Semantic Web Health Care and Life Sciences Interest Group Teleconference -- 03 Oct 2017

Intros

hank: Software engineer, working on CCDs and FHIR. Haven't yet persuaded my employer to use RDF.

eric: There's a CCD rep of RDF also.

Hank Ratzesberger

I2B2 Experiences (Ken Lord)

ken: Shameless plug for MDMI. We've created an OMG RFP for version 2.0 of the standard. Put it through the OMG arch board; in process now.
... Key feature desired: extend std for the Referent Index, though being renamed to Semantic Element Exchange Repository (SEER), 11170 compliant.
... Classification metadata for elements in ref index. It will use two parts of 11179 (classification scheme), which becomes an ont, and using data description passage to define a metamodel for MDMI.
... Part 2 is to take that metamodel and create an example in the healthcare industry for people to work on it. Want Yosemite Project people to work with that, to come up w closed ont with what everyone is doing. Trying to track FHIR and RIM.
... Our I2B2 work was done 4-5 years ago.
... (disclaimer)
... Work done w Sean Murphy.

(Ken demos MDMI Runtime tool)

ken: Loaded a CCDA .xmi file as the source. Generated from MDMI map editor.
... Target goal was to take v2 XML format and convert it to CCDA, then convert that CCDA to I2B2.
... So on the target side we load an I2B2 metamodel.
... The runtime takes the CCDA and converts to I2B2, which is very different structure.

harold: Core I2B2 is the Observation Fact table.
... It joins patient ID (internal)
... There's also a Visit table, with event ID
... And there's a Provider dimension.
... Observation Fact takes patient event and maps to concept code, optional modifier code.
... Concept code takes you down to Concept dimension, which defines what the concept means, with a bunch more info that goes insite the concept. Most importantly, there's an ont table down there that organizes this stuff to make it browsable.
... Organizes concepts hierarchically.

ken: What came out for us was to understand automation. Problem we had at the time: they didn't have an ont at all. Codes in medication were whatever you wanted to use.

harold: Even today, it isn't just the I2B2 space. The research tooling space tends to allow a researcher to create their own models and identifiers usefiul to that researcher.
... But need to settle on common ontologies, but that's a tricky problem. Without that, without FHIR community agreeing on concept codes, studies are not easy to share.

ken: Are you finding that anything you did in CTS2 would provide a mechanism to enable that to happen?

harold: Focus now in the process of building FHIR RDF we had to produce a catalog of FHIR predicates.
... That seems to give us an interesting common level. We're proposing to the I2B2 community is to create a set of concept codes and modifier codes that represent all of the FHIR codes (URIs).
... Then researcher-specific ontologies can speak in terms of those codes.
... In order to make it useful to researchers, who don't want to have to understand what a FHIR Observation is, they don't want to say fhir.Observation.foo code is something.
... They want to auto select hematocric. We think that's a two-step operation: 1. Loading I2B2 2. Set of transforms to get from I2B2 to researcher-specific ontologies.
... When you start using I2B2 from a particular study, you don't publish only a study-specific ont but also a set of rules to convert it to standard I2B2 model.

dbooth: Big +1 from me!

ken: We found that rules needed become complicated.
... They're not really human readable unless you spend time understanding relationship between children, parents and how they get populated into different selection.

dbooth: We've been experimenting with using shex to transform data from one model to another.

harold: We've first been working with FHIR, then plan to figure out where to go with the ont.

eric: Part of the utility of using ShExMap with FHIR, is that the FHIR schema is already defined in ShEx.

dbooth: The Shex schemas for FHIR are already downloadable.

ken: We have associated metadata with the business elements, in a classification scheme. Would be interested in the ont that are going around. We dont' want to invent one. Wnat to grab them and harmonize around one for interop. Is there a way, looking at the classification scheme to uniquely describe what something means.
... Also, will this (MDMI) help you with FHIR and I2B2?

harold: One part -- sorry I don't know more about how it works -- I'm assuming that it's OMG, so it is MOF-to-MOF?

ken: yes. Alll open source.
... We do a lot of model-to-model transformation, but those tend to be design-time.

harold: You had an xml schema for the source and target?

ken: MDMI has a core interop model, which is the OMG standard, and it's MOF compliant.
... We created the CDA model from the metamodel.
... We do many-to-many.

(Ken shows MDMI tool more)

scribe: You can mix and match them.

harold: What preprocessing do you do to get a FHIR target?

ken: We build a FHIR map. Could be a Structure Def (in theory), but right now we build our own.

harold: How do you go from FHIR to XMI?

ken: We do a model-model transformation to get from FHIR model into XMI.

harold: The challenge is that FHIR itself, FHIR has a model that competes with UML -- it does similar things.

(Ken shows design environment for MDMI)

<inserted> Screenshots from Ken's demo: https://lists.w3.org/Archives/Public/www-archive/2017Oct/0000.html

ken: This is MDHT -- Model Driven Health Tools.
... Dave Carlson and Sean Muir built this.
... We import this model, annotate it, and then create both the XMI and the computable model.
... We've done STU2 and now working on STU3.

harold: Would be interesting to see a couple of map rules.

ken: We also generate reports, e.g., longitudinal report. Take a CDA map, generate for analyists.

dbooth: Are the rules written in a standard language?

ken: No. We use javascript

dbooth: Could these tools import rules written in a standard, such as shexmap?

ken: Absolutely. We're trying to attach metadata that others could use.

dbooth: One goal of Yosemite Project is to encourage the sharing of model translation rules.

ken: We do a lot of model-model transformation. We are importing both XML and XSDs into this underlying MOF model.
... We're using QVT - a model-model transformation language.

harold: QVT is an OMG std.
... Curious about the relationship between this work and the FHIR mapping language.

dbooth: Separate efforts so far, as far as I know.
... What followup would people like to see?

ken: After review by Elisa Kendall, we'll be able to show how we've approached 11179 classifcation scheme and share that with others to see how it may be leveraged or use what you have done to incorporate.

harold: Would like to see it!

ADJOURNED

s/Ratzesberge /Ratzesberger /

- DRAFT -

Semantic Web Health Care and Life Sciences Interest Group Teleconference

03 Oct 2017

Attendees

Contents

Intros

I2B2 Experiences (Ken Lord)

Summary of Action Items

Summary of Resolutions

Scribe.perl diagnostic output