13:53:20 RRSAgent has joined #hcls 13:53:20 logging to http://www.w3.org/2017/05/30-hcls-irc 13:53:22 RRSAgent, make logs world 13:53:22 Zakim has joined #hcls 13:53:24 Zakim, this will be HCLS 13:53:24 ok, trackbot 13:53:25 Meeting: Semantic Web Health Care and Life Sciences Interest Group Teleconference 13:53:25 Date: 30 May 2017 14:02:54 Present: David Booth, Thomas Beale, Nikolai RyzhikovNikolai Ryzhikov (Health-Samurai) 14:03:04 Topic: Introductions 14:03:26 nikolai: Implementing FHIR standard. Small company. 14:03:41 ... Also working on FHIRBase 14:04:03 ... an open source solution. Also working on RDF -- right direction for FHIR. FHIR core shold be expressed as RDF. 14:05:48 thomas: Main architect of OpenEHR 14:06:35 claude: With Cognitive Medical Systems 14:07:15 Topic: Mapping OpenEHR ADL to RDF 14:08:19 claude: Met with Eric, discussed mapping from CIMI. Clarified goals. 14:08:43 ... Formal work on CIMI RDF is still premature -- CIMI itself still being developed. 14:08:57 ... Two aspects of what we're doing: 1. representing CIMI as an ontology. 14:09:42 ... Also representation through OpenEHR. But that's more an effort with Thomas and the OpenEHR community, and CIMI is merely one model using openEHR. 14:10:28 ... One posibility: represeent logical model itself. Another is through OpenEHR if there's an RDF version of that. 14:10:34 ... Through FHIR. 14:10:52 ... Third: a rep of CIMI logical model as an ontology., regardless of OpenEHR. 14:11:05 ... I'm most interested in that last open. 14:11:20 ... Re CIMI as an OpenEHR version of RDF, we need to ask Thomas. 14:12:03 eric: Would be useful to take a represeentative archetype and show how it looks in OpenEHR, and how an ont would represent it, then also in FMA. 14:12:37 thomas: We don't have a burning project to produce RDF form of archetypes. 14:12:52 ... If we did, we would probably work on the terminology bindings first. 14:13:01 ... Discussed that a little on FHIR zulip server. 14:13:26 ... That binding stuff is probably the hardest. Structural part of ADL is probably the easiest. 14:14:11 eric: In support of that approach, the RIM v3 kept all of its definitive artifacts, and as a consequence users had a lot of line noise. 14:14:29 ... Having the OpenEHR side be used, but a skinnier thing for serialization would be preferable. 14:14:38 thomas: RDF-based transform of RIM? 14:15:33 eric: No, we have the ADL for defining something, then an RDF ont built out of it for serialization, having those separate from the serialization would be in keeping with not making too much v3 traffic that was unnecessary for messages. 14:15:43 ... They went too far with keeping the schema in the data. 14:16:01 ... If we separate the two then that removes the tension about how much detail to put in the model. 14:16:43 thomas: One important point: how terms, post-coord, are presented. I've always thought we'd use compositional expressions inside archetypes, and they'd appear in RDF. 14:17:12 ... But in a number of recent projects, carrying around those expressions is nice theoretically, but it makes everything else hard. 14:18:23 ... Better is to have compositional form, then work out the structural equivalent in terms of multiple data points -- multiple fields. 14:18:56 ... The equivalent of snomed structures. 14:19:08 ... The problem with snomed is you cannot mix in things that are not coded. 14:20:13 best to rely on the axis-specific attributes in the information model 14:20:28 ... For hi fi clinical you'll need hierarchical multiple data points. 14:20:46 ... Using post-coord terms like that in operational EHRs probably isnt' goiug to fly. 14:20:59 ... But they're not the full picture that clinical modelers want. 14:21:09 ... They're also not good for most tools/systems. 14:21:25 ... Nothing that works with compositional expressions. 14:21:59 eric: Counter argument: compositional is more the fundamental rep, then info models have some set of axes you can pull out, but not all. 14:22:14 thomas: As an exception, laterality should always be composed. 14:22:58 eric: owl rep such as IHTSDO's of owl would allow you to mix other ont (if they exist). 14:23:39 ... But majority of uses case that people see would best be met by axis-specific info models, because they have been extended for those use cases. 14:23:57 ... The other side is inference you want to do, where you represent it in triples instead of a big string. 14:24:17 ... If grahame and linda are making good progress we'll have a good way to go back and forth. 14:24:30 thomas: gold standard model of any domain content, if it 14:24:49 ... has non-snomed attributes, then the snomed comp term expr can't be the gold standard model. 14:25:11 eric: You could use snomed's owl and mix in others. 14:25:43 claude: snomed has some limits. partial models with many attributes missing, but it does have the fundamentals. 14:26:06 ... and it allows you to provide context for complex model. 14:26:23 ... The rep of info in a pt rec 14:27:12 ... And we can build on that -- supplement what you need. Not only biomed concepts but other info pertaining to the pt. 14:28:14 thomas: to establish good qual med semantics, the archtetypes are the gold std. 14:28:34 ... General case: snomed with larger models that have other things mixed in. 14:29:04 ... Ideally agreement by all in archetypes to use the same concept model attributes. 14:29:26 eric: Seems like you want, for an info model, here are the expressions you can pull out. 14:30:03 ... For any model, there will be an ont view of it, and ways to parcel out bits into particular info models. 14:31:01 ... Could be compositions where the laterality is nested, and makes it disingenuous to express in the info model. 14:31:38 thomas: Keep in mind the juxtaposition of sibling nodes, sometimes you can see what they mean (laterality, etc), but sometimes not. 14:31:57 ... Could look into ont to find out what they mean -- legislation reasons, etc. 14:32:19 ... Lots of data items are surrogates for what you want to measure. 14:33:02 ... Looking at BP 1-5, we collect 1 and 5, sometimes 1 and 4, but not others typically. 14:35:56 dbooth: what use case or burning need do people have? 14:36:43 claude: limited right now on time, because of VA project. But it overlaps a little on what we are discussing. Discussing SOLR -- like an uber ont to allow them to be related in expressions. 14:37:08 ... SOLR unlike snomed would support concrete domains. Need to look at how to represent them in CIMI. 14:37:20 ... They start to look a lot like this ont we're discussing. 14:37:36 ... I'm not exploring RDF representation yet, but it will eventually be part of SOLR. 14:38:24 ... A lot is premature for deep dives, but I can give eric material to use when the time comes. But my time is limited. 14:40:01 claude: Before we start on an example, I want some guidance from eric on the VA work I'm doing. 14:40:21 ... Then as part of that involvement, eric could see some of that stuff and it could motivate an RDF approach. 14:40:41 ... After that happens we could discuss an early pilot. 14:42:54 dbooth: Plan on meeting again around June 27? 14:43:14 claude: Sounds ok. We need to have something within a month, even if prelim. 14:44:47 dbooth: Same time, 10am Eastern. 14:45:54 Present+ Claude Nanjo, EricP 14:46:01 ADJOURNED 14:46:08 rrsagent, draft minutes 14:46:08 I have made the request to generate http://www.w3.org/2017/05/30-hcls-minutes.html dbooth 14:50:28 Chair: David Booth 14:50:38 rrsagent, draft minutes 14:50:38 I have made the request to generate http://www.w3.org/2017/05/30-hcls-minutes.html dbooth 16:50:04 Zakim has left #hcls