CIMI MTF Minutes 20121004

Jump to: navigation, search
CIMI Modelling Taskforce - Meeting Minutes
Thursday 4th October 2012 @ 20:00-22:00 UTC


Linda Bird

Dave Carlson

Sarah Ryan

Mark Shafarman

Ian McNicoll

Harold Soldrig

Thomas Beale

Gerard Freriks

Joey Coyle

Jay Lyle

Michael van der Zel

Daniel Karlsson

Rahil Qamar Siddiqui

Mike Lincoln

Hendry Wijaya

Graham Grieve

Stan Huff

Stephen Chu

Peter Hendler

Eithne Keelaghan


  • Weekly news & updates
  • FHIR Connect-a-thon (Grahame)
  • Intermountain Lab Models
  • Laboratory Report - Modelling Patterns

Weekly News & Updates

  • Amsterdam Meeting - waiting for final confirmation of venue availability from William. Tentatively scheduled for Sunday 2nd to Tuesday 4th.
  • CIMI MT Terminology SIG - Doodle poll indicates Tuesday 20:00 UTC
    • SemanticHealthNet - Collaboration?
  • Minutes -> Eithne to send to Michael van der Zel to add to wiki
  • Tom Beale - ADL tutorials:
  • TASK: Create and maintain ADL 1.5 version of models. Volunteers required.
    • Ian McNicoll/ Tom Beale
    • IMH representative (perhaps Doug Mitchell)
    • Michael van der Zel -> autogenerate ADL 1.5 from UML (which indicates the patterns and reference models)
  • Models submitted:
    • MOHH (Singapore)
    • IMH
    • NEHTA/openEHR
    • EN13606 Association - Lab Test example
    • Results4Care - Lab Report
    • HL7 (CDA documents)
  • Gerard -
    • ISO13606-3 Semantic links and state model
    • Will look at lab modelling patterns which align with Contsys
  • Peter - OWL Pizza tutorial
  • Michael van der Zel: RM Issues ?
  • Ian McNicoll

A couple of links to resources that might help explain the workflow and quality guidance we use in CKM.

Currently we would expect Draft models (pre-review) to be pretty well modelled and documented. I have just uploaded a Visual Acuity archetype as an example

We are however, also about to add a new facility called Incubators that will allow more immature resources to be uploaded and discussed.

FHIR Connect-a-thon inputs

Deliverables for Connect-a-thon

  • Lab Report Profile(s) (XML instance of FHIR Profile model)
  • Example instances of Lab Report

Intermountain Lab Models

Laboratory Results - Modelling Patterns

Discussed last week:

Sheet 1

Brainstorming on modelling patterns:

Sheet 2

Detailed Minutes

Michael: I will be on line only 1 hour.

Linda: Updates.

Spoke with Virginia - confirmed with Michael re: venue available from 2nd to 4th.

Comment: Will work out details with William tomorrow.

Linda: Terminology Subgroup - sent out email.

Tuesday at 22:00 UTC is most appropriate, but need Daniel and Rahil's response. So will first confirm with them and then finalize.

Gerard: Today - phone conference on Semantic Healthnet and discussed CIMI. Semantic Healthnet is going to discuss... and want to try and get synergy. They wish to think about this as well.

Linda: Thank you. Perhaps some of us can be invited to Semantic Healthnet meeting. Might be useful, especially Terminology area.

Gerard: Will be doodle in Semantic HealthNet...

Linda: Thank you. Anyone - Please respond to CIMI doodle - small group...

Linda: Next - Tom Beale send information on ADL tutorials - YouTube Link...

Michael: I don't remember seeing minutes of last week.

Linda: Eithne sent me detailed notes. I combined these with my brief notes - plan to get up on the wiki... Need to find someone she can send these to...

Michael: She can send to me.

Linda: I will do quick review. If anyone has any changes, please make them. Also - I want to acknowledge Galen's help with taking minutes... [he is not at the meeting]

Linda: I encourage all to go to site [from Tom Beale]. You will have a good understanding of ADL after viewing this. For us to produce implementation we need to produce ADL 1.5. Need volunteer to take Mindmap and write and maintain ADL 1.5. So we can talk through models in ADL found... Anyone interested in taking this on?

Ian: I am happy to help

Michael: I want to try to generate... UML's from...

Linda: UML needs what ref... Style of UML - Not sure if that will be a fast job...

Dave C.: The archetypes don't intend to include reference patterns like observation... Lab archetypes derived from observation. Is that it?... from DCM - how they derive from reference pattern?

Linda: Yes...

Michael: I was planning to do some UML stuff... not DCM...

Dave C.: Great - UML representation from ADL from that...

Linda: As develop - would be great to see...

Dave: I want to see... partial... want to see if UML structures are same as ADL... want to see UML then...

Tom: Probably do something similar to Intermountain. Need a kind of reference target - hand built in time for Amsterdam meeting... in repository. Quick for Ian and me to build.

Linda: Can I put you down?

Tom: Have 1 version - most faithful version in ADL. We want some version that provides a reference point -- target -- in formalism.

Dave: I agree. Use to create UML models...

Tom: Yes - that is the way to go.

Stan: We'll have someone... Douglas Mitchell - from Intermountain - staff modelers.

Linda: The more we have writing ADL, the better.

Rahil: Could I volunteer my time to get involved in modeling? If enough for ADL, perhaps help in UML representation.

Linda: Join UML calls... or join meetings re: MindMap...

Dave: Great to have you... still trying to get doodle...

Linda: Thank you all. If anyone goes through ADL tutorial and wants to write me, we can use more people.

Singapore submitted MOHH... Temperature, Medication.. Thank you Hendry.

Linda (cont'd): Thanks to Stan for Intermountain...

EN13606 - lab test example - is Josh online? He mentioned clarity of models. Michael - thanks for lab and translation...

Stephen: For NEHTA - we also use med, medical instruction...


Linda: Yes - thanks - need to get up on Google docs.

Stephen: It is there.

Linda: Good. Go to Google docs website with submissions folder. Ask if don't know...

Rahil: Is call for models just lab result models?

Linda: No.

Rahil: I sent back Body Temperature...

Linda: Perhaps not on broader group... mailing list. Want Lab Report, Body Temp, Immunizations... You need to get on mailing list - send email to Virginia.

Hopefully we'll have more models in next couple of weeks.

Mark: I sent in HL7 CDA... late last night.

Linda: Thanks - I saw it.

Mark: I can show some things, if there is time today... small number of basic patterns for giving a type of model... and templates on... constraints... interesting spreadsheet from CDA... Also, similarity of patters... CDA, FHIR, and emerging CIMI. So we have patterns - don't have to re-invent...

Gerard: I am in the process of developing... system of concepts for Continuity of Care... I will write about this...

Ian: I will make sure models up in next few days. Not surprised OpenEHR models alligned with CDA...

Mark: Let's make use of the parallels...

Linda: Yes - over next 2 weeks, need to make sure we make use of all these models...

Gerard sent out ISO 13606. Please read (it is a PDF).

Rahil - please respond to terminology subgroup doodle poll.

Rahil: I did.

Linda: Alan Rector...?

Rahil: ...Alan Rector... probably at conference... I will let you know when hear from him.

Linda: Michael - reference models - any issues?

Michael: I still have to do... creating issues...

Linda: Ian - thanks for sending links for CKM...

First talk about FHIR Connect-a-thon. I put Lab Report... UML diagram... Want to discuss what artifacts... MIF should be producing.

Phone (Graham?): Look at Lab Report... We think has capacity to produce all Lab Reports, but... engineering... work with lab reports. How do you use the group and particular test structure e.g. profiles on lab report... ...and if you go to "Resources Profiles" on the left... Another resource - applies set of rules... constraints on resource... functionally equivalent to ADL, although not as expressive.

Sheet 3

Sheet 4

Key is Profile... has 3 parts... set of bindings... references that you can define extensions that are used in the profile, and then define profiles on the resources. So for each resource - name profile... constraints... use Lab Report... then are a set of things you can say additionally... is a flat model.

Content using structure inside path... Things can say... constraints... formalism... Take constraints expressed on underlying resources and those constraints can be expressed...

A number of different modeling paradigms for expressing constraints... Resource profile... XML(?)... The computer will ?... framework... expressing constraints because that is what is easy to do. Produce profiles... spreadsheets... We are not saying a great way to produce profiles, but it is simple.

Sheet 5

Linda: So, Connect-a-thon is the goal. Produce our own Lab Report in MindMap and ADL... then XML... and if worked, then generate resource in...

Gerard: If that works... If can attract enough attention to Connect-a-thon... but we agreed... we need profiles and that is where CIMI's interest is - in profiles... and in as much, is important to participate in Connect-a-thon... a pathway to test... It is an independent modeling platform, so is worth participating in...

Stan: Yes - we would want to recapitulate the elements in the resource, but then produce profiles on resource - lipid panel, electrolyte panel. Joey and I have been having interactions with Graham about this. It might be useful to show a profile. Can you go to Lipid profile?

Graham: Yes - example of Lipid profile.

Sheet 6

This is a rendering... You can see some elements. The profile does not say anything about them... profile says this is a lab report... Has a code -> fixes this... fixes names, results - result code must be particular... Down at bottom - more details... Describes how lab report is used for standard lipid profile... From here, we list the elements - explaining what this is.

Gerard: Not a statement of constraints. Typically... differences... using profile.

Stan: Don't need details... There is a level of modeling... the value of CIMI is making profiles... define lipid panel and cholesterol, LDL, and other. With Intermountain, we have quite a few CDL definitions... would go from these to ADL versions of our panel definitions. Would wait to automate... from resource of value to FHIR connect-a-thon.

Linda: Graham - what is granularity of 1 resource? Is it a single panel?

Graham: What do you mean? Granularity is a single lab report. If multiple panels in single report then... but most labs... think of one lab in lab report.

Stan: So - can have CBC and white cell differential. CBC = hematocrit, red blood cells, white blood cells... differential - number of lymphocytes, PMNs... So - we don't have models for these yet. We don't even have ASN1 models. But to match lab, would have super-panel... CBC with differential -> CBC and items from WBC count, so collections from collection. Linda - was that your question?

Linda: Yes - that is my question. ...profile group - profile each panel, or...?

Graham: Partly question is how things map into FHIR, but also asking how things work. But labs vary wildly... In Australia, full blood exam - some pulled differential out into separate section... The point of profiles - to try and get labs to standardize their reports. Grouping and categorizing test results...

Stan: So - you can allow collections in terms of models and style... it is a different question.

Graham: ... Level of categorization... But when we did... for NEHTA, no need to do that - arbitrary...

Stan: ... Makes good sense to define what we call panels... Create model so someone can reason when get back panel... member of 2 different panels. Other - making data entry screens... like form definition... data on screen... some unique name... only use form to collect data. Don't know if we need to get into this level of detail.

Mark: Could look at... has standard name for panel... If fairly standardized, probably has name for it. Goes back to idea of need to find things that are independent even if in panel. Don't want to lose... in container.

Ian: I did analysis of LOINC panels - pretty messy. Complex and messy. Does sometimes make sense to model... Ultimately we get from labs what they give us.

Stan: Speaking for LOINC, I am co-chair with Clem McDonald. We are like CIMI and OpenEHR... trying to make things that are useful. We will add useful content. [Over] 4-5 years - discussion about panels. Haven't even gotten into calculated values... So Linda - up to you about how much you want to discuss... I've gotten a better idea about FHIR. Need to be coordinated, so we make profiles that conform to and are subtypes of resource.

Linda: Yes - must discuss. Granularity. What level we come into with FHIR resource. Graham - CIMI different layers... Can you profile... on profile in FHIR?

Graham: [can't hear]

Linda: Because they are flat?

Graham: Perfectly reasonable to cascade constraints...

Linda: In ADL, we could have generic CIMI lab report that is flattened and ... Lab Report.

Graham: Yes - there would be tooling. Tell you where profiles are consistent... One thing - the profile ?... is not intended to be an authoring format.

Linda: CIMI models - would say what connection was?

Graham: Yes - meta-data... profile.

Linda: So that is... generic profile and other flattened profile.

Linda: Any other comments for Graham?

Stan: Question: If someone orders lab report with single result, still done as lab report... If only want hematocrit, still do panel?

Question #2: For Connect-a-thon, want to agree on things we want to generate in connect-a-thon? ...or use your Use-case... or high-volume results?

Graham: I think what would be useful... I worked on... I have limited resources - can't do 100's of profiles. I am interested in 3 or 4... If anyone expresses interest...

Stan: We work on producing one first... Want to take advantage of things already defined at InterMountain. We can get back to you. Thanks.

Linda: Any other questions for Graham?

Thanks Graham.

Linda (cont'd):

Next - finish looking at InterMountain models.

So last week - StandardLab Obs Qualitative.

My question: How to define a panel in CBC... and what is granularity of models?

Stan: Core StandardLabObs (screen)

Sheet 7

Same level of specificity as Graham...

Is the parent of all that you see...

Next level under Core - have coded quantitative, narrative, or... interval...

Next pattern down... can see core models...

Secondary use models - models [we are] making to support Sharp Grant activity.

Sheet 8

StandardAlbObs - in production use... ...are different - secondary use are like Open EHR templates. We have taken away... not part of secondary use... No need to match orders... These are essentially a template of core.

Same is true of Standard model.

Sheet 9

Next level down - jump into... glucose - would see models specific to glucose - can click on any - can see style issues popping up...

2-hour glucose tolerance test...

We are doing terminology binding to set key correspond to LOINC code... set to mg/dL. Other constraints we could make - example: what is absolute magnitude...

Sheet 10

So - start observation... became standard core lab result, and that has 5 subtypes (...) - and each of those become parent of 10-100... glucose...

Talking with Peter Hendler to understand 5 models vs 200 models... if can argue... define subtypes of models vs additional...

...LOINC names

...Proper units

...Proper interpretation - can represent as knowledge object.


Tom: Might not be aware, but form on right is like the compiled flattened form - source models down - inheritance-tree. If go past comments - get what is in any given model rather than parent. That tree tab... is result of the way things are - Look at CDL tab - is a lot smaller than what you might have thought.

Sheet 11

Stan(?): Not trying to model things explicitly, but LOINC... combine the two together.

When I was modeling devices, thought this sort of information was better in reference dictionary rather than modeling explicitly.

Linda: Granularity and modeling pattern... Want to set for future discussion. We previously talked about having layers and building on them.

Sheet 12

Linda: Last week we talked about FHIR model and looked at Composition, Entries, Clusters... "Lab Results Report Comparisons (MOHH Holdings)"

Last week we talked about possible CIMI structure... exactly what can be represented... Single Lab result - CBC & cluster within. But what Stan was talking about - each lab result - single result, single observation - and grouped. Want people's feedback on this. Stan - about individual CBC - entry vs links?

Stan: Not sure I understand entries, clusters, items in ADL. I can describe what we are doing - maybe Tom can explain.

We have statements = grouping of data elements that would be a statement about observation finding about patient. So in lab - simple - you have the code for the finding - value associated with codes and abnormal flag and ref-ranges. That would be statement... Grouping would be bigger collection - include who did this, etc. qualifying information... Statement = this group of things together... make statement about patient, so create statements - and those are building blocks you create other things from.

Linda: This is important to explore further. In the models we looked at to date, laboratory result - could have test result in it - e.g. CBC - what are modeling layers.

Sheet 13

One option - composition level - might represent documents... into summary. Intermediate layer - then specialization - either Lipid Report or...

Linda: In terms of administration entries...

Clinical - looking at order -> specialized into...

If panel -> cluster - single... Lots of different compositions.

Mark: I know this is a common way to look at clusters... events... in Section... or subsection... Somewhat arbitrary... People organize documents differently. We may have something we want to say in clinical statement... Investigate separating document paradigm from set of clinical model and type of model -> map into a particular document. I don't know if we will ever agree.

Peter: Truck - Branches - Smaller branches - entries... I don't use for naming the different branches. Something is branch of something else. No use in cluster - entries - section...

Linda: Comes back to questioning the Ref Model.

Ian: Entries are equivalent of clinical statement. Sections - we would regard as navigational. Cluster are sub-entry elements... Not stand alone. No clinical meaning in isolation. Everything sits around entry. Fits pretty well with CD(?) clinical statement... however we would group...

Mark: Clinical statement has arbitrary nesting within a clinical statement... The viewpoint of a clinician... data within clinical statement has standalone meaning... The clinical statements have one view of data... across clinical meaning... So if I have WBC and have high WBC, I will say test for leukemia... Must be identifiable outside of path in clinical statement. Although I agree... it has meaning at that level, but also have meaning at another level. Need to look at WBC as part of CBC and... change in WBC irrespective of... independent...

Stan: In the way we use the term, the clinical statement = Body Temperature, Hemoglobin, WBC... So CBC --> not a clinical statement, but is a collection.

Mark: So what you are stating is what I am saying... independent statement.

Peter: an independent triplet... arrays... you can have trees, lists... Can't see we would divide into clusters... Only tree - subtree - You can say hematocrit is exactly this, but collection... Arbitrary names for well-known... the kinds of collections sets, lists... Don't understand... sets, clusters...

Ian: Understand in context of Reference Model support...

Linda: So is in terms of where context...

Peter: Context for entry and... inherits the context.

Stan: Ian is saying... everything to us is a collection. Statements = collection. Other thing - Component. Component = part of collection, not standalone. So, for example, Body location -> make it a component and then can use in BP or... Not that it is structurally different ... is only a collection... Purpose of component = dependent on context used and is reusable. Statement = standalone... Peter - you are right... collection...

Peter: So if we think what is... and context = who, what, where, when, why... So... where the context lives at level of entry and section = arbitrary bags of... But who, what, when, where... Does not have to work this way...

Stan: I think you are right. This is where iso-semantic... comes in again. There are other iso-semantic representations that are different.

Peter: ... is at entry... what you are saying, only useful is... If context is same -> put it there. OK - you can call it component...

Linda: In Rockville, we began to talk about this...

Mark: Is a panel = group of things that inherit, but are independent... Is this a section or recursive...

Linda: Also panels - what is a single observation...

Stan: Both collections - sodium, potassium, glucose... should have information. Sodium in Chem 7, same as in Chem 20, same as in electrolyte panel... Different from medication order - dose, amount, frequency - all elements in collection. Not standalone. Dose - dose of what? So there are collections of panels -> each are independent. Other collections - you gain meaning from knowing they are all part of consistent statement. We call them compound statement... different from panel... sodium, potassium... independent but part...

Mark: Are they both entries?

Stan?: I think they are... whether we want two or not, we have a typology. I can only say if I know type of entry then I need all of these or use...

Linda: Same as time series...

Mark: Not just context. I have entries that are well understand collections. So drug order is an entry... Bunch of stuff must be named or not meaningful, whereas Lab Panel is different. So in addition to entries - need type of entries.

Ian: I don't understand. Don't have to label. The clinical model is king.

Mark: Legal question - if I have a system that collects drug information... but no amount... If I don't name collection, can't run quality check on it...

Ian: I agree, but

Mark: Will have to deal with this...

Tom: It could be a long discussion. If you look at OpenEHR archetypes or ... from VA, you'll see grouping of things. In OpenEHR - eg LFT's - natural grouping because those elements form a governance unit. Need to document together. Difficult to imagine a whole review separately for LDL & HDL & Triglycerides & different electrolytes... So lots of groupings... People look at archetype.... When I looked at Stan's - or InterMountain - ordered group - panel - could be anything. Could order specific [electrolyte], or Chem 7... New machine could spew out... In OpenEHR, you design archetype as governance units because they make sense... If want to build panel, build template... Have to build... 4 or 5 archetypes. So template has Use-Case Set of things like... order... business... Value is governance... not have to be done on every... Don't want governance write-up on every... Stan's 6000 data points, but in more open community based governance environment... Need economy of scale from governance...

Mark: I think we are in agreement... I was saying there were patterns... All I am saying is what we are collecting entry is... There is a level where, if I am doing drug administration, I am doing pattern... Not different from Stan's... It is unclear where that pattern is... and looking across patterns for similarity.

Gerard: I like picture... Need to define rules. Each clinical statement always starts at entry level. Everything that will be model... one generic pattern... all other archetypes...

Linda: So individual test result would be an entry? How group?

Gerard: In section. One attribute make it...Governance must not be an issue in finding patterns...

Linda: Important for conclusion on CIMI preferred...

Stephen: Mark and others mentioned arbitrary nature of clinical statement, but not as arbitrary... Atomicity. CBC - where have several members - WBC, etc. Each can be isolated. CBC = collection of clinical statements. Can be separated -> each on its own. But medication order cannot be divided - must be together for it to be medically safe and... That is what a clinical statement is...

Linda: Individual test result can be treated as separate entry.

Stephen: [Voice fading - Not able to hear all of what is said] ... in context of CBC, renal function test, electrolytes... but when it comes to discerning or understanding... component can exist on its own without much loss of its meaning.

Linda: Need to wrap up discussion and come to conclusion. Need to look at specific examples. Other final comments before conclude? OK. Thanks.