Semantic Web Health Care and Life Sciences Interest Group Teleconference

11 Apr 2017

See also: IRC log


EricP, David_Booth, Rob_Hausam, Ken_Lord
David Booth


Applying FHIR RDF

eric: OWL is good at subsumption reasoning -- classes and subclasses. So you end up asking if something is a member of a class.
... E.g., a CodeableConcept is an instance of a SNOMED CT class. Ont weird, but allows you to do what you want to do.
... Wanted to use ValueSets, but could only do inference if something was a member of a class.
... Ont weird because the SNOMED term describes something like serum creatinine level. But its not really a CodeableConcept.
... Would be more ont correct to say that the Observation is an instance of that SNOMED concept, rather than the code.

rob: Question is whether that SNOMED connection applies to the element itself or the observation.

eric: Lloyd wanted to propagate the concept up to the Observation.
... A resource with multiple CodeableConcepts in it . . . if we had Observation.code and Observation.status ... We need to find a resource that has CodeableConcepts on their own, and figure out if there is a principled rule that would allow the CodeableConcepts to propagate up to be applied to the resource as a whole.
... There will be people who will oppose us if we do not fix this.

rob: But FHIR describes records -- not diseases.

ken: If the binding is at the Observation level, how would that work with detailed clinical models like CIMI?

eric: A detailed clinical model is a combination of observations and interventions.
... If an APGAR score is a constellation of 5 observations then that would still be consistent with any one of those observations have the SNOMED type apply to the observation itself.
... We want to apply the code to something so that we can apply inference.
... Another screw case: medical records have SNOMED identifiers, and SNOMED gurus model things in nice ways, but when people use the codes they can use an Observation code, or a Level code. Probably a lot of confusion about using the right one, because historically there was no machinery to tell them the were doing it wrong.

dbooth: That's a benefit of doing this work, because inference could tell them they've made a mistake -- used the wrong code.

eric: Disease vs diagnosis.
... There are useful things you want to look up about the disease. Need coding to be right in order to use SNOMED inference.

rob: That may be a bad example, because diagnosis and disease are not different in snomed.
... It has been impossible to make a reliable distinction between Disease and Finding, so we need to remove that distinction.

eric: In med school you learn disease, but the finding applies to a particular patient.

ken: I think a Concern can be tied to disease also. Is there a SNOMED diferentiation of a Concern?

rob: There's not a notion of Concern in snomed.
... But there is a difference between Finding and Observable.

eric: screw case: Using snomed relationships, searching, might get false negatives if people use the wrong nature of code.
... If that happens, you could instead look for something with a relationship to what you want.

rob: Example: if you're recording an allergy, do you record a substance, or a precoordinated code as an allergy to that substance.
... Hardly anyone uses post-coordinated expressions. Not common.

eric: I think sometimes you'll see the snomed commas, but never the equals.
... If there's a pre-coord term for laterality-left, and one for hairline fracture, they'll write them both (with a comma), but not the official post-coord expression (with equals)

rob: I don't ever see people doing that these days.

eric: Who does use the compositional grammar?

rob: People like Cecil Lynch

eric: Once people can see the benefits of what we can do, then there will be more incentive for more structured info. But still a pain to put in all that structure info.

dbooth: But Natural Language Processing (NLP) is coming along very well. It can help bridge the gap of capturing structured info.
... I envision a conversational interface between the system and the healthcare provider.

eric: Build the whole snomed expression in RDF, or use either the pre-coord term or equivalent expression.

rob: Jim Campbell, U of Nebraska Med Center, is doing that.

dbooth: Should we record this as an issue?

eric: More of an opportunity. Will someone pay me to do it?
... Would be cool to be able to take a post-coord expr and turn it into a blob of RDF as an expression of the snomed terms.
... Or from pre-coord.
... This would harmonize: pre-coord, compositional terms (post-coord), and atomic terms sprinkled throughout specialized attributes in the info model.

<inserted> See https://github.com/w3c/hcls-fhir-rdf/issues/50

eric: Need a microparser for parsing apart the compositional expressions

Terminology Licensing

dbooth: SNOMED licensing is still a barrier, even within the USA, which has a blanket free license to use, because the license restrictions do not allow people to transform and post portions of snomed freely on the web.

eric: a billionaire could buy out IHTSDO and provide SNOMED maintenance for 20 years.

<ericP> http://gforge.hl7.org/gf/project/fhir/tracker/?action=TrackerItemBrowse&feedback=Successfully+added+%5B%2313162%5D&tracker_id=677


Summary of Action Items

Summary of Resolutions

[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.152 (CVS log)
$Date: 2017/04/11 16:16:29 $

Scribe.perl diagnostic output

[Delete this section before finalizing the minutes.]
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: s/ken/rob/
Succeeded: i/rrsagent/See https://github.com/w3c/hcls-fhir-rdf/issues/50
Succeeded: s/Topic: Harmonizing pre-coordinated and compositional terms in the information model//
Succeeded: s/Present+ Ken/Present+ Ken Lord/
Succeeded: s/Ken Lord/Ken_Lord/
Succeeded: s/Licensing/Terminology Licensing/
Present: EricP David_Booth Rob_Hausam Ken_Lord
No ScribeNick specified.  Guessing ScribeNick: dbooth
Inferring Scribes: dbooth
Found Date: 11 Apr 2017
Guessing minutes URL: http://www.w3.org/2017/04/11-hcls-minutes.html
People with action items: 

[End of scribe.perl diagnostic output]