See also: IRC log
grahame: Implemented, but I'll turn it on next week.
grahame: ballot status discussion. planning to ballot the next round of FHIR for normative content. XML and JSON formats are part of what we are taking normative. We do not propose to take the RDF to normative.
AGREED: RDF format for FHIR is not ready for normative release
grahame: We published the JSON-LD
format, asked at connectathon, and did not find implementers
that were aware of it. They wondered why it existed. FHIR Core
asked me to convey that existence of JSON-LD format is a
negative because of potential for confusion, so we're asking
the committee to remove it.
... Paul thought that it's too early to decide. David thought
it would not be effective until release 4. Under what criteria
should we keep it?
harold: I can see non-value in it. If the JSON-LD exists we need to keep it synchronized with the Turtle, which seems counter productive.
grahame: the code for both is similar.
eric: Bigger danger is that you
end up being . . . .
... Value that could come of JSON-LD would come if you were
using it sometimes as JSON and sometimes as RDF. But that's a
long shot. Also, JSON-LD 1.1 is likely to address content model
issues that prevented us from simply putting a @context on FHIR
JSON to make JSON-LD.
michael: We did round-trip test to test the sync.
grahame: One other issue that Josh asked: what is valid JSON-LD if it looks like that JSON and has the correct @context , or has the same triples?
dbooth: The answer has to be "same triples", otherwise we're going in a very dangerous direction.
eric: I think the value of JSON-LD is that it can be processed as JSON.
grahame: If the JSON matters, then you might as well use FHIR JSON.
dbooth: There is purely marketing
value in JSON-LD --- that it is familiiar.
... I agree with dropping it. Also would be better to do
conversion to JSON-LD using a standard RDF tool such as Jena
instead of building it into a FHIR referenceimplementation,
<inserted> AGREED: Drop FHIR JSON-LD format
michael: .NET API is now updated to STU3. Now working on adding the RDF to STU3. Also tested running under Mono. Created a console app to do FHIR format conversion to/from RDF.
grahame: Planning to add to standard .NET FHIR API?
michael: Not sure. Ewout is not sure of including the dependency on the RDF library.
christiaan: dependencies are easier now in .NET
michael: Plan is to release it as
a separate release, but will discuss with Ewout.
... But there are some RDF changes that I have not yet
made.
dbooth: ETA?
michael: Before September.
harold: No more progress yet.
grahame: As product director, I
set the overall agenda for FHIR -- big picture changes. I've
published things I want the product to improve. One item: get
better ontology bindings at the base of FHIR.
... Originally I saw FHIR as a set of resources and supporting
plank: conformance, REST API, ont bindings, implementation
support around libraries/packaging.
... We've nailed all those except the ont part.
... Want to bind all of our definitions to underlying
ont.
... Added RDF format. Two kinds of lower ont: methodology ont,
like Dublin Core, for mapping definitions.
... The other is content -- clinical ont, to enable reasoning.
Because we have not done any mappings to those, there is only
so much we can achieve with the RDF.
... Want the community to pick an example ont and produce a
good set of bindings as an example of what we want to do.
... Want to do a useful ont mapping, both methodological and
clinical. Want them expressed in RDF.
harold: My demo is a start. It's
taking fhir.ttl and classifying them using SNOMED, with
interesting examples. That's square 0.
... Need to open to more things than SNOMED. Need to publish
all the built-in FHIR valuesets in OWL so that instead of
dealing with data properties we can deal with object
properties.
... E.g., DiagnosticReport has several data properties that
could count as Final, but aren't currently as data
properties.
... Also the work with Linda Bird on FHIR templates is related,
but has a drawback.
... The drawback is that the SNOMED templates can map to a
limited set of OWL, but not into the disease onto or gene ont
or any non-SNOMED resource that is OWL-able.
... The other thing that needs discussion: There's a huge
difference between fhir.ttl, which deals with FHIR clinical
records, and what the clinical records are about.
grahame: I was hoping we would
target something other than SNOMED, because it isn't what they
think of when they think of ont.
... What you identify as the limit to fhir.ttl is exactly what
I want to change -- want to have those mappings that allow you
to cross those bridges.
harold: We've been discussing where to put those bridges.
eric: Harold's SNOMED demo shows
a really big win.
... If we wanted to capture W5 in a FHIR info model ont, that
would enable a few use cases around processing, saying that
there is a generic handler for all orders.
... If you want to bind to disease ont or something like that,
SNOMED will have by far the most description of properties that
appear in SNOMED resources, such as body site, laterality,
etc.
... FMA might have some of those. We could bind them
individually to FHIR, but we'd be saying at the same time that
they corresponded between FMA and SNOMED.
grahame: Work is happening around
SNOMED, and it has a habit of diappearing down a rabbit hole of
complexity.
... Perhaps we want to try to represent what is being said in
the templates as triples.
harold: NOt sure that a set of
triples is the right idiom, but we need to be able to state it
in OWL. We may have a good use case at Mayo.
... Davide Sottaro has a good use case.
... Keith Campbell gripe about SNOMED is that they do their own
stuff instead of using what exists. I'd like to encourage
convergence.
... I'd like to fire up a parallel project for this.
... Only one example in FHIR core of FHIR compositional
grammar, and we converted it to OWL. I'd like to have an
automatic conversion to turtle.
... Does the code system necessarily determine the
compositional grammar? Or do we need to include a marker in
it?
grahame: The pattern we have is
that we have a coding (system + code) and a URL representation,
which links directly into the coding system.
... Where's that example?
ken: Would SOLR work be of interest?
grahame: Harold is saying that SNOMED is sufficient.
harold: No, but it's a
start.
... As soon as you get into genetics and all, SNOMED isn't
used.
grahame: That's why I was hoping to jump beyond SNOMED.
harold: At the moment we're only generating URIs for LOINC and SNOMED.
grahame: I looked in Barry Smith's work, but could not find matching concepts. It's so different.
<ericP> 71341001:272741003=7771000
harold: Olivier Bodenreider has
done a good write-up on the difference between SNOMED and the
disease ont.
... Exmaple of observation: http://build.fhir.org/observation-example-bmd.ttl.html
... Here is the use of observation:
https://github.com/BD2KOnFHIR/BLENDINGFHIRandRDF/blob/master/observation-example-bmd.ttl
grahame: I've screwed up the representation of the SNOMED expr.
harold: Michael Lawley proposed
URI expression for compositional grammar --
microparsable.
... To process that in SNOMED, it needs to be in the form I
posted in the URI above.
... Here is the relevant fragment:
[[
fhir: Observation.bodySite [ fhir:CodeableConcept.coding [ fhir:index 0; a sct:71341001, rdfs:subClassOf [a owl:Restriction; owl:onProperty sct:272741002; owl:someValuesFrom sct:7771000]; fhir:Coding.system [ fhir:value "http://snomed.info/sct" ]; fhir:Coding.code [ fhir:value "71341001:272741003=7771000" ] ]; fhir:CodeableConcept.text [ fhir:value "Left Femur" ] ] .
]]
grahame: SNOMED expression is discussed in zulip.
harold: Laterality is one of the
few things in SNOMED that is sensible, but most things get
stuck in the role-group world, which really adds
problems.
... I'll talk with Quogian to see what ideas he has. Finishing
getting the FHIR internal valuesets into OWL will help
also.
... Working on non-SMOMED templates is interesting also.
grahame: Vocab group agreed to move everything into FHIR valuesets, including v2 v3 CDA.
harold: For folks to see, was
going to generate a separate codesystem.owl to put them into a
reasoner.
... Also considering an OWL format for the terminology
services.
grahame: Let's start with the first step.
harold: Need to set up content negotiation also, to avoid .ttl suffix on the ont name.
grahame: Content is hosted on a server that only does conneg based on file extensions.
harold: If we had no extension, can it redirect?
grahame: Yes, with 302. I already
do that depending on the content requested. I can extend that
for turtle.
... Need the canonical URL to get it right the first time.
harold: We want to drop the .ttl extension.
<scribe> ACTION: Eric will work out canonical URIs and content types with Grahame, to redirect to Turtle. [recorded in http://www.w3.org/2017/05/09-hcls-minutes.html#action01]
<trackbot> Created ACTION-85 - Will work out canonical uris and content types with grahame, to redirect to turtle. [on Eric Prud'hommeaux - due 2017-05-16].
robert: Have mapped to OpenEHR,
but not to RDF.
... Wanted to show you mappings from OpenEHR to FHIR.
... FML is "push" style.
... FTE maps both source and target to a common class model,
and then you can get bidirectional transforms.
... That's more 'pull' in character.
... FTE mappings are more fine-grained than FML.
... But they are interconvertable between FML and FTE
mappings.
... Now want to show you how a FHIR transform engine has been
applied to convert OpenEHR to FHIR.
... You have a source structure, and you get down to substance,
for example, in allergyIntolerance.
... I mapped it to a FHIR bundle.
... You get the source and target models by reading in files,
then manually map between them.
... It's a visual mapping editor that allows you to pick
equivalent nodes.
... Once you have made mappings, you can test them
easily.
... And you can test round-tripping.
... Synergy between FML and FTE? FML to FTE is easy; FTE to FML
is harder, but I have a demo.
... The ability to convert between them allows for lots of
possibilities in tools.
grahame: FML is declarative presentation of a process. You could represent that process as RDF, but that's very different from representing the relationships between them.
eric: want to understand the differences between what can be done with ISO 11179, RDF, Concept Maps, etc.
<inserted> christiaan: Licensing? robert: Free for open source efforts, but paid for commercial efforts.
ADJOURNED
This is scribe.perl Revision: 1.152 of Date: 2017/02/06 11:04:15 Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00) Succeeded: i/Topic: Status of C/AGREED: Drop FHIR JSON-LD format Succeeded: i/ADJOURNED/christian: Licensing? robert: Free for open source efforts, but paid for commercial efforts. Succeeded: s/Knapp/Knaap/g Succeeded: s/Christian/Christaan/g Succeeded: s/christian/christiaan/g Present: David_Booth Michael_van_der_Zel Christaan_Knaap Grahame_Grieve Robert_Worden EricP Harold_Solbrig No ScribeNick specified. Guessing ScribeNick: dbooth Inferring Scribes: dbooth Found Date: 09 May 2017 Guessing minutes URL: http://www.w3.org/2017/05/09-hcls-minutes.html People with action items: canonical content eric grahame out types uris will with work[End of scribe.perl diagnostic output]