harold: Revisiting
fhir.schema.org idea
... Is fhir.schema.org conflicting with medical schemas? I
think not. fhirschema is about medical records.
Harold
Harold's slides: https://docs.google.com/presentation/d/1_yqpKWBhT8nTcPmRVm4nLtPIzdQ_nj8CpO9e8jIlNb4/edit#slide=id.p
harold: medical institutions such
as Mayo Clinic annotate their web pages with medical info, and
google uses it. They use micro-schema-like data for google to
produce info boxes.
... That info is harvested both from Mayo clinic and NLM.
... This brings us to an interesting alignment question.
Grahame has started to mark up FHIR with semantics. Some work
pointing at SNOMED.
... Health and medical types potential are an alternative
markup language for putting semantics into the FHIR
language.
... Another one is bioschemas.org , which seems to be the same
domain as FHIR. Not so much semantics, but what is in the
medical record.
... FHIR schemas abut each other in schema.org space just as
they abut each other in real life.
... interesting to look at a page:
https://hangouts.google.com/_/elUi/chat-redirect?dest=http%3A%2F%2Ffhir.fhirschema.org%2F
here is the real URL: http://fhir.fhirschema.org/
harold: these are all the FHIR
resources defined.
... Here is a URL for Observation.BodySite . There's a place
for the Observation.code value, etc. Everything you can
represent in FHIR, it gives you a set of URLs for it.
... skip to slide 6
<ericP> FHIRschema.org Obervation http://fhirschema.org/Observation
harold: FDA has structured
product labels
https://docs.google.com/presentation/d/1_yqpKWBhT8nTcPmRVm4nLtPIzdQ_nj8CpO9e8jIlNb4/edit#slide=id.g5f38253709_0_0
... We grabbed the HTML for their labels, and we're showing a
use of fhirschema.org, and it has info about medical
knowledge.
... WE put in the medical knowledge properties using a google
testing tool for schema.org, and we show that you can pull the
FHIR info, if you mark it up, from this medical knowledge
database.
... Somebody is working on a direct transformation from the
structured product label to the FHIR knowledge.
... Another use case on the slide 7 is a lipid panel. Not as
pretty as on the Mayo web page. I can view it online. Today
underneath is only HTML. But on the next slide we can include
FHIR markup in the HTML.
harold: Mayo Clinic probably has this info directly in FHIR, but here we have a human-readable resource that is simultaneously FHIR data.
mark: Last I knew people in Mayo
Clinic were ... They attached to each, a code for exchange with
other systems.
... Are you thinking of adding such things?
harold: If Mayo puts in test codes they will put LOINC codes in.
mark: Most countries are mapping to SNOMED. There is semantic in LOINC code.
harold: Mark is saying that Mayo
is talking about putting LOINC codes in published clinical
records. If they do, that will give an interop component.
... If they do, the LOINC codes can be put into the microdata
or RDFa markup, so that the patient does not have to see the
codes. Most patients don't want to see them, but having them in
the records enables others to know exactly which trigliceride
test was used.
... Marc also says that wikipedia and other sources use the
LOINC code as common anchor, so that gives a shared semantic
link.
eric: Isn't FHIR planning to do
the same thing?
... Everytime you have a codeable concept you have a bunch of
codings, and each one has a codesystem and code, and in RDF
they equate to URLs, and there was no resistence to using URLs
there. That would give a gateway.
harold: And LOINC is leading the
pack in having a sanctioned FHIR terminology with examples of
how to get it. They are providing URLS for them.
... We want similar things to happen in the other common code
systems that are used in FHIR. And as communities move into
bioinformatics we're looking to the successor of the disease
ontologies and various OBO ontologies.
... Back to slide 5:
https://docs.google.com/presentation/d/1_yqpKWBhT8nTcPmRVm4nLtPIzdQ_nj8CpO9e8jIlNb4/edit#slide=id.g5f38253709_0_0
... fhirschema.org allows FHIR translation for medical
knowledge
... Could also provide provenance. All FHIR examples have a
text representation of what is in there.
... This could provide provenance of where it came from. If you
do that, then the rest of the JSON could be superfluous. This
could lead to an exchangeable format, which is interesting.
marc: HTML provides human readable. If we could use this in FHIR to make the text human readable.
harold: mayo, cleveland clinic
and washington U are working on a joint project to make
clinical records machine readable. Marc sees fhirschema.org as
a big big jump from human readable to adding machine readable
to it.
... Nice thing about schema.org idiom, as opposed to a formal
FHIR resource with everything formalized, it allows things to
be incrementally formalized.
... Good example was patient record. Inititially it did not lab
codes. Gives an incremental entry to FHIR world.
... NLP markup ... having worked with Hung Fung Lee (sp?), they
have ad hoc NLP to FHIR, and it requires annotating natural
language text. fhir.schema.org would provide a standard for
that.
... it gives a series of tags that are already defined in the
FHIR space, and allows use of tools by google and others to
extract the FHIR resources.
... Another area: IOP devices, and annotating the compact data
records that come from heart rate monitors and smart watches,
etc.
... There's a very active effort in IOT community to do a FHIR
mapping.
david: Who is doing that?
lusha: Microsoft
harold: MS is working on maps
bewteen the standard space and FHIR.
... Would be good to do a paper on this.
... slide 4:
https://docs.google.com/presentation/d/1_yqpKWBhT8nTcPmRVm4nLtPIzdQ_nj8CpO9e8jIlNb4/edit#slide=id.g5f38253709_0_0
... We discussed with danbri about dots in identifiers.
Microformats bans them. FHIR uses them everywhere.
... But banning them is kind of counter to the microdata spec.
Danbri said microdata spec is outdated and suggested
RDFa.
... On slide 7
https://docs.google.com/presentation/d/1_yqpKWBhT8nTcPmRVm4nLtPIzdQ_nj8CpO9e8jIlNb4/edit#slide=id.g5f38253709_0_0
... we only got dose form, not everything. So it looks like
google tools handle the dots.
marc: But they will not
display.
... It was discussed and they will keep the dots.
harold: Can you send a link to that discussion?
marc: If we could avoid the dots and make everyone happy . . . .
harold: draft spec on FHIR site
there is a draft RDFa spec that uses dots. We played with it
and looked at using CamelCase, but FHIR already uses CamelCase.
We thought about hyphens, but that's an issue that will take
work.
... Need to work with FHIR/RDF people. Hate to have two RDF
specs.
david: How to follow up on that? What forum or people?
harold: I think FHIR/RDF spec
will be at risk. Grahame said so far it is okay, but if things
don't start happening it will eventually get pruned. First task
is to show value.
... If we expand on these slides I think that is doable.
marc: Regarding discussion with danbri, what was final conclusion?
harold: Nothing final. danbri
said he wasn't sure what the google parser would do. There was
discussion but it got bogged down in the WG. He suggested using
RDFa.
... I'd like to respond to danbri with more complete slides, to
better outline the problem and maybe he can help.
... Another issue has to do with the JSON-LD @context.
schema.org builds an @context that it uses.
... We have looked at JSON-LD [a few years ago] but it was not
adequate at that time. JSON-LD 1.1 theoretically would address
some of the key deficiencies. We run the risk of having two
JSON-LD @contexts. One comes from fhir.schema.org, you get it
for free. And it produces a different graph from what FHIR/RDF
sanctions.
... Another thing that could bootstrap and ensure that RDF
takes, if we could get JSON-LD @context, then the cost goes way
down because it would only be a matter of pulling in a library
for processing FHIR as RDF.
eric: The new JSON-LD 1.1 is context sensitive, which is one of the first things needed for FHIR/RDF. But another problem is it will never match what comes from fhir.schema.org
harold: Right, because fhir.schema.org is a path notation.
eric: There is no expectation in FHIR that the name fully identifies the item. The context is needed.
harold: We desperately need to work on that, the JSON-LD harmonization.
eric: Could work the microdata angle by giving danbri a copy of the fhir.schema.org with dots?
harold: Yes, we can do that.
<scribe> ACTION: Harold to respond to danbri on dots issue
marc: Is there a plan for an API that someone can use?
harold: Turning medical records
into FHIR is hard if they are not marked up from the
beginning.
... Our hope is that google wiil have done the bulk of turning
marked-up HTML into structured data.
... But the hard part is marking it up from the
beginning.
... This BBC thing mentions google's rich snippets: https://www.bbc.co.uk/ontologies/fo#terms_Food
... The top of that page mentions Google's Rich Snippets. How
could FHIR inform Google's Rich Snippets?
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: Lusha_Cao EricP Harold_Solbrig MarcT 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: harold 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.)[End of scribe.perl diagnostic output]