have bacteria as the target thing of the archetype? The analyte, as you like... The fluid or tissue - whatever makes sense for that. And the method - whatever makes sense there. You would end up with 1 archetype rather than...

Stan: Let me insert one thing. This is fundamental, from my perspective...

Tom: Oh - the last time we got to this... you completely rightly said something in CSF has a different implication than something in some other body fluid. Was that what you wanted to say?

Stan: Sort of. CIMI modeling - iso-semantic models, models in... preferred and models in a Use-case but not preferred. Then we have to make those pre-coordinated anyways, if someone is using... My understanding with iso-semantic models... But CIMI, within scope, would still have the responsibility to make those pre-coordinated models.

Tom: So - my question... If we were to treat the archetype-library as... Imagine we did correct... archetype... some parts of the body... let's just say we did the proper clinical analysis... But also smaller... for a good reason... represents a good theoretical representation... Then I would go... template... mark the archetypes... and those archetypes or templates can... reconstitute all of those 25 pre-coordinated on... So have... all of those... in bone marrow...

Stan: So - explain if we represent those 25 instances as archetypes or alternatively as templates - what are the implications? What is the advantage of templates vs archetypes? What is the diff between those 2 approaches?

Tom: You want the archetype library to be as small as it can be rather than... rather than copy a coding system. And I think the archetype-library needs to be... best practices... the possibilities we need to represent...

Stan: I agree, but I assume we would do that by saying which is preferred. So, if preferred were... the 6... And would only see the other - the 25 - when your case was... Tom(?): ... the 6... mark as preferred... So, bacteria... all are pre-coordinated... no?

Stan: But that is because... how Patrick did hits. Those are possible... are implied by other LOINC codes that may or may not have been in the set that Patrick was working with... because the XXX codes imply that...

Tom: So - these... bacteria in unspecified specimen... is that an XXX?

Stan: Yes - that is where XXX came from.

Gerard: ... A template I reserved for a set of archetypes that... system... most often the 136060 composition class... this is how I use the term archetype and template... So, need another word for template...

Tom: Think of them as a second level of archetype. All I am saying is - we could potentially use... Let's say this group of archetypes... collapsed to.... Say Patrick changed something in the converter... which is totally pre-coordinated ones... I would be inclined to mark in some way... only the 6 that have design sense. So, imagine we do - some final transformation... BMM library. Which ones do you work on? I would argue... it is the 6... and the others... generate-able within the CIMI environment...

Stan: Yes - well... this... I don't know if relevant, but this comes to mind. The situation is... as per our sort-of previous discussions, from modeling perspective, the post-coordinated are more often the most elegant. Would be most elegant if never pre-coordinated the... specimen or... timing into the LOINC Code... Most elegant from modeling view. But if look at what is in systems, there are certain common things that people never post-coordinate... and always pre-coordinate... Serum glucose and serum sodium... Never seen "sodium" and then second element which is "serum". But there are other areas... get into less common areas... fluids... we do cell counts... We can do on acidic fluid or wound drainage... Or they have some test that is a generic cell-count... always post-coordinate... and say whether it was acidic fluid or synovial fluid or whatever... So - the work always does certain things pre-coordinated and other that it always does post-coordinated. So if you adopt a standard mechanism and say always do post-coordinate... that is elegant... but will... just to be consistent and elegant.

Tom: I would not advocate that, Stan.

Stan: I am not advocating a certain position myself. Just trying to understand our options.

Tom: Could we generate the post-coordinate-able form - the archetype that corresponds to the 20-30... And then those things that are pre-coordinated in the world... make a specialization... So if there is a specialization... than use it... or otherwise...

Stan: Yes - a good approach. And what comes into that... We have options... about making everything archetypes or everything preferred... Some of those are archetypes and some templates... and not clear to me what the tradeoffs are in that... Are some less work or...? Are things that are templates represented differently than archetypes in terms of representation of model... Are talking about a particular type of template... Templates that are a constrained subtype of a ... Not templates that are aggregators and... I am talking about templates that... have a parent type and are making multiple parent subtypes... If all of those constrained things were archetypes or all of those were templates - would they be different?

Tom: Would be the same structurally - I am thinking now - that ... make them archetypes... All of the things... bacteria... they are all archetypes... post-coordinate ...be way of doing it... pre-coordinated in the real world... if all looked like archetypes... equally... talk about templates if we like. I think we should be able to easily spot the things that are the sort of designed-parent archetypes... because they define the actual possibilities... even if have... serum sodium... and only talk about serum-sodium... So what is used is what is there... Parent type... more important... Define a... number of archetypes.

Stan: Yes - that is where you want to put your effort... properly... those parent types...

Tom: Kind of pre-coordinated... attached to ... Make sense?

Stan: That is what is going on in relationship between LOINC and SNOMED. The observable concept model - not only post-coordinates the parts we are talking about, but goes to the extent of... the other parts... the adheres in... and the substance... and all of those kinds of things.

Next CIMI MTF Meeting on January 8

Stan: So, we are overtime. I will take the next 2 weeks off. Next time I will plan to meet would be... Jan 8th. Seem alright to people?

Linda: I have a couple of extra weeks off - back on the 19th of January.

Stan: We will try to make progress, but will give you time to object when come back.

Stan: These are the kind of style issues... [end-of-meeting]