Using OWL to connect ANF (SNOMED, LOINC, etc.) and FHIR/RDF - Ken Lord

Cory: working with Ken on MDMI, and OMG

Daniel Stone: Working at Mayo with Dr Jiang

Dahzi: Working at Johns Hopkins on FHIRCat project and terminology ser4vices

EricP: Works W3C, lots of RDF geekery with Sem web and healthcare.

harold: Prof at Johns Hopkins, working on FHIR

Ken Lord: Working for MDIS, project lead for MDMI project (model driven message interop), and OMG standard, working on enhancement with Cory Davide Sottoro, Elisa Kendall, Robert Lario

Sajad Hussein: Ontario hospitals, RDF, V2 to FHIR use cases, and FHIR to I2B2.

Slides from Ken Lord: https://lists.w3.org/Archives/Public/www-archive/2020Jan/att-0000/Yosemite_OMG_MDMI_2.pdf

ken: RDF and OWL experts are other members of the team -- Cory and Elisa.

(slide 3)

ken: It's all about transformation from source message to target message
... Not a research project.
... Partners Research was transforming CCDA data to I2B2.

(slide 4)

ken: Starts w an interop model - metamodel. Deconstruct a file into its semantic elements. Then uses syntax components to reconstruct the target message.
... The semantic component describes the semantics in that specific format. Structure is captured in the model, such as containership.
... The mapping is an isosemantic relationship between a semantic element in the model for the format, with a central respository, that contains the equivalent semantic concepts. Relationship betweeen a data element in a format with adictionary in this repisotory. About 4000 right now.
... A business concept has a meaning: how that thing has been placed into a context. Like a triple.

Q: Central repository?

(slide 5)

ken: input message A is transformed using Map A and Map B to message B.
... Maps are created independently.

(slide 6)

scribe: SEER is a repository, but really a dictionary of concepts at the same granularity level. No structure. The only structure is that there is as flat as possible a datatype. Simplest one is patient ID or patient name.
... If we get into clinical data, medication administered code, which indicates what was administered. Or Med administered dosage, which is different. Small atomic concepts, unambiguous.
... The SEER does not contain synonyms, though it can have multiple names.
... We've been doing this for 8-9 years, and applying it to healthcare for 6 years.
... The problem: the meaning is described by unstructured text, and this makes them ambiguous.
... Our group is trying to extend the MDMI standard to provide a formalism for each MDMI business element.
... We don't want to reinvent anything. A principle of interop is you can't invent it, you have to work with what exists.
... Want to leverage existing healthcare ontologies and terminologies.

(slide 7)

ken: Selected an OWL tech stack. Had issues with 11179, with linkage. Want to link to existing semantic foundations.
... Want to link to reference models. Primary model we're working with: ANF, which just went to ballot.
... ANF is an information model, but it's built on a semantic foundation of the SOLR concept, to harmonize the underlying terminologies and ontologies.

harold: Use of SOLR is interesting because of the SNOMED component. Also want to undertand what the 11179 issues were.

david: Process for coming up with these formalisms?

ken: OMG issues an RFP, out of the healthcare domain task force. That was for an enhancement of MDMI 2.0.
... OMG creates open standards, and various proposal teams come together to work on it. Proposal team has submitted a proposal, and we've submitted a proposal. Trying to make it as open as possible.
... once the standard has been approved, it becomes open. But if people want to become a part of the team, they are welcome.

david: Do you have to come up with these formalisms?

ken: yes

(slide 8)

ken: We've currently scoped it initially to deal with clinical info exchange. This is informative, not normative.

cory: To clarify formalisms: they are being defined in the std, and they define how these business elements in the MDMI model reference ontologies. Not standardizing those ontologies.. Not making any assertions about how to model healthcare.
... We're doing the modeling in a profile of UML, and then generate OWL from that.
... OWL that we're using is quite restricted -- basically RDFS plus property type restrictions and property chains. Chains are important because in flattening the concepts we're typically taking a path through the model and making it single defined element.
... Also important to restrict properties in a context.
... Mostly subclasses and subproperties other than that.

eric: "redefines" = onproperty all values from?

cory: Yes.
... Not intending to do inference.
... Layers are color coded. Orange things are mid-upper level scaffolding to put in place basic categories.
... one important one: separation of concerns between speech act of a message from what the message is describing.
... As part of that we get that the statement is abouit something and things have participants.
... Blue layer is a stand-in for where a ref ont would go.
... These are examples to explain the bottom level, which are business elements (concepts) of MDMI
... Want to determine that they are unique.
... So the only logic is to look at how they're derived from more general concepts, using those restrictions and property chains.
... That green wire is what would be specific to MDMI. Want to determine the uniqueness of that concept.
... That's the only purpose of that ref ont.

eric: How does that work in practice? Example: When someone quits smoking, that's written in the EMR and it gets exchanged. Would that be a special item in the ref model?

cory: Something like "patient quit smoking <date>"

ken: it probably would have an end date, associated with the act of smoking.
... Even though it might be represented very differently in the info model.
... Very fine grained, but want to keep it at the level of information exchange.
... Green is what we want. Orange is what we need to get it through OMG and be robust. Blue is where we'd like to make sure there's an understanding of what those things mean, to do mappings between them.

harold: Curious about Observation here. Is this guided by the SNOMED work on Observables?

ken: I think so, but it's driven by Keith and Davide. We want their input.

cory: Challenge is finding abstractions that work well across other ways it has been done.
... This entire model is non-normative. Upper structure might move forward in some other effort. Want to understand generically whjat works acrross multiple approaches.

david: FHIR has 5-Ws ontology

harold: Yes, but it's a model of data, not the the information. This is a general model of how one gets a data model with what it is about.

cory: Lot of work I do is in finance, with FIBO model. Generic parts of this should be able to work with any ref model.
... It can be hard with a broken model, but with a reasonably vconstructed model it works.

harold: Many models blur the model of the data from what it is about. SNOMED for a long time did not distinguish between a def of appendicitis vs a finding of appendicitis, so could not ask how many documented appendicitis were diagnosed.
... Half the purpose of the doc is human doc.

cory: Distinguishing situation from definition. Also differentiation between roles of person as patient or doctor, so a doctor can operate on himself/herself.
... Trying to get a start in a non-normative way.

Next Steps

david: what would you like to see happen?

ken: Want to provide this group with what we're doing, on a regular schedule. Get comments. Especially need to be certain that what we describe is understood by a machine in a level of precision that people need.

harold: FHIR/RDF that we're doing is to define a third exchange format besides XML and JSON. FOcus now is to use JSON-LD 1.1 to combine JSON with RDF.
... FHIR now has 7 metamodels.
... Also want to seamlessly connect things into FHIR space. Out of scope: for a while we called the definition the FHIR ontology.
... We're modeling documents about patients. FHIR is beginning to add semantics. 5 Ws ont is the beginning of that.

eric: First step is to recognize the same entities between different resources. How that's documented is almost orthogonal.

harold: They're beginning to put links to concept codes in FHIR, though I'm not sure they're getting the right ones.
... Such as, about appendicitis vs a finding of appendicitis.

ken: How can we help that effort?

harold: First step is to look closer at the model to see what semantics are being added to FHIR. Also want to see what's going on w semantic elements repository.

ken: We've already mapped to FHIR. We have a FHIR map.
... Every attribute in a FHIR resource, to the business elements. We can exchange CCDA docs with FHIR.

harold: That sounds exciting, because we have official URIs for everything in the model. Might be really useful to put those together.

cory: Are you looking it FHIR as being a semantic ref, or would it ref something else?

harold: FHIR is a data model. Semantics are largely tacit. History is from people figuring out the enterprise or info model they needed.
... FHIR describes a pt record, but formal def is a block of text. We have no idea if we get a record from a different space, whether it refs the same thing or not. Can a pt be an animal, for example?
... The only way to find out is to read the text and hope that someone wrote it down.

ADJOURNED

This is scribe.perl Revision: 1.154  of Date: 2018/09/25 16:35:56  
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00)

Present: David_Booth EricP Harold_Solbrig Dazhi Daniel_Stone Cory Ken_Lord Sajad_Hussein
No ScribeNick specified.  Guessing ScribeNick: dbooth
Inferring Scribes: dbooth

WARNING: No date found!  Assuming today.  (Hint: Specify
the W3C IRC log URL, and the date will be determined from that.)
Or specify the date like this:
<dbooth> Date: 12 Sep 2002

People with action items: 

WARNING: IRC log location not specified!  (You can ignore this 
warning if you do not want the generated minutes to contain 
a link to the original IRC log.)