CIMI MTF Minutes 20120927

From CIMI
Jump to: navigation, search
CIMI Modeling Task Force
Phone Conference
Thursday, September 27, 2012, 20:00-22:00 UTC/GMT

Attending:

Linda Bird

Daniel Karlsson

Galen Mulrooney

Josh Mandel

Stan Huff

Hendry Wijaya

Mike Lincoln

Gerard Freriks

Peter Hendler

Virginia Riehl

Jay Lyle

Dave Carlson

Rahil Qamar Siddiqui

Mark Shafarman

Michael van del Zel

Stephen Chu

Joey Coyle

Thomas Beale

Harold Soldrig

Eithne Keelaghan


Agenda

  • Amsterdam meeting
  • Weekly news & updates
  • FHIR Connect-a-thon inputs (if Grahame is available)
  • MDHT/Eclipse UML inputs
  • Laboratory Results Model Analysis

Weekly News & Updates

  • Taskforce Secretary - Eithne Keelaghan
  • Rahil: Semantic relationship links OWL file
  • Peter Hendler: Semantic Node Labelling
  • Gerard: SemanticHealthNet
  • Michael van der Zel: RM Issues
  • John Gutai: Separating bindings from the model
  • Harold: DNS and IHTSDO namespace application

MDHT/Eclipse UML inputs

  • ADL archetype definitions + terminology bindings
  • Designing UML models for archetypes (UML style guide)
  • Transformation between the 2:
    • ADL -> UML 2.2 (XMI)
    • UML 2.2 (XMI) -> ADL
  • Transformation to the multiple implementation targets - e.g.
    • XML Schema (full/green, canonical/clinical)
    • Schematron / Java Libraries (full conformance testing)

Laboratory Results Model Analysis

Focus: Lab Report Composition, Entries & Clusters; plus Value Sets & Terminology Binding

Initial models to start with:

  • FHIR resource
  • NEHTA/openEHR Lab Results [Stephen to send out link]
  • IMH Lab Reports [Stan to identify the models that are relevant]
  • MOHH Lab Report [Hendry to distribute]
  • HL7 CDA Observation Reporting (constraints on Lab Model) [Mark to send link]
  • VA/DoD Lab Report Model [Mike, Dave and Galen to followup]
  • DCM Lab Results [Michael to followup]
  • Salus Lab Results model [Gerard]

Please refer to presentation for further details.


Detailed Minutes


Virginia: I apologize - error in spreadsheet led me to believe that Linda was not available on other dates. Thanks to those who corrected me. Possible dates - Dec 1, 2, 3, and maybe the 4th. All are available on the 3rd. Question to everyone - what is preference?


...Discussion...


Virginia: Is anyone opposed to Dec 2, 3 and 4th? If we hear from William that not possible then... ...otherwise, Dec 2, 3 & 4th.


Linda: Thanks to Virginia. Let's confirm with William and then send confirmation to all.


Agenda for today:

  • Amsterdam meeting dates
  • Weekly news & updates
  • FHIR - Grahame will try to attend next week
  • MDHT/Eclipse UML inputs
  • Laboratory Results Model Analysis

Anything else?


Weekly News & Update

Task Force Secretary - Eithne Keelaghan


Linda: A few documents were sent out.

(Linda To Peter) - semantic node labeling in next few weeks(?)

Peter: Paper is posted on Ringholm.


(Linda To Gerard) - early in week - discussion on Semantic HealthNet(?)

Gerard: Will be discussing... same topic as Peter's paper.


Linda: Rahil submitted links in ... Can everyone review?


Rahil: I missed (?) relationship between classes in ontology and ...


Linda: I'll give you the screen, Rahil.


.--------------------------------- RAHIL ------------------------------------------


Rahil: [Shows ontology.] Not part of official... artifacts... used to help with modeling to assure link relationships down logically and not break hierarchies. The benefits are... it has been reasoned with so... inferences are correct. It did help us to do link relationships. The CR_ prefix - relationships - it is not local to one set. LRA (?) is used...

The CR properties are the most important for relationships... The relation between two entry classes...

People familiar with how to read ontology:

  • specimen -> defined as subject for specimen...
  • Prefix by LRA - element classes...
  • This says "Material entity-element"... has a relationship with ...
  • There is an inverse to "Has - specimen".
  • We have these relationships defined...
  • One left - have an equivalent...
  • CT-has-specimen.

How do we know which...(?) this relates to? I missed this part but sent out doc explaining this: CT-has-specimen.

[Looks for Document]

To tell you how to link documents together... Material entity constraint - prefixed by SNOMEDCT - belongs to any one of hierarchies in SNOMED.


Linda: Which ones are created in attribute hierarchy and which in ...?


Rahil: Not have attribute hierarchy. We have an expression constrained... Proposing to break up into several attributes...


Linda: So are these link assertions?


Rahil: Yes - since we don't have attributes...


Linda: Implication of this... link assertions have...


Rahil: Not sure of implications on querying...


Peter: Question: Link assertion is universal?


Linda: No - with attributes, all concepts must have relationship, but with attributes....


Josh: Sounds like 2 different sets of rules. Could represent in (OWL?)... What assumptions?


Rahil: This ontology would need to be redone if... attribute relations... Not tied to requirements. I would not recommend using... Could be used on feeder class or reference ontology...


Harold: Stated as they were... But in OWL would... anything in domain of (CRX?)... Try to resolve ontology consistently based on that.


Peter: Will take what you are trying to constrain and make opposite true...


Peter: The reason for our paper... will burn you. You think limiting, constraint... but will not do this.


Rahil: Due to word-choice or ontology...?


Harold: No - important - not words. Declare as subclass - put in domain as ... and will assume ... element(?) But will come back and... Error comes out where you least expect it, so understand - not permissible values...


Peter: Not the words. It is what reasoners will do with it.


Josh: Some do prefer to... constraints in form of OWL. So if talk about reasoning - this is dangerous.


Rahil: We have not... has been used as reference point to assure within definitions... the one place we could refer to... to assure modeling is correct. We stopped at point where we didn't get around to it. This ontology - whether it can be reasoned over...


Stan: I would be helped by seeing where elements in ontology connect to information model.


Linda: One example - relationship between instruction and action.


Rahil: For example, if we are modeling a medication instruction, and activity with an LRA... Reason -> correctly use "has reason" because have med-instruction = activity -> is in domain. So, using this ontology - we can see that we can use this.


Linda: Do you have med order and med-dispensing?

Rahil: Not do... but...


Linda: Generic relationship between order and action?


Rahil: No - but let me show you what we have... will show you medication-activity. We did not have separate models for a plan or procedure contexts. We used the distinction to differentiate between (?) ...and flavor of activity. I think - general preference for "Has indication" rather than "Has reason"...


[Rahil shows diagram flow]

Activity class -> indication -> reason or has a relationship with any of these... could be a procedure... activity... or property investigation.


Rahil: That is how we used this. There is a reason why we did not use "Has-reason"-clause... used "Has-indication" because was in context of a clinical... General preference was to indicate clinical context relationship... why activity occurred. Is that what you were looking for?


Linda: Not really...


Rahil: Not have requirements for order-request-supply model. Status of activity - ...(?) in transition model.


Linda: Going to be important that we define... defined by meaning... to be assured the logic behind links... A bit of work.


Mark: HL7 v3 - has pretty complete classification between...(?) In CIMI.... I can take you through that.


Linda: Great - Please send link.


Peter: Should we have small meeting with Rahil offline - to see if words vs ... other?


Daniel: I think the way these axioms... it is the open world thing - being a problem. With a good dose of ... could work... constrained. An example in the...(?)... to look this up how...


Peter: "Pizza Ontology" - in OWL tutorial. It is exactly this - using OWL as constraints - this is where it will blow-up.


Rahil: I was a little skeptical about sharing - for reference only - to understand links of relationships - only used for this. If it will be used in ... environment - then work will be done. This is not a published document... So... how I was saying... activity could be in progress or planned. Where we have used... I want to say... a model.


Linda: What do people want? Have focused discussion about this - when? Smaller group?


Peter: Small group - need to understand pizza ontology...


Linda: Who: Rahil, Daniel, Peter, Harold, Mark, Mike, Josh, Galen.


Stephen: Open to listeners?


Linda: I will send out meeting details.


Peter: Background - domain range in OWL. I could find 4 pages in OWL tutorial... When ice cream becomes a pizza... I will send to you, Linda.


Linda: Logic statements between attribute statements and universal... Link assertion - "some x have y"...


Pete: Only Universal and existention (extentional?)...


Daniel: It's documented in ...(?) Guide and is not a part of SNOMED CT object property. It is stated that is used by NHS HL-messages (?). Not a format... what link-assertions are?


Linda: Might be useful - to reach out to Kent Spackman...? Anyone talk to him?


Peter: No - hard to catch him.


Daniel: I will email him.


Linda: OK. Thanks Daniel.


Linda: Next on agenda - inputs from modeling task force to ... Eclipse task...? Need to develop ADLs for... and then... group develop Eclipse XMI.


[Phone] - Dave? Begin in parallel. Look at drafts - design models for archetypes. Lab result and... other. I did work with Tom Beal - prototyping of Archetypes... Would like to start work on this. Like to start with design of models... profiles... So-start work on models, then implement... flattening archetypes.


Linda: ADL ->  XMI to know what XMI will look like.


Dave: Need UML Style Guide. Need drafts archetypes. Then start on UML Models. Latter part of exercise... onto ADL -> UML. Need each of elements of end-to-end chain of artifacts ->  then automate.


Linda: Modeling task force with ADL.


Dave: Also, what role to terminology binding going to play. Also - value sets. How to pull...(?) from terminologies?


Michael: What I did in face-to-face meeting... my style of transform... to...(?) I thought this is extension.


Dave: Important for us to have examples of both styles -> what do implemented artifacts look like? Both are useful. Both DCM(?) Style and architectural style of UML... Both approaches...


Michael: I am happy to help.


Stephen: If talk about DCM Style ... (?) not Universal...


Linda: Goal is to move to... style of UML...


Linda: MDHT/Eclipse UML inputs ADL archetype definition and terminology binding. Ultimate...round trip.

XMI -> ADL...

ADL ->  UML

UML ->  ADL


Mark: UML 2.2 or 2.0?


Dave: UML 2.2. Most tools support 2.2


Linda: Any other discussion on this?


[Phone]: Produce archetypes... I would use archetypes workbench and work on UML representation in parallel.


Stan: This gives us 1 step closer... I would assume from UML archetype need to get to what VA/DOD will implement in system. So from here produce VA/DOD archetype. Is this JAVA or ...?


[Phone]-Dave?: I don't know. XML Schema is probably the first. Once get to JAVA - Serialization... I don't know what would be most helpful to FHIR...


Linda: So if ... workbench is producing XML ...


Dave: Multiple outputs. Goal - I hope - general language. Questions as to Style... CDA - generic-use. Names... algorithms for flattening... left to discretion of implementation. Tooling... If we can document, should be able to produce equivalent outputs. Might be different set of rules. Still interest in MEME... federal agencies want this... want to produce from same archetype model.


Linda: In Singapore, XML is based on Logical Model, but it assumes that the receiving system knows the clinical model - so can strip out all information from the XML that is in the model.


Linda: Generic round trip between canonical representation of XML schema and clinically-relevant names in tags. So can be populated using clinically-relevant names... and round trip to canonical.


Dave: Important to pick goals. Canonical and clinical - relevant names to show both possible from same model...


Michael: XLM Schema... limited... Also need Schematron.


Dave: MDHT pilot generates JAVA libraries from the models, which can be used in conformance-testing for validation


Linda: Thanks. That gives is a fantastic goal to work towards.

Next week - FHIR.


Linda: Next - lab results model analysis. Thanks Stephen Chu for lab results model, Hendry for ... lab results model. The rest... please submit so we can discuss. If we can look at Intermountain on web, if Stat tells us... Mark - please send link to... Mark, ..., Galen -> Can you send in lab report model this week?


Comment: Probably need another week.


Michael: Most... are in Dutch. I can send as is or can translate...


Linda: Would be useful to have main bits translated.


Gerard: I will send...


Linda: The ones we have so far: FHIR, NEHTA, MOHHoldings, Intermountain... Looking at the 3 so far submitted, different in granularity of models... FHIR model - is a draft, up on a server. Lab Results Report - Comparison to FHIR. Links to patient-resource, admission-resource, laboratory, report_id, name, specimen, conclusion. Lab report - equivalent to composition - not directly contain data elements. Has 0-to-many request-detail, 0-to-many result-group, 0-to-many Result, 0-to-many Reference Range. If we look at how would be translated to semi-reference model...

1) Treat Lab Report at composition

Hierarchy:

Lab Reports:

Details ->

Request-details ->

Result-group ->


Linda: Composition -> ... or entries, and entries contain -> clusters...


Gerard: ...? group in my group will be a section because each lab result = entry


Linda: Differrent approaches. Would be interesting to compare. While these are iso-semantic - need to come up with preferred model. I have the NEHTA model up on CKM...


Stephen: Is available on DCM. NEHTA... /CKM web site. But also more direct link... I'll send between now and next week. HL7, FHIR Lab Results - based on NEHTA architecture. Look at ... hierarchy - it is grouped in similar way except lab-results.

Why is it in protocol? I would prefer details to be moved back to data hierarchy. Apart from that - if go to data hierarchy tab and scroll down, it is pretty much... so you find that this part is... in FHIR... in group - similar to FHIR.result.class and have result details. And Name->... and result-value... and comments about result and result-status. Result-detail as a slot - different from FHIR... unique to Australia. Pathologies and lab people - the way it is rendered to clinician - a matter of safety. Result-Details is about specimen... Test-result - representation. The way result-details are rendered is of clinical importance in Australia.

"Test result representation" - so there is no mistake as to how it should be interpreted. This is all I think is different from FHIR.


Linda: Test Request details...


Stephen: Are in protocol, but I don't like this... Another issue I need to discuss internally, we don't have a way to indicate it is a refreshed test -> not the original. Need to tidy up internally. We also have a test request.


Linda: This needs to be wrapped in a composition. Is there a standard composition?


Stephen: Yes - we have an old version. Need to produce a new one. Old one is deprecated due to some issues.


Linda: Would be a composition wrapper?


Stephen: Yes.


Linda: Another comment - Lab Report - is it intended to be one observation per report? If you look at data hierarchy, results can be grouped... hematology, renal tests... in self-grouping.


Stephen: Allows for individual... to be an individual result.


Linda: I meant other way: "be reported back as 1 component" - but I noticed... multiple result groups. Does that mean observations... or panels? So all based on single test result?


Stephen: So test result - can be taken as... full blood count. But within group - can be... multiple.


Stan: So - assumption is that all results were done using same protocol?


Stephen : Yes - from what I remember - this is because of test methods... protocol contained here... In our composition representation the test method is related to individual tests.


Stan: Is there a way in CKM to see as linear...?


Stephen: Unfortunately - No, but at composition level we have re-organized... protocol is...


Linda: Tag-View and Single-View. Append onto itself.


Stan: I was thinking more of a View you get in archetype workbench - can open/close nodes. Can walk down a tree...


Stephen: Unfortunately... can't... but I will try to come back with... representation..


Linda: I looked at FHIR model and relationship. NEHTA node - lab report composition document may or may not include... and many pathology tests - each observation... so this represents... more hierarchical - NEHTA - shows protocol with clusters. Stephen - does this represent...?


Stephen: Yes - admission dates - included. If important... e.g. In test for Clostridium - important to have admission date/time. But not necessarily included in each one.


Linda: Moving on the Singapore model... Lab Report Composition entry for header info... patient info - in which lab request: Lab order, Lab result, Test Result. I'll show you the spreadsheet model.


Linda: So we have an XML... version of this, but the spreadsheet is easier. With the composition: entry_pt_event_context... section for... groups together investigation order and investigation

[STATIC ON THE LINE]


Linda: (cont'd) Within investigation composition - crosses lab and radiology templates - narrows down to specific lab or radiology. Generic investigation -> order, results, pt event.

In patient event - healthcare

  • Start date/time - > into admission
  • Pt-location
  • Confirmation-entry
  • Document -> doc id and version

Investigation order

  • Clinician
  • Facility

Investigation item

  • Test name
  • Specimen details -> can be pre-coordinated into test-name or ... into structure. Can pre-coordinate or...

I noticed this mentioned in NEHTA as well. Need to look at for iso-semantic model.


Status

  • Investigation result -> reporting facility..
  • Investigation dates
    • Go across lab and...


Status

  • Within Investigation report
  • 0 to many...

Test Result -> Cluster

  • Describe individual tests
  • Result has a value

Reference Range Set

  • Can have a value or a Summary -> if not structure
  • If is structured -> a set of...

Test Result Values -> Hierarchy

Investigation result statuses...


To summarize:

  • Lab Report Composition
    • Doc info-entry
    • Pt Event-entry
    • Lab Report Item (section)
      • Lab Order (Entry)
      • Lab Result (entry)
        • Test result (cluster)
        • Reference Range Set (cluster)


Need to decide where to put Request info and result info. Order-info -> separated from...(?) Need links between entries...

Thoughts on whether this structure is appropriate?


Rahil: I think if we are using these relationships... it would be good... especially if query. One question on references range set - would it not be common to all test results?


Linda: No - might have test results for Iron levels - in a panel and set - sample 0-1 means very low, if 1-2 low, if 2-3... so each range is part of test.


Rahil: OK


Galen: Reference range only makes sense for (?)... ex) Staphylococcus in blood - never good. But others: results with reference range. Might be different dependent on age, race, sex...


Stan: Yes - you will see reference range in our models, too, and is different from 2 examples. Ours is a single reference range. 120-160 is normal. So our assumption is reference range in lab, but use reference range for this person at time occurred. So I only put 1 reference range for patient.


Galen: So - that is what is confusing. I need to understand all reference ranges. But if I am reporting results -> I only need results and reference value for this patient. If goal of CIMI is to define all possible reference ranges or only to report for this patient?


Linda: I don't think we have decided. Sometimes reference range is provided and sometimes not.


Stan: Could be both of those things, Galen. Want interoperability. Use Case - transferring data. We could also... trying to exchange reference range between models? Labs? So the first case - patient data focus. But we can do what people find useful. So first - around aggregation of patient data -> know how to ask for date regardless if from Singapore or VA or Intermountain...


Stephen: I agree...


Stan: Not applied to all? Big table - age-specific, gender-specific... All variables are taken into consideration, so instance does not have to have all possible reference ranges - only that which pertains to patient.


Mark: Question - often an abnormal flag in lab result - like a judgment -> generated by another set of tables. Question: Have 1-to-1 between order and results. What if have multiple results. Not clear if I repeat structure or...


Galen: Micro - multiple results from same order. Gram stain, then Culture, then Sensitivity... so preliminary report, then another...


Linda: Depends on how granular. Mark -> we also have abnormal indicator. Has, for example, critically low... We did make decision where... Your second point - 1-to-many - each lab order can have many results.


Mark: Yes.


--- Look at Intermountain ---


Stan: So - what you see - we've made 3 versions. Standard labs - Core - secondary use... Secondary use and standard lab without qualifier/qualification(?) are (?)... of core standard lab. Secondary use = constrained subset of these models. Used in Sharp grant activity. Needed for EHR, but not for Sharp grant Collaboration. So Top = Core standard Lab Observation. [Click on...] See parent type of any lab results whether is numeric or coded. Next = core standard lab observation coded - this is still a pattern - data is type CD - set of values are from large domain of lab values. So - we go a step further - we make hematocrit, hemoglobin and individual lab results. Children... urine color = key code indicates urine color and lab set to allowable codes.


Galen: This is what I was getting at - I [asked?] about EHR model and what CIMI is trying to do. "Here is all we want to know about urine color separate from patient-observation." So if... pt_results -> Urine is brown - then could feed into Clinical Decision Support to inference on this...


Stan: Yes - comes back to Peter's comment... We're trying to capture the information that allows someone to use it... for whatever reason... might be clinical decision.


Galen: So if wanted to build widget for... more complicated than just 1 widget description. So in some respect, easier for self-contained widget, and sometimes is harder.


Stan: Exactly - now we can talk about "Style Issues" -> not one right away, but we will choose for CIMI.


Peter: Let's say Kaiser has RIM... DB with depository for all data for all patients. If we were to query DB for something... Out of huge DB, what I want is ...this template.


Galen: I assume DB of HER.


Peter: But we have... SNOMED difference and Units difference... but I have this -> I only want those things...


Stan: That is exactly right. Difference is how you Query model. If Query "Get me serum sodium where value > 135:. Another way is "Get me standard lab, SNOMED - Hematocrit and data>135". So that is a question. You can put knowledge... There is more knowledge than measure and key code. If quantitative, have delta-flag and... and in numeric... also more knowledge about what max is. So can put into representation... They are iso-semantic. We found this to be useful - especially when we tried to... to have additional knowledge.


Galen: One criticism to open-EHR archetypes. Could be modeled differently... CIMI could apply series of patterns whether urine (?) or... All follow this pattern, so individual knowledge bits will have additional bits filled out... will follow same pattern...


Stephen: That is why I asked at meeting... My view is we should be defining patterns for people to benchmark...


Galen: ... patterns ... top... and... Once we figure those out, we are golden.


Stan: We can talk about this another time.


Linda: Next week, we'll continue with Intermountain model. Also, Galen... Also, Gerard and Rahil... And I'd encourage all to submit models... and if all could familiarize yourselves with models...