W3C home > Mailing lists > Public > public-semweb-lifesci@w3.org > February 2018

Re: Questions about FHIR RDF

From: David Booth <david@dbooth.org>
Date: Mon, 26 Feb 2018 17:28:33 -0500
To: public-semweb-lifesci@w3.org
Cc: João Moreira (UTwente) <j.luizrebelomoreira@utwente.nl>
Message-ID: <6d7870dd-9444-f7c6-e7ab-bc8fec444a60@dbooth.org>
Hi João,

Some initial responses below, but we can discuss further on tomorrow's 
call . . . .

On 02/26/2018 06:05 AM, João Moreira (UTwente) wrote:
> Hi all,
> 
> Last Tuesday we decided that I'd send some questions regarding FHIR RDF.
> The list is below, sorry in advance if there are "stupid questions":
> 
> - Regarding documentation of the ontology, I noticed that there are two 
> W3C specs:
> http://w3c.github.io/hcls-fhir-rdf/spec/ontology.html
> http://w3c.github.io/hcls-fhir-rdf/spec/
> 
> What is the difference between them?

Very good question, and this reminds me that we should probably clean 
this up a little.

The first one at
http://w3c.github.io/hcls-fhir-rdf/spec/ontology.html
was our initial exploration of designing an ontology for FHIR, and is 
now pretty much obsolete.  Tony Mallia did the bulk of this work -- 
Thanks Tony! -- and we learned quite a lot from it.  However, as we got 
farther along, we decided that the approach that we should take with 
FHIR/RDF should strictly mirror the XML and JSON formats of FHIR, both 
to facilitate mechanical round-tripping between formats and to make the 
correspondence to those formats simple and clear.  We then added 
FHIR/RDF to the official FHIR specification, largely through the efforts 
of Harold Solbrig and Grahame Grieve.  The FHIR specification is 
generated by an automated build process in order to ensure that all of 
the parts stay consistent, so adding FHIR/RDF required augmenting this 
build process, among other tasks.

The second document at
http://w3c.github.io/hcls-fhir-rdf/spec/
is intended to reflect the current FHIR/RDF specification at HL7, though 
I'm not sure that it is entirely up to date.  This was created through 
the efforts of Eric Prud'hommeaux and Harold Solbrig, who developed a 
formal schema for FHIR/RDF using ShEx.  This ShEx schema is now also 
incorporated into the FHIR specification build process, so that the 
downloadable ShEx always reflects the current FHIR spec.

Eric and Harold, please speak up if I got anything wrong here!

> 
> - I missed a history section (a timeline) explaining when FHIR RDF was 
> created, what was the problem, the goals, the versions published, etc.

FHIR RDF grew out of the conjunction of three efforts:

  - The Yosemite Project - http://YosemiteProject.org/

  - The W3C Healthcare and Life Sciences Interest Group

  - The HL7 FHIR project.

The group was chartered as a collaboration between HL7 and W3C, via an 
RDF subgroup of the HL7 Implementable Technology Specifications (ITS) group:
http://wiki.hl7.org/index.php?title=ITS_WG
This RDF subgroup,
http://wiki.hl7.org/index.php?title=RDF_for_Semantic_Interoperability
was chartered more broadly "to establish semantic interoperability of 
structured healthcare information by using W3C Resource Description 
Framework (RDF) and related standards to express machine-processable 
meaning."
http://wiki.hl7.org/images/3/31/Rdf-semantic-interop-subgroup-v9.docx
Thus far the focus has been on FHIR/RDF because there seemed to be a lot 
of momentum behind FHIR, and we thought we could get the most mileage 
from our work by joining that effort.

> 
> - Is there a comparison study of FHIR RDF vs other ontologies that share 
> equivalent principles (records x facts and non-monotonicity)? E.g. FHIR 
> RDF vs OpenEHR approach or vs UniversAAL ontology?

There was some discussion with Thomas Beale about OpenEHR and how it 
compares with FHIR/RDF, and there has also been some discussion about 
mapping between FHIR/RDF and OpenEHR, but thus far people have not had 
the bandwidth to pursue it.  Claude Nanjo may have looked at this the most.

Regarding records vs facts, this is definitely a chasm that would be 
great to bridge.  FHIR/RDF is based on records -- not facts -- because 
FHIR is based on records, and FHIR is based on records because that is 
how healthcare EHRs have operated for decades.  Some ontologists have 
been advocating for a shift to fact-based ontologies, and I think there 
is fairly wide consensus among RDF folks that this would nice to 
achieve.  But the reality is that existing EHR systems are universally 
records based.  This means that we either need to bridge between those 
worlds or figure out how to convince the entire healthcare IT industry 
to change their way of thinking to *our* way of thinking, which might 
take a little while.  ;)

I will note that the need to bridge between ontological worlds is a 
natural result of bottom-up ontology design, because different 
communities will inevitably model things differently.  It is just part 
of the overall work that needs to be done, to make all this RDF stuff 
work the way we eventually want it to work.  But it *is* a lot of 
effort, and we are not there yet.  In fact, we do not even yet have 
well-established mechanisms in the RDF world for expressing and 
performing ontology mappings.  (Side note: anyone who thinks these 
mappings can be done solely with simple ontology mechanisms like 
owl:sameAs, rdfs:subPropertyOf and rdfs:subClassOf is being woefully 
naive.  In real life, when ontologies are designed independently, with 
very different world views, the modeling differences are much more 
complex than what can be expressed with those properties.)

This need to bridge between different ontological worlds is one of the 
reasons why some of us are so interested in ShEx.  ShEx has an 
extension, ShExMap that in essence allows one form of RDF to be mapped 
to a different form of RDF, in fairly arbitrary ways.   SPIN also 
provides this capability, though it differs in style, being based on SPARQL.

Anyway, that's a brief intro.  Summary: there's still a lot of work to 
be done in this area.

> 
> - Is there an official agenda (roadmap) for FHIR RDF (e.g. ongoing and 
> future work, expected results, etc)?

Our weekly meeting agenda are posted here:
http://wiki.hl7.org/index.php?title=ITS_RDF_ConCall_Agenda

We drafted requirements early on, which I believe we have largely met:
http://wiki.hl7.org/index.php?title=FHIR_Ontology_Requirements

We have a "Deliverables and Editors" list, which needs to be updated:
http://wiki.hl7.org/index.php?title=RDF_for_Semantic_Interoperability#FHIR_RDF_and_Ontology

And we have a github issues list, which we also need to review and update:
https://github.com/w3c/hcls-fhir-rdf/issues

We are now at the point where FHIR/RDF is a part of the official HL7 
FHIR spec and usable!   However, I think it is also time for us to 
review and take stock of what still remains to be done.   We are 
entirely volunteer driven, so it depends on who has time and energy to 
do what, but I think it would very good to discuss.

> 
> - What is the process between W3C and HL7 to evolve this ontology? Is 
> there a workflow for decision making?

We have adopted the HL7 FHIR decision-making process, which depends on 
the maturity of the spec.  We make decisions on our weekly 
teleconferences, and sometimes on zulip:

The process can be fairly informal when we are initially drafting 
something, but once it becomes a part of the official HL7 Draft Standard 
for Trial Use (DSTU) then the process for making changes becomes more 
formal, with gForge use for tracking
https://gforge.hl7.org/gf/account/
and the official HL7 balloting process used for decision-making.

Thus far, FHIR/RDF is in the FHIR spec but -- thankfully -- is not yet 
subject to the formal balloting process, so we can still make changes 
relatively easily if we need to.  FHIR/RDF is actually pretty stable at 
this point, but if you or others see things that you think should 
change, please do bring them up so that we can consider them.

> 
> - I think that Eric mentioned that IoT was one of the use cases that the 
> workgroup is interested for applying FHIR RDF. I noticed that the 
> section "3.0.5 Common use cases" 
> (http://build.fhir.org/linked-data-module.html) is stamped with "TODO" 
> yet. Is there a list of use cases, listing the past ones and the 
> intended ones?

We had some discussion of Use Cases early on, but did not finish 
drafting them.  Here are links:
http://wiki.hl7.org/index.php?title=FHIR_Ontology_Requirements#USE_CASES
http://wiki.hl7.org/index.php?title=RDF_for_Semantic_Interoperability#Use_Cases
https://www.w3.org/wiki/HCLS/ClinicalObservationsInteroperability#Use_Cases

> 
> - The prior question brings to this: what are the case studies that used 
> FHIR RDF and/or are using it now and/or planned to near future? I saw 
> the presentation from Harold and Eric about Rheumatoidarthrits 
> (https://www.w3.org/2016/Talks/1205-swat4ls-egp/reasoning#(1)), is this 
> a case study being developed for healthcare system?

Hopefully Harold can address that.  We do not have a list of projects 
using or planning to use FHIR/RDF, but maybe we should try to make one. 
We certainly do want to hear from people who are using or plan to use 
FHIR/RDF.

> 
> - We talked about the WP5 document, but I'd like to know what is WP5, is 
> this a working package within a project? Which project? What is the goal 
> of WP5?

I believe Grahame Grieve designed it, so perhaps he could explain more. 
Or Harold Solbrig might know.

> 
> - Did I understand correctly that the goal of this current FHIR RDF is 
> to thoroughly representation of FHIR standard (the data model)? If yes, 
> why not having two ontologies: this one and another one as a lightweight 
> FHIR RDF to be used by semantic healthcare systems (e.g. the IoT ECG 
> system I'm working on)? This second ontology could be derived from and 
> mapped to the first.

I think that's an idea worth considering.  Let's discuss it on the 
teleconference.

Thanks!
David Booth

> 
> Thanks in advance.
> 
> Best regards,
> 
> Joao
> 
> 
Received on Monday, 26 February 2018 22:28:59 UTC

This archive was generated by hypermail 2.3.1 : Monday, 26 February 2018 22:28:59 UTC