20160405 arb minutes

ARB - Meeting (Date in Title)

logistics

Teleconferences are held on Tuesday at 4:00pm U.S. Eastern Schedules may be found at HL7.org Conference Call Center

Please join my meeting from your computer, tablet or smartphone.

1. Join the meeting:
   • Join the online meeting: Online Meeting Link:https://join.freeconferencecall.com/arb97
   • Online Meeting ID:arb97
2. If you cant use voip then capture the PIN from the screen for the above action, then
   • Dial into the conference: Dial-in Number:(605) 472-5556 - United States
   • enter the access code 332183
   • enter the pin.
   • International Dial-in Numbers:https://www.freeconferencecall.com/wall/arb97/#international

• Weekly conference call.
• For 24/7 customer service please call (844) 844-1322.

Agenda

1. Call to order
2. Roll Call
3. Approval of Agenda and Minutes
4. Management
5. Governance
6. Methodology (60 minutes)
   1. Substantivity
   2. Clinical Ontology
7. Other business and planning
8. Adjournment

Meeting Information

HL7 ArB Work Group Meeting Minutes Location: Telcon		Date: 20160405 Time: 4:00pm U.S. Eastern
Facilitator	Julian, Tony	Note taker(s)	Julian, Tony
Attendee	Name	Affiliation
X	Bond,Andy	NEHTA
X	Constable, Lorraine	Constable Consulting Inc.
.	Dagnall, Bo	HP Enterprise Services
.	Hufnagel, Steve	?????
.	Hyland, Mario	AEGIS
X	Julian, Tony	Mayo Clinic
.	Knapp, Paul	Pknapp Consulting
X	Kubick, Wayne	HL7 CTO
.	Loyd, Patrick	ICode Solutions
X	Lynch, Cecil	Accenture
X	Milosevic, Zoran	Deontik Pty Ltd
X	Stechishin,Andy	CANA Software and Service Ltd.
Guests
.	Sottara, Davide	Mayo Clinic
.	Grow, Richard	U.S. Department of Veterans affairs
.
Legend
X	Present
.	Absent
R	Regrets
Quorum Requirements (Co-chair + 3) Met: Yes

Minutes

Andy S, Tony, Wayne, Lorraine, Cecil, Zoran, Andy B Guests:Davide Sottara Agenda: Ok Minutes: Approve(Zoran/Andy S) Vote: 6-0-0

1. Canonical Ontology
   1. Problem:
      1. There are several information modelling and standardisation approaches in eHealth, all of which deal with the same conceptual space, i.e. healthcare and its delivery. Despite sharing a common business domain, there are often many discussions about differences and ambiguities between these approaches, which are both syntactic (i.e. how information is represented) and semantic (i.e. what kind of information is represented, and its exact meaning). Recent FHIR and openEHR discussions are case in point, although the discussions of the past regarding CDA and V2 standards have faced similar challenges.
      2. Each of these efforts attempted to provide a common model “of record” - that is, a way to standardize how clinical data is conceptually captured in an electronic health record, with the intent to enable the sharing of the data across institutions and/or applications. For this purpose, the traditional HL7 strategy is to develop “logical” information models, provide some more or less normative terminology binding, and recommend one or more canonical serialization, e.g. using XML, JSON or RDF.
      3. In an ideal world, a model “of record” should be aligned with a model “of meaning” that conceptualizes the specific business domain. The latter should define the semantics of the domain in terms that would resonate with subject matter experts, while the former should focus on the strategies for the representation and the communication of that semantics.
      4. The advantages of such an alignment are manifold: it allows to verify the semantic consistency of clinical records, it affords for a unique and unambiguous interpretation of a record entry, and generally provides guidelines for the information modelling.
      5. Instead, the semantics of information models is often loosened in favor of other criteria such as minimizing the number of data structures to be implemented (“economy of representation”); the interpretation is based on the intended or recommended usage, as defined in normative but relatively informal natural language implementation guides; the (lack of) adherence to the recommended terminology bindings may lead to “noisy” documents which are still interpretable by humans but hardly usable by information processing software.
      6. Nevertheless, there are several ongoing attempts to formalise conceptual models of meaning. Consistent effort has been focused on the modeling of diseases or genomes: for example, the Basic Foundation Ontology (BFO) has been used as a basis for developing ontologies in the biomedical domain (Gene Ontology and Open Biomedical Ontology). IHTSDO has been evolving SNOMED-CT along the lines of a formal ontology, covering some aspects of healthcare. The Veterans Health Administration is sponsoring the SOLOR project, to use standards based, formal ontologies to blend SNOMED with other major terminology systems such as RxNORM and LOINC. Likewise, additional VA-sponsored work is looking to formalise care delivery, with particular attention to goal-driven plans.
      7. Under the direct aegis of HL7, there is an “ontologic” approach to FHIR, but it’s centered on defining the canonical RDF serialization of resources and enabling some form of schema-like validation, rather than being a declarative, axiomatic formalisation of FHIR as a model of record (even less so trying to be a model of meaning). Similarly, there was an attempt to define an ‘ontological’ version of the RIM, named O-RIM, which followed similar approaches..
   2. Assessment
      1. In healthcare, logical information models focus on records of data, but they are very loosely connected (using valueset bindings), if not disconnected completely, from the models of meaning which should provide a formal interpretation for the data.
      2. Communities look at information models (FHIR, RIM, vMR, OpenEHR archetypes, CEMs, DCIMs) at different levels of detail and granularity. However, as per above, the intent and semantics around the use of those models lives mostly in the documentation and implementation guides. This makes it hard, if not impossible, to leverage or ensure the desired level of consistency and correctness without resorting to implementation specific approaches. The lack of a common formal and sufficiently strict specification, as well as the lack of a common conformance testing framework, require every institution to implement the standards in ways that are hardly guaranteed to be compatible. In fact, it has been proven that to date no major EHR system, while being FHIR compliant, represents information as simple as vitals and medications using common patterns. This commonality between iso-semantic, but syntactically “chiral” representations, is not formalized in a computational way, limiting the definition of model driven translation frameworks, on top of the already cited validation and interpretation ones.
      3. Previous attempts to align e.g. FHIR and RIM have shown that, without a clear underlying semantics, translations based on structural matches (e.g. “both models have a structure called Observation, so any fhir:Observation can be translated into a rim:Observation”) may be partial, ambiguous and result in lossy and imperfect representations.
   3. Recommendation:
      1. Create a common ontology, as a conceptual, domain specific “model of meaning”, and bind the information models of record to it.
         1. This ontology is not “yet another overarching schema”, but a common way to formally establish the meaning of health records.
         2. This is considered a necessary step towards
            1. 1) non ambiguous interpretations,
            2. 2) formal mechanisms to detect inconsistencies computationally and
            3. 3) a common space for translations between information models.
         3. This ontology should follow some desirable principles:
            1. It should be consumable computationally, to enable semantic model-driven frameworks.
            2. It should be architected to be modular, for maintenance and evolution purposes
            3. It should be grounded in some foundational ontology, to provide formal definitions even for the basic, primitive concepts
            4. It should leverage preexisting ontologies addressing the same conceptual space, such as the Clinical Care Ontology developed by Mitre Corp. on behalf of the VA
            5. It should leverage (an ontological expression of) the concepts described by major terminology systems
      2. Zoran: We looked a ontologies and the way they have developed for healthcare, such as gene ontology. This is an enormous piece of work - is there a business case. On discussion with Davide we believe that there is a used.
      3. Cecil: Using SNOMED or LOINC, or owl to provide formal extensions only goes so far. It does not give you a semantically formal model. While the RIM is not a semantic model, it is closer than FHIR. FHIR is a step backwards. We built a OWL model of a tuberculosis domain with the reporting requirements of the CDC. It is being deprecated for one less semantically sufficient because it wass too difficult to maintain.
      4. Zoran: Grahame agreed that an ontology is required it would not be used by the implementors. It is a big piece of work.
      5. Davide: Tooling that facilitates implementation the state of the art advanced over ten years ago.
      6. Cecil: I stay up on the tooling, and semantic web. Even if you get it modeled in RDF and OWL you still lack semantics. The first step would be to build tooling around common logic.
      7. Davide: Interesting you mentioned common logic. The intent would be to formally capture from an implementation guide. Have not made a commitment to OWL only.
      8. Zoran: What is 'SOLOR'?
      9. Davide: 'SOLOR' is to use semantic web languages (OWL-EL) to express the formal definitions is a system. OWL is used for the sheer size of the models. Healthcare delivery ontology must be able to bind all of the terms, and be expressive as needed.
      10. MOTION To go forward (Zoran/Andy S)
      11. Andy B: Wary of Cecils comment - complexity and time makes this a difficult job. Need a compelling limited domain that will provde a compelling argument beyond that that FHIR has. Who would be the stakeholders? A very broad foundation for FHIR - everyone agrees it would be wonderful, but it would be unlikely to carry forward because of how FHIR operates.
      12. Davide: When the scope is narrowed to Observation in general, specifically vitals and Labs. Vitals I can bring in a collaberator who know many systems. None of the vendors is representating the vitals in the same way.
      13. Andy B: Need to convince that it is maintaiable, implementable, and compatible with the technology investments. Making a broad based ontology as a goal state will fail. Making a smaller set success would be accepted/
      14. Davide: It could be proven in a smaller space.
      15. Zoran:Need to prepare a PSS taking into account of all of the concerns. Cecil re your CIMI involvement would others be interested?
      16. Davide: Will talk to the HPCC folks.
      17. Zoran: There have been discussion on the FHIR channels.
      18. Davide: Mayo thinks FHIR is easily implementable, but without an underlying ontology it is not consistently applied for a large enterprised. There has to be some way to validate and find consistency.
      19. Davide: Short PSS, and add attachments?
      20. Andy S: Yes, good idea.
      21. Lorraine: Will need to reference to the requirements page.
      22. Vote 5-0-0
      23. On the agenda for April 19, 2016.
2. Other business and planning
   1. April 12 - Substantivity
   2. APril 19 - Continue on Ontology discussion