RIMBAA 201309 Agenda

From HL7Wiki
Revision as of 17:30, 9 October 2013 by Rene spronk (talk | contribs)
Jump to navigation Jump to search

This is the agenda of the RIMBAA WG for the September 2013 WGM in Cambridge.

Monday Q3 (13:45-15:00)

  1. Administrative items (max 30 minutes)
    • Agenda approved, with additional May out of cycle discussion
    • Approval of Atlanta WGM Minutes
      • Motion to approve minutes of the Atlanta meeting (Andy/Peter), 3-0-1 (y/n/abst)
    • Approval of November Out of Cycle meeting
    • Discussion of a May out-of-cycle
    • RIMBAA name change - discussion did not occur in Atlanta (time permitting, max 10 minutes)
  2. HL7v2 and CDA Implementation Bundles (Rene Spronk)
    • FHIR is published in such a way, and combined with reference implementations etc. that appeal to software implementers. On hindsight, what would a HL7v2 (or: CDA) Implementation Bundle look like ? Should we as HL7 offer such 'bundles' for download?
    • Example: The HL7v2 Implementation Bundle could consist of the HL7v2 spec (in HTML) format, v2 examples, the HAPI/nHAPI toolkits, the v2 XML schema, the v2 database (with definition of message structures).
    • Discussion: what exactly would we like to see in a v2 bundle? Or a CDA bundle?

Wednesday Q4 (15:30-17:00)

Attendance:

  • Charlie McCay, Michael van der Zel, Peter Hendler, Rene Spronk, Gordon Raup, Brian Peck, Carmela Coudeau, Micheal Rossman, Rob Hausam, Willem Dijkstra, David Markwell, Justin Fyfe, William Goossen, Duane Bender, Lisa Spellman, Andy Stechishin.
  1. Administrative
    • Rene calls to order at 15:31
    • Agenda approval
    • Announcements
      • Rene: the board has created a User Group taskforce to come up with ideas on how to set up user groups. There are no real plans yet, but they intent to pilot the mechanism/structure with 1 or 2 user groups. I suggested they'd pilot their work with two user groups: a user group that has yet to be created from scratch, and one that effectively already is up and running (but can always be improved upon): AID (ex-RIMBAA). I haven't committed us to anything, but I feel it's an opportunity to improve/expand our offerings.
    • MOTION "To revisit the motion related to a new name for this WG made on Monday, and to follow the recommendation of the FTSD steering division in renaming this WG to Application Implementation and Design (AID)"
      • Moved by Peter/Justin, accepted 9-0-2.
  2. Kaiser's Clinical Ontology (OWL) Modeling KCOM (Peter Hendler/Mike Rossman, see[ http://www.hl7.org/documentcenter/public/wg/java/2013.ppt http://www.hl7.org/documentcenter/public/wg/java/2013.ppt] for presentation).
    • Topics coveredWhat is OWL and how is it related to UML and OO?
    • What do you gain with Open World Description Logic, Reasoners and Inferencing?
    • What happens if you try to move the HL7 RIM to OWL?
    • Topics: How do you capture clinical specialists domain knowledge in an Ontological OWL model? How can you represent these models in UML and in tabular Excel format?
    • Questions/discussion points:
      • Rene: also see Peter's introduction on inference and reasoners, available as a streaming video.
      • William: I'm a UML cat as you know, why if you can do this in OWL are you showing models in UML? Peter: implementation strategy reasons, e.g. performance, and to help convince UML cats (and programmers - tools dealing with OWL are far between).
      • Peter: (slide 76) dangling box is a datatype definition
      • David: you point out a known issue related to procedures and observables, esp. in other languages this seems to be a problem. LOINC (with whom IHTSDO has a cooperation) is about observables, but names sound like procedures.
      • Mike R: Time is also an issue, OWL is not a temporal logic. It gets awkward when yopu talk about 'no change in meds'.
      • Charlie: there's a problem with cardinalities, expressing in OWL 'you can have 3 out of 5 ...'. Peter: didn't run into that.
      • Peter: if your system uses clinical models, have them based on OWL, great added value.
      • Gordon: putting in all the is-a's is hard work. Peter: lots of them are inferenced. Rules are triplets.
      • Peter: not implemented yet. Mike: we'd like as much of it to be open sourge, Protege is open source, UML tooling is open source.
      • Gordon: when can I get hold of this? Peter: if people start expressing interest - who knows.
      • David: is this tractable at a large scale? Peter: I wouldn't do it.
      • Willem: we have a database that supports SNOMED inferences, but not OWL.
      • Peter: we also have excel versions of these models.
      • Charlie: there was work in the BRIDG model of reneding models in OWL, RIM, ... Mike R: has been one of the influences. Charlie: could your approach be bridged with that? Peter: they started with the assumption of putting the RIM in OWL. We preserved only the RIM-backbone. Charlie: reason for full RIM is to avoid different models in different projects, different ways of representing things. Stitching these OWL projects together, and using SNOMED makes this easier, do you still need a common language or is the SNOMED ontology enough? Mike R: RIM provides a structure for on-the wire formats. From our models there are 1000 ways to get to the wire format. Those rules are not part of this. This is more abstract than the RIM.
      • Charlie: sementantic health web project that I'm working on also touches upon this space, would be interested in putting those together
      • Peter: Take away message/to summarize: how do you model clinical data - RIM is more structured/organized than database. In futire we go in the direction to make clinical mdoels in OWL, not bacuse we expect that to be implemeted at OWL up to the wire, but because of the additional options/power it offers.
  3. Adjourned at 16:54

Thursday Q3 (13:45-15:00) - joint with Tooling WG

RIMBAA Hosting Tooling

Attendance: Peter Hendler KP, Gordon Raup, A Stechishin, George de la Torre, Alexander Henket, Rene Spronk, Willem Dykstra, Yeb Havinga, Michael Van der Zel, Michael Rossman, Wendy Huang

  1. Administrative
    • Peter calls to order.
  2. How Domain Driven Design and Event Souring (CQRS) uses RIM (George de la Torre, Harvard Medical School, US, see http://www.hl7.org/documentcenter/public/wg/java/20132609_RIMBAA_CQRS.ppt for slides).
    • Will explain the difference between a transactional model like RIM with, for example data warehouse model.
    • George: this is an update/continuation of my talk held in San Antonio (see RIMBAA 201201 Minutes San Antonio, Thursday Q1, item 5 of the agenda)
    • Vocabulary model is a bounded context, external to the core models (the RIM), but interacts with it. It is not persisted, it doesn't exist anywhere, only in-memory.
    • Architecture consists of two core part
      1. Use event oriented OLTP database with persisted v3 models (Domain Transaction Model, RIM based, e.g. NoSQL based). Archive and audit log of events. Events could in turn lead to publication of a view-specific update to a OLAP data warehouse, with specific views suitable for the domain.
      2. View/Query model: use anything you want, optimized for querying, mostly ER database, SQL.
    • Used at Harvard Medical School, and CTSA at Harvard (NIH grant for research), Fresenius healthcare.
    • Develop custom clinical apps then discovered RIM “long time ago”
    • Domain Driven Design. Apply RIM to DDD.
      • Refer to online book on DDD (see slides).
    • DDD is OO
      • Use RIM as it is “the design” of apps.
      • Already has business needs for this domain.
      • Deep insights into domain knowledge
    • Start every model with looking at overall core RIM diagram
    • RIM is only in memory, not in persistance.
    • Anti Corruption Layer
    • Event Sourcing
      • Only the Aggregate (D-MIM) is persisted.
      • The command is saved
      • Publish Events
      • Command Query Responsibility Segregation (CQRS)
      • transaction mdoel and query model.
    • The aggregate may be XML or JSON or RDBM or NoSQL.
      • NoSQL works well as you can query the aggregates.
    • From the Events and aggregates you “publish to” a query model that is persisted.
      • The Query model is not the RIM but in the right shape for queries.
      • The RIM is used as the transactional model not the persistance model.
      • The Query Model is what is persisted.
    • Only doing writes on the transaction so no concurrency problems.
      • The Query model is read only so that is also concurrency safe.
      • Did not persist the RIM.
    • This project done with Harvard Med School. Scheduling for CTSA Clinical Translational Science Awards at Harvard.
      • Fresenius Health Care Kidney Dialysis
      • Messages and Events from Hospitals, Clinics, Home Heath.
  3. Using a RIM based database (Willem, Yeb, MGRID, the Netherlands, see for http://www.hl7.org/documentcenter/public/wg/java/mgrid-mda.pdf slides)
    • MGRID SQL
      • Integrating, storing and retreiving large amounts of medical data
      • Puts HL7 V3 datatypes into database.
      • Does full UCUM
      • Time and Intervals QSET(TS)
    • Using Core MIF generate DDL and make RIM table structure.
      • Message and CDA parsers
    • Even the message parsers are auto generated.
    • Can shard ( on patient) and parallel query.
    • CDA generated made in SCALA
    • Demo
      • Auto generate 1000 CDA
      • then generate new RIM database
      • Then import them all
      • Then load to data ware house database
      • Then query
    • Can use mobile BP cuff to blue tooth to phone to CDA to DB.
    • Measured my blood pressure, Device to phone to server as jSON to CDA to database to data warehouse. SNOMED is attached by the server with template.
  4. Adjournement