CIMI – SemanticHealthNet Meeting Brussels 2014 Presentation of WP 4: "Harmonised Resources for Semantic Interoperability"

CIMI – SemanticHealthNet Meeting Brussels 2014

Presentation of WP 4:

"Harmonised Resources for Semantic Interoperability"

Activity domain: Tangible evidence

Activity domain:Generalisability & scalability and sustainability

WP 1 Patient care

exemplar: chronic heart failure

WP 3 Stakeholder

validation Additional conditions

and patient populations

Professional communities

Clinical governance Citizen communities Health authorities Global public health

WP 2 Public health

exemplar: cardio-vascular prevention

Workstream I

Health-directed requirements, success criteria, governance

Workstream II

SDO-harmonised and tailored resources, tools and methods

WP 4 Harmonised

resources for EHRs/PHRs & aggregation

WP 6 Industrial

engagement Integration into

clinical/public health information systems

Industrial exploitation

Recommendations to SDOs

WP 5 Infostructure &

tools Artefact

governance, certification & testing

WP 7 Adoption & sustainability

strategies Business success factors Sustainability models

Workstream III

Sustainability / Network co-ordination

WP 9 Project management, dissemination, promotion

WP 8 European Virtual

Organisation Community building Organisational

governance Liaison with EU initiatives Liaison with national

bodies Education, training

…planned follow up of SemanticHealthNet

Activity domain: Tangible evidence

Activity domain:Generalisability & scalability and sustainability

WP 1 Patient care

exemplar: chronic heart failure

WP 3 Stakeholder

validation Additional conditions

and patient populations

Professional communities

Clinical governance Citizen communities Health authorities Global public health

WP 2 Public health

exemplar: cardio-vascular prevention

Workstream I

Health-directed requirements, success criteria, governance

Workstream II

SDO-harmonised and tailored resources, tools and methods

WP 4 Harmonised

resources for EHRs/PHRs & aggregation

WP 6 Industrial

engagement Integration into

clinical/public health information systems

Industrial exploitation

Recommendations to SDOs

WP 5 Infostructure &

tools Artefact

governance, certification & testing

WP 7 Adoption & sustainability

strategies Business success factors Sustainability models

Workstream III

Sustainability / Network co-ordination

WP 9 Project management, dissemination, promotion

WP 8 European Virtual

Organisation Community building Organisational

governance Liaison with EU initiatives Liaison with national

bodies Education, training

Workpackage 4

5

Mission: Provide an intermediate semantic layer

able to deal with the unavoidable heterogeneity which arises when clinical information is represented across or within the same

medical domain.

• Leader: Medical University of Graz• Participants:

- Ocean Informatics- EN 13606- HL7 International- Eurorec

External experts: Daniel Karlsson, Rahil Qamar, Ronald Cornet, Alan Rector, Rong Chen, Jesualdo Tomás, Diego Boscá, Mathias Brochhausen, Bill Hogan, Mar Marcos

- Geneva Univ. Hospital- INSERM, Paris- IHTSDO- WHO

6

Basic assumption of WP 4– Plurality of Information Model approaches exists:

• openEHR, 13606 / SIAMM / HL7 v3 / CIMI, …• Local schemas are still largely predominant• Information model like structures in existing terminology systems:

context model of SNOMED CT• Free text (on purpose out of scope in SHN)

– Plurality of representations within one specification exists– WP4's relation to information models

• does not develop "yet another" information model• maintains equidistance and neutrality • does not contribute to the development of new information models or

model variants• looks at content and not at structure

– WP4 wants to explore formal approaches to improve interoperability

15 Mar 2014

7

Role of Ontology and Logic– Transform existing resources (terminologies,

clinical models) into “semantically enhanced” ones, using ontology-based formalisms

– Rationales for using formal ontology: • Possibility to detect equivalences across different

distributions of content between information models and terminologies using logic-based reasoning

• Advanced exploitation of clinical information by means of semantic query possibilities

also• Terminologies like SNOMED CT increasingly using

ontology languages such as OWL• Fuzziness of terminology / information model boundary

15 Mar 2014

Overlap Terminologies / Information Models

Clinical Terminologies

Clinical Information Models

• Clinical Information models to be used without or with inexpressive terminologies

• Terminologies to be used without information models

• Contextual statements (negation, plans, beliefs…) within terminologies– SNOMED CT context model– ICD 11 content model

• Local terminology within IMs• Postcoordination within IMs

Consequence: Plurality of isosemantic expressions

Consequence: Plurality of isosemantic expressions

• Information model representation (no binding)

confirmed

cancer

It provides some semantics but does not distinguish between information and what it represents. Reference to terminological / ontological standard is optional

Consequence: Plurality of isosemantic expressions

• Information model representation (no binding)

• Terminology representation395099008 |cancer confirmed|

confirmed

cancer

It provides some semantics but does not distinguish between information and what it represents. Reference to terminological / ontological standard is optional

Everything packaged in one codeNo separate information model needed

12

Consequence: Plurality of isosemantic expressions

• Information model representation (no binding)

• Terminology representation395099008 |cancer confirmed|

15 Mar 2014

• Information model / Terminology representation

confirmed

cancer

Cancer

confirmed

cancer

Terminology expression

NO SEMANTIC INTEROPERABILITY

binding

It provides some semantics but does not distinguish between information and what it represents. Reference to terminological / ontological standard is optional

Everything packaged in one codeNo separate information model needed

13

Ontologies – chances and difficulties

• Our hypothesis: ontologies can act as a “semantic glue” to create an unambiguous representation by relating information model entities and clinical terminologies

• Ontologies will help us to distinguish between:– Clinical entities “what does Heart Failure exactly mean?”

– Information entities “what is documented about a specific heart failure instance?”

– Epistemic entities “how sure am I whether it is heart failure?”

– Clinical process entities “what is done to acquire the knowledge I need?”

• Known limitations– expressiveness limited if computable (subset FOL)– the difficulty of "thinking ontologically"

15 Mar 2014

Ontologies re-used and created in SemanticHealthNet

"Clinical entities" (findings, disorders, procedures, substances, organisms...)

ClinicalProcesses

InformationArtifacts

"Clinical entities" (findings, disorders, procedures, substances, organisms...)

ClinicalProcesses

InformationArtifacts SNOMED CT

Domain Ontology

Ontologies re-used and created in SemanticHealthNet

Toplevel Categories

Basic relations Constraining axioms

"Clinical entities" (findings, disorders, procedures, substances, organisms...)

ClinicalProcesses

InformationArtifacts

BioTopLiteUpper LevelOntology

SNOMED CTDomain Ontology

Constraining axioms

Ontologies re-used and created in SemanticHealthNet

"Clinical entities" (findings, disorders, procedures, substances, organisms...)

ClinicalProcesses

InformationArtifacts

existence can be taken for granted existence of concrete instances in a real patient may be hypothetic

????

Basic representational pattern for terminology binding

• Example: Diagnosis (statement about clinical situation)

EHR

WHAT?

WHO?

WHEN?

Neoplasia

"this is an information entity of a certain type (e.g. diagnostic statement) which has an attribute (e.g. "suspected") , which is created by a health professional at a given time and is about some type of clinical entity (e.g. neoplasia)…"

DemographicsTime stampsMetadata

annotation of an information item Patient X

Example: “Suspected heart failure caused by ischaemic heart disease”

One code or postcoordinated expression in SNOMED CT Reference to two kinds of disorders

(ontological types / concepts) Semantic relation between both Epistemic context: represents state of knowledge about a

clinical situation Not clear whether there is really some heart failure at all! • Many entries in EHRs must not be interpreted as factual statements• Blending of ontological and epistemic information in one code

characteristic for many clinical terminologies

20

• Three heterogeneous representations of the same statement• Three different atomic information entities

“Suspected heart failure caused by ischaemic heart disease”

Organ Failure Diagnosis

Organ Heart

Status Suspected

Caused by ischaemic heart disease

Yes NoUnknown

Diagnosis

Suspected heart failure caused by ischaemic heart disease

x

Diagnosis

Heart Failure

Status

Suspected

Cause

Ischaemic heart disease

21

“Suspected heart failure caused by ischaemic heart disease” Annotation 1

OrganFailureDiagnosis

Organ Heart

Status Suspected

Causedbyischaemicheartdisease

YesNoUnknown

x

is a diagnosisabout organfailure

is adiagnosisabout heartfailure

is asuspectedorgan failurediagnosis

is a organ failure diagnosisabout a disordercausedbyischaemicheartdisease

22

“Suspected heart failure caused by ischaemic heart disease” Annotation 1

23

“Suspected heart failure caused by ischaemic heart disease” Annotation 2

24

“Suspected heart failure caused by ischaemic heart disease” Annotation 3

25

One diagnosis instance for each model

Query 1

26

All three information instances found

27

All three information instances found

Query 2

28

How do we apply that?

ClinicalModel

ISO 13606

ClinicalModel

openEHR

ClinicalModel

HL7 CDA ClinicalModel

…

SEMANTICPATTERNS

SHNOntology

Framework

annotated with

annotated with

annotated with

annotated with

compliant with

SNOMED CT

ContextModel

Use cases: heart failure and

cardivascular health

How do we apply that?

29

ClinicalModel

ISO 13606

ClinicalModel

openEHR

ClinicalModel

HL7 CDA ClinicalModel

…

SEMANTICPATTERNS

SHNOntology

Framework

annotated with

annotated with

annotated with

annotated with

compliant with

SNOMED CT

ContextModel

Use cases: heart failure and

cardivascular health

CIMI