RIMBAA 201105 Minutes Orlando

Minutes of the May 2011 WGM in Orlando FL USA

May 16 (Monday Q3)

Workgroup	Date/Time	Location	Chair/Scribe
RIMBAA WG	2011-05-16, 13:45-15:00	Orlando FL, US	Chair: Peter Hendler, Scribe: Peter Hendler, Michael van der Zel

Attendance

At	Name	Email Address
	Peter Hendler	peter@javamedical.com
	Rik Smithies	rik@nprogram.co.uk
	Amnon Shabo	shabo@il.ibm.com
	Michael Van Der Zel	m.van.der.zel@umcg.nl
	Astrid Broere	asbroere@diaconessenhuis.nl
	Gordon Raup	graup@datuit.com
	Anil Luthra	aluthra@alfisconsultants.com
	Chirag Bhatt	chirag.bhatt@feisystems.com
	Assaf Halevy	assaf.halevy@dbmotion.com
	Stacy Berger	sberger@coh.org
	Yunwei Want	yunwei.ant@seimens.com
	Chris Melo	chris.melo@philips.com
	Dave Shaver	dave.shaver@carepointhealth.com
	Brian Pech	bpech1@gmail.com
	Dale Nelson	dale.nelson@squaretrends.com
	Markbool Hussain	marqbool.hussain@seecs.edu.pk
	Lee Coller	lee.coller@oracle.com
	Nick Radov	nradov@axolotl.com
	Vassel Paytchev	vassil@epic.com

Minutes

1. Peter called to order at 13:43
2. Administrative agenda items
   • Approval of agenda for the week
     • MOTION to approve the agenda for the week (Amnon/Michael, 16-0-0 Y/Abst/no)
   • No announcements
   • Approval of the minutes of the Washington meeting.
     • MOTION to approve of the minutes of the Washington meeting (as present on the hl7.org website) (Amnon/Michael, 16-0-0 Y/Abst/no).
   • Planning of the WGM in San Diego (september 2011)
     • Skipped this item; just let us know if want to present.
3. dbMotion (Assaf Halevy, founder of the company, see http://www.hl7.org/documentcenter/public/wg/java/dbMotionOrlandoMay2011.ppt for his presentation) - see also RIMBAA: dbMotion
   • dbMotion is a RIM based integration platform used on a very large scale at U of PItt, Israel and elsewhere. Practical experience making RIMBAA work on a large scale for integrating disparate hospitals and facilities.
   • This is an example of using the RIM as an (semantic) integration layer.
   • DbMotion is used in Israel with nearly 4 million patients. It is also used at the University of Pittsburg Medical Center where various hospitals that use Cerner, Allscripts, Epic and other E.H.R.s are integrated with dbMotion. DbMotion has what is called a Unified Medical Schema. It was designed by Abdul Malik Shakir and it is like a universal RMIM or SMIRF that is a patient centric representation of the entries in a patient record. It is very close to the RIM but has had a few minor simplifications. For example, the place of birth is not represented as an entiry in a role, but was instead added as an attribute to the Person entity class. It is easily translatable to pure RIM.
   • The RIM database is the central integration repository for all clinical systems.
   • If you are in on E.H.R system, you have an extra little button from an extra listener. The button will indicate if there is other clinical information on the patient that you don't know about. You can then use the dbMotion viewer to see the information from the other systems, and optionally, you can choose to import the new information into your current E.H.R. System.
     • Idea is to share without loosing control & ownership of data.
     • Data is federated, distributed but centralized virtually. Hide this complexity for the user. The presenter calls this “Centributed”
     • Each Clinical Data Repository of each node integrated will have only his own dataset, no duplication of data.
     • Business layer will expose Business oriented Services. Standards based services and data IHE, CCD, etc.
     • dbMotion is not an EHR-System, it doesnot create data.
     • Often applications claim to be standard but are not. DBMotion shields / adapts those.
     • dbMotion provides a SDK and a platform
   • Also interesting is the use of SNOMED. All local codes are mapped to SNOMED. This way when the two local codes for example Warfarin and Coumadin, are found to be semantically the same, you can change your view to collapse all the medicaitons or observaitons that are semanically equivalent.
4. Hybrid approaches to RIMBAA (Amnon Shabo, see http://www.hl7.org/documentcenter/public/wg/java/20110517_IBM_Hypergenes.pdf for his slides)
   • Amnon Shabo presented the IBM research project on human genomics. They collect genotypes including genome-wide scan of each subject and they also collect phenotypes. The data is in a hybrid warehouse with RIM-based database for clinical data snd relational database for the genomic data (1M SNP/subject). They are able to find new correlations between genotype and phenotypes, and if the correlation is already known, they can be used in decision support. For example, a given genotype may suggest drug resistance for a given patient and conditon. The system can warn the clinician not to prescribe it - based on the genotype. The RMIM's used for the warehouse XML database are the Pedigree and Genetic Variation (of HL7 Clinical Genomic Domain) and CDA. These RMIM's were constrained to meet the project requirements. The constrained RMIM's were interrelated under one infromation model that govern the warehouse XML database.
   • Native-XML RIMBAA for semantic warehousing with XQUERY exploration. Exported data marts in RDF or relational formats for analytics and optimization. RDF-based promotion layer facilitates the definition of data marts. Data mart schemas are user-defined and mapped to the XML warehouse and/or promotion layer.
     • Stage data as is fully-indexed XML and not as blob
     • Separate databases for normal-data and mass-data (e.g. genotype data)
     • MDHT tool used to create templates with UML/OCL.
     • XML in generic layer = core model, query using XQuery.
     • IHE QED implemented and tested on Connecthathon, QED connected to RIM database (XML, RIMon)
5. Adjournment at 15:15

Appendix A: Summary of Motions

The table below captures all substantial motions.

Motions
MOTION to approve of the minutes of the Washington meeting (as present on the hl7.org website) (Amnon/Michael, 16-0-0 Y/Abst/no).

Wednesday Q6

Attendance

At	Name	Affiliation	Email Address
	Abdul Malik Shakir	COH, US	ashakir@coh.org
	Amnon Shabo	IBM, IL	shabo@il.ibm.com
	Andy Stechislin	GordonPoint, CA	andy.stechishin@gmail.com
	Ann Wrightson	NHS Wales, UK	ann.wrightson@wales.nhs.uk
	Brian Pech	KP, US	brian.pech@kp.org
	Grahame Grieve	AU	grahameg@gmail.com
	Jane Curry	HIS inc, CA	JaneCurry@healthinfostrategies.com
	Michael van der Zel	Groningen University Hospital, and Results4Care, NL	m.van.der.zel@umcg.nl
	Mike Rossman	KP, US	michael.k.rossman@kp.org
	Muhammad Afzal	SEECS, PK	muhammad.afzal@seecs.edu.pk
	Lloyd McKenzie	Lmcenzie	lloyd@lmckenzie.com
	Peter Hendler	KP	peter@javamedical.com
	Pradeep Chowdhury	KP	pradeep.k.chowdhury@kp.com
	Stan Huff	Intermountain Health	stan.huff@imail.org
	Nick Radov	Azolotl	nradov@axolotl.com
	Galen Mulrooney	VA	galen.mulrooney@va.gov
	William Goossen	Results4care	wgoossen@results4care.nl
	Allen Hobbs	KP	Allen.Hobbs@kp.org
	Dale Neslon	SquareTrends	dale.nelson@squaretrends.com
	Gordon Raup	Datuit, LLC	graup@datuit.com
	Anneke Goossen	Results4Care	agoossen@results4care.nl
	Hugh Glover	Blue Wave Informatics	hugh_glover@bluewaveinformatics.co.uk
	Chirag Bhatt	Fei Systems	Chirag.Bhatt@feisystems.com
	John Hatem	Oracle	john.hatem@oracle.com
	Lee Coller	Oracle	lee.coller@oracle.com

Minutes

1. Peter calls to order
   • Agenda for this quarter
   • Announcements: None.
2. HL7 V3 New Look Taskforce and PIIMs
   • An in person continuation of our lively online discussion (on the RIMBAA e-mail list) of what the Fresh Look Taskforce may mean to V3 implementers.
     • See Grahame's blog on the HL7 Fresh Look Taskforce.
     • One of the concepts to come out of discussions prior to this meeting is the creation of PIIMs, an artefact which will be of key importance to HL7 v3 implementers, and to increased adoption and uptake of HL7 v3. A PIIM is a PIM, a Platform Independent Implementable Model. All v3 static models are to be expressed as UML models, which makes life easy for the implementers when using MDA. All ITSs may be deprecated, and no new ones defined, HL7 should not be in de business of defining ITSs at all.
   • Discussion:
     • It has become apparent that there are many implementors who feel that HL7 V3 RIM based integration puts too much burden on the developers to learn too many unnecessary model details in order to exchange a simple set of data. Although the RIM is needed if you are going to be able to integrate systems in different healthcare domains, you do not need all of the complexity if you only want to trade a limited let of data within one healthcare domain to a partner.
     • This may even be the most typical situation. The current HL7 V3 implementation process requires either understanding of RIM, the MIF, and vocabulary bindings to name just a few.
     • The published XML schemas can not be generated without “magic”. For example the RMIMs are missing some Infrastructure Root attributes and vocabulary binding details and other hidden things that must be added before you can go from the design model (RMIM MIF) to an implementable model (which might be XMI for example).
     • Graham Greive, Lloyd McKenzie, suggested one part of the possible solution. This would be Platform Independent Implementation models (maybe in XMI or even eCore). These would be able to be directly transformed to XML or Objects.
     • Galen Mulrooney suggested better implementation guides. We also brought up the Green CDA approach which hides the complexity and the “magic” in the XSLT transform.
     • Ann Wrightson had a different but similar suggestion of use in a special case. In the case that a green CDA was to be only traded with a partner that you had pre agreements with, you would not need to transform the simple format to the universal valid CDA but you could just exchange the simple format on the wire.
     • In summary there were four suggestions discussed but this does not preclude brain storming or completey new approaches.
     • We have created a new page on the RIMBAA wiki (Fresh Look) where all are encouraged to put up items for a wish list and near term activities.
     • (added after the meeting took place, and as such not part of the official minutes): see also Grahame's comments at http://www.healthintersections.com.au/?p=314.
3. Adjournment

THU Q3

Workgroup	Date/Time	Location	Chair/Scribe
RIMBAA WG	2011-05-19, 13:45-15:00	Orlando FL, US	C/S: Peter Hendler

Attendance

At	Name	Email Address
	Nick Radov	nradov@axolotl.com
	Phil Pochan	Phill.pochon@covance.com
	Joyce Hernandez	joycehernandez@merck.com
	Amnon Shabo	shabo@ilibm.com
	Peter Hendler	peter@javamedical.com
	Daryl Thomas	darly.thomas@lifetech.com
	Geoff Hyde	ghyde@1life.com
	Grand Wood	grant.wood@imail.org
	Chirag Bhatt	Chirag.bhatt@feisystems.com
	Loraine Constable	loraine@constable.ca
	Michael Van Der Zel	m.van.der.zel@umcg.nl
	Stan Huff	stan.huff@imail.org
	Pradeep K Chowdhury	pradeep.k.chowdhury@kp.org
	Michael Rossman	michael.rossman@kp.org

Minutes

1. Peter calls to order
   • Agenda for this quarter
   • Announcements: None.
2. Report on v3 implementation at Intermountain (Stan Huff, see http://www.hl7.org/documentcenter/public/wg/java/20110519_RIMBAA_Stan_Huff.ppt for his slides)
   • There is more than one way to express a clinical statement in RIM. If you are using a RIM model to trigger clinical decision support rules, how do you know if two different expressions represent the same thing? Prior to the RIM and prior to OWL, Dr Huff was faced with the problem of representing clinical events in a "normalized" way so that equivalent expressions are known to be equivalent. He will discuss the "normalized" model that he and his group developed to address this problem.
   • Stan Huff and the Clinical Element Model (CEM):
     • Stan Huff has used the Clinical Element Model or CEM at Intermountain Health for many years.
     • This is a constrained model that consists of the following design. Everything is a Clinical Element. The clinical Element consists of sections. There is a type attribute that is human readable and tells what the element is.
     • There is a Key which is usually a LOINC code that is the exact and definitive way (not just for humans to read) way of saying what the element represents.
     • There is a value section that is a choice between a data element if there is only one value, or in the case where there are multiple values there is a list of items, each of which is a data value.
     • There is another attrubute for a collection of “quals”. And example of a qual would be “sitting” if the element was a blood pressure. It is an additional qualifiying bit of information. Finally there is an optional set of mods. An example of a mod is “father of subject” which would modify the meaning of the element to indicate it was the fathers blood pressure.
     • Each clinical element is modeled and put in a repository, so that there will not be more than one model for a given clinical concept.
     • This system is used to unambiguate the semantic meaning of different instances. For example, in many clinical models, the RIM included, there is more than one syntactic way to express the same semantic element. This CEM approach is a way to normalize the way things are represented so they can be checked for equality.
     • The CEM models also are used in the design of the GUI input screens for the E.H.R.
3. Kaiser Permanente, a preliminary trial of using RIMBAA (Java SIG) to aggregate data from Epic and NHIN data sources.
   • Using MySQL and the original version of the JAVA SIG API, and also developing a Clinical Data Extraction Framework (CDEF) to extract data from Epic without having to go through Clarity (Epics relational projection), Pradeep and his group at Kaiser Permanente created a system that can capture data in real time as it is being entered into Epic, and persist it in a pure RIM database (MySQL). They have a web interface that can show the data minutes later. So far this demo project us capturing Demographic data but it will be part of a wider effort to capture all clinical data.
   • Pradeep Chowdhury from Kaiser Permanente on the EMRLD apprach for a RIMBAA data integration layer for Kaiser Permanente (see http://www.hl7.org/documentcenter/public/wg/java/20110517_KP_EMRLD.ppt for his slides):
     • EMRLD stands for an approach for handling the data of a large enterprise.
       • E is extract. In our example them means from Epic, Pharmacy, ATD, CDA (from VA etc), Clincal devices, and various other sources. The data is extracted from all of these proprietary or native systems.
       • M is for Merge. The data is compared using mappings to standard terminology such as SNOMED so things which may appear to be different can be discovered to be the same (Warfarin and Coumadin).
       • R is for RIMify. Once we have the merged data, they are placed into a CDA like format (uses the CDA MIF file).
       • L is for Load, the data is then loaded into a pure RIM database.
       • D is for Deploy. Once we have all the data in a single RIM based persitence layer, we can use the data for many use cases. Quality reporting, decision support, research, creating new CDAs and many other things one would want pooled data for.
     • The project was built upon the HL7 Java SIG API. It sits on top of a MySQL database. The data extraction was custom built to extract data from Epic.
   • This pilot system is not in production and we hope to find funding and come back and report later on a production version.
4. A templated RIMBAA CDR for a Hospital Wide Continuity of Care Record (Michael van der Zel, see http://www.hl7.org/documentcenter/public/wg/java/RIMBAA_Impl_Aspects_2011_may_WGM_MZ.pdf for his presentation.)
   • Simple CRUD CDR at UMCG, static storage is easy, the dynamic behaviour is where the fun/trouble starts, report on issues we ran into creating the CDR store. And the hybrid SQL+XML database solution applied. Plus, a discussion type of presentation were he wants to point out the Research specific issues @ LifeLines/Target
   • Michael created a RIM based system for a large (about one million patients) system in the Netherlands. (UMCG). It is a continuity of care system. He didn't start with the Java SIG but developed the system from scratch. They were required to use SQL server. At first they used a pure RIM relational database but found it was too complex for the developers (who were not familiar with RIM), and it did not perform well.
   • On a second attempt they used a more constrained RMIM specific database which was adequate and more understandable to the developers. Later experimentation proved that a third method of persistence was even better for the developers and for performance. This is an XML approach where the entire continuity of care instance is stored as an XML blob in the database. There are indexes that allow for querying of sub parts fo the whole document. This solution scales and is in production in this fairly large system.
     • Peter: Our presentation from dbMotion on Monday indicates that the approach of a pure RIM database does scale. This is done at U of Pitt and the country of Israel. It would be interesting to see what the problems were in this present case and what the differences were in the implementions of the RIM databases. Maybe we'd see certain patterns to avoid that effect performance?
5. Adjournement