See also: IRC log
<jerven> I might be ??P30 on the phone as well
<HarryH> Zakim - 412.623.. is me.
<HarryH> Harry Hochheiser
<HarryH> Who is 215?
<iker> is me
<iker> Zakim +1.215.230.. is me
<iker> thx Scott
<JoanneLuciano> is there a skype way into zakim (is skype on zakim)?
<mscottm> you can use skype out to call it but you knew that :)
<kalpana> this is kalpana i am on skype out
<JoanneLuciano> yes, was aiming to save my pennies
<mscottm> talked to ericP about it a while back. Problem is with Skype and proprietary APIs, no open source, etc.
<ericP> FHIR debrief
<jerven> there are some free sip software that you can use to call in for free
<jerven> such as jitsi
<StuartTurner> You're fine (a bit of background noise)
<StuartTurner> Yes, better
<JoanneLuciano> @jerven thanks! I could probably use google voice too, now that i think of it.
<scribe> ScribeNick: Michel
<mscottm> btw, Stian Soiland-Reyes (Manchester) made a nice gadget which provides a URL to a skype chat: http://s11.no/s/
ericp: i2b2 - controls access to data to make sure that patients remain anonymous
<JoanneLuciano> i2b2 folks the ones (only ones) able to get the access to the clinical data.
ericp: shrine (sp?) interface to i2b2 data
... need large data to uncover associations between genotype-phenotype, or treatments; huge cost to patient data protection
... in SW modeling, pick schema(s) and encode data accordingly; form that readable as "sentences"; bob has blood pressure bp1; bp1 has a systolic component, etc, and a value with unit of. ..
... every piece of infrastructure has a sophisticated map from the data they capture to the representation;
... refer to other organizations for coding data - LOINC for lab data, SNOMED, etc
<ericP> FHIR debrief
<bobP> wrong, sorry!
<HarryH> aaaa is HarryH
slide 4: observation - LOINC code for test and results
slide 5: for early TMO, we straightforward represented the test result
scribe: huge infrastructure depends on this explicit coding; RDF's scaling good for references to a number of terminologies
slide 6: first pass at an RDF version of FHIR (pronounced fire; HL7)
slide 7: HL7 produced a message passing for patient data, without specifying the nature of the data; coded terminologies handle that
scribe: small upper level ontology - patients, clinical encounters, labs, lab results, etc; constrained
... mood code - whether something was a query, completed, order, medication prescription, - very complex
... v2 (ascii separated by pipe) while v3 more sophisticated
... number of people who could understand and manipulate the RIM was very small - needed a model to accommodate every use case; so it became large and abstract
... only 3 countries adopted rim / v3; refactoring now - more webby oriented - "based on REST"
... not going to be transporting the record with all the details, but transactions now more targeted/focused using identifiers
<ericP> FHIR lab report instance
slide 6: rdf version of the above xml file
<StuartTurner> FYI - Units of Measure (UOM) code system refers to UCUM (http://unitsofmeasure.org/)
slide 7: rdf version of the v3 might look like; more data, but says the same as slide 6 in terms of content
scribe: the rest is template like; pretty easy to write SPARQL construct from one to the other
... need bi-directional conversion because i) infrastructure similarity ii) this is the language of HL7; e.g. this is a certain kind of event with mood code
... so need to show how the abbreviated descriptions maps to the full entry
... validation - open world-ness useful for knowledge representation, but these records contain current determinants, which need access to
... need to find out what might be incompletene
<StuartTurner> And also the reason for the ubiquitous use of NullFlavor in HL7
<StuartTurner> Great overview
stuart: a null flavor captures the state (unknown, unavailable, pending, etc)
<JoanneLuciano> NullFlavor [2.16.840.1.113883.5.1008] Description: A collection of codes specifying why a valid value is not present.
ericp: different flavors to specify the null
<JoanneLuciano> Name of speaker?
jerven: software developer for swissprot / uniprot
<StuartTurner> Joanne: Recent HL7 discourse has been quoting Donald Rumsfeld's quote regarding "known unknowns" @ http://en.wikipedia.org/wiki/There_are_known_knowns
jerven: validation of RDF : i) is the message structure complete (are all values supplied); ii) are the supplied reasonable or correct
... SPIN allows you to do both - i) does a person have a birthday, ii) does the birthday fall within a certain range
... SPIN can be seen as a collection of SPARQL queries to check with the records are complete or correct
... e.g. a protein that contains a homeobox should have a DNA binding function; if find homeobox that doesn't have function, could be an error ; exceptions (non-functional homeoboxes)
... SPARQL construct for violations - generates human-readable error message
... SPIN rules as a syntax are not just SPARQL, but saves them as RDF. You can then have SPIN rules about SPIN rules; e.g. organism dependence; if the organism id changes, then the rules will trigger an error
... HL7 -> same kind of problems: want to check that all the values that you need are there; that the ones that are there are within a reasonable range
... if a human can derive the check, it can be encoded as a SPIN rule
... 150 rules for uniprot - checking curation quality
... infrastructure: interactive editor (assert homeobox - complains of lack of functional annotation)
... checking identifiers cross-links to external databases
<matthias_samwald> Light-weight alternative to SPIN: 'Simple SPARQL Rules' http://purl.org/zen/ssr.ttl
<matthias_samwald> Example of such rules: http://safety-code.org/ont/cds.ttl
ericp: validation languages make use of grammar conditions - can specify the number of (one or more, etc) to check structure; is it possible to have such a grammar to generate the SPARQL rules?
jerven: for message structure, I would write them as OWL; SPIN takes OWL in the closed world assumption and generates SPARQL; schema validation through OWL
<jerven> This is the basic schema ontology we use http://www.uniprot.org/core
<kalpana> I have to run now but great discussion
<jerven> I also have to go
<jerven> feel free to send questions via twitter to jervenbolleman or email@example.com
<kalpana> i will do that, we will have quite a we questions
<StuartTurner> Wanted to share that in the veterinary informatics realm, we have significant concerns about moving forward with HL7 (especially regarding the Clinical Document Architecture) in a number of projects. Much of this relates to IP/licensing issues as well as technical.
<StuartTurner> We are interested in FHIR and its direction since it's nascent and there maybe opportunity to add and mitigate some of the modeling issues in HL7 regarding non-human animals (clinical as well as research).
<StuartTurner> We'll be meeting at our annual veterinary informatics symposium in August, so I may have more to report if interested. I'll be making a pitch for SemWeb approaches. Much of the emerging work is relatively unknown to that audience.
michel: recap on OSCAR discussion from last week - two goals - real-time clinical narratives to parse out coded content for patient-problem tracking and visualization; and CDS (some discussion around SPARQL, JBOSS drools)
link to a genetic profile a person from a 23andme result (above)
<JoanneLuciano> nice update - unfortunate that Warfarin is late alphabetically