Meeting minutes
Introducttions
dbooth
ericP: Worked w Josh Mandel on FHIR RDF. Was W3C contact a long time.
jim: At U NC. Work on bio ont, and applying OWL.
Gaurav: Also at UNC. Working on sem web stuff for 5-6 years.. General research software dev. int in how to represent data in RDF and make it easier to connect data. Want to do it in large scale data like wikipedia.
Allen Flynn: Prof U mich. Was in pharma, then EHR, Trinity health. 2018 PHD in CS, got me into sem web. Working on algo that are evidence based for healthcare. Did work on metadata, packaging algo. Lots of work informed by Linked data ideas.
… Published one ont on describe things w parts. Working on joint project w C mental health in toronto.
avanti: informatic student, working w Allen.
adeel: Work at Ctr for addiction and mental health in toronto. In ontario there's a push to use FHIR, so we're trying to create a data sharing platforme, for EHR, imaging, wearable, use FHIR and RDF and other vocabs.
phhilippe: U of Oxford, prev astra zeneca. OBO foundry.
U Mich & Ctr for Addict and Mental health
allen: want to deploy functions in the graph env that can compute over it and add value.
… W're in the middle of that process now. Multiple sources of data, already in a graph DB.
… Inside the dashed line is our work. Gray pipeline Adeel is building.
… Current focus is 3 resources: 1. library resource from FHIR standard. Using that to describe individ computer functions, eg, medical calculators, predictive models, etc.
… Inside the graph DB we're adding descriptions of fns. Assuming they're already in there, the pipeline extracts one pt data and relevant params, then calls a fn to gen new pt features as the outcome, eg, risk score.
… Two things come along w that: want to describe the fn call itself. Landed on an exploration of the service request FHIR resource -- not originally intended for that use.
… Then onthe output we're using the FHIR Observation resource. Then it gets pushed back to the graph DB.
… We keep a record of the request and the output.
dbooth: Nice clean computing flow model.
eric: Did you use SMART?
allen: I don't think so.
eric: e.g. using it as the executoin env, such as auth and auth.
adeel: The flow we're exploring is more automated, not user driven.
(slide 2)
allen: the library resource is close to what we want to describe. Good fit for purpose.
… The FHIR RDF Observation is okay. Status field isn't too painful. Not aware of anything better, so we're using it for now.
… FHIR RDF service request is not the best fit. Prob not the originalintent for that.
… in the lower part of this slide, want to use URLs everywhere. @contexts are getting longer.
… Helped a lot by the paper from Solbrig and all. Playground has been a super resource.
… We Started saying we were using FHIR. Earlier were using the Function ont and the Prov ont. Still looing to include them, even though we're most focuse on FHIR so far.
eric: re service request, you mentioned the terminology doesn't line up well. Want to say "I want a resource w a value in this terminology"
… A diagnosis, allergy, etc tries to be more specific than Observation.
… If you were doing some coding that was producing a bunch of values...
allen: Questionnaire and response has aome relevance here, but need something a little more general.
eric: When you talk about FHIR RDF lib, what's your input. If there were a svc req for this, it will find code for it?
allen: Yes. Want fn assoc w those things. We have a record and triples. That's why the slide shows iterations.
… Anticipate having many hudreds of fns, and when it runs, point back to what was run.
eric: There's some prov stuff in Domain resource I think. You want to use the Service Request to use DRY principles.
eric: Is writing that in FHIR land useful, because you can interop w yourself, but not others.
allen: Might have gone to strongly on FHIR.
dbooth: Re Harold's observations about R4 problems, Struggled w trying to get rid of the extra bnode, but concluded that we couldn't.
eric: Working on NHS project, using SOLID. Similar motivation to your work. Also working w Claude Nanjo on U Utah project for CDS.
… Goal of SOLID is for all pers data that can be shared or not, under the individuals control -- FHIR in and FHIR out.
gaurav: For this group, I'm working on Concept IRIs. Working to get FHIR UTG group to include that in their repo.
jim: Gaurav and I get involved in this through Emily Pfaff on NIH grant. Helped on R5 changes, and implementation and java code. Emily has plans to apply RDF to integrate ont w diseases. Export to FHIR RDF and then query and aggregate.
claude: At U of Utah, clinical informaticist. Working on dev CDS appllication for disease mgmt, used in production. Also int in the FHIR RDF work and ShEx and ShexMap work. Part of our knowledge base, looking at boundary of rule-based knowledge vs KB that should not be written as rules.
HAPI
eric: It's upgraded to R5, passing tests, but have not yet added switches to allow you to switch between R4 and R5.
Concept IRIs
gaurav: now have 16 votes, but not sure we have all the quorum req met yet. On the other one, Reuben hasn't replied yet to my response.
https://
https://
rob: We discussed the UTG tickets on Term infra co-chairs call, for planing agenda. Agreed to put thse tickets on the agenda, try to get them over the finish line this afternoon, 3;30pm Eastern.
What works for RDF reasoning?
eric: What works? What doesn't? What Arch features would you want to provide the ecosystem you want?
… I interpreted the U Mich work as simillar to SMART on FHIR: doing computes and injecting things back in.
… And Claude is injecting things back, like med request.
… So far U Mich work doestn' need Auth and Auth, becuaes it's internal.
allen: Yes, but if we want to open beyond firewall is a good one.
eric: And when you're doing computes, are you infrancing over term. Any machine learning?
allen: No maching learning yet. Fixed computing fn. But it's picked up separately.
adeel: Currently the sys doesn't do inferencing. Fns are writtin in JS or python.
allen: Blue brain has been used in community to add reasoning. Think that's coming, but we haven't used it yet.
eric: What did crystal daniele use? Claude: IDK.
… She used CCDA on her own.
ADJOURNED