Using SNOMED CT with openEHR Archetypes to model clinical requirements

1
Using SNOMED CT with openEHR Archetypes to model
clinical requirements

• 10 June 2008, London, U.K.

Dr. Rahil Qamar Siddiqui Business Analyst BT
Health, London
2
Talk Outline
Structuring data capture requirements
Current State
Different data entry screens and values for same
information
openEHR Archetypes
Terminologies for providing semantics to clinical
requirements
Bringing together structure and semantics
Issues and way forward
SNOMED CT
3
Basics

• Brief introduction to openEHR Archetypes and
  SNOMED CT. Detailed information to be
  investigated individually.
• Concentrate on how to use Archetypes and SNOMED
  together to model clinical requirements.
• By the end,
• Should have general background of Archetypes and
  SNOMED,
• Should be aware of how to work with the two
  models collaboratively,
• Should be aware of some of the general issues
  when attempting this collaborative or integrated
  work.

4
Current State

• Different systems use different screen capture
  designs and vocabularies to record similar data

Taken from NHS Northern Network of Cardiac Care
Taken from Blue Bay website
5
Is this data interoperable?

• Data from the two fields stored in local
  repositories.

SYSTEM 1
SYSTEM 2
?
No interface mechanism to help System 1 and 2 to
achieve semantic interoperability
6
What do we want to achieve?

• A framework that can guide controlled and
  structured capture of clinical requirements.

SYSTEM 1
SYSTEM 2
I N T E R F A C E
?
7
Structured Clinical Data Modelling

• Enables capture of data accurately at an adequate
  level of granularity.
• Specify context and constraints on the data to be
  captured.
• Standard models reference some underlying
  information model that provides a logical view of
  the physical data models.
• E.g. openEHR Archetypes, HL7 CDAs.

8
openEHR Archetypes

• Archetypes are essentially nested hierarchies of
  data elements required for recording a particular
  clinical scenario. These models specify
  pre-defined constraints on the data recorded.
• A computable expression of a domain content model
  in the form of structured constraint statements,
  based on a reference (information) model.
  Archetypes are all expressed in the same
  formalism. In general, they are defined for wide
  re-use, however, they can be specialized to
  include local particularities. They can
  accommodate any number of natural languages and
  terminologies. (Taken from Archetype Definitions
  and Principles Rev 1)
• The in-built structuring of the data elements
  required for a particular clinical scenario
  provides context to its immediately underlying
  sub-elements. Each data element is defined as
  being either mandatory 1..1 or optional 0..1.
  Any constraints to be placed on the element
  values are also modelled in the archetypes.
• The data models conform to the openEHR Reference
  Model.

9
openEHR Archetypes

• Archetype Definition Language (ADL) is a formal
  language for expressing archetypes. It is one
  possible serialisation of an archetype.
• The current version of the Archetypes
  specification is the same used for the European
  Health Record standard (CEN 13606) and is called
  ADL 1.4 in future it is planned to move to ADL
  2. Archetypes and ADL comply to an archetype
  reference model, which is in fact the model
  behind ADL 2. XML archetypes comply with the ADL
  2.0 model. (Taken from openEHR website)
• Check http//www.openehr.org/ for latest
  publications on openEHR specifications.

10
openEHR Archetypes

• Archetypes sit between knowledge resources in a
  computing environment, such as terminologies and
  ontologies, and runtime data in production
  systems. Their primary purpose is to provide a
  reusable, interoperable way of managing data
  creation, validation and querying, by ensuring
  that data conform to particular structures and
  semantic constraints.

Archetype Model High-level Architecture
11
openEHR Archetypes

• Templates
• A directly locally usable definition which
  composes archetypes into larger structures often
  corresponding to a screen form, document, report
  or message. A template might add further local
  constraints on the archetypes it references,
  including removing or mandating optional
  sections, and might define default values. (Taken
  from Archetype Definitions and Principles Rev
  1.0)
• Archetypes can be composed, specialised, and
  rendered as templates for local use.

12
openEHR Archetype Sample Archetype (BP)
13
openEHR Archetype Sample Template (Antenatal)
14
Translating legacy data to archetypes
openEHR Standardised Knowledge Environment
Archetypes
Templates
Generate from Tools
Template Data Schema
Template Data Document
Validates
Implementation Standardised XML Environment
Archetype based standard XSL Transform fragments
openEHR
CEN 13606
openEHR Display
Taken from Hugh Leslies slide, Ocean Informatics
15
Therefore
SYSTEM 1
SYSTEM 2
I N T E R F A C E
?
16

• But how do we deal with variations in the
  vocabularies and semantics used in the different
  legacy systems?
• How do we know Systolic blood pressure in the
  Clinical Measurements table in System 1 is
  semantically similar to the first reading of BP
  in the ECG Recordings table in System 2?

17

• With the help of terminologies chosen as
  standards for providing semantics and controlled
  vocabularies used and stored by different legacy
  systems.

18
Clinical Terminologies

• It is a structured list of terms for use in
  clinical practice.
• These terms describe the care and treatment of
  patients over a large area of medicine such as
  diseases, operations, treatments, drugs, and
  health care administration. (Taken from NHS CfH
  website)
• Need for terminologies in the field of medicine
  arose due to the desire to re-use clinical data.
• Data re-use is required either to integrate
  systems, to link patient records to decision
  support and knowledge management, or to re-use
  information collected in the course of
  patient-care for management, remuneration,
  quality assurance or research. (A.Rector
  Description Logic Handbook Medical Informatics)

19
Clinical Terminologies

• In medicine, coding data is required to allow
  some degree of consistency in recording clinical
  data.
• Therefore, terminologies should be able to meet
  the need to expand the term coverage for an
  increasing number of medical domains.
• They should make explicit all information
  required to define, describe, and categorise a
  concept.
• It is important to explicitly state whether a
  concept is a kind of the other, or whether they
  differ so that computers making use of the
  terminologies can correctly infer the information
  and work with it.
• Since terminologies are built primarily for
  computer systems rather than for the human eye,
  such requirements should be intrinsic features of
  any terminology aiming to be widely and safely
  used.

20
Clinical Terminologies Examples

• LOINC Logical Observation Identifier Names and
  Codes provides a universal code system mainly for
  reporting laboratory observations.
• ICD International Classification of Diseases
  provides codes to classify diseases and a wide
  variety of signs, symptoms, abnormal findings,
  complaints, social circumstances and external
  causes of injury or disease.
• FMA Foundational Model of Anatomy is an
  ontology of the phenotypic structure of the human
  body/anatomy.
• SNOMED CT Systematised Nomenclature of
  Medicine Clinical Terms provides clinical
  content and expressivity for clinical
  documentation and reporting.

21
SNOMED CT

• SNOMED Clinical Terms (2002) SNOMED RT
  Clinical Terms V3 (Read Codes)
• Original SNOMED work on medical nomenclature
  began with focus on pathology (SNOP) in 1965, and
  progressed to include other medical specialities
  such as surgery and autopsy (SNOMED) from 1974
  onwards.
• Intention is to improve sharing health care
  knowledge through the use of scientifically
  validated terminologies. (Taken from SNOMED
  website)
• Merger of Read Codes and SNOMED RT (2000) was
  aimed to improve and safeguard patient care by
  using an agreed terminology.
• SNOMED CT aims to index all aspects of clinical
  care and has greater depth and coverage of
  healthcare than the individual versions.
• NHS in UK has adopted SNOMED CT as the standard
  computerised terminology for coding patient data
  (Taken from CfH website)

22
SNOMED CT

• All development and management of SNOMED CT is
  now governed by IHTSDO since Apr 2007, while CAP
  continues to serve as the commissioned support
  organisation.
• NHS CfH acts as the host organisation of the
  IHTSDO National Release Centre in the UK known as
  the UK Terminology Centre (UKTC).
• SNOMED CT has over 370,000 unique concepts and
  over 1.4million triples (Object-Attribute-Value).

23
SNOMED CT

• SNOMED content divided into 19 semantic categories

SNOMED CT Concept (138875005)
24
SNOMED CT

• Searching for concepts and concept hierarchies

Fully Specified Name (FSN)
Search term
Preferred Term (PT)
Synonym (SYN)
Search results
Concept definition (Primitive/Fully Defined)
Legacy codes
Taken from CliniClue 2006 Browser
25
So..

• How do we bring together the structure provided
  by Archetypes with the controlled vocabulary and
  semantics provided by SNOMED CT?
• AND
• Why?

26
Integrating Archetypes and SNOMED

• Why?
• To enable semantic interoperability i.e. the
  ability to transfer data to and use data in any
  conforming system such that the original
  semantics of the data are retained irrespective
  of its point of access.
• Subsequently, this will result in more efficient
  sharing and interpretation of clinical data
  across different systems and subsystems resulting
  in better use of data and better patient care.

27
Integrating Archetypes and SNOMED

• How?
• Develop Integrated Models

Mapping of elements/values to concepts
28
Integrating Archetypes and SNOMED The BP
example continued

• Mapping archetype elements to SNOMED concepts to
  provide semantic clarity against a chosen
  terminology

Taken from openEHR website
29
Integrating Archetypes and SNOMED The BP
example continued

• System 1 and 2 conform to the same semantics as
  defined in SNOMED CT thereby enabling semantic
  interoperability.

O/E - Systolic BP reading (163030003 )
O/E - Systolic BP reading (163030003 )
O/E - Diastolic BP reading (163031004 )
O/E - Diastolic BP reading (163031004 )
SYSTEM 1
SYSTEM 2
I N T E R F A C E
?
30
However,

• Its not that simple .
• Manual mapping processes take far too long.
  Therefore, such efforts are infrequent and
  incomplete. (Check http//www.openehr.org/clinical
  models/archetypes.html)
• Automated systems are not readily available. Lot
  of research still being carried out in developing
  such systems.
• The few systems that have been developed as part
  of research projects vary in complexity and
  intelligence in implementing mapping procedures
  (lexical vs. semantic searches).
• It is difficult to derive semantic inferences
  when either archetype elements/values or SNOMED
  concepts or both are ambiguous.
• Therefore, quality and coverage of archetypes
  determine quality of SNOMED mappings and vice
  versa.

31
Demos

• Demo 1
• Using an ambiguous archetype model and working
  through initial set of SNOMED mappings.
• Discussion of issues with the archetype.
• Redoing the archetype
• Remapping to SNOMED with a view of the revised
  results.
• Demo 2
• Using an unambiguous archetype model and working
  through the initial set of SNOMED mappings.
• Initial results sufficient requiring no redoing
  of the archetype

32

• DEMO 1

33
The Histology Pap Archetype
Data model view of the archetype from the
Archetype Editor with all elements
Terminology view in Archetype Editor with all
elements and values
34
Initial SNOMED mappings Histology Pap Archetype
Lexical and Semantic Automated search filtering
system
35
Initial mapping results Histology Pap archetype

• MAPPING RESULTS
• Total archetype elements/values (or terms) 41
• Mapping coverage 48.6
• Trust score (scale 0 to 1) 0.45
• Inter-rater reliability 55.5

36
Issues with the histology pap archetype .. (1)

• Ambiguous categorisation of top level data terms

openEHR Observation Archetype
Observation
Histology Pap (record cytological findings of a
pap test)
Finding
openEHR Instruction Archetype
Procedure
openEHR Action Archetype
37
Issues with the histology pap archetype .. (2)

• Ambiguous naming of data terms
• 1) Clinical (clinical notes sent with request)
• Misinterpretation Clinical specimen collection
  process
• Correction Clinical Notes
• 2) Macroscopic (Macroscopic findings)
• Child Appearance (Macroscopic appearance of
  tissue)
• 3) Microscopic (Microscopic findings)
• Child Findings (Microscopic findings)
• 4) Other terms Process, Identification, Comment,
  Laboratory, Preservative

Archetype term definitions
38
Issues with the histology pap archetype .. (3)

• Similar or duplicate labeling of terms with
  inadequate definitions
• 1) Specimen
• Used as Specimen (Details of specimen)
• Specimen (Problem with specimen) and ALL
  children
• Specimen collection (Initial reporting event)
• 2) Image
• Used as Image (Images of macroscopic
  appearance)
• Image (Images associated with microscopy)

39
Issues with the histology pap archetype .. (4)

• Use of post-coordinated terms adequately and in
  the right place
• Post-coordination Composition of 2 or more codes
  to represent a concept.
• E.g. Severe dyslexia Dyslexia59770006
  Severity246112005 Severe24484000
• 1) Inappropriate post-coordination of specimen
  and description
• Specimen Description (Details of specimen)

Pre-coordinated match for specimen description
40
Issues with the histology pap archetype .. (4)

• 2) Need for post-coordinated SCT mapping rather
  than single pre-coordinated match.
• Low grade epithelial abnormality preferred
  results.

41
Issues with the histology pap archetype .. (5)

• Insufficient separation of meta data information
  from core clinical recording information
• Meta/general data terms Specimen collection,
  satisfactory assessment, quality, labeling,
  preservative.
• Specific data terms Low grade, high grade
  epithelial abnorm., negative for dysplasia or
  malignancy, specific cytological finding
• NOTE Encourage reuse of archetype nodes for meta
  data

42
Issues with SNOMED w.r.t Histology Pap archetype
..(1)

• Concepts missing in SCT considered important for
  inclusion
• Technically unsatisfactory technical problems
  with tests performed
• Appearance specific to histopathology
  examinations. Also microscopic and
  macroscopic appearance.
• Adequate no. of squamous cells present specific
  to PAP smear. OR general concept adequate no. of
  cells present squamous epithelial
  cell80554009.

43
Issues with SNOMED w.r.t Histology Pap archetype
..(2)

• 4) Slide unlabelled, container unlabelled,
  specimen unlabelled instead of general sample
  unlabelled125159002.
• 5) Qualifiers such as insufficient, damaged,
  contaminated to qualify the quality a specimen
  (as pre-coordinated concepts).
• 6) Preservation of specimen along with
  qualifiers insufficient, incorrect, to code
  problems with specimen (as pre-coordinated
  concepts).
• 7) Specimen image, macroscopic image, and
  microscopic image as placeholders to upload
  image files.

44
Revision of Histology Pap archetype
Revised archetype
45
Changes to the Histology Pap archetype

• Archetype remodelled to describe procedure
  elements and its related findings (rather than
  observations)
• Procedure/test meta data separated from
  recordings of PAP smear.
• Remodelled to reflect general pattern of any
  clinical procedure/test with PAP smear specific
  data.
• Labels and descriptions modified to conform to
  SCT standard of labelling.

46
Improved SNOMED mapping results
MoST results for revised histology pap archetype
47
New mappings results Revised Histology Pap
archetype

• NEW MAPPING RESULTS
• Total archetype terms 41
• Mapping coverage 80.55
• Trust score 0.8
• Inter-rater reliability 93.1

(48.6)
(0.45)
(48.57)
48
Synopsis of Histology Pap archetype mappings

• Main intent of histology pap mapping exercise
• (i) to highlight the importance of creating
  unambiguous data models for data capture.
• (ii) modeling data models and terminologies are
  NOT two completely independent activities.
• Good modeling techniques required to ensure
  source of the integration process results in
  good quality mapping results.
• However, data model terms also require good
  coverage in the terminologies to which mappings
  are searched.
• Any integration work with structurally and
  compositionally different models needs to focus
  on CONTENT.

49

• DEMO 2

50
The Tendon Babinski Archetype
Data model view of the archetype from the
Archetype Editor with all elements
Terminology view in Archetype Editor with all
elements and values
51
Initial SNOMED mappings Tendon Babinski
archetype
52
Initial SNOMED mappings Tendon Babinski
archetype

• Top ten scoring SNOMED concepts in the Tendon
  Babinski archetype

53
Initial mapping results Tendon Babinski
archetype

• MAPPING RESULTS
• Total archetype elements/values (or terms) 45
• Mapping coverage 90.9
• Trust score (scale 0 to 1) 0.9
• Inter-rater reliability 86

54
Synopsis of Tendon Babinski archetype mappings

• Of the four archetypes tested Histology pap,
  Visual acuity, Body weight, and Tendon Babinski
  the last archetype scored the highest.
• The observation was that not only was the tendon
  babinski archetype model the least ambiguous but
  SNOMED CT had the best coverage of the tendon
  reflexes.
• The babinski response archetype terms were not as
  well covered in SNOMED leading to a slight
  lowering of the results.
• This proves that good quality archetypes and good
  quality SNOMED concepts together with a good
  coverage are essential to achieve high level
  quality of mappings.

55
Issues with SNOMED - Quick Overview .. (1)

• INCONGRUENT HIERARCHIES

Similar SNOMED concepts are modelled differently
into their respective hierarchies. The
subsumption relationship between reflex normal
and O/E reflex normal concepts differs for
SUPINATOR reflexes.
56
Issues with SNOMED - Quick Overview .. (2)

• DISJOINTNESS-RELATED HIERARCHY ISSUES

Plantar Grasp Reflex Finding
Plantar Reflex Finding
57
Plantar Grasp Reflex Observable Entity
Plantar Reflex Observable Entity
Are these both disjoint? Are they the same?
58
Simple Heuristics
Research in this area continues
59
I can go on ..

• However, times limited !
• Papers published on general issues with SNOMED CT
• Papers published on the lexical and semantic
  search and filter tool used i.e. Model
  Standardisation using Terminology (MoST) system.
• Thesis available online http//www.openehr.org/sha
  red-resources/publications/archetypes.html

60
Conclusion

• It is essential to understand the objectives of
  both Archetypes and SNOMED when attempting to
  work with these two models.
• Ideal approach to creating archetype models is to
  iteratively map to SNOMED concepts to ensure
  100 semantic mapping of all archetype terms to
  be coded.
• Not all archetype terms will require to be coded.
• Deficiencies in the SNOMED model should be noted
  and fed back to the SNOMED working group (CMWG)
  or alternative measures adopted.
• Adopt general heuristics when building rules for
  mappings archetype terms to SNOMED concepts.

61
Thank you

• Dr. Rahil Qamar Siddiqui, BT Health
• rahil.qamar_at_bt.com