Meeting minutes
Ken Lord's interest in an ontology for semantics of mapping
ken: My company provides healthcare data interop. As a basic principle, to have accurate transformations btwn data in different data models, it requires semantic equivalence.
… So we do bi-directional mappings between FHIR, X12, CCDA, and proprietary structures like CSV.
… Want to have an OWL ont that supports RDF implementations, and this can be linked to data models such as FHIR.
… This group has done a lot of good work on syntactic side. Interest or plans for semantics and classes in OWL?
ericp: I played a while ago connecting SNOMED to FHIR RDF and doing classification inference. It worked well.
… But it was hard not to overwhelm the machine, because of the size of SNOMED.
… Had to use bits of SNOMED, which means I kinda had to know the answer.
… Another useful thing would be process mgmt, such as substance request and procecure request, and make sure they don't get too old.
… Want use cases that would drive the OWL ont.
… Most would mix terminilogy w info model.
ken: THis would provide a foundation for someone doing the inference.
… Would be valuable to my work. Hope that having an ont that could span muliple data models would have value.
tim: There are multiple use cases OWL could help for.
… But your primary problem is translation or mapping data?
ken: yes. We segregate translations (at valueset level), but at the concept level it's at the transmission of one concept to another in different data models.
tim: For mapping valuesets, FHIR has concept map.
ken: We use that. But bindings of concepts to data. Need to transform the data from one concept in one data model to another concept in another data model.
… Very useful, but there are ambiguities involved. Also going across data models, having an ont at a concept level, w equiv concepts, gives you confidence in the transformation.
gaurav: I wonder if you could use LinkML to model your data, map permissible values to concepts, and use that to transform your data into RDF: https://
… It also has a mapping format: https://
ken: In 2008 OMG developed a model-driven message interop std.
… It was a generic approach for modeling any data model. We've adapted that for healthcare industry.
… Very confident in our operational approach. Want to add an upper ont to give people confidence.
… OMG std was MDMI
EricJahn: Re LinkML, W3C CSV for the Web could also be used to link CSV columns to RDF contexts
… We use that. Simple way to go.
dbooth: SSSOM: mapping-commons/
ken: Lots of implementation approaches can be used, but an ont could help
… We used BFO
tim: I would suggest we generate equivalence/subsumption mappings from FHIR ConceptMaps
… I mapped BFO to concept maps using prov ont.
… Planning to do that w FHIR. We have a FHIR ont.
ken: I've looked at fhir.ttl, but when it gets to semantics, e.g., BackBone class leads you to nowhere semantically.
… If there's interest in expanding that, I'd be interested.
tim: We could potentially map it to BFO. But why?
ken: For data quality. There are no computable ways to provide quality assurance around transformations.
tim: We can validate data, according to the BackBone elements. Also there are axioms for the BackBone element classes.
ericp: You want a checksum. A human said there's a mapping from x to y, and you want to verify that there is.
tim: If you're mapping X12 to FHIR, and there's something in FHIR that you've mapped to a BFO object, and you've mapped x12 to a process, and those don't link up BFO, but you would already have to have mapped x12 and FHIR to BFO.
ericp: Another use case is to drive the dev of the mapping.
… having a human in the loop, the computer could help guide the search space.
ken: But humans introduce errors, though you require them to sign off.
gaurav: I would suggest we generate equivalence/subsumption mappings from FHIR ConceptMaps
IRI stems
gaurav: Want David and other review on it
R6
ACTION: tim to do step 0: Run our R6 kindling-turtle generator and R6 kindling-shex generator using R5 structure defs and R5 XML examples to produce R5+6 turtle example and R5+6 shex shapes.
dbooth: How do you run kindling to gen a single example?
tim: It's at the bottom of our repo: w3c/
… I point it to a directory. Whatever's in there, it will try to create a Turtle version of it.
ericp: You'll want to work on Master from the FHIRcat repo.
tim: What's R6 progress? Chance for more changes?
rob: Nov 1 deadline shifted. Grahame tried to close it out in the next day or two. Locked now for Jan ballot.
… But there will definitely be another ballot.
… Two more opportunities: If there's something missing it can be a ballot comment and it can be fixed in ballot reconciliation process. But also another round of balloting.
tim: This is fixing something that should be there.
dbooth: It's a deficiency in shex, not a spec change, right?
tim: yes
tim: I figured out how to do shex for CodeableConcepts
tim: I can implement issue 198 after Jan 1
tim: w3c/
… And it works for any descendent of the SNOMED concept
ADJOURNED