to tackle structure from the point of implementers

SNOMED Bound to (Information) Model
Putting terminology to work…
HL7 NZ Workshop, 24 Jul 2015
Koray Atalag MD, PhD, FACHI
[email protected]
Senior Research Fellow (NIHI & ABI)
openEHR Management Board Member
Vice Chair HL7 New Zealand
Member HISO / SAG
Information Models
• Definitely NOT reference information models;
– as in HL7 v3 RIM, openEHR/ISO13606 or FHIR resource ontology!
• Define how to capture/represent health information, not the
actual concepts (that’s epistemology/ontology!)
– e.g. FHIR Patient resource is concerned with how to capture patient
information, not define what patient means per se!
• Use terminology to define clinical meaning or bind valuesets
• They may employ a number of formats and methods; inc.
– Spreadsheets, data dictionaries, mindmaps, UML, XSD, Archetypes, v3
& FHIR and even programming languages
• Examples: FHIR (HL7), Archetypes (openEHR), Detailed Clinical
Models (DCM), Clinical Element Models (CEM-Intermountain)
Dealing with forms, Access databases
Dealing with forms, Access databases
From 2001!!!
http://pathos-web.sf.net
Using ASTM healthcare
DTDs…
No CDA, no archetype!
First class IM
Where in the interop stack?
Network & Data
Standards
Terminology
Standards
Content
Standards
Exchange
Standards
• Technical/Format
TCP/IP, HTTP
HTML, XML
RDF, JSON, DDL
• Syntax/Semantics
SNOMED, ICD, ICPC
LOINC, NZMT/NZULM
ATC / GMDN
• Structure/Advanced semantics
HL7 FHIR, CDA, ASTM CCR
openEHR/13606/CIMI
DICOM SR
• Process/Integration
UN/EDIFACT
HL7 v2, v3, FHIR
openEHR/13606 Extract
Structure with terminology: SNOMED
Inconsistencies due to different post-coordination of concepts
In a vasculitis physical examination: “Vascular exams: Carotid Right/Tender”
247348008 | tenderness (finding) | :
363698007 | finding site | = 69105007 | Carotid artery structure (body structure) | :
24028007 | Right (qualifier value) |
_____________________________________________________________________________
301390006 | tenderness of cardiovascular structure | :
363698007 | finding site | = 69105007 | Carotid artery structure (body structure) |:
272741003 | laterality | = 24028007 | Right (qualifier value) |
_______________________________________________________________________________
309655006 | On examination-artery (finding) | :
69105007 | Carotid artery structure (body structure) | :
24028007 | Right (qualifier) |:
247348008 | tenderness |
_______________________________________________________________________________
401050002 | Carotid artery finding (finding) | :
363698007 | finding site | = 69105007 | Carotid artery structure (body structure) | :
272741003 | laterality | = 24028007 | Right (qualifier value) | :
247348008 | tenderness |
The Sensible Way
to tackle structure from the point of implementers
The Sensible Way
to tackle structure from the point of implementers
The Sensible Way
to tackle structure from the point of implementers
The Sensible Way
to tackle structure from the point of implementers
 Open specs, tools and content for representing health
records & building EHR Systems
–
–
–
–
20+ years of international experience (Good European Health Record - GEHR)
Origin and superset of ISO/CEN 13606 Standard
An MDA/MDD Software Engineering Paradigm (e.g. “Inside Systems” vs HIE)
Learning curve for implementers BUT easy on clinicians / modellers
 Governed by openEHR Foundation (not-for-profit)
 I’m one of 4 elected Management Board members (end of 2014)
 What’s different?
 Modelling method: separation of clinical and technical worlds
 Mature tooling, scientific research & reference implementations in
almost all programming languages
• Once in Gartner Hype Cycle!
– Steady international community, maturing now 
– Underpins many national EHR programs (most Nordic, Brazil, Slovenia)
The Sensible Way
cont.
The Sensible Way
cont.
The Sensible Way
cont.
The Sensible Way
cont.
The Sensible Way
cont.
The Sensible Way
cont.
IM & Terminology
Terminology: Labels/codes attached to atomic concepts
(mostly without clinical context)
– Diabetes Mellitus, ear ache, left hip, CT scan etc.
Some have hierarchy (ICD) & relationships (SNOMED)
Boundary Problem
(overlap)
 Terminology binding
Information Model: structure and semantics of concrete
clinical concepts with clinical context/provenance
– Health condition, lab test result, discharge summary,
adverse reaction, prescription etc.
Terminology Binding
‘A formally expressible connection between information model representation
and terminology representation of clinical statements recorded in the EHR’
Two different types of terminology bindings:
1) linking a data item to external terminology/ontology for
the purpose of defining its real-world meaning
2) Linking data element values to external terminology
(e.g. a RefSet or terminology query)
 Examples define a terminology subset
^ 1111000000132 |allergy event|:
246075003 |causative agent| =
< 373873005 |pharmaceutical / biologic product|
OR
< 105590001 |substance|
Blood Pressure Measurement
1) Linking items to SNOMED to define clinical meaning
mindmap representation of openEHR Archetype
NZ Cardiac Registry: Medication
2) Linking data element values to external terminology (NZULM)
Clinical Information Modeling Initiative
• Led by Stan Huff to address profiling needs
• Develop a single curated repository of models
• Using a single modelling formalism
– Selected Archetypes as starting point
– Will harmonise with UML > Archetype Modeling Language
(OMG is on it!)
&
• FHIR resources and Archetypes are closely related
– should avoid reinvention at all costs!
– Also FHIR extensions/profiles <> openEHR Templates
• Archetype  FHIR resource conversion is expected to
be seamless
– Archetypes are usually more detailed; as opposed to
• FHIR resources include most commonly used items (80/20 rule)
with an option to extend as needed
• An opportunity exists for FHIR to leverage openEHR
content, tooling and expertise
– FHIRman says base resources are not enough!
• Oh did I just forget SNOMED?
– Not really, both FHIR and openEHR allow for terminology
bindings (for data items which make sense to encode)
Conclusions
• Information Models key for putting SNOMED to work
– Package structure, syntax and semantics in one go
– Easy to transform and consume for implementers
• There’s no point coding every data element!
– But SNOMED (and other terminology) play key role in:
• Advanced decision support (using ontology-like features/inferencing)
• Analytics, research and reporting
• It requires effort, tools and training
• Good news is (for implementers)
– They don’t have to be concerned (terminologists and
modellers’ job!)
– FHIR (inc. SNOMED bindings) sufficient for HIE
Questions?
www.taniwha.org.nz
[email protected]