ericP: mike is working on Integrated EHR for IPO (integrated program office) for DOD/VA
... integrating the DOD's EHR wtih Vista
... originally expected to be a monolothic architecture
... now moving towards FHIR for integration layer
... interested in Semantics in FHIR
ericP: FHIR specs come with examples in both XML and JSON
... Josh translated diagnostic-report and order into RDF
... (by hand)
... ericP created and XSLT which generates that ideal RDF (mostly)
... goal was to generated a library of e.g. Resource Reference, Coded Value etc.
<JoshMandel> https://github.com/jmandel/fhir-rdf/tree/master/generated
JoshMandel: we wanted a generic way to translate the examples that come fro the FHIR spec to RDF
... using XPath on the FHIR spec, we get types for each XML element what type it should be
... (example above)
<JoshMandel> https://github.com/jmandel/fhir-rdf/blob/master/generated/diagnosticreport-example-lipids.xml
JoshMandel: if you do a GET on a FHIR resource, you can get linked resources in a <contained/> element
... e.g. a diagnostic report links to individual refs to observations
... you'd think that the observations would be their own web entities
... but if some service doesn't "host" entities, it sticks them in a <contained/> element
... you can also use this when you wnat to embed references for convenience
<JoshMandel> https://github.com/jmandel/fhir-rdf/blob/master/generated/patient-example-a.xml
<JoshMandel> https://github.com/jmandel/fhir-rdf/blob/master/generated/patient-example-a.ttl
ericP: want to tour examples of what worked and what didn't
... ?
JoshMandel: sure.
... look at the top of patient-example-a.xml
... we see an extension, text, contained, idenfifier, ...
... the turtle file (patient-example-a.ttl) mirrors the XML
... two things aren't well-translated: contained and extension
... every element in FHIR can have an <extension/> inside it
... your patient can have an extension, but so can the string value for your patient's first name
... the extension is basically a key-value pair
ericP: iirc, one possible way to handle extensioons was to just put then in RDF properties which aren't in the FHIR namespace
mike: did the HL7 WG come to you about RDF support?
ericP: the wanted to use and OWL representaiton of MIF and generated the O-RIM
JoshMandel: he would like to have RDF examples along side the XML and JSON
... (graham)
... also like to see if they can use OWL to catch errors
emory: davide (ASU) has been using Empire
... he was looking at the FHIR XSD
ericP: HL7 would like to use (RDF) semantics to help with governance.
... They're sort of re-starting with FHIR, with lots of WGs likely to develop specialized resources.
... With some mapping from bits of FHIR resources to some normalized form like the O-RIM (or the XML ITS), they'd likle to use the machine to detect events like two FHIR Resources covering the same bit of clinical information.
mike: good for exceeding CCDA [expressivity] and importing [linking to] other terminologies
emory: connor and i met with mustafa and his group
... they are doing similar things with vocab sets what we are doing in DOD
... they used ICD9, we used ICD10, we used RxNorm, they used ...
... both groups used a triple store for sameAs links between vocabs
... then looking for out DL can help us with many-to-one, etc relationships
... we had started with UMLS, which means any licensee ...
... we don't have a business model for maintaining our SKOS assertions
... SALUS folks do have a business model but are interested in sharing their mappings
... (we used Jena, they used OpenVirtuoso)
... we'd like to see what we can do within W3C to continue this
... connor and i think we should start limited to the disease portion of SNOMED (disorders)
... creating a solid ontology of diseases
... next step is to represent the SALUS work in a publicly accessible way
... i went to WebProtege presenation at SemTech. plan to load our SKOS into a public webprotoege instance
... socializing the idea with some cliniciains (e.g. Foster Carr)
ericP: FDA also needs a disease ontology
mike: webprotege gives you good outreach
emory: my reason for using a non-Stanford store is for buidling plugins