CIMI MTF Minutes 20131017

Jump to: navigation, search

CIMI Modeling Taskforce - Meeting Minutes

Thursday 17 Oct 2013 @ 20:00-22:00 UTC


  • Linda Bird
  • Stan Huff
  • Daniel Karlsson
  • Gerard Freriks
  • Michael van der Zel
  • Galen Mulrooney
  • Rahil Siddiqui
  • Sarah Ryan
  • Joey Coyle
  • Patrick Langford
  • Eithne Keelaghan - Secretary

Draft Agenda

  1. Plans for joint meeting with SemanticHealthNet
  2. Brief update on binding discussion from last week.
  3. Entries in entries discussion - See attached files for background

Detailed Meeting Minutes

SemanticHealthNet - Joint Meeting with CIMI in Feb/Mar 2014 in Europe

Stan: Planning joint meeting with SemanticHealthNet and CIMI... in February or March in Europe... Duration not known... But those who could possibly attend, let us know your availability. Galen - you responded to poll, and Gerard... I am excited to meet with SemanticHealthNet and see how we can work together.

Daniel: One idea - have presentations from SemanticHealthNet before meeting... So people will know more about SemanticHealthNet and vice versa.

Stan: Great idea. Would make our meeting more productive...

Gerard: Also true vice-versa...

Stan: Yes - I think he mentioned that... Could be CIMI presentations...

Goal is to Identify Starter Set of Examples to use as Patterns

Stan (cont'd): Last week we worked through examples... Hopefully people looked at this and can comment... Those were 6-8 months old. Goal is to come back with starter set of examples and use as patterns for more complicated lists. Anyone have time to work on those - to improve items?

Linda Sorry - I didn't get a chance, but I am feeling better, so ... will send them out and hopefully will stimulate discussion.

ACTION ITEM: Review last week's examples for next week's discussion

Stan: Thanks, Linda. We appreciate all you do, and I am not going to complain... but thank you so much. That's great - can follow up on next meeting... We are getting close to consensus and approval on it.

Stan (cont'd): OK - any carry-over form last week?

[No Response]

Entries in Entries

Stan: OK - Entries in Entries. Will have to leave in one hour. OK - Linda - Start with your examples - illustrate better. Let's look at these.

Stan: So there are examples that Linda created a while ago - good examples. At first I did not understand... Best illustration of choices we have around representing choices... You want to present or me, Linda?

Linda: I will present.

Option 1 - Hematocrit Example

The slide of option 1

Linda: Option 1 based on openEHR... Represents clinical statement on patient... Panel performed - recorded as an entry... panel... So in example - single Hematocrit - recorded as entry in own context... and Complete Blood Count (CBC) - see clusters... One Hemoglobin, another Hematocrit... and... context recorded as entry. Any questions?

[No response]

Linda: When query over starter(?)... The path to Hematocrit result value is different depending on if going through single entry or panel, and how panel made up in a lab.

Stan: Would be good if Thomas could confirm that this is an accurate representation of what is in openEHR.

Linda: Thing that... subject would be... (?) but overall structure is equivalent.

Daniel: Which of these alternatives...? Well - did he say this cluster is according to openEHR clusters?

Linda: Yes - if single Hematocrit, then in entry... If panel - then cluster.

Daniel: I am surprised. I would expect there are different entries in the section.

Linda: No - this is based on what Thomas said.

Stan: We will confirm.

Gerard: I agree that this is openEHR pattern. They are rather eclectic... and this is an example... path of left-hand-side not same as right-hand-side.

Linda: Yes - raises challenges for querying.

Rahil: I was... if date would be a separate element or attribute... Wondering if there would be elements... Except for that, cluster element alright...

Daniel: I am still checking openEHR CKM and it is as Linda says - they look like this - cluster and element in Entry. So full CBC contains Hemoglobin and Red Blood Cell count and White blood cell count and other...

Linda: Yes - clusters.

Daniel: Yes, sometimes cluster, sometimes elements.

Linda: Anything else?

[No Response]

Option 2 - Lab Test Example

The slide of option 2

Linda: OK - Option 2. Some entries represent a panel - records the panel as whole test... Link at whole blood count... with entry for Hematocrit result, another for Hemoblogin... So think of as individual statements, and the compound one Links to them... Single path to get to... entry level.

Linda (cont'd): Another comment - different types of Links - may want to distinguish compositional Link where CBC has Hematocrit and Hemoglobin... and distinguish from request that originally made it... separate on different date. Stan - anything to add? And Gerard - this is based on... your approach.

Gerard: One of the possibilities. In case of lab test panel, will never be an entry... will most likely be a reference... stored somewhere as a resource... stored separately... as see on right-hand-side.

Linda: Oh - so no separate entry for CBC?

Gerard: No - not in... Better example is BMI and Links to calculate the BMI...

Stan: So - I guess I am trying to separate out your preferred style to what is legal. So - legal - this does not cause entries in entries.

Gerard: Yes - but reserve entry to atomical clinical... Reserve entries for this [Option-2]. So for BMI, is atomical as well, but calculated based on Option-2...

Linda: So is Blood Pressure different?

Gerard: Systolic, Diastolic - 2 separate entries.

Linda: OK.

Stan: People recognize... as this panel... HL7 message... panel... Bind into Data - Display - all want to (?) to what came at the same time. So Gerard - how do you make it easy for people to aggregate?

Gerard: Not a concern - how display. Only store.

Stan: But need to make it possible for people to display. I want to display a complete blood panel...

Gerard: We have the date and order# - can be constructed.

Galen: I am wondering if... step back for a moment... an anology... I use quicken... main interface is check book registry. If I write a check, there is a single entry for what it went to. But sometimes I need to split this out to mortgage or taxes or other... Check is analogous to... But the report is organized by the account... This is how much paid for groceries or other... So, analogy - we are conflating the container - who, what, where... single result fits nicely in there, but maybe the result does not belong in container... Maybe instead of 2 Use-Cases, we treat them the same way.

Stan: Gerard's answer was clear as well. I create panel as query... based on individual... But all entries are in same table, if you will... I don't treat... The only difference is... the container - sometimes contains single and sometimes many... The structure of the data does have an impact on how you display.

Stan (cont'd): I do not understand... Do you agree with Gerard - the left-hand-side does not need to exist, or are you saying something else?

Galen: I am leaning toward Option-2. Not only use for panels, but single... So lab Hematocrit test on left - Link to another Hematocrit test on right-hand-side... Oh - now I see - information subject... duplicated... Oh - I was thinking would Link to Left... Like a Left-Outer-Join in SQL...

Stan: Things on Left-hand-side need to be complete... So context of virtual panel... need to be in panel on right-hand-side.

Galen: Yes... need both left and right to make sense of data.

Stan: I think... right in physical implementation. I don't want to have to do the logical join to get back... Don't want to logically go through... on left-hand-side to get...

Galen: I am suggesting... you have to... All... explicit on right-hand-side... but context on left-hand-side... stuff on right-hand-side needs to depend on left-hand-side.

Stan: That is what I am trying to avoid.

Galen: OK - a thought.

Daniel: The panel has a name. There is data associated with the panel and... individual test... Additionally - I assume you would allow panels of panels. So panel with CBC and Liver test... Not only a two-level structure, but multi-level structure... Probably some kind of practical limit...

Gerard: Number of permutations are finite, but can be large.

Linda: Because so many variations of how can group panels, probably important to query on single result... So don't need to query path... So direct to Hematocrit without going through all those groupings.

Gerard: That is the thinking we have... Another rule...

Daniel: So using BMI - would have its own value. Could consist of its own components of BMI as well... But still, BMI result and interpretation... So left-outside-panel - a (?) of right-hand-side.

Gerard: Yes - with BMI - a number or a calculation using archetype... Although don't advise... or out of queries...

Stan: I think good to separate as two different Use-cases. #1 - same source on same patient. Other - BMI, calculator with weight from 1 point in time and height from other and take these and calculate with own interpretation. Both are valid Use-cases. Could have policies - use one with BMI stuff and other... all panels... But I am... that panels have attributes of their own and not only qualities of measurement.

Gerard: ...Must be complete in describing context so would be repetition.

Stan: So, Gerard - what would be your answer to... Daniel... overall panel interpretation... copy that into Hematocrit and Hemoglobin as well? What is your strategy?

Gerard: You mean - there is a way to express the blood count as normal?

Stan: Yes.

Gerard: At this time... not... I must think about it.

Daniel: I would like to challenge your statement of difference between BMI and this lab test [Option-2]. BMI... small, but lab test could be... So the machine would do in sequence... Assume the relevant point in time is sampling time and not... But same idea in BMI. If have an old value of body mass, and see this patient has gained or lost weight, then takes new one [measurement]. But... not matter. Still representative of the condition... at the time of BMI calculation...

Stan: There are a lot of similarities. You can argue that BMI more... and can fit panel into BMI... But I am saying... the common cases - we don't get the actual analysis time... We only get the sampling time and overall... But not time Hematocrit came off instrument or Hemoglobin came off the instrument... So - internal to machine, things happen in order... These things appear together and show up at same time. And I am saying - in that situation... and resulting time is common for all. BMI model is more general... simpler... Don't have to exercise the option to report each thing separately at its own time... I am not sure if this is central...

Daniel: My main point - having an entry for the panel... Name, instrument, anything specific to panel... synergistic results.

Stan: OK. Ready for Option-3?

Option 3 - Complete Blood Count (CBC) Example

The slide of option 3

Linda: Option 3 - similar - but link... Link is compositional Link. Hematocrit and Hemoglobin recorded as part of CBC... share some context... But want to query Hematocrit in same way whether recorded as panel or single... So record as atomic entry... The content is query-able within each entry... Context is desirable... Individual atomic entries... Can create rule when query over Hematocrit you can derive from...

Stan: The assumption is... Want to be able to query directly on Hematocrit whether part of direct Blood Count or part of Hemoglobin... Hematocrit... Could be... panels and want to query regardless of how... So want to make each whole. So anyone can query data - have date and... as attributes of result... So Hematocrit result has to be complete with its value... and contextual information along with subject and date... as Gerard record... Atomic... Needs to be complete and query-able as a thing...

Michael: You are talking about query-able. Do you take into account the structure or...

Linda: One answer is that there are different ways may want to query. CBC as a whole - then would want to bring back all recorded as part of CBC - that chunk. Or... as a result of query... value... Hematocrit as atomic entry... Path would start as Hematocrit result... Depends on your granularity... But both... there is a Use-case for...

Michael: We assume in our Detailed Clinical Models that things are together... So - Blood Pressure - things that belong together... Might want to query on parts, but get back the whole thing. You could query on the whole thing, and Links... what you do afterwards... So I am saying - all three options are possible, but mean different things.

Stan: I don't think we want to support all. First option - actual structure of option is different. Hematocrit as Entry and as Cluster, and I don't like for it to be either as Entry or Cluster. People will implement their own systems this way but CIMI... Hematocrit as Entry or Hematocrit as Cluster makes me uncomfortable.

Rahil: Want to query Hematocrit independently. But for it to be in context as CBC... So independent entries... That's OK. To be able to query on different levels, and entry(c) on left-hand-side... as CBC... includes both on right-hand-side. Option-3 with 2 separate entries... inherits the context of parent... So Hematocrit or Hemoglobin, if queried independently, perhaps not correct, but... If queried... not independently. If that is the case, then Option-3 is not the way to do it. Looks bad... The CBC is the parent entry within which is constrained... Not to represent as individual entries... is incorrect. If that is the intention, then should be option-1... Want to record Hematocrit results or Hemoglobin... If context is CBC - then should be option-1. If recorded separately - Option-2. Option-3 is bizarre... I don't think should allow independent querying... which option-3 allows.

Stan: When we talk about a panel - there are independent measurements done on the same patient at the same time. There are situations similar to this that are not the case... If this set of results - these two results - are part of a treadmill test, and test Hemoglobin and Hematocrit at different parts of test... If tested, each had Hematocrit and Hemoglobin as components... or if had CBC before surgery and after surgery - knowing that one was before and one after - would be important to know this. When I use panel or... done on same patient at same time. Their use is independent of whether in panel. So option 3 - the things I can query are entries. So I can query on 42 without knowing if Hematocrit or... Not a useful query unless comes back with... "I have a Hematocrit of 42"... Could be a Systolic Blood Pressure or... Need to know the atomic thing. So, Rahil - Use-case... only when completely independent whether part of CBC or... panel...

Rahil: OK - another example. Vital signs with patient - nurse with machine... Blood Pressure, Pulse, Body Temp - these 4 things on machine - on same patient at same time. How model - chuck all in entry as same time... vs.... in cluster with each having own context with... So - back to what Gerard said earlier... Do we say not matter if... or whether we want to introduce some level of context so makes sense. So my question - where is context being enforced so I can ask what is Blood Pressure? But at some level, need to stop those. Whether cluster or... But has to be some level of context. Introduced at some level... Level of enforcement...

Stan: Stat level of enforcement is at entry... Your question - how to model - is the subject of this call. Entries have to be query-able and have to contain all context so is safe... That is what makes it sane and do-able. Option-2 and Option-3 are acceptable to me. Option 3 works - the way we have done it for 20 years... not some philosophical flaw... it works. But what are the trade-offs? So continue - I look forward to you coming up with the right answer.

Stan leaves meeting

Rahil: I want to discuss further. Would be useful to know the underlying reference... If using openEHR then... If simplifying so all are entries or cluster... If simplifying the Reference Model, then...

Gerard: The problem I have with Option 3 - blood count - might get agreement, but there are countries... each has visions of what each is... is limited - we can agree on... Liver test... cannot agree on. Local decision... So can't have as an entry if is ad hoc... Entry is true forever and not dependent on subjective...

Linda: I agree, Gerard. Difficult to get agreement... But as far as blood count being an entry, I prefer to think as... irrespective of whether this lab is 3 tests and other is 2... This is just what tests were performed at that time... So individual tests capture their own context... Tests recorded at that time... or here are Hematocrit recorded...

Option-3 - would require change to Ref Model

David: I don't remember general CIMI strategy, but Option 2 - still be composition which all these entries are part of? So - Left-hand-side - could be panel entry, and then individual lab test... All would be part of composition. So I think option 2 and option 3 are alike. Except option 3... quite a bit of change in Reference Model... Are we really willing to accept this change at this time? I am skeptical. Probably made to work, but would have to define our strategy... Nothing is impossible, but not my first choice.

Linda: Yes - Option 3 would require change to Reference Model...

Gerard: (?)

Daniel: Still - a big change in relation to what we decided not to do long ago... would bring questions about stability of strategy.

Linda: Yes - but we decided if we found problems with...

Rahil: There is no issue with Option 1 or 2... and it is a choice to go with Option-3.... Not that there is a fault with existing style of modeling... Need to be consistent... Whether support general style of modeling done by 13606 and openEHR... CIMI is an international organization... there is no flaw in the model for this change.

Daniel: I see Option 2 and Option 3 very much alike. And I don't see any reason to go with the more problematic Option 3 when Option 2 is basically the same. If could... would have something that is indistinguishable from Option-3.

Linda: Main thing with Option 2 - would need to categorize which are... Links. There are Links that would refer to lab reports that led to... and if distinguish between those different types of Links... they are the same.

Rahil: (?)

Linda: To record Hematocrit... can be performed independently of whether in panel or individual...

Rahil: Yes, but... ad hoc...

Daniel: I think most panels are a list based on the clinical reasoning. So these... had the same use... Liver Test... But actual (?) in each panel could vary... depending on instruments or local conditions.

Linda: I agree.

Daniel: Gerard - what would happen if tests in the same panel and having specific entry for information that is specific to panel as a whole?

Gerard: Well... 4 panels... clear that can use the section, but BMI - cannot. Panel - no meaning on its own...

Daniel: I don't agree - Not completely, but different degrees between BMI and panels.

Gerard: I agree. Section names are meaningless. People give it meaning... Not recordable - so there is an entry that orders a procedure - panel of lab tests defined outside the world we talk about. And when reported - right-hand-side - they are linked with that order. Procedure is executed... has facility to collect... procedure... where I would look for overall result. But all individual items in panel are reordered as individual items as well. So - is the order... context for individual tests.

Linda: Further comments? I suggest we end call and continue next week when Stan can join us.

Comment: Yes.