W3C

- DRAFT -

Semantic Web Health Care and Life Sciences Interest Group Teleconference

28 Apr 2015

See also: IRC log

Attendees

Present
Tony, DBooth, pknapp, lloyd, Marc, Rafael, rhausam, ericP
Regrets
Chair
David Booth and Paul Knapp
Scribe
dbooth

Contents


<trackbot> Date: 28 April 2015

david: this is how to scribe :)

<Tony> DBooth: instructions for recording

<inserted> Scribe: dbooth

<inserted> Scribe: dbooth

Approve Minutes of previous meetings

Apr 21 minutes: http://wiki.hl7.org/index.php?title=ITS_RDF_Concall_Minutes_20150421

RESOLUTION: Apr 21 minutes approved!

FHIR RDF and Validation/Translation Task Force

http://www.w3.org/2015/04/22-hcls-minutes.html

<Tony> David: 3 issues that are in agenda should be reviewed

ISSUE: How to represent CodeableConcept, Coding and Code? Should all three be constraints on a common RDF type?

<trackbot> Created ISSUE-10 - How to represent codeableconcept, coding and code? should all three be constraints on a common rdf type?. Please complete additional details at <http://www.w3.org/2014/HCLS/track/issues/10/edit>.

Upcoming HL7 Meeting - Paris May 10-15

Paul: Probably will miss the call next week.

<Tony> Paul: Will not be available next week

FHIR RDF Project Scope Statement (Claude Nanjo) - Status?

Paul: Will go to Foundations Steering Division for approval next.
... If approved, will go to TSC for next week or the following week approval.

Side-by-side example of two FHIR RDF approaches (Tony Mallia)

http://wiki.hl7.org/images/1/19/FHIR_RDF_Sample_side_by_side_comparisons_v2.pdf

ISSUE-7 - Should we treat observation.code as an instance of a class?

Tony: See 2.1 and 2.2
... In 2.2 the type is declared for each of the blank nodes. Each blank node is an instance of a class.

david: What if it were not represented as an instance of a class?

tony: Put the LOINC type on the observation itself: :resource a <http://loinc.org/5544...>

lloyd: if you throw all the semantics on the observation then it would not be clear what is being discussed.

tony: agree

lloyd: Grahame had an alternative in mind: we define our own predicate instead of saying it is a subtype. He's nervous about using a subtype because there are ont that do silly things.
... E.g., SNOMED says something is copyright INSTDO, so if you say that something is an instance of that then you end up saying that your instance is copyright of IHSTDO.

Marc: For LOINC it's a class, but when we go to SNOMED and we go to low level concepts, do we consider it also a class?

lloyd: Yes, you'd be saying this instance I'm recording is an instance of diabetes or alzheimers.

Marc: Then we would not have a hierarchy problem, but this concept belongs to other concepts in the hierarchy I'm afraid we may be overruling SNOMED somehow.
... We're saying that this element belongs to both SNOMED classes.
... We have classes with parents that do not contain the concept below. SNOMED clinical finding ... an element of observation is not necessarily a clinical finding. If we use the low level we end up with a problem with hierarchy.
... I'd be more comfortable using a higher level concept.

tony: I don't think that would give enough precision.

lloyd: If you have a SNOMED code in your instance -- diabetes -- then you're referencing every parent in they hierarchy. It is a clinical finding, a condition of the lung, copyright IHSTDO. That's how inheritance works.
... The fact that IHSTDO puts copyright on there is unfortunate, but it's what they'ave done.
... In theory it matters; in practice nobody will care that the copyright is there.
... But if they do it, what can we expect of others?

marc: Another point: When you look at fhir coding.code, you can use something like skos for display (prefLabel)?

rob: Lloyd's statement of the copyright issue may be overstated -- need more research. Copyright applies to codes, identifiers, descriptions. But in the ont the way it is represented I think it's an annotation which does not inherit.

tony: Yes, it's an annotation, so it does not inherit.

<Marc_Twagirumukiza> if we use the skos:notation /skos:prefLabel we can skip the information here in the graph and axpect it from a terminology server

<Marc_Twagirumukiza> at customer side

lloyd: Glad that SNOMED copyright is correct. But the premise that ont designers can do stupid things, but not a convincing reason not to use the proper mechanism for using RDF on the grounds that someone might do something stupid.
... If they do, then that ont may need to be excluded from reasoning until it is fixed.
... But the is-a relationship is semantically correct and what we should use

rob: At what level to indicate semantics? Okay for coding to be an instance of SNOMED CT class, but at the CodeableConcept level that's not the best thing to do, particularly if there are multiple concepts.

tony: Instance is pointing at the level of CodeableConcept.

rob: To get the power of reasoning that's probably what is needed, but need to be cautious about making that strong a link.

<Zakim> dbooth, you wanted to agree with lloyd. Bad data will also appear sometimes similarly.

lloyd: To expand, if you have a snomed code for bacterial pneumonia, and an ICD10 concept saying pneumonia, and a mapping saying this ICD10 code is equiv to the snomed concept of viral pneumonia, then you'll have a conflict in your reasoning.
... The conflict will show up because somewhere the data is wrong, and it will be aggravating, but the tools are doing what they are supposed to do.
... And if you want to use RDF tools, then you need to be sure the data is clean enough to use it.

+1 to what lloyd is saying :)

<rhausam> I agree with Lloyd on that

tony: Lloyd has made an argument for using RDF to validate data. :)

+1 to tony :)

paul: No different than any other data analysis situation. You need to filter out the bad data.

rafael: As another use case, could use RDF as analytics platform. This will be our methodology for analyzing, using RDF as a metadata standard. Valuable use case for validation/auditing before exchanging it.

lloyd: To clarify, it is not quite the same as other circumstances, because with RDF, if you have the bad data and you try to reason around how many patients exist within a particular age range and geographic area, the reasoner would still blow up, even if that inconsistency is unrelated.

tony: But you can condition the way the reasoner works based on the ontologies that you import.

<Zakim> dbooth, you wanted to say that you can partition your data using named graphs to help get around inconsistency problems.

david: Are we agreed that we should represent Observation.code as an instance of one or more classes?

tony: yes.

<Lloyd> +1

<rafael> +1

paul: +1

RESOLUTION: agreed that we should represent Observation.code as an instance of one or more classes

ISSUE-8 - If we treat observation.code as an instance of a class, should that class represent the disease itself or a *description* of the disease?

Kerstin suggests: https://code.google.com/p/ogms/

tony: there is real world occurrence of the disease, and the information record about it. Two sorts of instances.
... I think the class or type in RDF is a description -- not the occurrence of the disease.

rob: Not sure what tony means by description. Information artifact? That's the traditional HL7 way of looking at it. But each ont may have a different way of looking at it.
... We may not be able to declare that it is only oen thing.

eric: Agree w tony. You would not have a temporal characteristic for appendicitis, but for someone's incidence of appendicitis.

david: Isn't that the difference between instance and class rather than between disease and its description?

eric: You could say there is a class of all incidences of appendicitis, but that's not what snomed is describing.

rob: Disagree w tony. Snomed is all classes. Pneumonia concept could represent all instances of pneumonia disease.
... But definitely snomed is not going to the instance level.
... if we put it into a fhir resource, it is a specific instance of disease.

<Zakim> dbooth, you wanted to say that I would be surprised if snomed did that. can we confirm?

<Zakim> ericP, you wanted to ask for proposals for tomorrow's V&T agenda

tony: Discuss blank node issue?

+1

tony: suggest we do update proposals for issues list?

ADJOURNED

Summary of Action Items

[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.140 (CVS log)
$Date: 2015/04/28 16:45:22 $

Scribe.perl diagnostic output

[Delete this section before finalizing the minutes.]
This is scribe.perl Revision: 1.140  of Date: 2014-11-06 18:16:30  
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: RRSAgent_Text_Format (score 1.00)

Succeeded: s/mintues/minutes/
Succeeded: i/Scribe:/Scribe: dbooth
Succeeded: s/ISSUE-7/Topic: ISSUE-7/
Succeeded: i/Approve/Scribe: dbooth
Succeeded: s/erstin/Kerstin/
Succeeded: s/Scribe: Tony//
Found Scribe: dbooth
Inferring ScribeNick: dbooth
Found Scribe: dbooth
Inferring ScribeNick: dbooth
Default Present: Tony, DBooth, pknapp, lloyd, Marc, Rafael, rhausam, ericP
Present: Tony DBooth pknapp lloyd Marc Rafael rhausam ericP
Found Date: 28 Apr 2015
Guessing minutes URL: http://www.w3.org/2015/04/28-hcls-minutes.html
People with action items: 

[End of scribe.perl diagnostic output]